components

type CheckResult interface {
	// ComponentName returns the name of the component that produced this check result.
	ComponentName() string

	// String returns a string representation of the data.
	// Describes the data in a human-readable format.
	fmt.Stringer

	// Summary returns a summary of the check result.
	Summary() string

	// HealthStateType returns the health state of the last check result.
	HealthStateType() apiv1.HealthStateType
	// HealthStates returns the health states of the last check result.
	HealthStates() apiv1.HealthStates
}

type CheckResultDebugger interface {
	// Debug returns a string representation of the check result.
	Debug() string
}

type Component interface {
	// Defines the component name,
	// and used for the HTTP handler registration path.
	// Must be globally unique.
	Name() string

	// Tags returns a list of tags that describe the component.
	//
	// The component tags are static, and will not change over time.
	// It is important to keep in mind that, the tags only represent
	// the component's functionality, but not the component's health state.
	//
	// This is useful to trigger on-demand component checks based on specific tags.
	//
	// e.g.,
	// GPU enabled components may return its accelerator manufacturer,
	// but does not report its health state via the tags.
	Tags() []string

	// IsSupported returns true if the component is supported on the current machine.
	// For example, this returns "false" if a component requires NVIDIA GPUs,
	// but the machine does not have NVIDIA GPUs.
	IsSupported() bool

	// Start called upon server start.
	// Implements component-specific poller start logic.
	Start() error

	// Check triggers the component check once, and returns the latest health check result.
	// This is used for one-time checks, such as "gpud scan".
	// It is up to the component to decide the check timeouts.
	// Thus, we do not pass the context here.
	// The CheckResult should embed the timeout errors if any, via Summary and HealthState.
	Check() CheckResult

	// LastHealthStates reads the latest health states of the component,
	// cached from its periodic checks.
	// If no check has been performed, it returns a single health state of apiv1.StateTypeHealthy.
	LastHealthStates() apiv1.HealthStates

	// Events returns all the events from "since".
	Events(ctx context.Context, since time.Time) (apiv1.Events, error)

	// Called upon server close.
	// Implements copmonent-specific poller cleanup logic.
	Close() error
}

type Deregisterable interface {
	// CanDeregister returns true if the custom plugin can be deregistered.
	CanDeregister() bool
}

type GPUdInstance struct {
	RootCtx context.Context

	// MachineID is either the machine ID assigned from the control plane
	// or the unique UUID of the machine.
	// For example, it is used to identify itself for the NFS checker.
	MachineID string

	KernelModulesToCheck []string

	NVMLInstance         nvidianvml.Instance
	NVIDIAToolOverwrites nvidiacommon.ToolOverwrites

	DBRW *sql.DB
	DBRO *sql.DB

	EventStore       eventstore.Store
	RebootEventStore pkghost.RebootEventStore

	MountPoints  []string
	MountTargets []string
}

type HealthSettable interface {
	// SetHealthy sets the health state to healthy.
	SetHealthy() error
}

type InitFunc func(*GPUdInstance) (Component, error)

type Registry interface {
	// MustRegister registers a component with the given initialization function.
	// It panics if the component is already registered.
	// It panics if the initialization function returns an error.
	MustRegister(initFunc InitFunc)

	// Register registers a component with the given initialization function.
	// It returns an error if the component is already registered.
	// It returns an error if the initialization function returns an error.
	Register(initFunc InitFunc) (Component, error)

	// All returns all registered components.
	All() []Component

	// Get returns a component by name.
	// It returns nil if the component is not registered.
	Get(name string) Component

	// Deregister deregisters a component by name, and returns the
	// underlying component if it is registered.
	// It returns nil if the component is not registered.
	// Meaning, it is safe to call it multiple times,
	// and it is also safe to call it with a non-registered name.
	Deregister(name string) Component
}