{ "description": "InferenceService is the Schema for the inferenceservices API", "properties": { "apiVersion": { "description": "APIVersion defines the versioned schema of this representation of an object.\nServers should convert recognized schemas to the latest internal value, and\nmay reject unrecognized values.\nMore info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#resources", "type": [ "string", "null" ] }, "kind": { "description": "Kind is a string value representing the REST resource this object represents.\nServers may infer this from the endpoint the client submits requests to.\nCannot be updated.\nIn CamelCase.\nMore info: https://git.k8s.io/community/contributors/devel/sig-architecture/api-conventions.md#types-kinds", "type": [ "string", "null" ] }, "metadata": { "type": [ "object", "null" ] }, "spec": { "additionalProperties": false, "description": "spec defines the desired state of InferenceService", "properties": { "affinity": { "additionalProperties": false, "description": "Affinity constrains inference Pod placement (node/pod affinity and\nanti-affinity). Passthrough to the Pod spec, for finer control than\nNodeSelector — e.g. preferring or avoiding nodes already running other\nGPU workloads.", "properties": { "nodeAffinity": { "additionalProperties": false, "description": "Describes node affinity scheduling rules for the pod.", "properties": { "preferredDuringSchedulingIgnoredDuringExecution": { "description": "The scheduler will prefer to schedule pods to nodes that satisfy\nthe affinity expressions specified by this field, but it may choose\na node that violates one or more of the expressions. The node that is\nmost preferred is the one with the greatest sum of weights, i.e.\nfor each node that meets all of the scheduling requirements (resource\nrequest, requiredDuringScheduling affinity expressions, etc.),\ncompute a sum by iterating through the elements of this field and adding\n\"weight\" to the sum if the node matches the corresponding matchExpressions; the\nnode(s) with the highest sum are the most preferred.", "items": { "additionalProperties": false, "description": "An empty preferred scheduling term matches all objects with implicit weight 0\n(i.e. it's a no-op). A null preferred scheduling term matches no objects (i.e. is also a no-op).", "properties": { "preference": { "additionalProperties": false, "description": "A node selector term, associated with the corresponding weight.", "properties": { "matchExpressions": { "description": "A list of node selector requirements by node's labels.", "items": { "additionalProperties": false, "description": "A node selector requirement is a selector that contains values, a key, and an operator\nthat relates the key and values.", "properties": { "key": { "description": "The label key that the selector applies to.", "type": "string" }, "operator": { "description": "Represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists, DoesNotExist. Gt, and Lt.", "type": "string" }, "values": { "description": "An array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. If the operator is Gt or Lt, the values\narray must have a single element, which will be interpreted as an integer.\nThis array is replaced during a strategic merge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchFields": { "description": "A list of node selector requirements by node's fields.", "items": { "additionalProperties": false, "description": "A node selector requirement is a selector that contains values, a key, and an operator\nthat relates the key and values.", "properties": { "key": { "description": "The label key that the selector applies to.", "type": "string" }, "operator": { "description": "Represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists, DoesNotExist. Gt, and Lt.", "type": "string" }, "values": { "description": "An array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. If the operator is Gt or Lt, the values\narray must have a single element, which will be interpreted as an integer.\nThis array is replaced during a strategic merge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": "object", "x-kubernetes-map-type": "atomic" }, "weight": { "description": "Weight associated with matching the corresponding nodeSelectorTerm, in the range 1-100.", "format": "int32", "type": "integer" } }, "required": [ "preference", "weight" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "requiredDuringSchedulingIgnoredDuringExecution": { "additionalProperties": false, "description": "If the affinity requirements specified by this field are not met at\nscheduling time, the pod will not be scheduled onto the node.\nIf the affinity requirements specified by this field cease to be met\nat some point during pod execution (e.g. due to an update), the system\nmay or may not try to eventually evict the pod from its node.", "properties": { "nodeSelectorTerms": { "description": "Required. A list of node selector terms. The terms are ORed.", "items": { "additionalProperties": false, "description": "A null or empty node selector term matches no objects. The requirements of\nthem are ANDed.\nThe TopologySelectorTerm type implements a subset of the NodeSelectorTerm.", "properties": { "matchExpressions": { "description": "A list of node selector requirements by node's labels.", "items": { "additionalProperties": false, "description": "A node selector requirement is a selector that contains values, a key, and an operator\nthat relates the key and values.", "properties": { "key": { "description": "The label key that the selector applies to.", "type": "string" }, "operator": { "description": "Represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists, DoesNotExist. Gt, and Lt.", "type": "string" }, "values": { "description": "An array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. If the operator is Gt or Lt, the values\narray must have a single element, which will be interpreted as an integer.\nThis array is replaced during a strategic merge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchFields": { "description": "A list of node selector requirements by node's fields.", "items": { "additionalProperties": false, "description": "A node selector requirement is a selector that contains values, a key, and an operator\nthat relates the key and values.", "properties": { "key": { "description": "The label key that the selector applies to.", "type": "string" }, "operator": { "description": "Represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists, DoesNotExist. Gt, and Lt.", "type": "string" }, "values": { "description": "An array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. If the operator is Gt or Lt, the values\narray must have a single element, which will be interpreted as an integer.\nThis array is replaced during a strategic merge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": "object", "x-kubernetes-map-type": "atomic" }, "type": "array", "x-kubernetes-list-type": "atomic" } }, "required": [ "nodeSelectorTerms" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" } }, "type": [ "object", "null" ] }, "podAffinity": { "additionalProperties": false, "description": "Describes pod affinity scheduling rules (e.g. co-locate this pod in the same node, zone, etc. as some other pod(s)).", "properties": { "preferredDuringSchedulingIgnoredDuringExecution": { "description": "The scheduler will prefer to schedule pods to nodes that satisfy\nthe affinity expressions specified by this field, but it may choose\na node that violates one or more of the expressions. The node that is\nmost preferred is the one with the greatest sum of weights, i.e.\nfor each node that meets all of the scheduling requirements (resource\nrequest, requiredDuringScheduling affinity expressions, etc.),\ncompute a sum by iterating through the elements of this field and adding\n\"weight\" to the sum if the node has pods which matches the corresponding podAffinityTerm; the\nnode(s) with the highest sum are the most preferred.", "items": { "additionalProperties": false, "description": "The weights of all of the matched WeightedPodAffinityTerm fields are added per-node to find the most preferred node(s)", "properties": { "podAffinityTerm": { "additionalProperties": false, "description": "Required. A pod affinity term, associated with the corresponding weight.", "properties": { "labelSelector": { "additionalProperties": false, "description": "A label query over a set of resources, in this case pods.\nIf it's null, this PodAffinityTerm matches with no Pods.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "matchLabelKeys": { "description": "MatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key in (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both matchLabelKeys and labelSelector.\nAlso, matchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "mismatchLabelKeys": { "description": "MismatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key notin (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both mismatchLabelKeys and labelSelector.\nAlso, mismatchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "namespaceSelector": { "additionalProperties": false, "description": "A label query over the set of namespaces that the term applies to.\nThe term is applied to the union of the namespaces selected by this field\nand the ones listed in the namespaces field.\nnull selector and null or empty namespaces list means \"this pod's namespace\".\nAn empty selector ({}) matches all namespaces.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "namespaces": { "description": "namespaces specifies a static list of namespace names that the term applies to.\nThe term is applied to the union of the namespaces listed in this field\nand the ones selected by namespaceSelector.\nnull or empty namespaces list and null namespaceSelector means \"this pod's namespace\".", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "topologyKey": { "description": "This pod should be co-located (affinity) or not co-located (anti-affinity) with the pods matching\nthe labelSelector in the specified namespaces, where co-located is defined as running on a node\nwhose value of the label with key topologyKey matches that of any node on which any of the\nselected pods is running.\nEmpty topologyKey is not allowed.", "type": "string" } }, "required": [ "topologyKey" ], "type": "object" }, "weight": { "description": "weight associated with matching the corresponding podAffinityTerm,\nin the range 1-100.", "format": "int32", "type": "integer" } }, "required": [ "podAffinityTerm", "weight" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "requiredDuringSchedulingIgnoredDuringExecution": { "description": "If the affinity requirements specified by this field are not met at\nscheduling time, the pod will not be scheduled onto the node.\nIf the affinity requirements specified by this field cease to be met\nat some point during pod execution (e.g. due to a pod label update), the\nsystem may or may not try to eventually evict the pod from its node.\nWhen there are multiple elements, the lists of nodes corresponding to each\npodAffinityTerm are intersected, i.e. all terms must be satisfied.", "items": { "additionalProperties": false, "description": "Defines a set of pods (namely those matching the labelSelector\nrelative to the given namespace(s)) that this pod should be\nco-located (affinity) or not co-located (anti-affinity) with,\nwhere co-located is defined as running on a node whose value of\nthe label with key \u003ctopologyKey\u003e matches that of any node on which\na pod of the set of pods is running", "properties": { "labelSelector": { "additionalProperties": false, "description": "A label query over a set of resources, in this case pods.\nIf it's null, this PodAffinityTerm matches with no Pods.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "matchLabelKeys": { "description": "MatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key in (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both matchLabelKeys and labelSelector.\nAlso, matchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "mismatchLabelKeys": { "description": "MismatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key notin (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both mismatchLabelKeys and labelSelector.\nAlso, mismatchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "namespaceSelector": { "additionalProperties": false, "description": "A label query over the set of namespaces that the term applies to.\nThe term is applied to the union of the namespaces selected by this field\nand the ones listed in the namespaces field.\nnull selector and null or empty namespaces list means \"this pod's namespace\".\nAn empty selector ({}) matches all namespaces.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "namespaces": { "description": "namespaces specifies a static list of namespace names that the term applies to.\nThe term is applied to the union of the namespaces listed in this field\nand the ones selected by namespaceSelector.\nnull or empty namespaces list and null namespaceSelector means \"this pod's namespace\".", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "topologyKey": { "description": "This pod should be co-located (affinity) or not co-located (anti-affinity) with the pods matching\nthe labelSelector in the specified namespaces, where co-located is defined as running on a node\nwhose value of the label with key topologyKey matches that of any node on which any of the\nselected pods is running.\nEmpty topologyKey is not allowed.", "type": "string" } }, "required": [ "topologyKey" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "podAntiAffinity": { "additionalProperties": false, "description": "Describes pod anti-affinity scheduling rules (e.g. avoid putting this pod in the same node, zone, etc. as some other pod(s)).", "properties": { "preferredDuringSchedulingIgnoredDuringExecution": { "description": "The scheduler will prefer to schedule pods to nodes that satisfy\nthe anti-affinity expressions specified by this field, but it may choose\na node that violates one or more of the expressions. The node that is\nmost preferred is the one with the greatest sum of weights, i.e.\nfor each node that meets all of the scheduling requirements (resource\nrequest, requiredDuringScheduling anti-affinity expressions, etc.),\ncompute a sum by iterating through the elements of this field and subtracting\n\"weight\" from the sum if the node has pods which matches the corresponding podAffinityTerm; the\nnode(s) with the highest sum are the most preferred.", "items": { "additionalProperties": false, "description": "The weights of all of the matched WeightedPodAffinityTerm fields are added per-node to find the most preferred node(s)", "properties": { "podAffinityTerm": { "additionalProperties": false, "description": "Required. A pod affinity term, associated with the corresponding weight.", "properties": { "labelSelector": { "additionalProperties": false, "description": "A label query over a set of resources, in this case pods.\nIf it's null, this PodAffinityTerm matches with no Pods.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "matchLabelKeys": { "description": "MatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key in (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both matchLabelKeys and labelSelector.\nAlso, matchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "mismatchLabelKeys": { "description": "MismatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key notin (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both mismatchLabelKeys and labelSelector.\nAlso, mismatchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "namespaceSelector": { "additionalProperties": false, "description": "A label query over the set of namespaces that the term applies to.\nThe term is applied to the union of the namespaces selected by this field\nand the ones listed in the namespaces field.\nnull selector and null or empty namespaces list means \"this pod's namespace\".\nAn empty selector ({}) matches all namespaces.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "namespaces": { "description": "namespaces specifies a static list of namespace names that the term applies to.\nThe term is applied to the union of the namespaces listed in this field\nand the ones selected by namespaceSelector.\nnull or empty namespaces list and null namespaceSelector means \"this pod's namespace\".", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "topologyKey": { "description": "This pod should be co-located (affinity) or not co-located (anti-affinity) with the pods matching\nthe labelSelector in the specified namespaces, where co-located is defined as running on a node\nwhose value of the label with key topologyKey matches that of any node on which any of the\nselected pods is running.\nEmpty topologyKey is not allowed.", "type": "string" } }, "required": [ "topologyKey" ], "type": "object" }, "weight": { "description": "weight associated with matching the corresponding podAffinityTerm,\nin the range 1-100.", "format": "int32", "type": "integer" } }, "required": [ "podAffinityTerm", "weight" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "requiredDuringSchedulingIgnoredDuringExecution": { "description": "If the anti-affinity requirements specified by this field are not met at\nscheduling time, the pod will not be scheduled onto the node.\nIf the anti-affinity requirements specified by this field cease to be met\nat some point during pod execution (e.g. due to a pod label update), the\nsystem may or may not try to eventually evict the pod from its node.\nWhen there are multiple elements, the lists of nodes corresponding to each\npodAffinityTerm are intersected, i.e. all terms must be satisfied.", "items": { "additionalProperties": false, "description": "Defines a set of pods (namely those matching the labelSelector\nrelative to the given namespace(s)) that this pod should be\nco-located (affinity) or not co-located (anti-affinity) with,\nwhere co-located is defined as running on a node whose value of\nthe label with key \u003ctopologyKey\u003e matches that of any node on which\na pod of the set of pods is running", "properties": { "labelSelector": { "additionalProperties": false, "description": "A label query over a set of resources, in this case pods.\nIf it's null, this PodAffinityTerm matches with no Pods.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "matchLabelKeys": { "description": "MatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key in (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both matchLabelKeys and labelSelector.\nAlso, matchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "mismatchLabelKeys": { "description": "MismatchLabelKeys is a set of pod label keys to select which pods will\nbe taken into consideration. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are merged with `labelSelector` as `key notin (value)`\nto select the group of existing pods which pods will be taken into consideration\nfor the incoming pod's pod (anti) affinity. Keys that don't exist in the incoming\npod labels will be ignored. The default value is empty.\nThe same key is forbidden to exist in both mismatchLabelKeys and labelSelector.\nAlso, mismatchLabelKeys cannot be set when labelSelector isn't set.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "namespaceSelector": { "additionalProperties": false, "description": "A label query over the set of namespaces that the term applies to.\nThe term is applied to the union of the namespaces selected by this field\nand the ones listed in the namespaces field.\nnull selector and null or empty namespaces list means \"this pod's namespace\".\nAn empty selector ({}) matches all namespaces.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "namespaces": { "description": "namespaces specifies a static list of namespace names that the term applies to.\nThe term is applied to the union of the namespaces listed in this field\nand the ones selected by namespaceSelector.\nnull or empty namespaces list and null namespaceSelector means \"this pod's namespace\".", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "topologyKey": { "description": "This pod should be co-located (affinity) or not co-located (anti-affinity) with the pods matching\nthe labelSelector in the specified namespaces, where co-located is defined as running on a node\nwhose value of the label with key topologyKey matches that of any node on which any of the\nselected pods is running.\nEmpty topologyKey is not allowed.", "type": "string" } }, "required": [ "topologyKey" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "args": { "description": "Args overrides the container arguments entirely.\nOnly used when Runtime is \"generic\". For llamacpp, use ExtraArgs instead.", "items": { "type": "string" }, "type": [ "array", "null" ] }, "autoscaling": { "additionalProperties": false, "description": "Autoscaling configures horizontal pod autoscaling for the inference service.\nWhen set, the controller creates and manages an HPA resource targeting the\ninference Deployment. Requires Prometheus Adapter for custom metrics.\nMutually exclusive with manual replica management: when autoscaling is enabled,\nthe Replicas field serves as the initial replica count only.", "properties": { "maxReplicas": { "description": "MaxReplicas is the upper limit for the number of replicas.", "format": "int32", "maximum": 100, "minimum": 1, "type": "integer" }, "metrics": { "description": "Metrics defines the scaling metrics and target values.\nIf empty, defaults to llamacpp:requests_processing with target average value of 2.", "items": { "additionalProperties": false, "description": "MetricSpec defines a single metric for HPA scaling.", "properties": { "name": { "description": "Name is the metric name (e.g., llamacpp:requests_processing).", "type": "string" }, "targetAverageUtilization": { "description": "TargetAverageUtilization is the target utilization percentage for Resource-type metrics.", "format": "int32", "type": [ "integer", "null" ] }, "targetAverageValue": { "description": "TargetAverageValue is the target per-pod average for Pods-type metrics.", "type": [ "string", "null" ] }, "type": { "description": "Type is the metric source type.", "enum": [ "Pods", "Resource" ], "type": "string" } }, "required": [ "name", "type" ], "type": "object" }, "type": [ "array", "null" ] }, "minReplicas": { "default": 1, "description": "MinReplicas is the lower limit for the number of replicas.", "format": "int32", "maximum": 10, "minimum": 1, "type": [ "integer", "null" ] } }, "required": [ "maxReplicas" ], "type": [ "object", "null" ] }, "batchSize": { "description": "BatchSize sets the token batch size for prompt processing.\nLarger values improve throughput but use more memory.