scheduler

type Config struct {
	// FarmID is the farm this scheduler instance manages. The assignment loop
	// only considers tasks that belong to this farm.
	FarmID string

	// AssignInterval is how often the assignment loop polls the store for ready
	// tasks. Lower values reduce latency; higher values reduce DB pressure.
	// Default: 1 s.
	AssignInterval time.Duration

	// AssignBatchSize is the maximum number of ready tasks fetched per
	// assignment loop tick. Limits burst DB load without capping throughput
	// over time. Default: 50.
	AssignBatchSize int

	// AssignWorkers is the size of the goroutine pool that processes assignment
	// decisions concurrently. Default: 4.
	AssignWorkers int

	// WorkerTimeout is the maximum time since a worker's last heartbeat before
	// the heartbeat sweep considers it offline. Default: 30 s.
	WorkerTimeout time.Duration

	// HeartbeatSweepInterval is how often the heartbeat sweep runs. Should be
	// well below WorkerTimeout so stale workers are caught promptly. Default: 15 s.
	HeartbeatSweepInterval time.Duration

	// AssignedTaskTimeout is the maximum time a task may sit in 'assigned'
	// without transitioning to 'running' before the reaper returns it to the
	// ready queue. It guards the brief leased→running window: a task a worker
	// leased but never reported running (e.g. the worker crashed between lease
	// and process start). The heartbeat sweep cannot recover such tasks while the
	// worker is still heartbeating, so this reaper runs independently on the
	// heartbeat sweep tick. Default: 30 s.
	AssignedTaskTimeout time.Duration

	// OfflineWorkerRetention is how long a worker may remain in
	// [store.WorkerStatusOffline] before the retention sweep hard-deletes it,
	// bounding the growth of the worker table on farms with ephemeral nodes.
	// Disabled and online workers are never auto-removed. A value <= 0 disables
	// the sweep entirely. Default: 24 h.
	OfflineWorkerRetention time.Duration

	// JobRetention is how long a terminal job is kept before the retention
	// sweep hard-deletes it and all of its data (steps, tasks, attempts, logs).
	// completed and canceled jobs are always eligible; failed jobs only when
	// JobRetentionIncludeFailed is set. A value <= 0 disables the sweep.
	// Default: 168 h (7 days).
	JobRetention time.Duration

	// JobRetentionIncludeFailed extends the retention sweep to failed jobs.
	// Default: false (failed jobs are kept for debugging).
	JobRetentionIncludeFailed bool
}

type HeartbeatMsg struct {
	WorkerID string    `json:"worker_id"`
	At       time.Time `json:"at"`
}

type RegisterMsg struct {
	// WorkerID is the stable unique identifier for this worker instance.
	// Workers MUST use the same ID across restarts so that re-registration
	// updates the existing record rather than creating a duplicate.
	WorkerID string `json:"worker_id"`

	// FarmID is the farm this worker belongs to. Required.
	FarmID string `json:"farm_id"`

	// QueueID restricts the worker to a single queue if non-empty.
	// An empty value means the worker accepts tasks from any queue in FarmID.
	QueueID string `json:"queue_id,omitempty"`

	// Name is the worker's human-readable display label. Persisted so the UI
	// can distinguish multiple workers running on a single host.
	Name string `json:"name,omitempty"`

	// Hostname and IPAddress are the worker's network identity.
	Hostname  string `json:"hostname"`
	IPAddress string `json:"ip_address,omitempty"`

	// ComputeLocation is the named compute location this worker belongs to
	// (e.g. "onprem_linux", "cloud_aws_us_east"). Used for path translation
	// and task affinity.
	ComputeLocation string `json:"compute_location,omitempty"`

	// OS and OSVersion are the worker's operating system identity, used for
	// capability matching.
	OS        string `json:"os"`
	OSVersion string `json:"os_version,omitempty"`

	// WorkerVersion is the sqi-worker build version the worker self-reports,
	// distinct from the protocol Version field above.
	WorkerVersion string `json:"worker_version,omitempty"`

	// CPUCount and RAMMb are the worker's hardware capacity, used for
	// resource-aware scheduling in future phases.
	CPUCount int `json:"cpu_count,omitempty"`
	RAMMb    int `json:"ram_mb,omitempty"`

	// GPUInfo describes any GPU(s) installed on the worker host.
	// omitempty is omitted intentionally: it has no effect on struct fields.
	GPUInfo store.GPUInfo `json:"gpu_info"`

	// Tags holds arbitrary key/value capability tags the worker self-reports.
	// Job requirements reference these tags when specifying worker constraints.
	Tags map[string]string `json:"tags,omitempty"`
}

type Scheduler struct {
	// contains filtered or unexported fields
}