swagger.yaml

basePath: /
definitions:
  github_com_leptonai_gpud_api_v1.ComponentEvents:
    properties:
      component:
        type: string
      endTime:
        type: string
      events:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.Event'
        type: array
      startTime:
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.ComponentHealthStates:
    properties:
      component:
        type: string
      states:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.HealthState'
        type: array
    type: object
  github_com_leptonai_gpud_api_v1.ComponentInfo:
    properties:
      component:
        type: string
      endTime:
        type: string
      info:
        $ref: '#/definitions/github_com_leptonai_gpud_api_v1.Info'
      startTime:
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.ComponentMetrics:
    properties:
      component:
        type: string
      metrics:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.Metric'
        type: array
    type: object
  github_com_leptonai_gpud_api_v1.ComponentType:
    enum:
    - custom-plugin
    type: string
    x-enum-varnames:
    - ComponentTypeCustomPlugin
  github_com_leptonai_gpud_api_v1.Event:
    properties:
      component:
        description: Component represents which component generated the event.
        type: string
      message:
        description: Message represents the detailed message of the event.
        type: string
      name:
        description: Name represents the name of the event.
        type: string
      time:
        description: Time represents when the event happened.
        type: string
      type:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.EventType'
        description: Type represents the type of the event.
    type: object
  github_com_leptonai_gpud_api_v1.EventType:
    enum:
    - Unknown
    - Info
    - Warning
    - Critical
    - Fatal
    type: string
    x-enum-varnames:
    - EventTypeUnknown
    - EventTypeInfo
    - EventTypeWarning
    - EventTypeCritical
    - EventTypeFatal
  github_com_leptonai_gpud_api_v1.HealthState:
    properties:
      component:
        description: Component represents the component name.
        type: string
      component_type:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentType'
        description: |-
          ComponentType represents the type of the component.
          It is either "" (just 'component') or "custom-plugin".
      error:
        description: |-
          Error represents the detailed error information, which will be shown
          as More Information to help analyze why it isn’t healthy.
        type: string
      extra_info:
        additionalProperties:
          type: string
        description: ExtraInfo represents the extra information of the state.
        type: object
      health:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.HealthStateType'
        description: |-
          Health represents the health level of the state,
          including StateHealthy, StateUnhealthy and StateDegraded.
          StateDegraded is similar to Unhealthy which also can trigger alerts
          for users or operators, but what StateDegraded means is that the
          issue detected does not affect users’ workload.
      name:
        description: |-
          Name is the name of the state,
          can be different from the component name.
        type: string
      raw_output:
        description: |-
          RawOutput represents the raw output of the health checker.
          e.g., If a custom plugin runs a Python script, the raw output
          is the stdout/stderr of the script.
          The maximum length of the raw output is 4096 bytes.
        type: string
      reason:
        description: Reason represents what happened or detected by GPUd if it isn’t
          healthy.
        type: string
      run_mode:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.RunModeType'
        description: |-
          RunMode is the run mode of the state.
          It can be "manual" that requires manual trigger to run the check.
          Or it can be empty that runs the check periodically.
      suggested_actions:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.SuggestedActions'
        description: SuggestedActions represents the suggested actions to mitigate
          the issue.
      time:
        description: Time represents when the event happened.
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.HealthStateType:
    enum:
    - Healthy
    - Unhealthy
    - Degraded
    - Initializing
    type: string
    x-enum-varnames:
    - HealthStateTypeHealthy
    - HealthStateTypeUnhealthy
    - HealthStateTypeDegraded
    - HealthStateTypeInitializing
  github_com_leptonai_gpud_api_v1.Info:
    properties:
      events:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.Event'
        type: array
      metrics:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.Metric'
        type: array
      states:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.HealthState'
        type: array
    type: object
  github_com_leptonai_gpud_api_v1.MachineCPUInfo:
    properties:
      architecture:
        type: string
      logicalCores:
        type: integer
      manufacturer:
        type: string
      type:
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineDiskDevice:
    properties:
      children:
        items:
          type: string
        type: array
      fsType:
        type: string
      model:
        type: string
      mountPoint:
        type: string
      name:
        type: string
      parents:
        items:
          type: string
        type: array
      partUUID:
        type: string
      rev:
        type: string
      rota:
        type: boolean
      serial:
        type: string
      size:
        type: integer
      type:
        type: string
      used:
        type: integer
      vendor:
        type: string
      wwn:
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineDiskInfo:
    properties:
      blockDevices:
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineDiskDevice'
        type: array
      containerRootDisk:
        description: ContainerRootDisk is the disk device name that mounts the container
          root (such as "/var/lib/kubelet" mount point).