\nMaps to llama.cpp --batch-size flag.", "format": "int32", "maximum": 16384, "minimum": 1, "type": [ "integer", "null" ] }, "cacheTypeCustomK": { "description": "CacheTypeCustomK sets a custom KV cache type for keys that is not in the\nstandard enum. Used for llama.cpp forks with additional cache formats such\nas TurboQuant (turbo3, turbo4, tbqp3, etc.). Maps to llama.cpp\n--cache-type-k. The runtime binary must understand the value or llama-server\nwill fail to start; LLMKube does not validate the string.\nTakes precedence over CacheTypeK when both are set.", "type": [ "string", "null" ] }, "cacheTypeCustomV": { "description": "CacheTypeCustomV sets a custom KV cache type for values that is not in the\nstandard enum. See CacheTypeCustomK for usage notes. Takes precedence over\nCacheTypeV when both are set.", "type": [ "string", "null" ] }, "cacheTypeK": { "description": "CacheTypeK sets the KV cache quantization type for keys.\nSupported values depend on the llama.cpp build version.\nMaps to llama.cpp --cache-type-k flag. Default: f16 (llama.cpp default).\nFor custom build types not in the enum (e.g. TurboQuant turbo3, tbqp3), use\nCacheTypeCustomK instead.", "enum": [ "f16", "f32", "q8_0", "q4_0", "q4_1", "q5_0", "q5_1", "iq4_nl" ], "type": [ "string", "null" ] }, "cacheTypeV": { "description": "CacheTypeV sets the KV cache quantization type for values.\nMaps to llama.cpp --cache-type-v flag. Default: f16 (llama.cpp default).\nFor custom build types not in the enum (e.g. TurboQuant turbo3, tbqp3), use\nCacheTypeCustomV instead.", "enum": [ "f16", "f32", "q8_0", "q4_0", "q4_1", "q5_0", "q5_1", "iq4_nl" ], "type": [ "string", "null" ] }, "command": { "description": "Command overrides the container entrypoint.\nOnly used when Runtime is \"generic\" or for advanced customization.", "items": { "type": "string" }, "type": [ "array", "null" ] }, "containerPort": { "description": "ContainerPort overrides the primary container port.\nEach runtime has its own default (llamacpp: 8080).", "format": "int32", "maximum": 65535, "minimum": 1, "type": [ "integer", "null" ] }, "contextSize": { "description": "ContextSize sets the context window size for the llama.cpp server (-c flag).\nLarger values allow processing longer inputs but require more memory.\nIf not specified, llama.cpp uses its default (typically 512 or 2048).\nThe upper bound covers Qwen 3.6 at 1M-via-YaRN with margin and accommodates\nnear-future hybrid-attention model architectures. KV cache memory is the\nuser's responsibility to size via spec.resources.memory or hostMemory.", "format": "int32", "maximum": 2097152, "minimum": 128, "type": [ "integer", "null" ] }, "disruption": { "additionalProperties": false, "description": "Disruption controls how the operator manages node-disruption annotations\non inference pods during the vulnerable startup window (model download +\nload). When ProtectStartup is true (the default), the operator sets\nkarpenter.sh/do-not-disrupt: \"true\" on the pod template while the\nInferenceService is not yet Ready, then removes it once the service\nreaches the Ready phase. Set ProtectAlways to true to keep the annotation\npermanently (equivalent to setting it via podAnnotations). User-provided\npodAnnotations always win on collision.", "properties": { "protectAlways": { "description": "ProtectAlways keeps the disruption-protection annotation on the pod\ntemplate permanently, regardless of the InferenceService phase. This is\nequivalent to setting karpenter.sh/do-not-disrupt: \"true\" via\npodAnnotations, but managed by the operator. Defaults to false.", "type": [ "boolean", "null" ] }, "protectStartup": { "default": true, "description": "ProtectStartup prevents node disruption (e.g., Karpenter consolidation,\nCluster Autoscaler scale-down) while the InferenceService is starting up.\nWhen true, the operator sets karpenter.sh/do-not-disrupt: \"true\" on the\npod template until the InferenceService reaches the Ready phase, then\nremoves it. Defaults to true.", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ] }, "endpoint": { "additionalProperties": false, "description": "Endpoint defines the service endpoint configuration", "properties": { "gateway": { "additionalProperties": false, "description": "Gateway opts this InferenceService into Envoy AI Gateway exposure. When\nset and Enabled, the operator generates the Backend / AIServiceBackend /\nAIGatewayRoute resources that front this service through a pre-installed\nEnvoy AI Gateway. nil (the default) preserves today's behavior (no\ngateway resources). The Envoy AI Gateway stack and the referenced Gateway\nare a documented prerequisite; LLMKube does not install or own them.", "properties": { "enabled": { "description": "Enabled is the opt-in switch. When false (or when Gateway is nil), the\noperator generates no gateway resources for this InferenceService.", "type": [ "boolean", "null" ] }, "gatewayRef": { "additionalProperties": false, "description": "GatewayRef identifies the pre-installed Gateway (gateway.networking.k8s.io)\nthe generated AIGatewayRoute attaches to. The Gateway typically lives in a\ndedicated gateway namespace; cross-namespace attachment requires the\nGateway listener's allowedRoutes.namespaces to permit this\nInferenceService's namespace (a documented prerequisite for the MVP; the\noperator does not generate ReferenceGrants or touch the listener).", "properties": { "name": { "description": "Name is the Gateway's name.", "type": "string" }, "namespace": { "description": "Namespace is the Gateway's namespace. Empty means the InferenceService's\nown namespace.", "type": [ "string", "null" ] } }, "required": [ "name" ], "type": "object" }, "modelName": { "description": "ModelName is the OpenAI \"model\" string clients send, matched by the\ngenerated route rule (the x-ai-eg-model header the gateway's ext_proc\npopulates from the request body). Defaults to ModelRef, falling back to\nthe InferenceService name when ModelRef is empty.", "type": [ "string", "null" ] } }, "required": [ "gatewayRef" ], "type": [ "object", "null" ] }, "nodePort": { "description": "NodePort is the specific NodePort to pin when endpoint.type is NodePort.\nIf set, the Service will use this exact port instead of auto-assigning\nfrom the 30000-32767 range. This provides a stable external endpoint\nacross redeployments.", "format": "int32", "maximum": 32767, "minimum": 30000, "type": [ "integer", "null" ] }, "path": { "default": "/v1/chat/completions", "description": "Path is the HTTP path for the inference endpoint", "type": [ "string", "null" ] }, "port": { "default": 8080, "description": "Port is the service port", "format": "int32", "maximum": 65535, "minimum": 1, "type": [ "integer", "null" ] }, "type": { "default": "ClusterIP", "description": "Type is the Kubernetes service type (ClusterIP, NodePort, LoadBalancer)", "enum": [ "ClusterIP", "NodePort", "LoadBalancer" ], "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "env": { "description": "Env adds environment variables to the inference container.\nUseful for HF_TOKEN, custom runtime config, etc.", "items": { "additionalProperties": false, "description": "EnvVar represents an environment variable present in a Container.", "properties": { "name": { "description": "Name of the environment variable.\nMay consist of any printable ASCII characters except '='.", "type": "string" }, "value": { "description": "Variable references $(VAR_NAME) are expanded\nusing the previously defined environment variables in the container and\nany service environment variables. If a variable cannot be resolved,\nthe reference in the input string will be unchanged. Double $$ are reduced\nto a single $, which allows for escaping the $(VAR_NAME) syntax: i.e.\n\"$$(VAR_NAME)\" will produce the string literal \"$(VAR_NAME)\".\nEscaped references will never be expanded, regardless of whether the variable\nexists or not.\nDefaults to \"\".", "type": [ "string", "null" ] }, "valueFrom": { "additionalProperties": false, "description": "Source for the environment variable's value. Cannot be used if value is not empty.", "properties": { "configMapKeyRef": { "additionalProperties": false, "description": "Selects a key of a ConfigMap.", "properties": { "key": { "description": "The key to select.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the ConfigMap or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "fieldRef": { "additionalProperties": false, "description": "Selects a field of the pod: supports metadata.name, metadata.namespace, `metadata.labels['\u003cKEY\u003e']`, `metadata.annotations['\u003cKEY\u003e']`,\nspec.nodeName, spec.serviceAccountName, status.hostIP, status.podIP, status.podIPs.", "properties": { "apiVersion": { "description": "Version of the schema the FieldPath is written in terms of, defaults to \"v1\".", "type": [ "string", "null" ] }, "fieldPath": { "description": "Path of the field to select in the specified API version.", "type": "string" } }, "required": [ "fieldPath" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "fileKeyRef": { "additionalProperties": false, "description": "FileKeyRef selects a key of the env file.\nRequires the EnvFiles feature gate to be enabled.", "properties": { "key": { "description": "The key within the env file. An invalid key will prevent the pod from starting.\nThe keys defined within a source may consist of any printable ASCII characters except '='.\nDuring Alpha stage of the EnvFiles feature gate, the key size is limited to 128 characters.", "type": "string" }, "optional": { "default": false, "description": "Specify whether the file or its key must be defined. If the file or key\ndoes not exist, then the env var is not published.\nIf optional is set to true and the specified key does not exist,\nthe environment variable will not be set in the Pod's containers.\n\nIf optional is set to false and the specified key does not exist,\nan error will be returned during Pod creation.", "type": [ "boolean", "null" ] }, "path": { "description": "The path within the volume from which to select the file.\nMust be relative and may not contain the '..' path or start with '..'.", "type": "string" }, "volumeName": { "description": "The name of the volume mount containing the env file.", "type": "string" } }, "required": [ "key", "path", "volumeName" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "resourceFieldRef": { "additionalProperties": false, "description": "Selects a resource of the container: only resources limits and requests\n(limits.cpu, limits.memory, limits.ephemeral-storage, requests.cpu, requests.memory and requests.ephemeral-storage) are currently supported.", "properties": { "containerName": { "description": "Container name: required for volumes, optional for env vars", "type": [ "string", "null" ] }, "divisor": { "description": "Specifies the output format of the exposed resources, defaults to \"1\"", "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" }, { "type": "null" } ], "x-kubernetes-int-or-string": true }, "resource": { "description": "Required: resource to select", "type": "string" } }, "required": [ "resource" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "secretKeyRef": { "additionalProperties": false, "description": "Selects a key of a secret in the pod's namespace", "properties": { "key": { "description": "The key of the secret to select from. Must be a valid secret key.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" } }, "type": [ "object", "null" ] } }, "required": [ "name" ], "type": "object" }, "type": [ "array", "null" ] }, "evictionProtection": { "description": "EvictionProtection marks this service as ineligible for memory-pressure\neviction by the metal-agent watchdog. Use this for production workloads\nthat should never be silently stopped under memory pressure, even when\nthey are the lowest-priority option. The agent's per-process pickEvictionTarget\nexcludes protected processes from the eviction-candidate set; the\nMemoryPressure status condition is still patched on protected services\nfor operator visibility.\n\nHas no effect when --eviction-enabled is unset on the metal-agent or\nfor non-llama-server runtimes (oMLX, Ollama). Defaults to false.", "type": [ "boolean", "null" ] }, "extraArgs": { "description": "ExtraArgs provides additional command-line arguments passed directly to the\nruntime process. Use for flags not yet supported as typed CRD fields.\nArguments are appended after all other configured flags.\nSupported by the \"llamacpp\" and \"vllm\" runtimes. Ignored by others.\nExample: [\"--seed\", \"42\", \"--log-disable\"]", "items": { "type": "string" }, "type": [ "array", "null" ] }, "extraVolumeMounts": { "description": "ExtraVolumeMounts mounts ExtraVolumes into the inference container,\nappended after the model-storage mounts. Names must match an entry in\nExtraVolumes (or a volume from another passthrough field).", "items": { "additionalProperties": false, "description": "VolumeMount describes a mounting of a Volume within a container.", "properties": { "mountPath": { "description": "Path within the container at which the volume should be mounted. Must\nnot contain ':'.", "type": "string" }, "mountPropagation": { "description": "mountPropagation determines how mounts are propagated from the host\nto container and the other way around.\nWhen not set, MountPropagationNone is used.\nThis field is beta in 1.10.\nWhen RecursiveReadOnly is set to IfPossible or to Enabled, MountPropagation must be None or unspecified\n(which defaults to None).", "type": [ "string", "null" ] }, "name": { "description": "This must match the Name of a Volume.", "type": "string" }, "readOnly": { "description": "Mounted read-only if true, read-write otherwise (false or unspecified).\nDefaults to false.", "type": [ "boolean", "null" ] }, "recursiveReadOnly": { "description": "RecursiveReadOnly specifies whether read-only mounts should be handled\nrecursively.\n\nIf ReadOnly is false, this field has no meaning and must be unspecified.\n\nIf ReadOnly is true, and this field is set to Disabled, the mount is not made\nrecursively read-only. If this field is set to IfPossible, the mount is made\nrecursively read-only, if it is supported by the container runtime. If this\nfield is set to Enabled, the mount is made recursively read-only if it is\nsupported by the container runtime, otherwise the pod will not be started and\nan error will be generated to indicate the reason.\n\nIf this field is set to IfPossible or Enabled, MountPropagation must be set to\nNone (or be unspecified, which defaults to None).\n\nIf this field is not specified, it is treated as an equivalent of Disabled.", "type": [ "string", "null" ] }, "subPath": { "description": "Path within the volume from which the container's volume should be mounted.\nDefaults to \"\" (volume's root).", "type": [ "string", "null" ] }, "subPathExpr": { "description": "Expanded path within the volume from which the container's volume should be mounted.\nBehaves similarly to SubPath but environment variable references $(VAR_NAME) are expanded using the container's environment.\nDefaults to \"\" (volume's root).\nSubPathExpr and SubPath are mutually exclusive.", "type": [ "string", "null" ] } }, "required": [ "mountPath", "name" ], "type": "object" }, "type": [ "array", "null" ] }, "extraVolumes": { "description": "ExtraVolumes adds additional Volumes to the inference Pod, appended\nafter the model-storage volumes built from ModelRef. Useful for a\nruntime-owned cache (e.g. a JIT kernel cache) that is unrelated to\nmodel weights and doesn't fit ModelCache's model-scoped PVC path.\nPair with ExtraVolumeMounts to actually mount it into the container.", "items": { "additionalProperties": false, "description": "Volume represents a named volume in a pod that may be accessed by any container in the pod.", "properties": { "awsElasticBlockStore": { "additionalProperties": false, "description": "awsElasticBlockStore represents an AWS Disk resource that is attached to a\nkubelet's host machine and then exposed to the pod.\nDeprecated: AWSElasticBlockStore is deprecated. All operations for the in-tree\nawsElasticBlockStore type are redirected to the ebs.csi.aws.com CSI driver.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#awselasticblockstore", "properties": { "fsType": { "description": "fsType is the filesystem type of the volume that you want to mount.\nTip: Ensure that the filesystem type is supported by the host operating system.\nExamples: \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#awselasticblockstore", "type": [ "string", "null" ] }, "partition": { "description": "partition is the partition in the volume that you want to mount.\nIf omitted, the default is to mount by volume name.\nExamples: For volume /dev/sda1, you specify the partition as \"1\".\nSimilarly, the volume partition for /dev/sda is \"0\" (or you can leave the property empty).", "format": "int32", "type": [ "integer", "null" ] }, "readOnly": { "description": "readOnly value true will force the readOnly setting in VolumeMounts.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#awselasticblockstore", "type": [ "boolean", "null" ] }, "volumeID": { "description": "volumeID is unique ID of the persistent disk resource in AWS (Amazon EBS volume).\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#awselasticblockstore", "type": "string" } }, "required": [ "volumeID" ], "type": [ "object", "null" ] }, "azureDisk": { "additionalProperties": false, "description": "azureDisk represents an Azure Data Disk mount on the host and bind mount to the pod.\nDeprecated: AzureDisk is deprecated. All operations for the in-tree azureDisk type\nare redirected to the disk.csi.azure.com CSI driver.", "properties": { "cachingMode": { "description": "cachingMode is the Host Caching mode: None, Read Only, Read Write.", "type": [ "string", "null" ] }, "diskName": { "description": "diskName is the Name of the data disk in the blob storage", "type": "string" }, "diskURI": { "description": "diskURI is the URI of data disk in the blob storage", "type": "string" }, "fsType": { "default": "ext4", "description": "fsType is Filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "kind": { "description": "kind expected values are Shared: multiple blob disks per storage account Dedicated: single blob disk per storage account Managed: azure managed data disk (only in managed availability set). defaults to shared", "type": [ "string", "null" ] }, "readOnly": { "default": false, "description": "readOnly Defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] } }, "required": [ "diskName", "diskURI" ], "type": [ "object", "null" ] }, "azureFile": { "additionalProperties": false, "description": "azureFile represents an Azure File Service mount on the host and bind mount to the pod.\nDeprecated: AzureFile is deprecated. All operations for the in-tree azureFile type\nare redirected to the file.csi.azure.com CSI driver.", "properties": { "readOnly": { "description": "readOnly defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "secretName": { "description": "secretName is the name of secret that contains Azure Storage Account Name and Key", "type": "string" }, "shareName": { "description": "shareName is the azure share Name", "type": "string" } }, "required": [ "secretName", "shareName" ], "type": [ "object", "null" ] }, "cephfs": { "additionalProperties": false, "description": "cephFS represents a Ceph FS mount on the host that shares a pod's lifetime.\nDeprecated: CephFS is deprecated and the in-tree cephfs type is no longer supported.", "properties": { "monitors": { "description": "monitors is Required: Monitors is a collection of Ceph monitors\nMore info: https://examples.k8s.io/volumes/cephfs/README.md#how-to-use-it", "items": { "type": "string" }, "type": "array", "x-kubernetes-list-type": "atomic" }, "path": { "description": "path is Optional: Used as the mounted root, rather than the full Ceph tree, default is /", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly is Optional: Defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.\nMore info: https://examples.k8s.io/volumes/cephfs/README.md#how-to-use-it", "type": [ "boolean", "null" ] }, "secretFile": { "description": "secretFile is Optional: SecretFile is the path to key ring for User, default is /etc/ceph/user.secret\nMore info: https://examples.k8s.io/volumes/cephfs/README.md#how-to-use-it", "type": [ "string", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef is Optional: SecretRef is reference to the authentication secret for User, default is empty.\nMore info: https://examples.k8s.io/volumes/cephfs/README.md#how-to-use-it", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "user": { "description": "user is optional: User is the rados user name, default is admin\nMore info: https://examples.k8s.io/volumes/cephfs/README.md#how-to-use-it", "type": [ "string", "null" ] } }, "required": [ "monitors" ], "type": [ "object", "null" ] }, "cinder": { "additionalProperties": false, "description": "cinder represents a cinder volume attached and mounted on kubelets host machine.\nDeprecated: Cinder is deprecated. All operations for the in-tree cinder type\nare redirected to the cinder.csi.openstack.org CSI driver.\nMore info: https://examples.k8s.io/mysql-cinder-pd/README.md", "properties": { "fsType": { "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nExamples: \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.\nMore info: https://examples.k8s.io/mysql-cinder-pd/README.md", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.\nMore info: https://examples.k8s.io/mysql-cinder-pd/README.md", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef is optional: points to a secret object containing parameters used to connect\nto OpenStack.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "volumeID": { "description": "volumeID used to identify the volume in cinder.\nMore info: https://examples.k8s.io/mysql-cinder-pd/README.md", "type": "string" } }, "required": [ "volumeID" ], "type": [ "object", "null" ] }, "configMap": { "additionalProperties": false, "description": "configMap represents a configMap that should populate this volume", "properties": { "defaultMode": { "description": "defaultMode is optional: mode bits used to set permissions on created files by default.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nDefaults to 0644.\nDirectories within the path are not affected by this setting.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "items": { "description": "items if unspecified, each key-value pair in the Data field of the referenced\nConfigMap will be projected into the volume as a file whose name is the\nkey and content is the value. If specified, the listed keys will be\nprojected into the specified paths, and unlisted keys will not be\npresent. If a key is specified which is not present in the ConfigMap,\nthe volume setup will error unless it is marked optional. Paths must be\nrelative and may not contain the '..' path or start with '..'.", "items": { "additionalProperties": false, "description": "Maps a string key to a path within a volume.", "properties": { "key": { "description": "key is the key to project.", "type": "string" }, "mode": { "description": "mode is Optional: mode bits used to set permissions on this file.