config

type AgentResourceManagerConfig struct {
	Scheduler                  *SchedulerConfig `json:"scheduler"`
	DefaultAuxResourcePool     string           `json:"default_aux_resource_pool"`
	DefaultComputeResourcePool string           `json:"default_compute_resource_pool"`
	NoDefaultResourcePools     bool             `json:"no_default_resource_pools"`
	// Deprecated: use DefaultAuxResourcePool instead.
	DefaultCPUResourcePool string `json:"default_cpu_resource_pool,omitempty"`
	// Deprecated: use DefaultComputeResourcePool instead.
	DefaultGPUResourcePool string `json:"default_gpu_resource_pool,omitempty"`

	RequireAuthentication bool   `json:"require_authentication"`
	ClientCA              string `json:"client_ca"`
}

type AssignWorkspaceCreatorConfig struct {
	Enabled bool `json:"enabled"`
	RoleID  int  `json:"role_id"`
}

type AuthZConfig struct {
	Type          string  `json:"type"`
	FallbackType  *string `json:"fallback"`
	RBACUIEnabled *bool   `json:"rbac_ui_enabled"`
	// Removed: this option is removed and will not have any effect.
	StrictNTSCEnabled      bool                         `json:"_strict_ntsc_enabled"`
	AssignWorkspaceCreator AssignWorkspaceCreatorConfig `json:"workspace_creator_assign_role"`
	StrictJobQueueControl  bool                         `json:"strict_job_queue_control"`
}

type CacheConfig struct {
	CacheDir string `json:"cache_dir"`
}

type Config struct {
	ConfigFile            string                            `json:"config_file"`
	Log                   logger.Config                     `json:"log"`
	DB                    DBConfig                          `json:"db"`
	TensorBoardTimeout    int                               `json:"tensorboard_timeout"`
	NotebookTimeout       *int                              `json:"notebook_timeout"`
	Security              SecurityConfig                    `json:"security"`
	CheckpointStorage     expconf.CheckpointStorageConfig   `json:"checkpoint_storage"`
	TaskContainerDefaults model.TaskContainerDefaultsConfig `json:"task_container_defaults"`
	Port                  int                               `json:"port"`
	Root                  string                            `json:"root"`
	Telemetry             config.TelemetryConfig            `json:"telemetry"`
	EnableCors            bool                              `json:"enable_cors"`
	LaunchError           bool                              `json:"launch_error"`
	ClusterName           string                            `json:"cluster_name"`
	Logging               model.LoggingConfig               `json:"logging"`
	Observability         ObservabilityConfig               `json:"observability"`
	Cache                 CacheConfig                       `json:"cache"`
	Webhooks              WebhooksConfig                    `json:"webhooks"`
	FeatureSwitches       []string                          `json:"feature_switches"`
	ReservedPorts         []int                             `json:"reserved_ports"`
	ResourceConfig

	// Internal contains "hidden" useful debugging configurations.
	InternalConfig InternalConfig `json:"__internal"`
}

type DBConfig struct {
	User        string `json:"user"`
	Password    string `json:"password"`
	Migrations  string `json:"migrations"`
	Host        string `json:"host"`
	Port        string `json:"port"`
	Name        string `json:"name"`
	SSLMode     string `json:"ssl_mode"`
	SSLRootCert string `json:"ssl_root_cert"`
}

type ExperimentConfigPatch struct {
	Name *string `json:"name,omitempty"`
}

type FairShareSchedulerConfig struct{}

type FluentConfig struct {
	Image string `json:"image"`
	UID   int    `json:"uid"`
	GID   int    `json:"gid"`
}

type InternalConfig struct {
	AuditLoggingEnabled bool                   `json:"audit_logging_enabled"`
	ExternalSessions    model.ExternalSessions `json:"external_sessions"`
	ProxiedServers      []ProxiedServerConfig  `json:"proxied_servers"`
}

type KubernetesResourceManagerConfig struct {
	Namespace string `json:"namespace"`

	// Deprecated: this can be per resource pool now on taskContainerDefaults.
	// This will always be the same as global
	// task_container_defaults.kubernetes.max_slots_per_pod so use that.
	MaxSlotsPerPod *int `json:"max_slots_per_pod"`