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineGPUInfo:
    properties:
      architecture:
        description: Architecture is "blackwell" for NVIDIA GB200.
        type: string
      gpus:
        description: GPUs is the GPU info of the machine.
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineGPUInstance'
        type: array
      manufacturer:
        description: Manufacturer is "NVIDIA" for NVIDIA GPUs (same as Brand).
        type: string
      memory:
        type: string
      product:
        description: Product may be "NVIDIA-Graphics-Device" for NVIDIA GB200.
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineGPUInstance:
    properties:
      boardID:
        type: integer
      busID:
        description: |-
          BusID is the GPU bus ID from the nvml API.
           e.g., "0000:0f:00.0"
        type: string
      minorID:
        type: string
      sn:
        type: string
      uuid:
        description: |-
          UUID is the GPU UUID from the nvml API.
          e.g., "GPU-46a3bbe2-3e87-3dde-b464-a03eba0c21d7"
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineInfo:
    properties:
      bootID:
        description: BootID is collected by GPUd.
        type: string
      containerRuntimeVersion:
        description: ContainerRuntime Version reported by the node through runtime
          remote API (e.g. containerd://1.4.2).
        type: string
      cpuInfo:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineCPUInfo'
        description: CPUInfo is the CPU info of the machine.
      cudaVersion:
        description: CUDAVersion represents the current version of cuda library.
        type: string
      diskInfo:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineDiskInfo'
        description: DiskInfo is the Disk info of the machine.
      gpuDriverVersion:
        description: GPUDriverVersion represents the current version of GPU driver
          installed
        type: string
      gpuInfo:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineGPUInfo'
        description: GPUInfo is the GPU info of the machine.
      gpudVersion:
        description: GPUdVersion represents the current version of GPUd
        type: string
      hostname:
        description: Hostname is the current host of machine
        type: string
      kernelVersion:
        description: Kernel Version reported by the node from 'uname -r' (e.g. 3.16.0-0.bpo.4-amd64).
        type: string
      machineID:
        description: MachineID is collected by GPUd. It comes from /etc/machine-id
          or /var/lib/dbus/machine-id
        type: string
      memoryInfo:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineMemoryInfo'
        description: MemoryInfo is the memory info of the machine.
      nicInfo:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineNICInfo'
        description: NICInfo is the network info of the machine.
      operatingSystem:
        description: The Operating System reported by the node
        type: string
      osImage:
        description: OS Image reported by the node from /etc/os-release (e.g. Debian
          GNU/Linux 7 (wheezy)).
        type: string
      systemUUID:
        description: SystemUUID comes from https://github.com/google/cadvisor/blob/master/utils/sysfs/sysfs.go#L442
        type: string
      tailscaleVersion:
        description: TailscaleVersion represents the version of the tailscale client
          binary (e.g. 1.76.6).
        type: string
      uptime:
        description: Uptime represents when the machine up
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.MachineMemoryInfo:
    properties:
      totalBytes:
        type: integer
    type: object
  github_com_leptonai_gpud_api_v1.MachineNICInfo:
    properties:
      privateIPInterfaces:
        description: PrivateIPInterfaces is the private network interface info of
          the machine.
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineNetworkInterface'
        type: array
    type: object
  github_com_leptonai_gpud_api_v1.MachineNetworkInterface:
    properties:
      interface:
        description: Interface is the network interface name of the machine.
        type: string
      ip:
        description: IP is the string representation of the netip.Addr of the machine.
        type: string
      mac:
        description: MAC is the MAC address of the machine.