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "path is the relative path of the file to map the key to.\nMay not be an absolute path.\nMay not contain the path element '..'.\nMay not start with the string '..'.", "type": "string" } }, "required": [ "key", "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "optional specify whether the ConfigMap or its keys must be defined", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "csi": { "additionalProperties": false, "description": "csi (Container Storage Interface) represents ephemeral storage that is handled by certain external CSI drivers.", "properties": { "driver": { "description": "driver is the name of the CSI driver that handles this volume.\nConsult with your admin for the correct name as registered in the cluster.", "type": "string" }, "fsType": { "description": "fsType to mount. Ex. \"ext4\", \"xfs\", \"ntfs\".\nIf not provided, the empty value is passed to the associated CSI driver\nwhich will determine the default filesystem to apply.", "type": [ "string", "null" ] }, "nodePublishSecretRef": { "additionalProperties": false, "description": "nodePublishSecretRef is a reference to the secret object containing\nsensitive information to pass to the CSI driver to complete the CSI\nNodePublishVolume and NodeUnpublishVolume calls.\nThis field is optional, and may be empty if no secret is required. If the\nsecret object contains more than one secret, all secret references are passed.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "readOnly": { "description": "readOnly specifies a read-only configuration for the volume.\nDefaults to false (read/write).", "type": [ "boolean", "null" ] }, "volumeAttributes": { "additionalProperties": { "type": "string" }, "description": "volumeAttributes stores driver-specific properties that are passed to the CSI\ndriver. Consult your driver's documentation for supported values.", "type": [ "object", "null" ] } }, "required": [ "driver" ], "type": [ "object", "null" ] }, "downwardAPI": { "additionalProperties": false, "description": "downwardAPI represents downward API about the pod that should populate this volume", "properties": { "defaultMode": { "description": "Optional: mode bits to use on created files by default. Must be a\nOptional: mode bits used to set permissions on created files by default.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nDefaults to 0644.\nDirectories within the path are not affected by this setting.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "items": { "description": "Items is a list of downward API volume file", "items": { "additionalProperties": false, "description": "DownwardAPIVolumeFile represents information to create the file containing the pod field", "properties": { "fieldRef": { "additionalProperties": false, "description": "Required: Selects a field of the pod: only annotations, labels, name, namespace and uid are supported.", "properties": { "apiVersion": { "description": "Version of the schema the FieldPath is written in terms of, defaults to \"v1\".", "type": [ "string", "null" ] }, "fieldPath": { "description": "Path of the field to select in the specified API version.", "type": "string" } }, "required": [ "fieldPath" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "mode": { "description": "Optional: mode bits used to set permissions on this file, must be an octal value\nbetween 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "Required: Path is the relative path name of the file to be created. Must not be absolute or contain the '..' path. Must be utf-8 encoded. The first item of the relative path must not start with '..'", "type": "string" }, "resourceFieldRef": { "additionalProperties": false, "description": "Selects a resource of the container: only resources limits and requests\n(limits.cpu, limits.memory, requests.cpu and requests.memory) are currently supported.", "properties": { "containerName": { "description": "Container name: required for volumes, optional for env vars", "type": [ "string", "null" ] }, "divisor": { "description": "Specifies the output format of the exposed resources, defaults to \"1\"", "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" }, { "type": "null" } ], "x-kubernetes-int-or-string": true }, "resource": { "description": "Required: resource to select", "type": "string" } }, "required": [ "resource" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" } }, "required": [ "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "emptyDir": { "additionalProperties": false, "description": "emptyDir represents a temporary directory that shares a pod's lifetime.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#emptydir", "properties": { "medium": { "description": "medium represents what type of storage medium should back this directory.\nThe default is \"\" which means to use the node's default medium.\nMust be an empty string (default) or Memory.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#emptydir", "type": [ "string", "null" ] }, "sizeLimit": { "description": "sizeLimit is the total amount of local storage required for this EmptyDir volume.\nThe size limit is also applicable for memory medium.\nThe maximum usage on memory medium EmptyDir would be the minimum value between\nthe SizeLimit specified here and the sum of memory limits of all containers in a pod.\nThe default is nil which means that the limit is undefined.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#emptydir", "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" }, { "type": "null" } ], "x-kubernetes-int-or-string": true } }, "type": [ "object", "null" ] }, "ephemeral": { "additionalProperties": false, "description": "ephemeral represents a volume that is handled by a cluster storage driver.\nThe volume's lifecycle is tied to the pod that defines it - it will be created before the pod starts,\nand deleted when the pod is removed.\n\nUse this if:\na) the volume is only needed while the pod runs,\nb) features of normal volumes like restoring from snapshot or capacity\n tracking are needed,\nc) the storage driver is specified through a storage class, and\nd) the storage driver supports dynamic volume provisioning through\n a PersistentVolumeClaim (see EphemeralVolumeSource for more\n information on the connection between this volume type\n and PersistentVolumeClaim).\n\nUse PersistentVolumeClaim or one of the vendor-specific\nAPIs for volumes that persist for longer than the lifecycle\nof an individual pod.\n\nUse CSI for light-weight local ephemeral volumes if the CSI driver is meant to\nbe used that way - see the documentation of the driver for\nmore information.\n\nA pod can use both types of ephemeral volumes and\npersistent volumes at the same time.", "properties": { "volumeClaimTemplate": { "additionalProperties": false, "description": "Will be used to create a stand-alone PVC to provision the volume.\nThe pod in which this EphemeralVolumeSource is embedded will be the\nowner of the PVC, i.e. the PVC will be deleted together with the\npod. The name of the PVC will be `\u003cpod name\u003e-\u003cvolume name\u003e` where\n`\u003cvolume name\u003e` is the name from the `PodSpec.Volumes` array\nentry. Pod validation will reject the pod if the concatenated name\nis not valid for a PVC (for example, too long).\n\nAn existing PVC with that name that is not owned by the pod\nwill *not* be used for the pod to avoid using an unrelated\nvolume by mistake. Starting the pod is then blocked until\nthe unrelated PVC is removed. If such a pre-created PVC is\nmeant to be used by the pod, the PVC has to updated with an\nowner reference to the pod once the pod exists. Normally\nthis should not be necessary, but it may be useful when\nmanually reconstructing a broken cluster.\n\nThis field is read-only and no changes will be made by Kubernetes\nto the PVC after it has been created.\n\nRequired, must not be nil.", "properties": { "metadata": { "description": "May contain labels and annotations that will be copied into the PVC\nwhen creating it. No other fields are allowed and will be rejected during\nvalidation.", "type": [ "object", "null" ] }, "spec": { "additionalProperties": false, "description": "The specification for the PersistentVolumeClaim. The entire content is\ncopied unchanged into the PVC that gets created from this\ntemplate. The same fields as in a PersistentVolumeClaim\nare also valid here.", "properties": { "accessModes": { "description": "accessModes contains the desired access modes the volume should have.\nMore info: https://kubernetes.io/docs/concepts/storage/persistent-volumes#access-modes-1", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "dataSource": { "additionalProperties": false, "description": "dataSource field can be used to specify either:\n* An existing VolumeSnapshot object (snapshot.storage.k8s.io/VolumeSnapshot)\n* An existing PVC (PersistentVolumeClaim)\nIf the provisioner or an external controller can support the specified data source,\nit will create a new volume based on the contents of the specified data source.\nWhen the AnyVolumeDataSource feature gate is enabled, dataSource contents will be copied to dataSourceRef,\nand dataSourceRef contents will be copied to dataSource when dataSourceRef.namespace is not specified.\nIf the namespace is specified, then dataSourceRef will not be copied to dataSource.", "properties": { "apiGroup": { "description": "APIGroup is the group for the resource being referenced.\nIf APIGroup is not specified, the specified Kind must be in the core API group.\nFor any other third-party types, APIGroup is required.", "type": [ "string", "null" ] }, "kind": { "description": "Kind is the type of resource being referenced", "type": "string" }, "name": { "description": "Name is the name of resource being referenced", "type": "string" } }, "required": [ "kind", "name" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "dataSourceRef": { "additionalProperties": false, "description": "dataSourceRef specifies the object from which to populate the volume with data, if a non-empty\nvolume is desired. This may be any object from a non-empty API group (non\ncore object) or a PersistentVolumeClaim object.\nWhen this field is specified, volume binding will only succeed if the type of\nthe specified object matches some installed volume populator or dynamic\nprovisioner.\nThis field will replace the functionality of the dataSource field and as such\nif both fields are non-empty, they must have the same value. For backwards\ncompatibility, when namespace isn't specified in dataSourceRef,\nboth fields (dataSource and dataSourceRef) will be set to the same\nvalue automatically if one of them is empty and the other is non-empty.\nWhen namespace is specified in dataSourceRef,\ndataSource isn't set to the same value and must be empty.\nThere are three important differences between dataSource and dataSourceRef:\n* While dataSource only allows two specific types of objects, dataSourceRef\n allows any non-core object, as well as PersistentVolumeClaim objects.\n* While dataSource ignores disallowed values (dropping them), dataSourceRef\n preserves all values, and generates an error if a disallowed value is\n specified.\n* While dataSource only allows local objects, dataSourceRef allows objects\n in any namespaces.\n(Beta) Using this field requires the AnyVolumeDataSource feature gate to be enabled.\n(Alpha) Using the namespace field of dataSourceRef requires the CrossNamespaceVolumeDataSource feature gate to be enabled.", "properties": { "apiGroup": { "description": "APIGroup is the group for the resource being referenced.\nIf APIGroup is not specified, the specified Kind must be in the core API group.\nFor any other third-party types, APIGroup is required.", "type": [ "string", "null" ] }, "kind": { "description": "Kind is the type of resource being referenced", "type": "string" }, "name": { "description": "Name is the name of resource being referenced", "type": "string" }, "namespace": { "description": "Namespace is the namespace of resource being referenced\nNote that when a namespace is specified, a gateway.networking.k8s.io/ReferenceGrant object is required in the referent namespace to allow that namespace's owner to accept the reference. See the ReferenceGrant documentation for details.\n(Alpha) This field requires the CrossNamespaceVolumeDataSource feature gate to be enabled.", "type": [ "string", "null" ] } }, "required": [ "kind", "name" ], "type": [ "object", "null" ] }, "resources": { "additionalProperties": false, "description": "resources represents the minimum resources the volume should have.\nUsers are allowed to specify resource requirements\nthat are lower than previous value but must still be higher than capacity recorded in the\nstatus field of the claim.\nMore info: https://kubernetes.io/docs/concepts/storage/persistent-volumes#resources", "properties": { "limits": { "additionalProperties": { "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true }, "description": "Limits describes the maximum amount of compute resources allowed.\nMore info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/", "type": [ "object", "null" ] }, "requests": { "additionalProperties": { "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true }, "description": "Requests describes the minimum amount of compute resources required.\nIf Requests is omitted for a container, it defaults to Limits if that is explicitly specified,\notherwise to an implementation-defined value. Requests cannot exceed Limits.\nMore info: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/", "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "selector": { "additionalProperties": false, "description": "selector is a label query over volumes to consider for binding.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "storageClassName": { "description": "storageClassName is the name of the StorageClass required by the claim.\nMore info: https://kubernetes.io/docs/concepts/storage/persistent-volumes#class-1", "type": [ "string", "null" ] }, "volumeAttributesClassName": { "description": "volumeAttributesClassName may be used to set the VolumeAttributesClass used by this claim.\nIf specified, the CSI driver will create or update the volume with the attributes defined\nin the corresponding VolumeAttributesClass. This has a different purpose than storageClassName,\nit can be changed after the claim is created. An empty string or nil value indicates that no\nVolumeAttributesClass will be applied to the claim. If the claim enters an Infeasible error state,\nthis field can be reset to its previous value (including nil) to cancel the modification.\nIf the resource referred to by volumeAttributesClass does not exist, this PersistentVolumeClaim will be\nset to a Pending state, as reflected by the modifyVolumeStatus field, until such as a resource\nexists.\nMore info: https://kubernetes.io/docs/concepts/storage/volume-attributes-classes/", "type": [ "string", "null" ] }, "volumeMode": { "description": "volumeMode defines what type of volume is required by the claim.\nValue of Filesystem is implied when not included in claim spec.", "type": [ "string", "null" ] }, "volumeName": { "description": "volumeName is the binding reference to the PersistentVolume backing this claim.", "type": [ "string", "null" ] } }, "type": "object" } }, "required": [ "spec" ], "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "fc": { "additionalProperties": false, "description": "fc represents a Fibre Channel resource that is attached to a kubelet's host machine and then exposed to the pod.", "properties": { "fsType": { "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "lun": { "description": "lun is Optional: FC target lun number", "format": "int32", "type": [ "integer", "null" ] }, "readOnly": { "description": "readOnly is Optional: Defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "targetWWNs": { "description": "targetWWNs is Optional: FC target worldwide names (WWNs)", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "wwids": { "description": "wwids Optional: FC volume world wide identifiers (wwids)\nEither wwids or combination of targetWWNs and lun must be set, but not both simultaneously.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "flexVolume": { "additionalProperties": false, "description": "flexVolume represents a generic volume resource that is\nprovisioned/attached using an exec based plugin.\nDeprecated: FlexVolume is deprecated. Consider using a CSIDriver instead.", "properties": { "driver": { "description": "driver is the name of the driver to use for this volume.", "type": "string" }, "fsType": { "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". The default filesystem depends on FlexVolume script.", "type": [ "string", "null" ] }, "options": { "additionalProperties": { "type": "string" }, "description": "options is Optional: this field holds extra command options if any.", "type": [ "object", "null" ] }, "readOnly": { "description": "readOnly is Optional: defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef is Optional: secretRef is reference to the secret object containing\nsensitive information to pass to the plugin scripts. This may be\nempty if no secret object is specified. If the secret object\ncontains more than one secret, all secrets are passed to the plugin\nscripts.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" } }, "required": [ "driver" ], "type": [ "object", "null" ] }, "flocker": { "additionalProperties": false, "description": "flocker represents a Flocker volume attached to a kubelet's host machine. This depends on the Flocker control service being running.\nDeprecated: Flocker is deprecated and the in-tree flocker type is no longer supported.", "properties": { "datasetName": { "description": "datasetName is Name of the dataset stored as metadata -\u003e name on the dataset for Flocker\nshould be considered as deprecated", "type": [ "string", "null" ] }, "datasetUUID": { "description": "datasetUUID is the UUID of the dataset. This is unique identifier of a Flocker dataset", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "gcePersistentDisk": { "additionalProperties": false, "description": "gcePersistentDisk represents a GCE Disk resource that is attached to a\nkubelet's host machine and then exposed to the pod.\nDeprecated: GCEPersistentDisk is deprecated. All operations for the in-tree\ngcePersistentDisk type are redirected to the pd.csi.storage.gke.io CSI driver.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#gcepersistentdisk", "properties": { "fsType": { "description": "fsType is filesystem type of the volume that you want to mount.\nTip: Ensure that the filesystem type is supported by the host operating system.\nExamples: \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#gcepersistentdisk", "type": [ "string", "null" ] }, "partition": { "description": "partition is the partition in the volume that you want to mount.\nIf omitted, the default is to mount by volume name.\nExamples: For volume /dev/sda1, you specify the partition as \"1\".\nSimilarly, the volume partition for /dev/sda is \"0\" (or you can leave the property empty).\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#gcepersistentdisk", "format": "int32", "type": [ "integer", "null" ] }, "pdName": { "description": "pdName is unique name of the PD resource in GCE. Used to identify the disk in GCE.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#gcepersistentdisk", "type": "string" }, "readOnly": { "description": "readOnly here will force the ReadOnly setting in VolumeMounts.\nDefaults to false.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#gcepersistentdisk", "type": [ "boolean", "null" ] } }, "required": [ "pdName" ], "type": [ "object", "null" ] }, "gitRepo": { "additionalProperties": false, "description": "gitRepo represents a git repository at a particular revision.\nDeprecated: GitRepo is deprecated. To provision a container with a git repo, mount an\nEmptyDir into an InitContainer that clones the repo using git, then mount the EmptyDir\ninto the Pod's container.", "properties": { "directory": { "description": "directory is the target directory name.\nMust not contain or start with '..'. If '.' is supplied, the volume directory will be the\ngit repository. Otherwise, if specified, the volume will contain the git repository in\nthe subdirectory with the given name.", "type": [ "string", "null" ] }, "repository": { "description": "repository is the URL", "type": "string" }, "revision": { "description": "revision is the commit hash for the specified revision.", "type": [ "string", "null" ] } }, "required": [ "repository" ], "type": [ "object", "null" ] }, "glusterfs": { "additionalProperties": false, "description": "glusterfs represents a Glusterfs mount on the host that shares a pod's lifetime.\nDeprecated: Glusterfs is deprecated and the in-tree glusterfs type is no longer supported.", "properties": { "endpoints": { "description": "endpoints is the endpoint name that details Glusterfs topology.", "type": "string" }, "path": { "description": "path is the Glusterfs volume path.\nMore info: https://examples.k8s.io/volumes/glusterfs/README.md#create-a-pod", "type": "string" }, "readOnly": { "description": "readOnly here will force the Glusterfs volume to be mounted with read-only permissions.\nDefaults to false.\nMore info: https://examples.k8s.io/volumes/glusterfs/README.md#create-a-pod", "type": [ "boolean", "null" ] } }, "required": [ "endpoints", "path" ], "type": [ "object", "null" ] }, "hostPath": { "additionalProperties": false, "description": "hostPath represents a pre-existing file or directory on the host\nmachine that is directly exposed to the container. This is generally\nused for system agents or other privileged things that are allowed\nto see the host machine. Most containers will NOT need this.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#hostpath", "properties": { "path": { "description": "path of the directory on the host.\nIf the path is a symlink, it will follow the link to the real path.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#hostpath", "type": "string" }, "type": { "description": "type for HostPath Volume\nDefaults to \"\"\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#hostpath", "type": [ "string", "null" ] } }, "required": [ "path" ], "type": [ "object", "null" ] }, "image": { "additionalProperties": false, "description": "image represents an OCI object (a container image or artifact) pulled and mounted on the kubelet's host machine.\nThe volume is resolved at pod startup depending on which PullPolicy value is provided:\n\n- Always: the kubelet always attempts to pull the reference. Container creation will fail If the pull fails.\n- Never: the kubelet never pulls the reference and only uses a local image or artifact. Container creation will fail if the reference isn't present.\n- IfNotPresent: the kubelet pulls if the reference isn't already present on disk. Container creation will fail if the reference isn't present and the pull fails.