descrybe/apps/api/internal/httpapi/health.go

package httpapi

import (
	"context"
	"net/http"
	"time"

	"github.com/descrybe/descrybe-v2/apps/api/internal/jobs"
)

const healthServiceName = "api"

// dbPinger is satisfied by *pgxpool.Pool; kept narrow for unit tests.
type dbPinger interface {
	Ping(ctx context.Context) error
}

func (s *Server) handleHealthz(w http.ResponseWriter, _ *http.Request) {
	JSON(w, http.StatusOK, map[string]any{
		"status":      "ok",
		"service":     healthServiceName,
		"maintenance": s.Config.MaintenanceMode,
		"read_only":   s.Config.ReadOnlyMode,
		"hypercare":   s.Config.HypercareMode,
	})
}

func (s *Server) handleReadyz(w http.ResponseWriter, r *http.Request) {
	var pinger dbPinger
	var prober jobs.HeartbeatQuerier
	if s.Pool != nil {
		pinger = s.Pool
		prober = s.Pool
	}
	s.writeReadyz(w, r, pinger, prober)
}

func (s *Server) writeReadyz(w http.ResponseWriter, r *http.Request, pinger dbPinger, prober jobs.HeartbeatQuerier) {
	ctx, cancel := context.WithTimeout(r.Context(), 2*time.Second)
	defer cancel()

	ok, check, errMsg := databaseReady(ctx, pinger)
	checks := map[string]string{"database": check}
	body := map[string]any{
		"status":      "ready",
		"service":     healthServiceName,
		"maintenance": s.Config.MaintenanceMode,
		"read_only":   s.Config.ReadOnlyMode,
		"hypercare":   s.Config.HypercareMode,
		"checks":      checks,
	}
	if !ok {
		body["status"] = "not_ready"
		body["error"] = errMsg
		JSON(w, http.StatusServiceUnavailable, body)
		return
	}

	probe := jobs.ProbeWorkerReadiness(ctx, prober, jobs.DefaultHeartbeatStaleAfter)
	checks["worker"] = probe.WorkerCheck
	checks["queue"] = probe.QueueCheck
	body["queue_pending"] = probe.PendingJobs
	if probe.LastSeenAgeS >= 0 {
		body["worker_last_seen_age_s"] = probe.LastSeenAgeS
	}
	if !probe.OK {
		body["status"] = "not_ready"
		body["error"] = probe.ErrMsg
		if probe.Reason != "" {
			body["reason"] = probe.Reason
		}
		JSON(w, http.StatusServiceUnavailable, body)
		return
	}

	JSON(w, http.StatusOK, body)
}

// databaseReady pings Postgres for readiness. check is "ok", "unavailable", or "fail".
// errMsg is empty when ok; never includes driver detail (safe for public probes).
func databaseReady(ctx context.Context, p dbPinger) (ok bool, check string, errMsg string) {
	if p == nil {
		return false, "unavailable", "database pool unavailable"
	}
	if err := p.Ping(ctx); err != nil {
		return false, "fail", "database ping failed"
	}
	return true, "ok", ""
}