        type: string
    type: object
  github_com_leptonai_gpud_api_v1.Metric:
    properties:
      labels:
        additionalProperties:
          type: string
        type: object
      name:
        type: string
      unix_seconds:
        type: integer
      value:
        type: number
    type: object
  github_com_leptonai_gpud_api_v1.RepairActionType:
    enum:
    - IGNORE_NO_ACTION_REQUIRED
    - REBOOT_SYSTEM
    - HARDWARE_INSPECTION
    - CHECK_USER_APP_AND_GPU
    type: string
    x-enum-varnames:
    - RepairActionTypeIgnoreNoActionRequired
    - RepairActionTypeRebootSystem
    - RepairActionTypeHardwareInspection
    - RepairActionTypeCheckUserAppAndGPU
  github_com_leptonai_gpud_api_v1.RunModeType:
    enum:
    - auto
    - manual
    type: string
    x-enum-varnames:
    - RunModeTypeAuto
    - RunModeTypeManual
  github_com_leptonai_gpud_api_v1.SuggestedActions:
    properties:
      description:
        description: Description describes the issue in detail.
        type: string
      repair_actions:
        description: A list of repair actions to mitigate the issue.
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_api_v1.RepairActionType'
        type: array
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.JSONPath:
    properties:
      expect:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.MatchRule'
        description: |-
          Expect defines the expected field "value" match rule.

          It not set, the field value is not checked,
          which means "missing field" for this query does not
          make the health state to be "Unhealthy".

          If set, the field value must be matched for this rule.
          In such case, the "missing field" or "mismatch" make
          the health state to be "Unhealthy".
      field:
        description: |-
          Field defines the field name to use in the extra_info data
          for this JSON path query output.
        type: string
      query:
        description: |-
          Query defines the JSONPath query path to extract with.
          ref. https://pkg.go.dev/github.com/PaesslerAG/jsonpath#section-readme
          ref. https://en.wikipedia.org/wiki/JSONPath
          ref. https://goessner.net/articles/JsonPath/
        type: string
      suggested_actions:
        additionalProperties:
          $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.MatchRule'
        description: |-
          SuggestedActions maps from the suggested action name,
          to the match rule for the field value.

          If the field value matches the rule,
          the health state reports the corresponding
          suggested action (the key of the matching rule).
        type: object
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.MatchRule:
    properties:
      regex:
        description: Regex is the regex to match the output.
        type: string
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.Plugin:
    properties:
      parser:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.PluginOutputParseConfig'
        description: |-
          Parser is the parser for the plugin output.
          If not set, the default prefix parser is used.
      steps:
        description: |-
          Steps is a sequence of steps to run for this plugin.
          Multiple steps are executed in order.
          If a step fails, the execution stops and the error is returned.
          Which means, the final success requires all steps to succeed.
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.Step'
        type: array
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.PluginOutputParseConfig:
    properties:
      json_paths:
        description: |-
          JSONPaths is a list of JSON paths to the output fields.
          Each entry has a FieldName (the output field name you want to assign e.g. "name")
          and a QueryPath (the JSON path you want to extract with e.g. "$.name").
        items:
          $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.JSONPath'
        type: array
      log_path:
        description: |-
          LogPath is an optional path to a file where the plugin output will be logged.
          If set, the raw plugin output will be appended to this file.
        type: string
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.RunBashScript:
    properties:
      content_type:
        description: |-
          ContentType is the content encode type of the script.
          Possible values: "plaintext", "base64".
        type: string
      script:
        description: |-
          Script is the script to run for this job.
          Assumed to be base64 encoded.
        type: string
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.Spec:
    properties:
      component_list:
        description: |-
          ComponentList is a list of component names for SpecTypeComponentList.
          Each item can be a simple name or "name:param" format.
          For component list, tags can be specified in the format "name#run_mode[tag1,tag2]:param"
        items:
          type: string
        type: array
      component_list_file:
        description: |-
          ComponentListFile is a path to a file containing component names for SpecTypeComponentList.
          Each line can be a simple name or "name:param" format.
          For component list file, tags can be specified in the format "name#run_mode[tag1,tag2]:param"
        type: string
      health_state_plugin:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.Plugin'
        description: |-
          HealthStatePlugin defines the plugin instructions
          to evaluate the health state of this plugin,
          which is translated into an GPUd /states API response.
      interval:
        allOf:
        - $ref: '#/definitions/v1.Duration'
        description: |-
          Interval is the interval for the script execution.
          For init plugin that only runs once at the server start,
          this value is ignored.
          Similarly, if set to zero, it runs only once.
      plugin_name:
        description: |-
          PluginName describes the plugin.
          It is used for generating the component name.
        type: string
      plugin_type:
        description: |-
          PluginType defines the plugin type.