\n\nThe volume gets re-resolved if the pod gets deleted and recreated, which means that new remote content will become available on pod recreation.\nA failure to resolve or pull the image during pod startup will block containers from starting and may add significant latency. Failures will be retried using normal volume backoff and will be reported on the pod reason and message.\nThe types of objects that may be mounted by this volume are defined by the container runtime implementation on a host machine and at minimum must include all valid types supported by the container image field.\nThe OCI object gets mounted in a single directory (spec.containers[*].volumeMounts.mountPath) by merging the manifest layers in the same way as for container images.\nThe volume will be mounted read-only (ro).\nSub path mounts for containers are not supported (spec.containers[*].volumeMounts.subpath) before 1.33.\nThe field spec.securityContext.fsGroupChangePolicy has no effect on this volume type.", "properties": { "pullPolicy": { "description": "Policy for pulling OCI objects. Possible values are:\nAlways: the kubelet always attempts to pull the reference. Container creation will fail If the pull fails.\nNever: the kubelet never pulls the reference and only uses a local image or artifact. Container creation will fail if the reference isn't present.\nIfNotPresent: the kubelet pulls if the reference isn't already present on disk. Container creation will fail if the reference isn't present and the pull fails.\nDefaults to Always if :latest tag is specified, or IfNotPresent otherwise.", "type": [ "string", "null" ] }, "reference": { "description": "Required: Image or artifact reference to be used.\nBehaves in the same way as pod.spec.containers[*].image.\nPull secrets will be assembled in the same way as for the container image by looking up node credentials, SA image pull secrets, and pod spec image pull secrets.\nMore info: https://kubernetes.io/docs/concepts/containers/images\nThis field is optional to allow higher level config management to default or override\ncontainer images in workload controllers like Deployments and StatefulSets.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "iscsi": { "additionalProperties": false, "description": "iscsi represents an ISCSI Disk resource that is attached to a\nkubelet's host machine and then exposed to the pod.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes/#iscsi", "properties": { "chapAuthDiscovery": { "description": "chapAuthDiscovery defines whether support iSCSI Discovery CHAP authentication", "type": [ "boolean", "null" ] }, "chapAuthSession": { "description": "chapAuthSession defines whether support iSCSI Session CHAP authentication", "type": [ "boolean", "null" ] }, "fsType": { "description": "fsType is the filesystem type of the volume that you want to mount.\nTip: Ensure that the filesystem type is supported by the host operating system.\nExamples: \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#iscsi", "type": [ "string", "null" ] }, "initiatorName": { "description": "initiatorName is the custom iSCSI Initiator Name.\nIf initiatorName is specified with iscsiInterface simultaneously, new iSCSI interface\n\u003ctarget portal\u003e:\u003cvolume name\u003e will be created for the connection.", "type": [ "string", "null" ] }, "iqn": { "description": "iqn is the target iSCSI Qualified Name.", "type": "string" }, "iscsiInterface": { "default": "default", "description": "iscsiInterface is the interface Name that uses an iSCSI transport.\nDefaults to 'default' (tcp).", "type": [ "string", "null" ] }, "lun": { "description": "lun represents iSCSI Target Lun number.", "format": "int32", "type": "integer" }, "portals": { "description": "portals is the iSCSI Target Portal List. The portal is either an IP or ip_addr:port if the port\nis other than default (typically TCP ports 860 and 3260).", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "readOnly": { "description": "readOnly here will force the ReadOnly setting in VolumeMounts.\nDefaults to false.", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef is the CHAP Secret for iSCSI target and initiator authentication", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "targetPortal": { "description": "targetPortal is iSCSI Target Portal. The Portal is either an IP or ip_addr:port if the port\nis other than default (typically TCP ports 860 and 3260).", "type": "string" } }, "required": [ "iqn", "lun", "targetPortal" ], "type": [ "object", "null" ] }, "name": { "description": "name of the volume.\nMust be a DNS_LABEL and unique within the pod.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": "string" }, "nfs": { "additionalProperties": false, "description": "nfs represents an NFS mount on the host that shares a pod's lifetime\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#nfs", "properties": { "path": { "description": "path that is exported by the NFS server.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#nfs", "type": "string" }, "readOnly": { "description": "readOnly here will force the NFS export to be mounted with read-only permissions.\nDefaults to false.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#nfs", "type": [ "boolean", "null" ] }, "server": { "description": "server is the hostname or IP address of the NFS server.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#nfs", "type": "string" } }, "required": [ "path", "server" ], "type": [ "object", "null" ] }, "persistentVolumeClaim": { "additionalProperties": false, "description": "persistentVolumeClaimVolumeSource represents a reference to a\nPersistentVolumeClaim in the same namespace.\nMore info: https://kubernetes.io/docs/concepts/storage/persistent-volumes#persistentvolumeclaims", "properties": { "claimName": { "description": "claimName is the name of a PersistentVolumeClaim in the same namespace as the pod using this volume.\nMore info: https://kubernetes.io/docs/concepts/storage/persistent-volumes#persistentvolumeclaims", "type": "string" }, "readOnly": { "description": "readOnly Will force the ReadOnly setting in VolumeMounts.\nDefault false.", "type": [ "boolean", "null" ] } }, "required": [ "claimName" ], "type": [ "object", "null" ] }, "photonPersistentDisk": { "additionalProperties": false, "description": "photonPersistentDisk represents a PhotonController persistent disk attached and mounted on kubelets host machine.\nDeprecated: PhotonPersistentDisk is deprecated and the in-tree photonPersistentDisk type is no longer supported.", "properties": { "fsType": { "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "pdID": { "description": "pdID is the ID that identifies Photon Controller persistent disk", "type": "string" } }, "required": [ "pdID" ], "type": [ "object", "null" ] }, "portworxVolume": { "additionalProperties": false, "description": "portworxVolume represents a portworx volume attached and mounted on kubelets host machine.\nDeprecated: PortworxVolume is deprecated. All operations for the in-tree portworxVolume type\nare redirected to the pxd.portworx.com CSI driver.", "properties": { "fsType": { "description": "fSType represents the filesystem type to mount\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "volumeID": { "description": "volumeID uniquely identifies a Portworx volume", "type": "string" } }, "required": [ "volumeID" ], "type": [ "object", "null" ] }, "projected": { "additionalProperties": false, "description": "projected items for all in one resources secrets, configmaps, and downward API", "properties": { "defaultMode": { "description": "defaultMode are the mode bits used to set permissions on created files by default.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nDirectories within the path are not affected by this setting.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "sources": { "description": "sources is the list of volume projections. Each entry in this list\nhandles one source.", "items": { "additionalProperties": false, "description": "Projection that may be projected along with other supported volume types.\nExactly one of these fields must be set.", "properties": { "clusterTrustBundle": { "additionalProperties": false, "description": "ClusterTrustBundle allows a pod to access the `.spec.trustBundle` field\nof ClusterTrustBundle objects in an auto-updating file.\n\nAlpha, gated by the ClusterTrustBundleProjection feature gate.\n\nClusterTrustBundle objects can either be selected by name, or by the\ncombination of signer name and a label selector.\n\nKubelet performs aggressive normalization of the PEM contents written\ninto the pod filesystem. Esoteric PEM features such as inter-block\ncomments and block headers are stripped. Certificates are deduplicated.\nThe ordering of certificates within the file is arbitrary, and Kubelet\nmay change the order over time.", "properties": { "labelSelector": { "additionalProperties": false, "description": "Select all ClusterTrustBundles that match this label selector. Only has\neffect if signerName is set. Mutually-exclusive with name. If unset,\ninterpreted as \"match nothing\". If set but empty, interpreted as \"match\neverything\".", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "name": { "description": "Select a single ClusterTrustBundle by object name. Mutually-exclusive\nwith signerName and labelSelector.", "type": [ "string", "null" ] }, "optional": { "description": "If true, don't block pod startup if the referenced ClusterTrustBundle(s)\naren't available. If using name, then the named ClusterTrustBundle is\nallowed not to exist. If using signerName, then the combination of\nsignerName and labelSelector is allowed to match zero\nClusterTrustBundles.", "type": [ "boolean", "null" ] }, "path": { "description": "Relative path from the volume root to write the bundle.", "type": "string" }, "signerName": { "description": "Select all ClusterTrustBundles that match this signer name.\nMutually-exclusive with name. The contents of all selected\nClusterTrustBundles will be unified and deduplicated.", "type": [ "string", "null" ] } }, "required": [ "path" ], "type": [ "object", "null" ] }, "configMap": { "additionalProperties": false, "description": "configMap information about the configMap data to project", "properties": { "items": { "description": "items if unspecified, each key-value pair in the Data field of the referenced\nConfigMap will be projected into the volume as a file whose name is the\nkey and content is the value. If specified, the listed keys will be\nprojected into the specified paths, and unlisted keys will not be\npresent. If a key is specified which is not present in the ConfigMap,\nthe volume setup will error unless it is marked optional. Paths must be\nrelative and may not contain the '..' path or start with '..'.", "items": { "additionalProperties": false, "description": "Maps a string key to a path within a volume.", "properties": { "key": { "description": "key is the key to project.", "type": "string" }, "mode": { "description": "mode is Optional: mode bits used to set permissions on this file.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "path is the relative path of the file to map the key to.\nMay not be an absolute path.\nMay not contain the path element '..'.\nMay not start with the string '..'.", "type": "string" } }, "required": [ "key", "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "optional specify whether the ConfigMap or its keys must be defined", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "downwardAPI": { "additionalProperties": false, "description": "downwardAPI information about the downwardAPI data to project", "properties": { "items": { "description": "Items is a list of DownwardAPIVolume file", "items": { "additionalProperties": false, "description": "DownwardAPIVolumeFile represents information to create the file containing the pod field", "properties": { "fieldRef": { "additionalProperties": false, "description": "Required: Selects a field of the pod: only annotations, labels, name, namespace and uid are supported.", "properties": { "apiVersion": { "description": "Version of the schema the FieldPath is written in terms of, defaults to \"v1\".", "type": [ "string", "null" ] }, "fieldPath": { "description": "Path of the field to select in the specified API version.", "type": "string" } }, "required": [ "fieldPath" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "mode": { "description": "Optional: mode bits used to set permissions on this file, must be an octal value\nbetween 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "Required: Path is the relative path name of the file to be created. Must not be absolute or contain the '..' path. Must be utf-8 encoded. The first item of the relative path must not start with '..'", "type": "string" }, "resourceFieldRef": { "additionalProperties": false, "description": "Selects a resource of the container: only resources limits and requests\n(limits.cpu, limits.memory, requests.cpu and requests.memory) are currently supported.", "properties": { "containerName": { "description": "Container name: required for volumes, optional for env vars", "type": [ "string", "null" ] }, "divisor": { "description": "Specifies the output format of the exposed resources, defaults to \"1\"", "oneOf": [ { "pattern": "^(\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\\+|-)?(([0-9]+(\\.[0-9]*)?)|(\\.[0-9]+))))?$", "type": "string" }, { "type": "integer" }, { "type": "null" } ], "x-kubernetes-int-or-string": true }, "resource": { "description": "Required: resource to select", "type": "string" } }, "required": [ "resource" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" } }, "required": [ "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "podCertificate": { "additionalProperties": false, "description": "Projects an auto-rotating credential bundle (private key and certificate\nchain) that the pod can use either as a TLS client or server.\n\nKubelet generates a private key and uses it to send a\nPodCertificateRequest to the named signer. Once the signer approves the\nrequest and issues a certificate chain, Kubelet writes the key and\ncertificate chain to the pod filesystem. The pod does not start until\ncertificates have been issued for each podCertificate projected volume\nsource in its spec.\n\nKubelet will begin trying to rotate the certificate at the time indicated\nby the signer using the PodCertificateRequest.Status.BeginRefreshAt\ntimestamp.\n\nKubelet can write a single file, indicated by the credentialBundlePath\nfield, or separate files, indicated by the keyPath and\ncertificateChainPath fields.\n\nThe credential bundle is a single file in PEM format. The first PEM\nentry is the private key (in PKCS#8 format), and the remaining PEM\nentries are the certificate chain issued by the signer (typically,\nsigners will return their certificate chain in leaf-to-root order).\n\nPrefer using the credential bundle format, since your application code\ncan read it atomically. If you use keyPath and certificateChainPath,\nyour application must make two separate file reads. If these coincide\nwith a certificate rotation, it is possible that the private key and leaf\ncertificate you read may not correspond to each other. Your application\nwill need to check for this condition, and re-read until they are\nconsistent.\n\nThe named signer controls chooses the format of the certificate it\nissues; consult the signer implementation's documentation to learn how to\nuse the certificates it issues.", "properties": { "certificateChainPath": { "description": "Write the certificate chain at this path in the projected volume.\n\nMost applications should use credentialBundlePath. When using keyPath\nand certificateChainPath, your application needs to check that the key\nand leaf certificate are consistent, because it is possible to read the\nfiles mid-rotation.", "type": [ "string", "null" ] }, "credentialBundlePath": { "description": "Write the credential bundle at this path in the projected volume.\n\nThe credential bundle is a single file that contains multiple PEM blocks.\nThe first PEM block is a PRIVATE KEY block, containing a PKCS#8 private\nkey.\n\nThe remaining blocks are CERTIFICATE blocks, containing the issued\ncertificate chain from the signer (leaf and any intermediates).\n\nUsing credentialBundlePath lets your Pod's application code make a single\natomic read that retrieves a consistent key and certificate chain. If you\nproject them to separate files, your application code will need to\nadditionally check that the leaf certificate was issued to the key.", "type": [ "string", "null" ] }, "keyPath": { "description": "Write the key at this path in the projected volume.\n\nMost applications should use credentialBundlePath. When using keyPath\nand certificateChainPath, your application needs to check that the key\nand leaf certificate are consistent, because it is possible to read the\nfiles mid-rotation.", "type": [ "string", "null" ] }, "keyType": { "description": "The type of keypair Kubelet will generate for the pod.\n\nValid values are \"RSA3072\", \"RSA4096\", \"ECDSAP256\", \"ECDSAP384\",\n\"ECDSAP521\", and \"ED25519\".", "type": "string" }, "maxExpirationSeconds": { "description": "maxExpirationSeconds is the maximum lifetime permitted for the\ncertificate.\n\nKubelet copies this value verbatim into the PodCertificateRequests it\ngenerates for this projection.\n\nIf omitted, kube-apiserver will set it to 86400(24 hours). kube-apiserver\nwill reject values shorter than 3600 (1 hour). The maximum allowable\nvalue is 7862400 (91 days).\n\nThe signer implementation is then free to issue a certificate with any\nlifetime *shorter* than MaxExpirationSeconds, but no shorter than 3600\nseconds (1 hour). This constraint is enforced by kube-apiserver.\n`kubernetes.io` signers will never issue certificates with a lifetime\nlonger than 24 hours.", "format": "int32", "type": [ "integer", "null" ] }, "signerName": { "description": "Kubelet's generated CSRs will be addressed to this signer.", "type": "string" }, "userAnnotations": { "additionalProperties": { "type": "string" }, "description": "userAnnotations allow pod authors to pass additional information to\nthe signer implementation. Kubernetes does not restrict or validate this\nmetadata in any way.\n\nThese values are copied verbatim into the `spec.unverifiedUserAnnotations` field of\nthe PodCertificateRequest objects that Kubelet creates.\n\nEntries are subject to the same validation as object metadata annotations,\nwith the addition that all keys must be domain-prefixed. No restrictions\nare placed on values, except an overall size limitation on the entire field.\n\nSigners should document the keys and values they support. Signers should\ndeny requests that contain keys they do not recognize.", "type": [ "object", "null" ] } }, "required": [ "keyType", "signerName" ], "type": [ "object", "null" ] }, "secret": { "additionalProperties": false, "description": "secret information about the secret data to project", "properties": { "items": { "description": "items if unspecified, each key-value pair in the Data field of the referenced\nSecret will be projected into the volume as a file whose name is the\nkey and content is the value. If specified, the listed keys will be\nprojected into the specified paths, and unlisted keys will not be\npresent. If a key is specified which is not present in the Secret,\nthe volume setup will error unless it is marked optional. Paths must be\nrelative and may not contain the '..' path or start with '..'.", "items": { "additionalProperties": false, "description": "Maps a string key to a path within a volume.", "properties": { "key": { "description": "key is the key to project.", "type": "string" }, "mode": { "description": "mode is Optional: mode bits used to set permissions on this file.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "path is the relative path of the file to map the key to.\nMay not be an absolute path.\nMay not contain the path element '..'.\nMay not start with the string '..'.", "type": "string" } }, "required": [ "key", "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "optional field specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "serviceAccountToken": { "additionalProperties": false, "description": "serviceAccountToken is information about the serviceAccountToken data to project", "properties": { "audience": { "description": "audience is the intended audience of the token. A recipient of a token\nmust identify itself with an identifier specified in the audience of the\ntoken, and otherwise should reject the token. The audience defaults to the\nidentifier of the apiserver.", "type": [ "string", "null" ] }, "expirationSeconds": { "description": "expirationSeconds is the requested duration of validity of the service\naccount token. As the token approaches expiration, the kubelet volume\nplugin will proactively rotate the service account token. The kubelet will\nstart trying to rotate the token if the token is older than 80 percent of\nits time to live or if the token is older than 24 hours.Defaults to 1 hour\nand must be at least 10 minutes.", "format": "int64", "type": [ "integer", "null" ] }, "path": { "description": "path is the path relative to the mount point of the file to project the\ntoken into.", "type": "string" } }, "required": [ "path" ], "type": [ "object", "null" ] } }, "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "quobyte": { "additionalProperties": false, "description": "quobyte represents a Quobyte mount on the host that shares a pod's lifetime.\nDeprecated: Quobyte is deprecated and the in-tree quobyte type is no longer supported.", "properties": { "group": { "description": "group to map volume access to\nDefault is no group", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly here will force the Quobyte volume to be mounted with read-only permissions.\nDefaults to false.", "type": [ "boolean", "null" ] }, "registry": { "description": "registry represents a single or multiple Quobyte Registry services\nspecified as a string as host:port pair (multiple entries are separated with commas)\nwhich acts as the central registry for volumes", "type": "string" }, "tenant": { "description": "tenant owning the given Quobyte volume in the Backend\nUsed with dynamically provisioned Quobyte volumes, value is set by the plugin", "type": [ "string", "null" ] }, "user": { "description": "user to map volume access to\nDefaults to serivceaccount user", "type": [ "string", "null" ] }, "volume": { "description": "volume is a string that references an already created Quobyte volume by name.", "type": "string" } }, "required": [ "registry", "volume" ], "type": [ "object", "null" ] }, "rbd": { "additionalProperties": false, "description": "rbd represents a Rados Block Device mount on the host that shares a pod's lifetime.\nDeprecated: RBD is deprecated and the in-tree rbd type is no longer supported.", "properties": { "fsType": { "description": "fsType is the filesystem type of the volume that you want to mount.