metrics

var (
	Up = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_up",
		Help: "1 while the scitq server is running. Also a scrape-target liveness signal — if this stops updating for > 30s, the server (or the refresh goroutine) has stalled.",
	})

	Workers = promauto.NewGaugeVec(prometheus.GaugeOpts{
		Name: "scitq_workers",
		Help: "Number of workers, labelled by status (R/O/I/…), provider, and whether they're permanent.",
	}, []string{"status", "provider", "permanent"})

	WorkerIdleSeconds = promauto.NewGaugeVec(prometheus.GaugeOpts{
		Name: "scitq_worker_idle_seconds",
		Help: "Seconds since this worker last finished a task (or, for born-idle workers, since the watchdog started tracking it). Non-permanent workers reading above 2× scitq.idle_timeout are the direct 09-09 worker-6629 leak signature.",
	}, []string{"worker_id", "worker_name", "permanent"})

	WorkerActiveTasksDrift = promauto.NewGaugeVec(prometheus.GaugeOpts{
		Name: "scitq_worker_active_tasks_drift",
		Help: "DB active-task count (status IN A/C/D/O/R AND NOT hidden) minus watchdog in-memory active-task count. Steady non-zero means the watchdog memory is out of sync with reality — the drift that let worker 6629 leak. Should be 0 in steady state after the 2026-09-10 retryTaskInternal + ResyncActiveTasks fixes.",
	}, []string{"worker_id", "worker_name"})

	WorkerRunningTasks = promauto.NewGaugeVec(prometheus.GaugeOpts{
		Name: "scitq_worker_running_tasks",
		Help: "Number of tasks the WORKER CLIENT itself reports as currently executing (container/exec launched, upload not started). Runtime ground truth, independent of any server-side DB status bucket — a stuck hidden R task (2026-09-09 worker-6629 leak) cannot inflate this because the client's executingTasks map only tracks its own live launches. Pair with scitq_worker_idle_seconds to build a leak-proof alert: idle_seconds > threshold AND running_tasks == 0.",
	}, []string{"worker_id", "worker_name", "permanent"})

	Tasks = promauto.NewGaugeVec(prometheus.GaugeOpts{
		Name: "scitq_tasks",
		Help: "Number of tasks by status (excludes hidden — hidden rows are superseded retry-parents, not queue state).",
	}, []string{"status"})

	TaskPendingSecondsMax = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_task_pending_seconds_max",
		Help: "Age (in seconds) of the oldest task stuck in P/W. Grows when the queue is starving for workers.",
	})

	WorkflowsActive = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_workflow_active",
		Help: "Number of workflows in Running or Debug status.",
	})

	DeletionJobsStuck = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_deletion_jobs_stuck",
		Help: "Number of worker-deletion jobs (action=D) stuck in R for >10 min AND whose target worker is still alive (worker.deleted_at IS NULL). Nonzero means a cloud-side delete is genuinely hanging — orphan VM risk. Delete jobs whose worker is already gone are excluded: they're bookkeeping drift (typically a server restart between the cloud API returning and the status-write), not an operational issue.",
	})

	DBConnectionsOpen = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_db_connections_open",
		Help: "Established connections in the DB pool (sql.DB.Stats().OpenConnections).",
	})

	DBConnectionsInUse = promauto.NewGauge(prometheus.GaugeOpts{
		Name: "scitq_db_connections_in_use",
		Help: "DB connections currently checked out of the pool (sql.DB.Stats().InUse). Sustained near the pool limit means the server is DB-bound.",
	})

	RecruiterLaunches = promauto.NewCounterVec(prometheus.CounterOpts{
		Name: "scitq_recruiter_launches_total",
		Help: "Cumulative worker deployments the recruiter has attempted, labelled by outcome (success/quota/error/blacklisted).",
	}, []string{"provider", "region", "outcome"})

	WatchdogReclaims = promauto.NewCounterVec(prometheus.CounterOpts{
		Name: "scitq_watchdog_reclaims_total",
		Help: "Cumulative times the watchdog has reset A/C/D/O tasks back to P because their worker went offline.",
	}, []string{"reason"})
)