	MasterServiceName        string                  `json:"master_service_name"`
	LeaveKubernetesResources bool                    `json:"leave_kubernetes_resources"`
	DefaultScheduler         string                  `json:"default_scheduler"`
	SlotType                 device.Type             `json:"slot_type"`
	SlotResourceRequests     PodSlotResourceRequests `json:"slot_resource_requests"`
	// deprecated, no longer in use.
	Fluent     FluentConfig `json:"fluent"`
	CredsDir   string       `json:"_creds_dir,omitempty"`
	MasterIP   string       `json:"_master_ip,omitempty"`
	MasterPort int32        `json:"_master_port,omitempty"`

	DefaultAuxResourcePool     string `json:"default_aux_resource_pool"`
	DefaultComputeResourcePool string `json:"default_compute_resource_pool"`
	NoDefaultResourcePools     bool   `json:"no_default_resource_pools"`
}

type ObservabilityConfig struct {
	EnablePrometheus bool `json:"enable_prometheus"`
}

type PodSlotResourceRequests struct {
	CPU float32 `json:"cpu"`
}

type PrioritySchedulerConfig struct {
	Preemption      bool `json:"preemption"`
	DefaultPriority *int `json:"default_priority"`
}

type ProxiedServerConfig struct {
	// Prefix is the path prefix to match for this proxy.
	PathPrefix string `json:"path_prefix"`
	// Destination is the URL to proxy to.
	Destination string `json:"destination"`
}

type ResourceConfig struct {
	ResourceManager *ResourceManagerConfig `json:"resource_manager"`
	ResourcePools   []ResourcePoolConfig   `json:"resource_pools"`
}

type ResourceManagerConfig struct {
	AgentRM      *AgentResourceManagerConfig      `union:"type,agent" json:"-"`
	KubernetesRM *KubernetesResourceManagerConfig `union:"type,kubernetes" json:"-"`
}

type ResourcePoolConfig struct {
	PoolName                 string                             `json:"pool_name"`
	Description              string                             `json:"description"`
	Provider                 *provconfig.Config                 `json:"provider"`
	Scheduler                *SchedulerConfig                   `json:"scheduler,omitempty"`
	MaxAuxContainersPerAgent int                                `json:"max_aux_containers_per_agent"`
	TaskContainerDefaults    *model.TaskContainerDefaultsConfig `json:"task_container_defaults"`
	// AgentReattachEnabled defines if master & agent try to recover
	// running containers after a clean restart.
	AgentReattachEnabled bool `json:"agent_reattach_enabled"`
	// AgentReconnectWait define the time master will wait for agent
	// before abandoning it.
	AgentReconnectWait model.Duration `json:"agent_reconnect_wait"`

	// If empty, will behave as if the value is resource_manager.namespace,
	// which in most cases will be the namespace the helm deployment is in.
	KubernetesNamespace string `json:"kubernetes_namespace"`

	// Deprecated: Use MaxAuxContainersPerAgent instead.
	MaxCPUContainersPerAgent int `json:"max_cpu_containers_per_agent,omitempty"`
}

type RoundRobinSchedulerConfig struct{}

type SSHConfig struct {
	RsaKeySize int `json:"rsa_key_size"`
}

type SchedulerConfig struct {
	FairShare              *FairShareSchedulerConfig  `union:"type,fair_share" json:"-"`
	Priority               *PrioritySchedulerConfig   `union:"type,priority" json:"-"`
	RoundRobin             *RoundRobinSchedulerConfig `union:"type,round_robin" json:"-"`
	FittingPolicy          string                     `json:"fitting_policy"`
	AllowHeterogeneousFits bool                       `json:"allow_heterogeneous_fits"`
}

type SecurityConfig struct {
	DefaultTask model.AgentUserGroup `json:"default_task"`
	TLS         TLSConfig            `json:"tls"`
	SSH         SSHConfig            `json:"ssh"`
	AuthZ       AuthZConfig          `json:"authz"`
}

type TLSConfig struct {
	Cert string `json:"cert"`
	Key  string `json:"key"`
}

type WebhooksConfig struct {
	BaseURL    string `json:"base_url"`
	SigningKey string `json:"signing_key"`
}