\nTip: Ensure that the filesystem type is supported by the host operating system.\nExamples: \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#rbd", "type": [ "string", "null" ] }, "image": { "description": "image is the rados image name.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "type": "string" }, "keyring": { "default": "/etc/ceph/keyring", "description": "keyring is the path to key ring for RBDUser.\nDefault is /etc/ceph/keyring.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "type": [ "string", "null" ] }, "monitors": { "description": "monitors is a collection of Ceph monitors.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "items": { "type": "string" }, "type": "array", "x-kubernetes-list-type": "atomic" }, "pool": { "default": "rbd", "description": "pool is the rados pool name.\nDefault is rbd.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly here will force the ReadOnly setting in VolumeMounts.\nDefaults to false.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef is name of the authentication secret for RBDUser. If provided\noverrides keyring.\nDefault is nil.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "user": { "default": "admin", "description": "user is the rados user name.\nDefault is admin.\nMore info: https://examples.k8s.io/volumes/rbd/README.md#how-to-use-it", "type": [ "string", "null" ] } }, "required": [ "image", "monitors" ], "type": [ "object", "null" ] }, "scaleIO": { "additionalProperties": false, "description": "scaleIO represents a ScaleIO persistent volume attached and mounted on Kubernetes nodes.\nDeprecated: ScaleIO is deprecated and the in-tree scaleIO type is no longer supported.", "properties": { "fsType": { "default": "xfs", "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\".\nDefault is \"xfs\".", "type": [ "string", "null" ] }, "gateway": { "description": "gateway is the host address of the ScaleIO API Gateway.", "type": "string" }, "protectionDomain": { "description": "protectionDomain is the name of the ScaleIO Protection Domain for the configured storage.", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly Defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef references to the secret for ScaleIO user and other\nsensitive information. If this is not provided, Login operation will fail.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": "object", "x-kubernetes-map-type": "atomic" }, "sslEnabled": { "description": "sslEnabled Flag enable/disable SSL communication with Gateway, default false", "type": [ "boolean", "null" ] }, "storageMode": { "default": "ThinProvisioned", "description": "storageMode indicates whether the storage for a volume should be ThickProvisioned or ThinProvisioned.\nDefault is ThinProvisioned.", "type": [ "string", "null" ] }, "storagePool": { "description": "storagePool is the ScaleIO Storage Pool associated with the protection domain.", "type": [ "string", "null" ] }, "system": { "description": "system is the name of the storage system as configured in ScaleIO.", "type": "string" }, "volumeName": { "description": "volumeName is the name of a volume already created in the ScaleIO system\nthat is associated with this volume source.", "type": [ "string", "null" ] } }, "required": [ "gateway", "secretRef", "system" ], "type": [ "object", "null" ] }, "secret": { "additionalProperties": false, "description": "secret represents a secret that should populate this volume.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#secret", "properties": { "defaultMode": { "description": "defaultMode is Optional: mode bits used to set permissions on created files by default.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values\nfor mode bits. Defaults to 0644.\nDirectories within the path are not affected by this setting.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "items": { "description": "items If unspecified, each key-value pair in the Data field of the referenced\nSecret will be projected into the volume as a file whose name is the\nkey and content is the value. If specified, the listed keys will be\nprojected into the specified paths, and unlisted keys will not be\npresent. If a key is specified which is not present in the Secret,\nthe volume setup will error unless it is marked optional. Paths must be\nrelative and may not contain the '..' path or start with '..'.", "items": { "additionalProperties": false, "description": "Maps a string key to a path within a volume.", "properties": { "key": { "description": "key is the key to project.", "type": "string" }, "mode": { "description": "mode is Optional: mode bits used to set permissions on this file.\nMust be an octal value between 0000 and 0777 or a decimal value between 0 and 511.\nYAML accepts both octal and decimal values, JSON requires decimal values for mode bits.\nIf not specified, the volume defaultMode will be used.\nThis might be in conflict with other options that affect the file\nmode, like fsGroup, and the result can be other mode bits set.", "format": "int32", "type": [ "integer", "null" ] }, "path": { "description": "path is the relative path of the file to map the key to.\nMay not be an absolute path.\nMay not contain the path element '..'.\nMay not start with the string '..'.", "type": "string" } }, "required": [ "key", "path" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "optional": { "description": "optional field specify whether the Secret or its keys must be defined", "type": [ "boolean", "null" ] }, "secretName": { "description": "secretName is the name of the secret in the pod's namespace to use.\nMore info: https://kubernetes.io/docs/concepts/storage/volumes#secret", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "storageos": { "additionalProperties": false, "description": "storageOS represents a StorageOS volume attached and mounted on Kubernetes nodes.\nDeprecated: StorageOS is deprecated and the in-tree storageos type is no longer supported.", "properties": { "fsType": { "description": "fsType is the filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "readOnly": { "description": "readOnly defaults to false (read/write). ReadOnly here will force\nthe ReadOnly setting in VolumeMounts.", "type": [ "boolean", "null" ] }, "secretRef": { "additionalProperties": false, "description": "secretRef specifies the secret to use for obtaining the StorageOS API\ncredentials. If not specified, default values will be attempted.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "volumeName": { "description": "volumeName is the human-readable name of the StorageOS volume. Volume\nnames are only unique within a namespace.", "type": [ "string", "null" ] }, "volumeNamespace": { "description": "volumeNamespace specifies the scope of the volume within StorageOS. If no\nnamespace is specified then the Pod's namespace will be used. This allows the\nKubernetes name scoping to be mirrored within StorageOS for tighter integration.\nSet VolumeName to any name to override the default behaviour.\nSet to \"default\" if you are not using namespaces within StorageOS.\nNamespaces that do not pre-exist within StorageOS will be created.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "vsphereVolume": { "additionalProperties": false, "description": "vsphereVolume represents a vSphere volume attached and mounted on kubelets host machine.\nDeprecated: VsphereVolume is deprecated. All operations for the in-tree vsphereVolume type\nare redirected to the csi.vsphere.vmware.com CSI driver.", "properties": { "fsType": { "description": "fsType is filesystem type to mount.\nMust be a filesystem type supported by the host operating system.\nEx. \"ext4\", \"xfs\", \"ntfs\". Implicitly inferred to be \"ext4\" if unspecified.", "type": [ "string", "null" ] }, "storagePolicyID": { "description": "storagePolicyID is the storage Policy Based Management (SPBM) profile ID associated with the StoragePolicyName.", "type": [ "string", "null" ] }, "storagePolicyName": { "description": "storagePolicyName is the storage Policy Based Management (SPBM) profile name.", "type": [ "string", "null" ] }, "volumePath": { "description": "volumePath is the path that identifies vSphere volume vmdk", "type": "string" } }, "required": [ "volumePath" ], "type": [ "object", "null" ] } }, "required": [ "name" ], "type": "object" }, "type": [ "array", "null" ] }, "flashAttention": { "description": "FlashAttention enables flash attention for faster prompt processing and\nreduced KV cache memory. Maps to llama.cpp --flash-attn flag.\n\nOn NVIDIA GPUs requires Ampere or newer (compute capability 8.0+).\nOn Apple Silicon (Metal agent path) the default is true when this field\nis unset, because the wired-collector + flash-attn combination prevents\nthe ~25% decode degradation observed at long context on Qwen-class\nmodels running on M-series chips.", "type": [ "boolean", "null" ] }, "image": { "description": "Image is the container image for the inference runtime.\nFor llamacpp runtime, defaults to ghcr.io/ggml-org/llama.cpp:server.\nFor generic runtime, this field is required.", "type": [ "string", "null" ] }, "imagePullSecrets": { "description": "ImagePullSecrets for pulling container images from private registries.", "items": { "additionalProperties": false, "description": "LocalObjectReference contains enough information to let you locate the\nreferenced object inside the same namespace.", "properties": { "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] } }, "type": "object", "x-kubernetes-map-type": "atomic" }, "type": [ "array", "null" ] }, "jinja": { "description": "Jinja enables Jinja2 chat template rendering for tool/function calling support.\nRequired when using the OpenAI-compatible API with tools. Maps to llama.cpp --jinja flag.", "type": [ "boolean", "null" ] }, "metadataOverrides": { "description": "MetadataOverrides overrides GGUF metadata key-value pairs at model load time.\nEach entry is passed as a separate --override-kv flag. Format: key=type:value\n(e.g., \"qwen35moe.context_length=int:1048576\" to extend context window, or\n\"tokenizer.chat_template.thinking=bool:false\" to tweak tokenizer behavior).\nMaps to llama.cpp --override-kv flag (one flag per entry).", "items": { "type": "string" }, "type": [ "array", "null" ] }, "mode": { "description": "Mode selects how the model is served: \"chat\" (default) for\nchat/completion, \"embedding\" for /v1/embeddings, \"rerank\" for /v1/rerank.\nFor the llamacpp runtime it auto-appends the required flags (embedding -\u003e\n--embedding --pooling last; rerank -\u003e --reranking --embedding --pooling\nrank); any flag already set in spec.extraArgs wins. When unset, the mode is\ninferred from spec.extraArgs / spec.endpoint.path. The resolved value is\nalways reported in status.mode.", "enum": [ "chat", "embedding", "rerank" ], "type": [ "string", "null" ] }, "modelCache": { "additionalProperties": false, "description": "ModelCache overrides where this InferenceService caches model weights:\nwhen claimName is set, the named user-owned PVC is mounted as the\nwritable model cache (prep + download init containers run against it)\ninstead of the operator's shared/perService cache PVC. When unset, the\noperator-global cache mode applies unchanged.", "properties": { "claimName": { "description": "ClaimName names a pre-existing PersistentVolumeClaim in the\nInferenceService's namespace to use as the writable model cache volume.\nWeights land under the usual \u003ccacheKey\u003e/ subdirectory of the claim, so\nRefreshPolicy and cache-key semantics are unchanged and multiple models\ncan share one claim without colliding. The claim must already exist:\nwhen it is missing the InferenceService is marked Degraded rather than\nsilently falling back to the shared cache. Ignored for pvc:// model\nsources (already staged, read-only, no download). Node alignment of\nRWO/local claims (via nodeSelector) is the user's responsibility.", "maxLength": 253, "minLength": 1, "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "modelRef": { "description": "ModelRef references the Model CR that contains the model to serve", "type": "string" }, "moeCPULayers": { "description": "MoeCPULayers sets the number of MoE layers to offload to CPU.\nWhen set, only the specified number of MoE layers run on CPU rather than all.\nMaps to llama.cpp --n-cpu-moe flag.", "format": "int32", "minimum": 0, "type": [ "integer", "null" ] }, "moeCPUOffload": { "description": "MoeCPUOffload offloads all MoE expert layers to CPU for reduced VRAM usage.\nEnables running large MoE models (e.g., Qwen3-30B, Mixtral) on VRAM-constrained\nhardware by keeping attention layers on GPU while expert weights use system RAM.\nMaps to llama.cpp --cpu-moe flag. Requires sufficient system RAM via resources.memory.", "type": [ "boolean", "null" ] }, "noKvOffload": { "description": "NoKvOffload keeps the KV cache in system RAM instead of VRAM.\nUseful for extended context windows when VRAM is constrained by model weights.\nMaps to llama.cpp --no-kv-offload flag. Requires sufficient system RAM via resources.memory.", "type": [ "boolean", "null" ] }, "noWarmup": { "description": "NoWarmup skips the llama.cpp startup warmup inference pass.\nReduces pod ready time at the cost of slightly higher first-request latency.\nUseful for scale-to-zero and quick redeployment patterns.\nMaps to llama.cpp --no-warmup flag.", "type": [ "boolean", "null" ] }, "nodeSelector": { "additionalProperties": { "type": "string" }, "description": "NodeSelector for pod placement (e.g., specific node pools)", "type": [ "object", "null" ] }, "parallelSlots": { "description": "ParallelSlots sets the number of concurrent request slots for the llama.cpp\nserver (--parallel flag). Each slot processes one request independently;\nhigher values use more KV cache memory. If not specified, the operator\nomits --parallel and llama.cpp picks an auto value (currently 4).", "format": "int32", "maximum": 64, "minimum": 1, "type": [ "integer", "null" ] }, "personaPlexConfig": { "additionalProperties": false, "description": "PersonaPlexConfig holds configuration for the PersonaPlex (Moshi) runtime.\nOnly used when Runtime is \"personaplex\".", "properties": { "cpuOffload": { "description": "CPUOffload enables model weight offloading to system RAM when GPU VRAM is insufficient.\nRequires the accelerate package in the container image.", "type": [ "boolean", "null" ] }, "hfTokenSecretRef": { "additionalProperties": false, "description": "HFTokenSecretRef references a Secret containing the HuggingFace token for model download.\nThe Secret key must be \"HF_TOKEN\".", "properties": { "key": { "description": "The key of the secret to select from. Must be a valid secret key.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "quantize4Bit": { "description": "Quantize4Bit enables NF4 4-bit quantization for reduced VRAM usage (~9.6 GB vs ~14 GB).\nRequires the bitsandbytes package in the container image.", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ] }, "podAnnotations": { "additionalProperties": { "type": "string" }, "description": "PodAnnotations are merged into the inference Pod's metadata.annotations.\nUse this to tag Pods for downstream tooling (cost attribution, service\nmesh routing, custom admission controllers) without those tools needing\nto know about LLMKube's CRD schema. Pure passthrough; the operator\nitself does not set any annotations on inference Pods today.", "type": [ "object", "null" ] }, "podLabels": { "additionalProperties": { "type": "string" }, "description": "PodLabels are merged into the inference Pod's metadata.labels alongside\nthe operator-managed labels (`app`, `inference.llmkube.dev/model`,\n`inference.llmkube.dev/service`). Operator-managed keys take precedence\non collision so the Deployment selector stays in sync with the Pods it\nowns. The Deployment selector itself uses only the operator-managed\nlabels and is immutable, so changing PodLabels later is safe.", "type": [ "object", "null" ] }, "podSecurityContext": { "additionalProperties": false, "description": "PodSecurityContext defines pod-level security attributes for inference pods.\nUse this to set fsGroup for volume permissions (required on OpenShift).", "properties": { "appArmorProfile": { "additionalProperties": false, "description": "appArmorProfile is the AppArmor options to use by the containers in this pod.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "localhostProfile": { "description": "localhostProfile indicates a profile loaded on the node that should be used.\nThe profile must be preconfigured on the node to work.\nMust match the loaded name of the profile.\nMust be set if and only if type is \"Localhost\".", "type": [ "string", "null" ] }, "type": { "description": "type indicates which kind of AppArmor profile will be applied.\nValid options are:\n Localhost - a profile pre-loaded on the node.\n RuntimeDefault - the container runtime's default profile.\n Unconfined - no AppArmor enforcement.", "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "fsGroup": { "description": "A special supplemental group that applies to all containers in a pod.\nSome volume types allow the Kubelet to change the ownership of that volume\nto be owned by the pod:\n\n1. The owning GID will be the FSGroup\n2. The setgid bit is set (new files created in the volume will be owned by FSGroup)\n3. The permission bits are OR'd with rw-rw----\n\nIf unset, the Kubelet will not modify the ownership and permissions of any volume.\nNote that this field cannot be set when spec.os.name is windows.", "format": "int64", "type": [ "integer", "null" ] }, "fsGroupChangePolicy": { "description": "fsGroupChangePolicy defines behavior of changing ownership and permission of the volume\nbefore being exposed inside Pod. This field will only apply to\nvolume types which support fsGroup based ownership(and permissions).\nIt will have no effect on ephemeral volume types such as: secret, configmaps\nand emptydir.\nValid values are \"OnRootMismatch\" and \"Always\". If not specified, \"Always\" is used.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "string", "null" ] }, "runAsGroup": { "description": "The GID to run the entrypoint of the container process.\nUses runtime default if unset.\nMay also be set in SecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence\nfor that container.\nNote that this field cannot be set when spec.os.name is windows.", "format": "int64", "type": [ "integer", "null" ] }, "runAsNonRoot": { "description": "Indicates that the container must run as a non-root user.\nIf true, the Kubelet will validate the image at runtime to ensure that it\ndoes not run as UID 0 (root) and fail to start the container if it does.\nIf unset or false, no such validation will be performed.\nMay also be set in SecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.", "type": [ "boolean", "null" ] }, "runAsUser": { "description": "The UID to run the entrypoint of the container process.\nDefaults to user specified in image metadata if unspecified.\nMay also be set in SecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence\nfor that container.\nNote that this field cannot be set when spec.os.name is windows.", "format": "int64", "type": [ "integer", "null" ] }, "seLinuxChangePolicy": { "description": "seLinuxChangePolicy defines how the container's SELinux label is applied to all volumes used by the Pod.\nIt has no effect on nodes that do not support SELinux or to volumes does not support SELinux.\nValid values are \"MountOption\" and \"Recursive\".\n\n\"Recursive\" means relabeling of all files on all Pod volumes by the container runtime.\nThis may be slow for large volumes, but allows mixing privileged and unprivileged Pods sharing the same volume on the same node.\n\n\"MountOption\" mounts all eligible Pod volumes with `-o context` mount option.\nThis requires all Pods that share the same volume to use the same SELinux label.\nIt is not possible to share the same volume among privileged and unprivileged Pods.\nEligible volumes are in-tree FibreChannel and iSCSI volumes, and all CSI volumes\nwhose CSI driver announces SELinux support by setting spec.seLinuxMount: true in their\nCSIDriver instance. Other volumes are always re-labelled recursively.\n\"MountOption\" value is allowed only when SELinuxMount feature gate is enabled.\n\nIf not specified and SELinuxMount feature gate is enabled, \"MountOption\" is used.\nIf not specified and SELinuxMount feature gate is disabled, \"MountOption\" is used for ReadWriteOncePod volumes\nand \"Recursive\" for all other volumes.\n\nThis field affects only Pods that have SELinux label set, either in PodSecurityContext or in SecurityContext of all containers.\n\nAll Pods that use the same volume should use the same seLinuxChangePolicy, otherwise some pods can get stuck in ContainerCreating state.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "string", "null" ] }, "seLinuxOptions": { "additionalProperties": false, "description": "The SELinux context to be applied to all containers.\nIf unspecified, the container runtime will allocate a random SELinux context for each\ncontainer. May also be set in SecurityContext. If set in\nboth SecurityContext and PodSecurityContext, the value specified in SecurityContext\ntakes precedence for that container.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "level": { "description": "Level is SELinux level label that applies to the container.", "type": [ "string", "null" ] }, "role": { "description": "Role is a SELinux role label that applies to the container.", "type": [ "string", "null" ] }, "type": { "description": "Type is a SELinux type label that applies to the container.", "type": [ "string", "null" ] }, "user": { "description": "User is a SELinux user label that applies to the container.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "seccompProfile": { "additionalProperties": false, "description": "The seccomp options to use by the containers in this pod.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "localhostProfile": { "description": "localhostProfile indicates a profile defined in a file on the node should be used.\nThe profile must be preconfigured on the node to work.\nMust be a descending path, relative to the kubelet's configured seccomp profile location.\nMust be set if type is \"Localhost\". Must NOT be set for any other type.", "type": [ "string", "null" ] }, "type": { "description": "type indicates which kind of seccomp profile will be applied.\nValid options are:\n\nLocalhost - a profile defined in a file on the node should be used.\nRuntimeDefault - the container runtime default profile should be used.\nUnconfined - no profile should be applied.", "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "supplementalGroups": { "description": "A list of groups applied to the first process run in each container, in\naddition to the container's primary GID and fsGroup (if specified). If\nthe SupplementalGroupsPolicy feature is enabled, the\nsupplementalGroupsPolicy field determines whether these are in addition\nto or instead of any group memberships defined in the container image.