          Possible values: "init", "component".
        type: string
      run_mode:
        description: |-
          RunMode defines the run mode of the plugin.
          Possible values: "auto", "manual".

          RunMode is set to "auto" to run the plugin periodically, with the specified interval.

          RunMode is set to "manual" to run the plugin only when explicitly triggered.
          The manual mode plugin is only registered but not run periodically.
          - GPUd does not run this even once.
          - GPUd does not run this periodically.

          This "auto" mode is only applicable to "component" type plugins.
          This "auto" mode is not applicable to "init" type plugins.

          The "init" type plugins are always run only once.
          This "manual" mode is only applicable to "component" type plugins.
          This "manual" mode is not applicable to "init" type plugins.
        type: string
      tags:
        description: |-
          Tags is a list of tags associated with this component.
          Tags can be used to group and trigger components together.
          For component list type, tags can also be specified in the run mode format.
        items:
          type: string
        type: array
      timeout:
        allOf:
        - $ref: '#/definitions/v1.Duration'
        description: |-
          Timeout is the timeout for the script execution.
          If zero, it uses the default timeout (1-minute).
    type: object
  github_com_leptonai_gpud_pkg_custom-plugins.Step:
    properties:
      name:
        description: Name is the name of the step.
        type: string
      run_bash_script:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.RunBashScript'
        description: RunBashScript is the bash script to run for this step.
    type: object
  github_com_leptonai_gpud_pkg_fault-injector.Request:
    properties:
      kernel_message:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_kmsg_writer.KernelMessage'
        description: KernelMessage is the kernel message to inject.
      xid:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_fault-injector.XIDToInject'
        description: XID is the XID to inject.
    type: object
  github_com_leptonai_gpud_pkg_fault-injector.XIDToInject:
    properties:
      id:
        type: integer
    type: object
  github_com_leptonai_gpud_pkg_kmsg_writer.KernelMessage:
    properties:
      message:
        description: Message is the message of the kernel message.
        type: string
      priority:
        allOf:
        - $ref: '#/definitions/github_com_leptonai_gpud_pkg_kmsg_writer.KernelMessagePriority'
        description: |-
          Priority is the priority of the kernel message.
          ref. https://github.com/torvalds/linux/blob/master/tools/include/linux/kern_levels.h#L8-L15
    type: object
  github_com_leptonai_gpud_pkg_kmsg_writer.KernelMessagePriority:
    enum:
    - KERN_EMERG
    - KERN_ALERT
    - KERN_CRIT
    - KERN_ERR
    - KERN_WARNING
    - KERN_NOTICE
    - KERN_INFO
    - KERN_DEBUG
    - KERN_DEFAULT
    type: string
    x-enum-varnames:
    - KernelMessagePriorityEmerg
    - KernelMessagePriorityAlert
    - KernelMessagePriorityCrit
    - KernelMessagePriorityError
    - KernelMessagePriorityWarning
    - KernelMessagePriorityNotice
    - KernelMessagePriorityInfo
    - KernelMessagePriorityDebug
    - KernelMessagePriorityDefault
  pkg_server.Healthz:
    properties:
      status:
        type: string
      version:
        type: string
    type: object
  pkg_server.SetHealthyStatesResponse:
    properties:
      code:
        type: integer
      failed:
        additionalProperties:
          type: string
        type: object
      message:
        type: string
      successful:
        items:
          type: string
        type: array
    type: object
  v1.Duration:
    properties:
      time.Duration:
        enum:
        - -9223372036854775808
        - 9223372036854775807
        - 1
        - 1000
        - 1000000
        - 1000000000
        - 60000000000
        - 3600000000000
        format: int64
        type: integer
        x-enum-varnames:
        - minDuration
        - maxDuration
        - Nanosecond
        - Microsecond
        - Millisecond
        - Second
        - Minute
        - Hour
    type: object
host: localhost:15132
info:
  contact:
    email: support@swagger.io
    name: API Support
    url: http://www.swagger.io/support
  description: GPU monitoring and management daemon API
  license:
    name: Apache 2.0
    url: http://www.apache.org/licenses/LICENSE-2.0.html
  termsOfService: http://swagger.io/terms/
  title: GPUd API
  version: "1.0"
paths:
  /healthz:
    get:
      consumes:
      - application/json
      description: Returns the health status of the gpud service
      operationId: healthz
      produces:
      - application/json
      responses:
        "200":
          description: Health status
          schema:
            $ref: '#/definitions/pkg_server.Healthz'
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Health check endpoint
      tags:
      - health
  /inject-fault:
    post:
      consumes:
      - application/json
      description: Injects a fault (such as kernel messages) into the system for testing
        purposes
      operationId: injectFault
      parameters:
      - description: Fault injection request
        in: body
        name: request
        required: true
        schema:
          $ref: '#/definitions/github_com_leptonai_gpud_pkg_fault-injector.Request'
      produces:
      - application/json
      responses:
        "200":
          description: Fault injected successfully
          schema:
            additionalProperties:
              type: string
            type: object
        "400":
          description: Bad request - invalid request body or validation error
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Fault injector not set up