\nIf unspecified, no additional groups are added, though group memberships\ndefined in the container image may still be used, depending on the\nsupplementalGroupsPolicy field.\nNote that this field cannot be set when spec.os.name is windows.", "items": { "format": "int64", "type": "integer" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "supplementalGroupsPolicy": { "description": "Defines how supplemental groups of the first container processes are calculated.\nValid values are \"Merge\" and \"Strict\". If not specified, \"Merge\" is used.\n(Alpha) Using the field requires the SupplementalGroupsPolicy feature gate to be enabled\nand the container runtime must implement support for this feature.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "string", "null" ] }, "sysctls": { "description": "Sysctls hold a list of namespaced sysctls used for the pod. Pods with unsupported\nsysctls (by the container runtime) might fail to launch.\nNote that this field cannot be set when spec.os.name is windows.", "items": { "additionalProperties": false, "description": "Sysctl defines a kernel parameter to be set", "properties": { "name": { "description": "Name of a property to set", "type": "string" }, "value": { "description": "Value of a property to set", "type": "string" } }, "required": [ "name", "value" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "windowsOptions": { "additionalProperties": false, "description": "The Windows specific settings applied to all containers.\nIf unspecified, the options within a container's SecurityContext will be used.\nIf set in both SecurityContext and PodSecurityContext, the value specified in SecurityContext takes precedence.\nNote that this field cannot be set when spec.os.name is linux.", "properties": { "gmsaCredentialSpec": { "description": "GMSACredentialSpec is where the GMSA admission webhook\n(https://github.com/kubernetes-sigs/windows-gmsa) inlines the contents of the\nGMSA credential spec named by the GMSACredentialSpecName field.", "type": [ "string", "null" ] }, "gmsaCredentialSpecName": { "description": "GMSACredentialSpecName is the name of the GMSA credential spec to use.", "type": [ "string", "null" ] }, "hostProcess": { "description": "HostProcess determines if a container should be run as a 'Host Process' container.\nAll of a Pod's containers must have the same effective HostProcess value\n(it is not allowed to have a mix of HostProcess containers and non-HostProcess containers).\nIn addition, if HostProcess is true then HostNetwork must also be set to true.", "type": [ "boolean", "null" ] }, "runAsUserName": { "description": "The UserName in Windows to run the entrypoint of the container process.\nDefaults to the user specified in image metadata if unspecified.\nMay also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "priority": { "default": "normal", "description": "Priority determines scheduling priority for GPU allocation.\nHigher priority services can preempt lower priority ones when GPUs are scarce.", "enum": [ "critical", "high", "normal", "low", "batch" ], "type": [ "string", "null" ] }, "priorityClassName": { "description": "PriorityClassName allows specifying a custom Kubernetes PriorityClass.\nTakes precedence over the Priority field if set.", "type": [ "string", "null" ] }, "probeOverrides": { "additionalProperties": false, "description": "ProbeOverrides allows replacing the auto-generated health probes.\nUseful for runtimes with non-HTTP health endpoints (e.g., TCP, WebSocket).", "properties": { "liveness": { "additionalProperties": false, "description": "Liveness overrides the liveness probe.", "properties": { "exec": { "additionalProperties": false, "description": "Exec specifies a command to execute in the container.", "properties": { "command": { "description": "Command is the command line to execute inside the container, the working directory for the\ncommand is root ('/') in the container's filesystem. The command is simply exec'd, it is\nnot run inside a shell, so traditional shell instructions ('|', etc) won't work. To use\na shell, you need to explicitly call out to that shell.\nExit status of 0 is treated as live/healthy and non-zero is unhealthy.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "failureThreshold": { "description": "Minimum consecutive failures for the probe to be considered failed after having succeeded.\nDefaults to 3. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "grpc": { "additionalProperties": false, "description": "GRPC specifies a GRPC HealthCheckRequest.", "properties": { "port": { "description": "Port number of the gRPC service. Number must be in the range 1 to 65535.", "format": "int32", "type": "integer" }, "service": { "default": "", "description": "Service is the name of the service to place in the gRPC HealthCheckRequest\n(see https://github.com/grpc/grpc/blob/master/doc/health-checking.md).\n\nIf this is not specified, the default behavior is defined by gRPC.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "httpGet": { "additionalProperties": false, "description": "HTTPGet specifies an HTTP GET request to perform.", "properties": { "host": { "description": "Host name to connect to, defaults to the pod IP. You probably want to set\n\"Host\" in httpHeaders instead.", "type": [ "string", "null" ] }, "httpHeaders": { "description": "Custom headers to set in the request. HTTP allows repeated headers.", "items": { "additionalProperties": false, "description": "HTTPHeader describes a custom header to be used in HTTP probes", "properties": { "name": { "description": "The header field name.\nThis will be canonicalized upon output, so case-variant names will be understood as the same header.", "type": "string" }, "value": { "description": "The header field value", "type": "string" } }, "required": [ "name", "value" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "path": { "description": "Path to access on the HTTP server.", "type": [ "string", "null" ] }, "port": { "description": "Name or number of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true }, "scheme": { "description": "Scheme to use for connecting to the host.\nDefaults to HTTP.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "initialDelaySeconds": { "description": "Number of seconds after the container has started before liveness probes are initiated.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] }, "periodSeconds": { "description": "How often (in seconds) to perform the probe.\nDefault to 10 seconds. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "successThreshold": { "description": "Minimum consecutive successes for the probe to be considered successful after having failed.\nDefaults to 1. Must be 1 for liveness and startup. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "tcpSocket": { "additionalProperties": false, "description": "TCPSocket specifies a connection to a TCP port.", "properties": { "host": { "description": "Optional: Host name to connect to, defaults to the pod IP.", "type": [ "string", "null" ] }, "port": { "description": "Number or name of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true } }, "required": [ "port" ], "type": [ "object", "null" ] }, "terminationGracePeriodSeconds": { "description": "Optional duration in seconds the pod needs to terminate gracefully upon probe failure.\nThe grace period is the duration in seconds after the processes running in the pod are sent\na termination signal and the time when the processes are forcibly halted with a kill signal.\nSet this value longer than the expected cleanup time for your process.\nIf this value is nil, the pod's terminationGracePeriodSeconds will be used. Otherwise, this\nvalue overrides the value provided by the pod spec.\nValue must be non-negative integer. The value zero indicates stop immediately via\nthe kill signal (no opportunity to shut down).\nThis is a beta field and requires enabling ProbeTerminationGracePeriod feature gate.\nMinimum value is 1. spec.terminationGracePeriodSeconds is used if unset.", "format": "int64", "type": [ "integer", "null" ] }, "timeoutSeconds": { "description": "Number of seconds after which the probe times out.\nDefaults to 1 second. Minimum value is 1.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] }, "readiness": { "additionalProperties": false, "description": "Readiness overrides the readiness probe.", "properties": { "exec": { "additionalProperties": false, "description": "Exec specifies a command to execute in the container.", "properties": { "command": { "description": "Command is the command line to execute inside the container, the working directory for the\ncommand is root ('/') in the container's filesystem. The command is simply exec'd, it is\nnot run inside a shell, so traditional shell instructions ('|', etc) won't work. To use\na shell, you need to explicitly call out to that shell.\nExit status of 0 is treated as live/healthy and non-zero is unhealthy.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "failureThreshold": { "description": "Minimum consecutive failures for the probe to be considered failed after having succeeded.\nDefaults to 3. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "grpc": { "additionalProperties": false, "description": "GRPC specifies a GRPC HealthCheckRequest.", "properties": { "port": { "description": "Port number of the gRPC service. Number must be in the range 1 to 65535.", "format": "int32", "type": "integer" }, "service": { "default": "", "description": "Service is the name of the service to place in the gRPC HealthCheckRequest\n(see https://github.com/grpc/grpc/blob/master/doc/health-checking.md).\n\nIf this is not specified, the default behavior is defined by gRPC.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "httpGet": { "additionalProperties": false, "description": "HTTPGet specifies an HTTP GET request to perform.", "properties": { "host": { "description": "Host name to connect to, defaults to the pod IP. You probably want to set\n\"Host\" in httpHeaders instead.", "type": [ "string", "null" ] }, "httpHeaders": { "description": "Custom headers to set in the request. HTTP allows repeated headers.", "items": { "additionalProperties": false, "description": "HTTPHeader describes a custom header to be used in HTTP probes", "properties": { "name": { "description": "The header field name.\nThis will be canonicalized upon output, so case-variant names will be understood as the same header.", "type": "string" }, "value": { "description": "The header field value", "type": "string" } }, "required": [ "name", "value" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "path": { "description": "Path to access on the HTTP server.", "type": [ "string", "null" ] }, "port": { "description": "Name or number of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true }, "scheme": { "description": "Scheme to use for connecting to the host.\nDefaults to HTTP.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "initialDelaySeconds": { "description": "Number of seconds after the container has started before liveness probes are initiated.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] }, "periodSeconds": { "description": "How often (in seconds) to perform the probe.\nDefault to 10 seconds. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "successThreshold": { "description": "Minimum consecutive successes for the probe to be considered successful after having failed.\nDefaults to 1. Must be 1 for liveness and startup. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "tcpSocket": { "additionalProperties": false, "description": "TCPSocket specifies a connection to a TCP port.", "properties": { "host": { "description": "Optional: Host name to connect to, defaults to the pod IP.", "type": [ "string", "null" ] }, "port": { "description": "Number or name of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true } }, "required": [ "port" ], "type": [ "object", "null" ] }, "terminationGracePeriodSeconds": { "description": "Optional duration in seconds the pod needs to terminate gracefully upon probe failure.\nThe grace period is the duration in seconds after the processes running in the pod are sent\na termination signal and the time when the processes are forcibly halted with a kill signal.\nSet this value longer than the expected cleanup time for your process.\nIf this value is nil, the pod's terminationGracePeriodSeconds will be used. Otherwise, this\nvalue overrides the value provided by the pod spec.\nValue must be non-negative integer. The value zero indicates stop immediately via\nthe kill signal (no opportunity to shut down).\nThis is a beta field and requires enabling ProbeTerminationGracePeriod feature gate.\nMinimum value is 1. spec.terminationGracePeriodSeconds is used if unset.", "format": "int64", "type": [ "integer", "null" ] }, "timeoutSeconds": { "description": "Number of seconds after which the probe times out.\nDefaults to 1 second. Minimum value is 1.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] }, "startup": { "additionalProperties": false, "description": "Startup overrides the startup probe.", "properties": { "exec": { "additionalProperties": false, "description": "Exec specifies a command to execute in the container.", "properties": { "command": { "description": "Command is the command line to execute inside the container, the working directory for the\ncommand is root ('/') in the container's filesystem. The command is simply exec'd, it is\nnot run inside a shell, so traditional shell instructions ('|', etc) won't work. To use\na shell, you need to explicitly call out to that shell.\nExit status of 0 is treated as live/healthy and non-zero is unhealthy.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "failureThreshold": { "description": "Minimum consecutive failures for the probe to be considered failed after having succeeded.\nDefaults to 3. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "grpc": { "additionalProperties": false, "description": "GRPC specifies a GRPC HealthCheckRequest.", "properties": { "port": { "description": "Port number of the gRPC service. Number must be in the range 1 to 65535.", "format": "int32", "type": "integer" }, "service": { "default": "", "description": "Service is the name of the service to place in the gRPC HealthCheckRequest\n(see https://github.com/grpc/grpc/blob/master/doc/health-checking.md).\n\nIf this is not specified, the default behavior is defined by gRPC.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "httpGet": { "additionalProperties": false, "description": "HTTPGet specifies an HTTP GET request to perform.", "properties": { "host": { "description": "Host name to connect to, defaults to the pod IP. You probably want to set\n\"Host\" in httpHeaders instead.", "type": [ "string", "null" ] }, "httpHeaders": { "description": "Custom headers to set in the request. HTTP allows repeated headers.", "items": { "additionalProperties": false, "description": "HTTPHeader describes a custom header to be used in HTTP probes", "properties": { "name": { "description": "The header field name.\nThis will be canonicalized upon output, so case-variant names will be understood as the same header.", "type": "string" }, "value": { "description": "The header field value", "type": "string" } }, "required": [ "name", "value" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "path": { "description": "Path to access on the HTTP server.", "type": [ "string", "null" ] }, "port": { "description": "Name or number of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true }, "scheme": { "description": "Scheme to use for connecting to the host.\nDefaults to HTTP.", "type": [ "string", "null" ] } }, "required": [ "port" ], "type": [ "object", "null" ] }, "initialDelaySeconds": { "description": "Number of seconds after the container has started before liveness probes are initiated.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] }, "periodSeconds": { "description": "How often (in seconds) to perform the probe.\nDefault to 10 seconds. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "successThreshold": { "description": "Minimum consecutive successes for the probe to be considered successful after having failed.\nDefaults to 1. Must be 1 for liveness and startup. Minimum value is 1.", "format": "int32", "type": [ "integer", "null" ] }, "tcpSocket": { "additionalProperties": false, "description": "TCPSocket specifies a connection to a TCP port.", "properties": { "host": { "description": "Optional: Host name to connect to, defaults to the pod IP.", "type": [ "string", "null" ] }, "port": { "description": "Number or name of the port to access on the container.\nNumber must be in the range 1 to 65535.\nName must be an IANA_SVC_NAME.", "oneOf": [ { "type": "string" }, { "type": "integer" } ], "x-kubernetes-int-or-string": true } }, "required": [ "port" ], "type": [ "object", "null" ] }, "terminationGracePeriodSeconds": { "description": "Optional duration in seconds the pod needs to terminate gracefully upon probe failure.\nThe grace period is the duration in seconds after the processes running in the pod are sent\na termination signal and the time when the processes are forcibly halted with a kill signal.\nSet this value longer than the expected cleanup time for your process.\nIf this value is nil, the pod's terminationGracePeriodSeconds will be used. Otherwise, this\nvalue overrides the value provided by the pod spec.\nValue must be non-negative integer. The value zero indicates stop immediately via\nthe kill signal (no opportunity to shut down).\nThis is a beta field and requires enabling ProbeTerminationGracePeriod feature gate.\nMinimum value is 1. spec.terminationGracePeriodSeconds is used if unset.", "format": "int64", "type": [ "integer", "null" ] }, "timeoutSeconds": { "description": "Number of seconds after which the probe times out.\nDefaults to 1 second. Minimum value is 1.\nMore info: https://kubernetes.io/docs/concepts/workloads/pods/pod-lifecycle#container-probes", "format": "int32", "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "reasoningBudget": { "description": "ReasoningBudget caps the number of reasoning tokens the model is allowed to\nemit per response. Zero disables visible thinking output entirely; the model\nstill reasons internally but does not emit thinking tokens. Critical for\nproduction agentic workloads on thinking models (Qwen 3.6, GLM-5) where\nrunaway reasoning can burn compute.\nMaps to llama.cpp --reasoning-budget flag.", "format": "int32", "minimum": 0, "type": [ "integer", "null" ] }, "reasoningBudgetMessage": { "description": "ReasoningBudgetMessage is injected when the reasoning budget is exhausted,\nforcing the model to conclude. Ignored unless ReasoningBudget is also set.\nMaps to llama.cpp --reasoning-budget-message flag.", "type": [ "string", "null" ] }, "replicas": { "default": 1, "description": "Replicas is the desired number of inference pods", "format": "int32", "maximum": 10, "minimum": 0, "type": [ "integer", "null" ] }, "resources": { "additionalProperties": false, "description": "Resources defines compute resources for inference pods", "properties": { "cpu": { "description": "CPU requests (e.g., \"2\" or \"2000m\")", "type": [ "string", "null" ] }, "gpu": { "description": "GPU count required per pod\nFor multi-GPU inference, each pod gets this many GPUs\nNote: Multi-GPU sharding config comes from Model CRD", "format": "int32", "maximum": 8, "minimum": 0, "type": [ "integer", "null" ] }, "gpuMemory": { "description": "GPUMemory specifies GPU memory limit per pod (e.g., \"16Gi\")\nUsed for scheduling and validation", "type": [ "string", "null" ] }, "hostMemory": { "description": "HostMemory specifies the system RAM required for hybrid GPU/CPU offloading (e.g., \"64Gi\").\nUsed when MoE expert weights or KV cache are offloaded to CPU via moeCPUOffload or noKvOffload.\nTranslated to pod resources.requests.memory, taking precedence over Memory when set.\nWithout this, the K8s scheduler has no visibility into the pod's actual RAM consumption,\nwhich can lead to OOM kills after model load.", "type": [ "string", "null" ] }, "memory": { "description": "Memory requests (e.g., \"4Gi\")", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "revisionHistoryLimit": { "description": "RevisionHistoryLimit caps how many old ReplicaSets the inference\nDeployment keeps for rollback. Unset uses the Kubernetes default (10);\n0 keeps none.", "format": "int32", "minimum": 0, "type": [ "integer", "null" ] }, "rolloutPolicy": { "additionalProperties": false, "description": "RolloutPolicy controls how deployment updates are applied. When waitForIdle\nis true, the controller will check backend slot idleness before updating\nthe Deployment pod-template. Idle detection support by runtime:\n - llama.cpp: native /slots endpoint (default)\n - vLLM: Prometheus metrics scrape (vllm:num_requests_running)\n - TGI: Prometheus metrics scrape (tgi_batch_current_size)\n - SGLang: Prometheus metrics scrape (sglang:num_running_reqs)\n - generic: optional AnnotationIdleEndpoint annotation for custom probe", "properties": { "force": { "description": "Force bypasses the idle check and proceeds with the rollout immediately.\nWhen true, WaitForIdle is ignored. Useful for emergency rollouts or when\nslots are stuck in a non-idle state.", "type": [ "boolean", "null" ] }, "idleTimeoutSeconds": { "default": 300, "description": "IdleTimeoutSeconds is the maximum time to wait for slots to become idle before\nproceeding with the rollout regardless of slot state. Defaults to 300 (5 minutes)\nwhen omitted or set to 0.", "minimum": 0, "type": [ "integer", "null" ] }, "waitForIdle": { "description": "WaitForIdle indicates whether to wait for all backend slots to report idle\nbefore applying a Deployment pod-template update. When true, the controller\nprobes each replica and defers the rollout until all replicas are idle or the\nidleTimeoutSeconds expires. Idle detection is runtime-specific: llama.cpp uses\n/slots, vLLM/TGI/SGLang scrape Prometheus gauges, and generic runtimes may set\nAnnotationIdleEndpoint for a custom HTTP probe. Runtimes without idle detection\nsupport proceed immediately with ReasonIdleCheckUnsupported.", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ] }, "ropeScaling": { "additionalProperties": false, "description": "RopeScaling configures RoPE-based context extension so a model can be\nserved past its native trained context (e.g. 128K served at 256K via\nYaRN). For the llamacpp runtime this maps to --rope-scaling /\n--rope-scale / --yarn-orig-ctx. Prefer this over raw spec.extraArgs:\nit is validated and discoverable via `kubectl explain`. If --rope-scaling\nis also present in spec.extraArgs, extraArgs wins and this is skipped.", "properties": { "factor": { "description": "Factor is the scale multiplier (--rope-scale), e.g. \"2.0\" to double the\nnative context. A string to avoid CRD float pitfalls; the runtime parses\nit as a float. Optional.", "pattern": "^[0-9]+(\\.[0-9]+)?$", "type": [ "string", "null" ] }, "originalContext": { "description": "OriginalContext is the model's native training context length\n(--yarn-orig-ctx), e.g. 131072 for a 128K model. Recommended with yarn.", "format": "int32", "minimum": 128, "type": [ "integer", "null" ] }, "type": { "description": "Type is the scaling method (--rope-scaling). \"yarn\" is the usual choice\nfor extending context (e.g. 128K to 256K).", "enum": [ "linear", "yarn", "longrope" ], "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "runtime": { "default": "llamacpp", "description": "Runtime selects the inference server backend.\n\"llamacpp\" (default): llama.cpp server with auto-generated args and /health probes.\n\"generic\": user-provided container with custom command, args, env, and probes.\n\"personaplex\": NVIDIA PersonaPlex (Moshi) speech-to-speech server.\n\"vllm\": vLLM OpenAI-compatible server with PagedAttention.\n\"tgi\": HuggingFace Text Generation Inference server.\n\"sglang\": SGLang OpenAI-compatible server with RadixAttention prefix caching.", "enum": [ "llamacpp", "personaplex", "vllm", "tgi", "sglang", "generic" ], "type": [ "string", "null" ] }, "runtimeClassName": { "description": "RuntimeClassName selects a Kubernetes RuntimeClass for the inference Pod.\nMost commonly set to \"nvidia\" on clusters where the NVIDIA Container\nRuntime is not configured as the cluster default. Without it, GPU pods\nschedule onto the GPU node but never get the device files bind-mounted,\nand the container fails at runtime with \"no CUDA-capable device is\ndetected\". Maps directly to PodSpec.RuntimeClassName.\n\nMost clusters running the NVIDIA GPU Operator with the default toolkit\nenv do not need this set; it is a safety hatch for clusters where the\nruntime configuration is non-default.", "type": [ "string", "null" ] }, "securityContext": { "additionalProperties": false, "description": "SecurityContext defines container-level security attributes for the inference container.", "properties": { "allowPrivilegeEscalation": { "description": "AllowPrivilegeEscalation controls whether a process can gain more\nprivileges than its parent process. This bool directly controls if\nthe no_new_privs flag will be set on the container process.\nAllowPrivilegeEscalation is true always when the container is:\n1) run as Privileged\n2) has CAP_SYS_ADMIN\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "boolean", "null" ] }, "appArmorProfile": { "additionalProperties": false, "description": "appArmorProfile is the AppArmor options to use by this container. If set, this profile\noverrides the pod's appArmorProfile.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "localhostProfile": { "description": "localhostProfile indicates a profile loaded on the node that should be used.\nThe profile must be preconfigured on the node to work.\nMust match the loaded name of the profile.\nMust be set if and only if type is \"Localhost\".", "type": [ "string", "null" ] }, "type": { "description": "type indicates which kind of AppArmor profile will be applied.\nValid options are:\n Localhost - a profile pre-loaded on the node.\n RuntimeDefault - the container runtime's default profile.\n Unconfined - no AppArmor enforcement.", "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "capabilities": { "additionalProperties": false, "description": "The capabilities to add/drop when running containers.\nDefaults to the default set of capabilities granted by the container runtime.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "add": { "description": "Added capabilities", "items": { "description": "Capability represent POSIX capabilities type", "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "drop": { "description": "Removed capabilities", "items": { "description": "Capability represent POSIX capabilities type", "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "type": [ "object", "null" ] }, "privileged": { "description": "Run container in privileged mode.\nProcesses in privileged containers are essentially equivalent to root on the host.\nDefaults to false.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "boolean", "null" ] }, "procMount": { "description": "procMount denotes the type of proc mount to use for the containers.\nThe default value is Default which uses the container runtime defaults for\nreadonly paths and masked paths.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "string", "null" ] }, "readOnlyRootFilesystem": { "description": "Whether this container has a read-only root filesystem.\nDefault is false.\nNote that this field cannot be set when spec.os.name is windows.", "type": [ "boolean", "null" ] }, "runAsGroup": { "description": "The GID to run the entrypoint of the container process.\nUses runtime default if unset.\nMay also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.\nNote that this field cannot be set when spec.os.name is windows.", "format": "int64", "type": [ "integer", "null" ] }, "runAsNonRoot": { "description": "Indicates that the container must run as a non-root user.\nIf true, the Kubelet will validate the image at runtime to ensure that it\ndoes not run as UID 0 (root) and fail to start the container if it does.\nIf unset or false, no such validation will be performed.\nMay also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.", "type": [ "boolean", "null" ] }, "runAsUser": { "description": "The UID to run the entrypoint of the container process.\nDefaults to user specified in image metadata if unspecified.\nMay also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.\nNote that this field cannot be set when spec.os.name is windows.", "format": "int64", "type": [ "integer", "null" ] }, "seLinuxOptions": { "additionalProperties": false, "description": "The SELinux context to be applied to the container.\nIf unspecified, the container runtime will allocate a random SELinux context for each\ncontainer. May also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "level": { "description": "Level is SELinux level label that applies to the container.", "type": [ "string", "null" ] }, "role": { "description": "Role is a SELinux role label that applies to the container.", "type": [ "string", "null" ] }, "type": { "description": "Type is a SELinux type label that applies to the container.", "type": [ "string", "null" ] }, "user": { "description": "User is a SELinux user label that applies to the container.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "seccompProfile": { "additionalProperties": false, "description": "The seccomp options to use by this container. If seccomp options are\nprovided at both the pod \u0026 container level, the container options\noverride the pod options.\nNote that this field cannot be set when spec.os.name is windows.", "properties": { "localhostProfile": { "description": "localhostProfile indicates a profile defined in a file on the node should be used.\nThe profile must be preconfigured on the node to work.\nMust be a descending path, relative to the kubelet's configured seccomp profile location.\nMust be set if type is \"Localhost\". Must NOT be set for any other type.", "type": [ "string", "null" ] }, "type": { "description": "type indicates which kind of seccomp profile will be applied.\nValid options are:\n\nLocalhost - a profile defined in a file on the node should be used.\nRuntimeDefault - the container runtime default profile should be used.\nUnconfined - no profile should be applied.", "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "windowsOptions": { "additionalProperties": false, "description": "The Windows specific settings applied to all containers.\nIf unspecified, the options from the PodSecurityContext will be used.\nIf set in both SecurityContext and PodSecurityContext, the value specified in SecurityContext takes precedence.\nNote that this field cannot be set when spec.os.name is linux.", "properties": { "gmsaCredentialSpec": { "description": "GMSACredentialSpec is where the GMSA admission webhook\n(https://github.com/kubernetes-sigs/windows-gmsa) inlines the contents of the\nGMSA credential spec named by the GMSACredentialSpecName field.", "type": [ "string", "null" ] }, "gmsaCredentialSpecName": { "description": "GMSACredentialSpecName is the name of the GMSA credential spec to use.", "type": [ "string", "null" ] }, "hostProcess": { "description": "HostProcess determines if a container should be run as a 'Host Process' container.\nAll of a Pod's containers must have the same effective HostProcess value\n(it is not allowed to have a mix of HostProcess containers and non-HostProcess containers).\nIn addition, if HostProcess is true then HostNetwork must also be set to true.", "type": [ "boolean", "null" ] }, "runAsUserName": { "description": "The UserName in Windows to run the entrypoint of the container process.\nDefaults to the user specified in image metadata if unspecified.\nMay also be set in PodSecurityContext. If set in both SecurityContext and\nPodSecurityContext, the value specified in SecurityContext takes precedence.", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] } }, "type": [ "object", "null" ] }, "sglangConfig": { "additionalProperties": false, "description": "SGLangConfig holds configuration for the SGLang runtime.\nOnly used when Runtime is \"sglang\".", "properties": { "attentionBackend": { "description": "Attention\nAttentionBackend selects the attention implementation. flashinfer is\nfastest on recent NVIDIA GPUs; flash_attn is portable; torch_native is\nthe fallback. Maps to SGLang --attention-backend flag.", "enum": [ "flashinfer", "flash_attn", "torch_native" ], "type": [ "string", "null" ] }, "chatTemplate": { "description": "ChatTemplate overrides the model's bundled chat template. Maps to\nSGLang --chat-template flag.", "type": [ "string", "null" ] }, "chunkedPrefillSize": { "description": "Batching\nChunkedPrefillSize sets the chunk size for chunked prefill (tokens).\nMaps to SGLang --chunked-prefill-size flag.", "format": "int32", "minimum": 512, "type": [ "integer", "null" ] }, "contextLength": { "description": "Memory \u0026 context\nContextLength sets the maximum model context length.\nMaps to SGLang --context-length flag.", "format": "int32", "minimum": 128, "type": [ "integer", "null" ] }, "dataParallelSize": { "description": "DataParallelSize sets the number of data-parallel replicas (SGLang-side controller).\nMaps to SGLang --dp flag. Not auto-derived; set explicitly.\n\nNOTE: at present this only sets the in-process SGLang `--dp` flag.\nMulti-replica rendezvous (SGLang's --dist-init-addr + a stable\nnetwork identity per pod, e.g. headless service + StatefulSet) is\nnot yet wired into the InferenceService controller and is tracked\nat https://github.com/defilantech/LLMKube/issues/1102. Setting\nthis on an InferenceService with replicas \u003e 1 will leave each\nreplica starting as its own DP-1 group; operators wanting true\nDP coordination should hold off on this flag until #1102 lands.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "enablePrefixCaching": { "description": "EnablePrefixCaching turns on RadixAttention automatic prefix caching.\nHeadline feature for agentic workloads with shared system-prompt +\ntool-definition + repo-context prefixes. Maps to SGLang --enable-prefix-caching.", "type": [ "boolean", "null" ] }, "expertParallelSize": { "description": "ExpertParallelSize sets the number of GPUs for expert parallelism (MoE models).\nMaps to SGLang --ep flag. Not auto-derived; set explicitly.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "hfTokenSecretRef": { "additionalProperties": false, "description": "HFTokenSecretRef references a Secret containing the HuggingFace token.\nInjected as HF_TOKEN env var.", "properties": { "key": { "description": "The key of the secret to select from. Must be a valid secret key.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "kvCacheCustomDtype": { "description": "KVCacheCustomDtype sets a custom SGLang KV cache type not in the standard\nenum. Maps to SGLang --kv-cache-dtype flag. Takes precedence over\nKVCacheDtype when both are set. LLMKube does not validate the string.", "type": [ "string", "null" ] }, "kvCacheDtype": { "default": "auto", "description": "KVCacheDtype selects the KV cache element type. auto follows dtype.\nfp8_e5m2 / fp8_e4m3 cut KV memory roughly in half. Custom values not\nin the enum (e.g., TurboQuant) go in KVCacheCustomDtype.\nMaps to SGLang --kv-cache-dtype flag.", "enum": [ "auto", "fp8_e5m2", "fp8_e4m3" ], "type": [ "string", "null" ] }, "logLevel": { "description": "LogLevel sets the SGLang server log level. SGLang accepts\n\"debug\"/\"info\"/\"warning\"/\"error\". Maps to SGLang --log-level flag.", "enum": [ "debug", "info", "warning", "error" ], "type": [ "string", "null" ] }, "loraAdapters": { "description": "LoraAdapters is a typed replacement for LoraModules. Each adapter has\na stable Name (SGLang-side handle) and Path (file mount). When both\nLoraAdapters and LoraModules are set, LoraAdapters wins on name\ncollision. Maps to SGLang --lora-paths flag (singular `lora_paths`,\nnot vLLM's --lora-modules — see\nhttps://github.com/sgl-project/sglang/blob/v0.5.15/python/sglang/srt/server_args.py).", "items": { "additionalProperties": false, "description": "SGLangLoRAAdapter names a single LoRA adapter for SGLang's --lora-paths\nflag (NOT vLLM's --lora-modules — see\nhttps://github.com/sgl-project/sglang/blob/v0.5.15/python/sglang/srt/server_args.py).\nName is the SGLang-side adapter handle; Path is the file mount where\nadapter weights live (typically backed by a PVC created via\nLoRAAdapter resources). Prefer this typed shape over the legacy\nLoraModules []string; both are merged with the typed form winning on name\ncollision.", "properties": { "name": { "description": "Name is the SGLang-side adapter handle used in inference requests.", "minLength": 1, "type": "string" }, "path": { "description": "Path is the path on disk inside the SGLang container where the\nadapter weights are mounted.", "minLength": 1, "type": "string" } }, "required": [ "name", "path" ], "type": "object" }, "type": [ "array", "null" ] }, "loraModules": { "description": "LoRA (basic)\nLoraModules is the legacy form of --lora-paths entries. Each\nelement is either `name=path` shorthand or a JSON object\n{\"name\":\"x\",\"path\":\"/p\"}. New callers should prefer the typed\nLoraAdapters field; the controller merges both, with\nLoraAdapters winning on name collision. Deprecated: use\nLoraAdapters instead.", "items": { "type": "string" }, "type": [ "array", "null" ] }, "loraTargetModules": { "description": "LoraTargetModules lists the modules LoRA adapters may target (e.g.,\n\"q_proj\", \"k_proj\"). Maps to SGLang --lora-target-modules flag.", "items": { "type": "string" }, "type": [ "array", "null" ] }, "maxLoraRank": { "description": "MaxLoraRank sets the maximum LoRA rank accepted at load time. Maps to\nSGLang --max-lora-rank flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "maxRunningRequests": { "description": "MaxRunningRequests caps concurrent in-flight requests. Maps to SGLang\n--max-running-requests flag. Spec.parallelSlots on the llama.cpp runtime\nis the analog; SGLang uses its own name.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "memFractionStatic": { "description": "MemFractionStatic sets the fraction of GPU memory used for static state\n(model weights + KV cache). Range 0.1-0.99. Requires GPU.\nMaps to SGLang --mem-fraction-static flag.", "maximum": 0.99, "minimum": 0.1, "type": [ "number", "null" ] }, "quantization": { "description": "Quantization \u0026 KV cache\nQuantization sets the quantization method. SGLang accepts fp8/awq/gptq/modelopt.\nMaps to SGLang --quantization flag.", "type": [ "string", "null" ] }, "reasoningParser": { "description": "ReasoningParser selects the reasoning-content extraction format. For\nthinking models (qwen3, deepseek-r1). Maps to SGLang --reasoning-parser.", "enum": [ "qwen3", "deepseek-r1" ], "type": [ "string", "null" ] }, "skipTokenizerInit": { "description": "SkipTokenizerInit skips tokenizer initialization at startup. Useful\nfor prefill-only disaggregation deployments. Maps to SGLang\n--skip-tokenizer-init flag. Omit to leave SGLang's default.", "type": [ "boolean", "null" ] }, "speculative": { "additionalProperties": false, "description": "Speculative configures speculative decoding (EAGLE / EAGLE3 / Medusa).", "properties": { "acceptThresholdAcc": { "description": "AcceptThresholdAcc sets the acceptance threshold for the bonus token\nin accepted-token-sequence verification (an accepted draft token's\nprobability must exceed p * accept_threshold_acc). Valid only when\nEnabled is true; surface a status condition when set otherwise.\nMaps to SGLang --speculative-accept-threshold-acc flag.", "maximum": 1, "minimum": 0, "type": [ "number", "null" ] }, "acceptThresholdSingle": { "description": "AcceptThresholdSingle sets the acceptance threshold for non-matched\ntokens in single-sequence decoding (a draft token is accepted when\nits probability exceeds p * accept_threshold_single). Valid only\nwhen Enabled is true; surface a status condition when set otherwise.\nMaps to SGLang --speculative-accept-threshold-single flag.", "maximum": 1, "minimum": 0, "type": [ "number", "null" ] }, "algorithm": { "description": "Algorithm selects the speculative algorithm (EAGLE, EAGLE3, Medusa).\nMaps to SGLang --speculative-algorithm flag.", "enum": [ "EAGLE", "EAGLE3", "Medusa" ], "type": [ "string", "null" ] }, "draftModelPath": { "description": "DraftModelPath is the path to the draft model weights (for EAGLE).\nMaps to SGLang --speculative-draft-model-path flag. Required when Enabled.", "type": [ "string", "null" ] }, "eagleTopK": { "description": "EagleTopK is the top-k sampling for EAGLE draft tokens.\nMaps to SGLang --speculative-eagle-topk flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "enabled": { "description": "Enabled toggles speculative decoding on. When false or nil, no flags\nare emitted regardless of other fields.", "type": [ "boolean", "null" ] }, "numDraftTokens": { "default": 4, "description": "NumDraftTokens is the number of draft tokens proposed per step.\nMaps to SGLang --speculative-num-draft-tokens flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "numSteps": { "description": "NumSteps is the number of draft steps per forward pass.\nMaps to SGLang --speculative-num-steps flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] }, "tensorParallelSize": { "description": "Sharding\nTensorParallelSize sets the number of GPUs for tensor parallelism.\nMaps to SGLang --tp flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "toolCallParser": { "description": "Agentic glue\nToolCallParser selects the tool-call extraction format. For foreman\ntool-loop workloads. Maps to SGLang --tool-call-parser flag.", "enum": [ "llama3", "qwen3", "qwen25", "hermes", "functionary", "mistral" ], "type": [ "string", "null" ] }, "trustRemoteCode": { "description": "TrustRemoteCode allows loading remote code from the HuggingFace Hub\nmodel repo. Mirrors the flag on other runtimes. Maps to SGLang\n--trust-remote-code flag. Omit to leave SGLang's default.", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ] }, "skipModelInit": { "description": "SkipModelInit disables the model-downloader init container.\nUse when the model is baked into the image or downloaded by the\ncontainer itself (e.g., via HF_TOKEN).", "type": [ "boolean", "null" ] }, "speculativeDecoding": { "additionalProperties": false, "description": "SpeculativeDecoding configures speculative decoding for the llama.cpp\nruntime using MTP (Multi-Token Prediction) or draft-model decoding.\nMaps to llama.cpp --spec-type and --draft-n-max flags. Only the\n\"llamacpp\" runtime supports this field; other runtimes must not set it.", "properties": { "nDraftMax": { "description": "NDraftMax is the maximum number of draft tokens to propose per step\n(--draft-n-max). Only emitted when set; llama.cpp uses its own default\notherwise.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "type": { "description": "Type is the speculative decoding method (--spec-type). \"mtp\" maps to\ndraft-mtp, \"draft\" maps to draft, and \"disabled\" (or omitting the\nentire SpeculativeDecoding block) means no speculative decoding.", "enum": [ "mtp", "draft", "disabled" ], "type": "string" } }, "required": [ "type" ], "type": [ "object", "null" ] }, "tensorOverrides": { "description": "TensorOverrides provides fine-grained tensor placement overrides for power users.\nEach entry specifies a tensor name and target device (e.g., \"exps=CPU\", \"token_embd=CUDA0\").\nMaps to llama.cpp --override-tensor flag (one flag per entry).", "items": { "type": "string" }, "type": [ "array", "null" ] }, "tgiConfig": { "additionalProperties": false, "description": "TGIConfig holds configuration for the TGI runtime.\nOnly used when Runtime is \"tgi\".", "properties": { "dtype": { "description": "Dtype sets the model data type (float16, bfloat16).", "enum": [ "float16", "bfloat16" ], "type": [ "string", "null" ] }, "hfTokenSecretRef": { "additionalProperties": false, "description": "HFTokenSecretRef references a Secret containing the HuggingFace token.", "properties": { "key": { "description": "The key of the secret to select from. Must be a valid secret key.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "maxInputLength": { "description": "MaxInputLength sets the maximum input token length.", "format": "int32", "type": [ "integer", "null" ] }, "maxTotalTokens": { "description": "MaxTotalTokens sets the maximum total tokens (input + output).", "format": "int32", "type": [ "integer", "null" ] }, "quantize": { "description": "Quantize sets the quantization method (bitsandbytes, gptq, awq, eetq).", "enum": [ "bitsandbytes", "gptq", "awq", "eetq" ], "type": [ "string", "null" ] } }, "type": [ "object", "null" ] }, "tolerations": { "description": "Tolerations for pod scheduling (e.g., GPU taints, spot instances)", "items": { "additionalProperties": false, "description": "The pod this Toleration is attached to tolerates any taint that matches\nthe triple \u003ckey,value,effect\u003e using the matching operator \u003coperator\u003e.", "properties": { "effect": { "description": "Effect indicates the taint effect to match. Empty means match all taint effects.\nWhen specified, allowed values are NoSchedule, PreferNoSchedule and NoExecute.", "type": [ "string", "null" ] }, "key": { "description": "Key is the taint key that the toleration applies to. Empty means match all taint keys.\nIf the key is empty, operator must be Exists; this combination means to match all values and all keys.", "type": [ "string", "null" ] }, "operator": { "description": "Operator represents a key's relationship to the value.\nValid operators are Exists, Equal, Lt, and Gt. Defaults to Equal.\nExists is equivalent to wildcard for value, so that a pod can\ntolerate all taints of a particular category.