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Inject fault into the system
      tags:
      - fault-injection
  /machine-info:
    get:
      description: Returns detailed information about the machine including hardware
        specifications
      operationId: getMachineInfo
      produces:
      - application/json
      responses:
        "200":
          description: Machine information
          schema:
            $ref: '#/definitions/github_com_leptonai_gpud_api_v1.MachineInfo'
        "404":
          description: GPUd instance not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get machine information
      tags:
      - machine
  /v1/components:
    delete:
      consumes:
      - application/json
      description: Deregisters a component from the system if it supports deregistration.
        Only components that implement the Deregisterable interface can be deregistered.
      operationId: deregisterComponent
      parameters:
      - description: Name of the component to deregister
        in: query
        name: componentName
        required: true
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Component deregistered successfully
          schema:
            additionalProperties: true
            type: object
        "400":
          description: Bad request - component name required or component not deregisterable
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error - failed to close component
          schema:
            additionalProperties: true
            type: object
      summary: Deregister a component
      tags:
      - components
    get:
      consumes:
      - application/json
      description: Returns a list of all currently registered gpud components in the
        system
      operationId: getComponents
      parameters:
      - description: Content type preference
        enum:
        - application/json
        - application/yaml
        in: header
        name: Accept
        type: string
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: List of component names
          schema:
            items:
              type: string
            type: array
        "400":
          description: Bad request - invalid content type
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get list of registered components
      tags:
      - components
  /v1/components/trigger-check:
    get:
      consumes:
      - application/json
      description: Triggers a health check for a specific component or all components
        with a specific tag. Either componentName or tagName must be provided, but
        not both.
      operationId: triggerComponentCheck
      parameters:
      - description: Name of the specific component to check (mutually exclusive with
          tagName)
        in: query
        name: componentName
        type: string
      - description: Tag name to check all components with this tag (mutually exclusive
          with componentName)
        in: query
        name: tagName
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Health check results with component states
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentHealthStates'
            type: array
        "400":
          description: Bad request - component or tag name required (but not both)
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
      summary: Trigger component health check
      tags:
      - components
  /v1/components/trigger-tag:
    get:
      consumes:
      - application/json
      description: Triggers health checks for all components that have the specified
        tag. Returns a summary of triggered components and their overall status.
      operationId: triggerComponentsByTag
      parameters:
      - description: Tag name to trigger all components with this tag
        in: query
        name: tagName
        required: true
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Trigger results with components list, exit status, and success
            flag
          schema:
            additionalProperties: true
            type: object
        "400":
          description: Bad request - tag name required
          schema:
            additionalProperties: true
            type: object
      summary: Trigger components by tag
      tags:
      - components
  /v1/events:
    get:
      consumes:
      - application/json
      description: Returns events from specified components within a time range. If
        no components specified, returns events from all components. Only supported
        components are queried.
      operationId: getEvents
      parameters:
      - description: Content type preference
        enum:
        - application/json
        - application/yaml
        in: header
        name: Accept
        type: string
      - description: Comma-separated list of component names to query (if empty, queries
          all components)
        in: query
        name: components
        type: string
      - description: Start time for event query (RFC3339 format, defaults to current
          time)
        in: query
        name: startTime
        type: string
      - description: End time for event query (RFC3339 format, defaults to current
          time)
        in: query
        name: endTime
        type: string
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Component events within the specified time range
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentEvents'
            type: array
        "400":
          description: Bad request - invalid content type, component parsing error,
            or time parsing error
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get component events
      tags:
      - components
  /v1/health-states/set-healthy:
    post:
      consumes:
      - application/json
      description: Sets specified components to healthy state if they implement the
        HealthSettable interface. If no components specified, attempts to set all
        components to healthy.