\nLt and Gt perform numeric comparisons (requires feature gate TaintTolerationComparisonOperators).", "type": [ "string", "null" ] }, "tolerationSeconds": { "description": "TolerationSeconds represents the period of time the toleration (which must be\nof effect NoExecute, otherwise this field is ignored) tolerates the taint. By default,\nit is not set, which means tolerate the taint forever (do not evict). Zero and\nnegative values will be treated as 0 (evict immediately) by the system.", "format": "int64", "type": [ "integer", "null" ] }, "value": { "description": "Value is the taint value the toleration matches to.\nIf the operator is Exists, the value should be empty, otherwise just a regular string.", "type": [ "string", "null" ] } }, "type": "object" }, "type": [ "array", "null" ] }, "topologySpreadConstraints": { "description": "TopologySpreadConstraints control how inference Pods are spread across\ntopology domains (e.g. one model server per GPU node). Passthrough to the\nPod spec; combine with PodLabels so the constraint's labelSelector can\nmatch sibling GPU workloads for a soft, cross-app spread.", "items": { "additionalProperties": false, "description": "TopologySpreadConstraint specifies how to spread matching pods among the given topology.", "properties": { "labelSelector": { "additionalProperties": false, "description": "LabelSelector is used to find matching pods.\nPods that match this label selector are counted to determine the number of pods\nin their corresponding topology domain.", "properties": { "matchExpressions": { "description": "matchExpressions is a list of label selector requirements. The requirements are ANDed.", "items": { "additionalProperties": false, "description": "A label selector requirement is a selector that contains values, a key, and an operator that\nrelates the key and values.", "properties": { "key": { "description": "key is the label key that the selector applies to.", "type": "string" }, "operator": { "description": "operator represents a key's relationship to a set of values.\nValid operators are In, NotIn, Exists and DoesNotExist.", "type": "string" }, "values": { "description": "values is an array of string values. If the operator is In or NotIn,\nthe values array must be non-empty. If the operator is Exists or DoesNotExist,\nthe values array must be empty. This array is replaced during a strategic\nmerge patch.", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" } }, "required": [ "key", "operator" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "matchLabels": { "additionalProperties": { "type": "string" }, "description": "matchLabels is a map of {key,value} pairs. A single {key,value} in the matchLabels\nmap is equivalent to an element of matchExpressions, whose key field is \"key\", the\noperator is \"In\", and the values array contains only \"value\". The requirements are ANDed.", "type": [ "object", "null" ] } }, "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "matchLabelKeys": { "description": "MatchLabelKeys is a set of pod label keys to select the pods over which\nspreading will be calculated. The keys are used to lookup values from the\nincoming pod labels, those key-value labels are ANDed with labelSelector\nto select the group of existing pods over which spreading will be calculated\nfor the incoming pod. The same key is forbidden to exist in both MatchLabelKeys and LabelSelector.\nMatchLabelKeys cannot be set when LabelSelector isn't set.\nKeys that don't exist in the incoming pod labels will\nbe ignored. A null or empty list means only match against labelSelector.\n\nThis is a beta field and requires the MatchLabelKeysInPodTopologySpread feature gate to be enabled (enabled by default).", "items": { "type": "string" }, "type": [ "array", "null" ], "x-kubernetes-list-type": "atomic" }, "maxSkew": { "description": "MaxSkew describes the degree to which pods may be unevenly distributed.\nWhen `whenUnsatisfiable=DoNotSchedule`, it is the maximum permitted difference\nbetween the number of matching pods in the target topology and the global minimum.\nThe global minimum is the minimum number of matching pods in an eligible domain\nor zero if the number of eligible domains is less than MinDomains.\nFor example, in a 3-zone cluster, MaxSkew is set to 1, and pods with the same\nlabelSelector spread as 2/2/1:\nIn this case, the global minimum is 1.\n| zone1 | zone2 | zone3 |\n| P P | P P | P |\n- if MaxSkew is 1, incoming pod can only be scheduled to zone3 to become 2/2/2;\nscheduling it onto zone1(zone2) would make the ActualSkew(3-1) on zone1(zone2)\nviolate MaxSkew(1).\n- if MaxSkew is 2, incoming pod can be scheduled onto any zone.\nWhen `whenUnsatisfiable=ScheduleAnyway`, it is used to give higher precedence\nto topologies that satisfy it.\nIt's a required field. Default value is 1 and 0 is not allowed.", "format": "int32", "type": "integer" }, "minDomains": { "description": "MinDomains indicates a minimum number of eligible domains.\nWhen the number of eligible domains with matching topology keys is less than minDomains,\nPod Topology Spread treats \"global minimum\" as 0, and then the calculation of Skew is performed.\nAnd when the number of eligible domains with matching topology keys equals or greater than minDomains,\nthis value has no effect on scheduling.\nAs a result, when the number of eligible domains is less than minDomains,\nscheduler won't schedule more than maxSkew Pods to those domains.\nIf value is nil, the constraint behaves as if MinDomains is equal to 1.\nValid values are integers greater than 0.\nWhen value is not nil, WhenUnsatisfiable must be DoNotSchedule.\n\nFor example, in a 3-zone cluster, MaxSkew is set to 2, MinDomains is set to 5 and pods with the same\nlabelSelector spread as 2/2/2:\n| zone1 | zone2 | zone3 |\n| P P | P P | P P |\nThe number of domains is less than 5(MinDomains), so \"global minimum\" is treated as 0.\nIn this situation, new pod with the same labelSelector cannot be scheduled,\nbecause computed skew will be 3(3 - 0) if new Pod is scheduled to any of the three zones,\nit will violate MaxSkew.", "format": "int32", "type": [ "integer", "null" ] }, "nodeAffinityPolicy": { "description": "NodeAffinityPolicy indicates how we will treat Pod's nodeAffinity/nodeSelector\nwhen calculating pod topology spread skew. Options are:\n- Honor: only nodes matching nodeAffinity/nodeSelector are included in the calculations.\n- Ignore: nodeAffinity/nodeSelector are ignored. All nodes are included in the calculations.\n\nIf this value is nil, the behavior is equivalent to the Honor policy.", "type": [ "string", "null" ] }, "nodeTaintsPolicy": { "description": "NodeTaintsPolicy indicates how we will treat node taints when calculating\npod topology spread skew. Options are:\n- Honor: nodes without taints, along with tainted nodes for which the incoming pod\nhas a toleration, are included.\n- Ignore: node taints are ignored. All nodes are included.\n\nIf this value is nil, the behavior is equivalent to the Ignore policy.", "type": [ "string", "null" ] }, "topologyKey": { "description": "TopologyKey is the key of node labels. Nodes that have a label with this key\nand identical values are considered to be in the same topology.\nWe consider each \u003ckey, value\u003e as a \"bucket\", and try to put balanced number\nof pods into each bucket.\nWe define a domain as a particular instance of a topology.\nAlso, we define an eligible domain as a domain whose nodes meet the requirements of\nnodeAffinityPolicy and nodeTaintsPolicy.\ne.g. If TopologyKey is \"kubernetes.io/hostname\", each Node is a domain of that topology.\nAnd, if TopologyKey is \"topology.kubernetes.io/zone\", each zone is a domain of that topology.\nIt's a required field.", "type": "string" }, "whenUnsatisfiable": { "description": "WhenUnsatisfiable indicates how to deal with a pod if it doesn't satisfy\nthe spread constraint.\n- DoNotSchedule (default) tells the scheduler not to schedule it.\n- ScheduleAnyway tells the scheduler to schedule the pod in any location,\n but giving higher precedence to topologies that would help reduce the\n skew.\nA constraint is considered \"Unsatisfiable\" for an incoming pod\nif and only if every possible node assignment for that pod would violate\n\"MaxSkew\" on some topology.\nFor example, in a 3-zone cluster, MaxSkew is set to 1, and pods with the same\nlabelSelector spread as 3/1/1:\n| zone1 | zone2 | zone3 |\n| P P P | P | P |\nIf WhenUnsatisfiable is set to DoNotSchedule, incoming pod can only be scheduled\nto zone2(zone3) to become 3/2/1(3/1/2) as ActualSkew(2-1) on zone2(zone3) satisfies\nMaxSkew(1). In other words, the cluster can still be imbalanced, but scheduler\nwon't make it *more* imbalanced.\nIt's a required field.", "type": "string" } }, "required": [ "maxSkew", "topologyKey", "whenUnsatisfiable" ], "type": "object" }, "type": [ "array", "null" ] }, "turboQuantBits": { "description": "TurboQuantBits sets the KV cache quantization bit width for the oMLX\nruntime (3, 6, or 8). Maps to oMLX --kv-cache-quant. When set, the\noMLX daemon uses TurboQuant to compress the KV cache, reducing memory\nusage by up to 67% with minimal speed impact (~7% overhead). Only\nmeaningful for the omlx runtime; ignored by llamacpp and other runtimes.\nRequires oMLX v0.3.4+ (which introduced 3-bit TurboQuant) or a later\ndev build (6-bit and 8-bit options).", "enum": [ 3, 6, 8 ], "format": "int32", "type": [ "integer", "null" ] }, "uBatchSize": { "description": "UBatchSize sets the micro-batch size for decoding.\nSmaller micro-batches reduce memory usage during generation.\nMaps to llama.cpp --ubatch-size flag.", "format": "int32", "minimum": 1, "type": [ "integer", "null" ] }, "vllmConfig": { "additionalProperties": false, "description": "VLLMConfig holds configuration for the vLLM runtime.\nOnly used when Runtime is \"vllm\".", "properties": { "attentionBackend": { "description": "AttentionBackend selects the attention implementation used by vLLM.\nFLASHINFER is typically fastest on recent NVIDIA GPUs (especially Blackwell);\nFLASH_ATTN is a solid default; XFORMERS and torch_sdpa are portability\nfallbacks. Requires a vLLM version that supports the chosen backend.\nBoth uppercase (vLLM's native form) and lowercase spellings are accepted\nfor backwards compatibility with earlier LLMKube releases.\nMaps to vLLM --attention-backend flag.", "enum": [ "FLASH_ATTN", "FLASHINFER", "XFORMERS", "flashinfer", "flash_attn", "xformers", "torch_sdpa" ], "type": [ "string", "null" ] }, "cpuOffloadGB": { "description": "CPUOffloadGB increases the GPU memory size. When set, passes\n--cpu-offload-gb to vLLM. Per-rank, so 4 on TP=2 means 4 GB of\nCPU RAM per GPU. Use when FP8 model weights don't fit VRAM.\nThroughput hit is 2-5x on the offloaded path.", "format": "int32", "minimum": 0, "type": [ "integer", "null" ] }, "dtype": { "description": "Dtype sets the model data type (auto, float16, bfloat16).", "enum": [ "auto", "float16", "bfloat16" ], "type": [ "string", "null" ] }, "enableChunkedPrefill": { "description": "EnableChunkedPrefill interleaves long prefills with decode steps so a\nlarge paste (e.g. a 32K-token file) does not starve concurrent decode\nstreams. Only emitted when explicitly set to true.\nMaps to vLLM --enable-chunked-prefill flag.", "type": [ "boolean", "null" ] }, "enableExpertParallel": { "description": "EnableExpertParallel distributes MoE experts across tensor-parallel ranks\ninstead of replicating them. Only meaningful for MoE models.\nMaps to vLLM --enable-expert-parallel flag.", "type": [ "boolean", "null" ] }, "enablePrefixCaching": { "description": "EnablePrefixCaching turns on vLLM's automatic prefix caching for repeated prompts.\nSignificantly reduces time-to-first-token for conversational and agentic workloads\nwhere requests share a common system prompt.\nOnly emitted when explicitly set to true — when nil or false, vLLM's own\ndefault is used (do not emit the flag).\nMaps to vLLM --enable-prefix-caching flag.", "type": [ "boolean", "null" ] }, "gpuMemoryUtilization": { "description": "GPUMemoryUtilization controls how much GPU memory each stage can use.\nWhen set, passes --gpu-memory-utilization to vLLM. Range from 0.1 - 0.99\nand default unset (vLLM uses 0.90).", "maximum": 0.99, "minimum": 0.1, "type": [ "number", "null" ] }, "hfTokenSecretRef": { "additionalProperties": false, "description": "HFTokenSecretRef references a Secret containing the HuggingFace token.", "properties": { "key": { "description": "The key of the secret to select from. Must be a valid secret key.", "type": "string" }, "name": { "default": "", "description": "Name of the referent.\nThis field is effectively required, but due to backwards compatibility is\nallowed to be empty. Instances of this type with an empty value here are\nalmost certainly wrong.\nMore info: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#names", "type": [ "string", "null" ] }, "optional": { "description": "Specify whether the Secret or its key must be defined", "type": [ "boolean", "null" ] } }, "required": [ "key" ], "type": [ "object", "null" ], "x-kubernetes-map-type": "atomic" }, "kvCacheCustomDtype": { "description": "KVCacheCustomDtype sets a custom vLLM KV cache element type that is not\nin the standard enum. Used for vLLM versions with additional cache\nformats such as TurboQuant 2-bit (turbo2, shipped in v0.20.0). Maps to\nvLLM --kv-cache-dtype. The runtime image must understand the value or\nvLLM will fail to start; LLMKube does not validate the string. Mirrors\nthe llama.cpp-side CacheTypeCustomK/V escape hatch.\nTakes precedence over KVCacheDtype when both are set.", "type": [ "string", "null" ] }, "kvCacheDtype": { "default": "auto", "description": "KVCacheDtype selects the KV cache element type. fp8_e5m2 and fp8_e4m3 cut\nKV cache memory roughly in half versus auto (which follows dtype), which\nis what unlocks 128K+ context on consumer VRAM for agentic workloads.\nMaps to vLLM --kv-cache-dtype flag.\nFor custom build types not in the enum (e.g. TurboQuant turbo2 from\nvLLM v0.20+), use KVCacheCustomDtype instead.", "enum": [ "auto", "fp8_e5m2", "fp8_e4m3" ], "type": [ "string", "null" ] }, "maxModelLen": { "description": "MaxModelLen sets the maximum model context length.", "format": "int32", "type": [ "integer", "null" ] }, "maxNumBatchedTokens": { "description": "MaxNumBatchedTokens sets the maximum number of tokens batched together\nper step. This is the main throughput knob: too low means prefill-bound,\ntoo high risks OOM on long context. No default — only emitted when set.\nMaps to vLLM --max-num-batched-tokens flag.", "format": "int32", "minimum": 512, "type": [ "integer", "null" ] }, "quantization": { "description": "Quantization method.\nawq, gptq, squeezellm are classic 4-bit formats. fp8 targets 8-bit FP\ncheckpoints (Qwen FP8, Llama FP8, etc.). nvfp4 is NVIDIA's Blackwell-native\n4-bit format. compressed-tensors is the neuralmagic/vLLM cross-format\nloader used by Unsloth and other recent releases.", "enum": [ "awq", "gptq", "squeezellm", "fp8", "nvfp4", "compressed-tensors" ], "type": [ "string", "null" ] }, "speculative": { "additionalProperties": false, "description": "Speculative enables draft-model speculative decoding. On single-stream\nagentic workloads this can be 30-60% faster than plain tensor-parallel\nexecution. Requires a second (smaller) Model CR to act as the draft.", "properties": { "enabled": { "description": "Enabled toggles speculative decoding on. When false or nil, no\nspeculative flags are emitted regardless of other fields.", "type": [ "boolean", "null" ] }, "model": { "description": "Model references the Model CR (in the same namespace as the\nInferenceService) to use as the speculative draft model.\nRequired when Enabled is true. If missing, speculative decoding is\nskipped and the InferenceService surfaces a SpeculativeInvalid\nstatus condition rather than failing the reconcile.\nMaps to vLLM --speculative-model flag.", "type": [ "string", "null" ] }, "numSpeculativeTokens": { "default": 4, "description": "NumSpeculativeTokens is the number of draft tokens proposed per step.\nTypical sweet spot is 3-5; higher values increase wasted work when the\ndraft disagrees with the target model.\nMaps to vLLM --num-speculative-tokens flag.", "format": "int32", "maximum": 16, "minimum": 1, "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] }, "tensorParallelSize": { "description": "TensorParallelSize sets the number of GPUs for tensor parallelism.", "format": "int32", "type": [ "integer", "null" ] } }, "type": [ "object", "null" ] } }, "required": [ "modelRef" ], "type": "object" }, "status": { "additionalProperties": false, "description": "status defines the observed state of InferenceService", "properties": { "conditions": { "description": "conditions represent the current state of the InferenceService resource.\nEach condition has a unique type and reflects the status of a specific aspect of the resource.\n\nStandard condition types include:\n- \"Available\": the resource is fully functional\n- \"Progressing\": the resource is being created or updated\n- \"Degraded\": the resource failed to reach or maintain its desired state\n\nThe status of each condition is one of True, False, or Unknown.", "items": { "additionalProperties": false, "description": "Condition contains details for one aspect of the current state of this API Resource.", "properties": { "lastTransitionTime": { "description": "lastTransitionTime is the last time the condition transitioned from one status to another.\nThis should be when the underlying condition changed. If that is not known, then using the time when the API field changed is acceptable.", "format": "date-time", "type": "string" }, "message": { "description": "message is a human readable message indicating details about the transition.\nThis may be an empty string.", "maxLength": 32768, "type": "string" }, "observedGeneration": { "description": "observedGeneration represents the .metadata.generation that the condition was set based upon.\nFor instance, if .metadata.generation is currently 12, but the .status.conditions[x].observedGeneration is 9, the condition is out of date\nwith respect to the current state of the instance.", "format": "int64", "minimum": 0, "type": [ "integer", "null" ] }, "reason": { "description": "reason contains a programmatic identifier indicating the reason for the condition's last transition.\nProducers of specific condition types may define expected values and meanings for this field,\nand whether the values are considered a guaranteed API.\nThe value should be a CamelCase string.\nThis field may not be empty.", "maxLength": 1024, "minLength": 1, "pattern": "^[A-Za-z]([A-Za-z0-9_,:]*[A-Za-z0-9_])?$", "type": "string" }, "status": { "description": "status of the condition, one of True, False, Unknown.", "enum": [ "True", "False", "Unknown" ], "type": "string" }, "type": { "description": "type of condition in CamelCase or in foo.example.com/CamelCase.", "maxLength": 316, "pattern": "^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$", "type": "string" } }, "required": [ "lastTransitionTime", "message", "reason", "status", "type" ], "type": "object" }, "type": [ "array", "null" ], "x-kubernetes-list-map-keys": [ "type" ], "x-kubernetes-list-type": "map" }, "desiredReplicas": { "description": "DesiredReplicas is the desired number of replicas", "format": "int32", "type": [ "integer", "null" ] }, "effectivePriority": { "description": "EffectivePriority shows the resolved priority value from the applied PriorityClass", "format": "int32", "type": [ "integer", "null" ] }, "endpoint": { "description": "Endpoint is the service URL where inference requests can be sent", "type": [ "string", "null" ] }, "gateway": { "additionalProperties": false, "description": "Gateway reports the result of Envoy AI Gateway exposure for this\nInferenceService. Populated only when spec.endpoint.gateway is enabled.\nnil means no gateway exposure was requested (or the gateway integration\nis disabled because the aigw CRDs are not installed; that case is also\nsurfaced via the GatewayReady condition).", "properties": { "authEnabled": { "description": "AuthEnabled indicates a SecurityPolicy enforcing JWT authentication was\ncompiled for this route (ModelRouter policy.auth.jwt). Set by the\nModelRouter dataPlane: Gateway path; false when no auth is configured.", "type": [ "boolean", "null" ] }, "endpoint": { "description": "Endpoint is the gateway address clients send OpenAI requests to. Set by\nthe ModelRouter dataPlane: Gateway path (resolved from the referenced\nGateway); empty for the InferenceService path.", "type": [ "string", "null" ] }, "modelName": { "description": "ModelName is the resolved model-name match value clients send as the\nOpenAI \"model\" string to reach this InferenceService through the gateway.\nSet by the InferenceService path; empty for ModelRouter (which fronts\nmany model names).", "type": [ "string", "null" ] }, "routeReady": { "description": "RouteReady indicates the AIGatewayRoute (and its backing Backend +\nAIServiceBackend) were reconciled successfully against the gateway.", "type": [ "boolean", "null" ] } }, "type": [ "object", "null" ] }, "lastUpdated": { "description": "LastUpdated is the timestamp of the last status update", "format": "date-time", "type": [ "string", "null" ] }, "mode": { "description": "Mode is the resolved serving mode (chat, embedding, or rerank): spec.mode\nwhen set, otherwise inferred from the runtime flags and endpoint path.", "type": [ "string", "null" ] }, "modelReady": { "description": "ModelReady indicates if the referenced Model is in Ready state", "type": [ "boolean", "null" ] }, "phase": { "description": "Phase represents the current lifecycle phase of the InferenceService.\nPossible values: Pending, Creating, Progressing, Ready, WaitingForGPU,\nStopped, Failed. Stopped is the terminal state when spec.replicas=0\nhas caused the agent to tear down the workload; tooling polling for\nreadiness should treat Stopped the same as Pending (the user\nintentionally took the service offline; this is not an error).", "enum": [ "Pending", "Creating", "Progressing", "Ready", "WaitingForGPU", "Stopped", "Failed" ], "type": [ "string", "null" ] }, "queuePosition": { "description": "QueuePosition indicates position among pending InferenceServices cluster-wide (0 = not queued)", "format": "int32", "type": [ "integer", "null" ] }, "readyReplicas": { "description": "Replicas tracks the number of ready vs desired pods", "format": "int32", "type": [ "integer", "null" ] }, "replicas": { "description": "Replicas is the current number of running inference pods", "format": "int32", "type": [ "integer", "null" ] }, "schedulingMessage": { "description": "SchedulingMessage provides details about scheduling issues", "type": [ "string", "null" ] }, "schedulingStatus": { "description": "SchedulingStatus indicates why pods cannot be scheduled (e.g., \"InsufficientGPU\")", "type": [ "string", "null" ] }, "waitingFor": { "description": "WaitingFor describes the resource constraint (e.g., \"nvidia.com/gpu: 1\")", "type": [ "string", "null" ] } }, "type": [ "object", "null" ] } }, "required": [ "spec" ], "type": "object" }