      operationId: setHealthyStates
      parameters:
      - description: Comma-separated list of component names to set healthy (if empty,
          sets all components)
        in: query
        name: components
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Components successfully set to healthy state
          schema:
            $ref: '#/definitions/pkg_server.SetHealthyStatesResponse'
        "400":
          description: Bad request - component does not support setting healthy state
            or failed to parse components
          schema:
            $ref: '#/definitions/pkg_server.SetHealthyStatesResponse'
        "404":
          description: Component not found
          schema:
            $ref: '#/definitions/pkg_server.SetHealthyStatesResponse'
      summary: Set components to healthy state
      tags:
      - components
  /v1/info:
    get:
      consumes:
      - application/json
      description: Returns comprehensive information including events, states, and
        metrics for specified components. If no components specified, returns information
        for all components. Only supported components are included.
      operationId: getInfo
      parameters:
      - description: Content type preference
        enum:
        - application/json
        - application/yaml
        in: header
        name: Accept
        type: string
      - description: Comma-separated list of component names to query (if empty, queries
          all components)
        in: query
        name: components
        type: string
      - description: Start time for query (RFC3339 format, defaults to current time)
        in: query
        name: startTime
        type: string
      - description: End time for query (RFC3339 format, defaults to current time)
        in: query
        name: endTime
        type: string
      - description: Duration string for metrics query (e.g., '30m', '1h') - defaults
          to 30 minutes
        in: query
        name: since
        type: string
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Component information including events, states, and metrics
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentInfo'
            type: array
        "400":
          description: Bad request - invalid content type, component parsing error,
            time parsing error, or duration parsing error
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get comprehensive component information
      tags:
      - components
  /v1/metrics:
    get:
      consumes:
      - application/json
      description: Returns metrics data for specified components within a time range.
        If no components specified, returns metrics for all components. Metrics are
        queried from the last 30 minutes by default.
      operationId: getMetrics
      parameters:
      - description: Content type preference
        enum:
        - application/json
        - application/yaml
        in: header
        name: Accept
        type: string
      - description: Comma-separated list of component names to query (if empty, queries
          all components)
        in: query
        name: components
        type: string
      - description: Duration string for metrics query (e.g., '30m', '1h') - defaults
          to 30 minutes
        in: query
        name: since
        type: string
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Component metrics data within the specified time range
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentMetrics'
            type: array
        "400":
          description: Bad request - invalid content type, component parsing error,
            or duration parsing error
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error - failed to read metrics
          schema:
            additionalProperties: true
            type: object
      summary: Get component metrics
      tags:
      - components
  /v1/plugins:
    get:
      consumes:
      - application/json
      description: Returns a list of all custom plugin specifications registered in
        the system
      operationId: getPluginSpecs
      parameters:
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: List of custom plugin specifications
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_pkg_custom-plugins.Spec'
            type: array
        "400":
          description: Bad request - invalid content type
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get custom plugin specifications
      tags:
      - plugins
  /v1/states:
    get:
      consumes:
      - application/json
      description: Returns the current health states of specified components or all
        components if none specified. Only supported components are included in the
        response.
      operationId: getHealthStates
      parameters:
      - description: Content type preference
        enum:
        - application/json
        - application/yaml
        in: header
        name: Accept
        type: string
      - description: Comma-separated list of component names to query (if empty, returns
          all components)
        in: query
        name: components
        type: string
      - description: Set to 'true' for indented JSON output
        in: header
        name: json-indent
        type: string
      produces:
      - application/json
      responses:
        "200":
          description: Component health states
          schema:
            items:
              $ref: '#/definitions/github_com_leptonai_gpud_api_v1.ComponentHealthStates'
            type: array
        "400":
          description: Bad request - invalid content type or component parsing error
          schema:
            additionalProperties: true
            type: object
        "404":
          description: Component not found
          schema:
            additionalProperties: true
            type: object
        "500":
          description: Internal server error
          schema:
            additionalProperties: true
            type: object
      summary: Get component health states
      tags:
      - components
swagger: "2.0"