Metrics

GET /metrics - Prometheus metrics for queue depth, backlog, delivery outcomes, and lateness.

GET /metrics

Returns Prometheus metrics in the text exposition format (version=0.0.4).

Unlike /healthz and /readyz, this endpoint sits behind the API key when SCHEDY_API_KEY is set - queue depth and backlog are operational detail.

curl http://localhost:8080/metrics -H "X-API-Key: your-secret"

What’s exported

Metric	Type	Meaning
schedy_tasks{status}	gauge	Tasks currently in each lifecycle status. Every status is exported, including zeroes.
schedy_tasks_overdue	gauge	Pending tasks whose execute_at has already passed.
schedy_deliveries_total{result}	counter	Delivery requests fired at task targets. Retries count individually.
schedy_tasks_finished_total{status}	counter	Tasks that reached a terminal delivery outcome, counted once each.
schedy_tasks_skipped_total{reason}	counter	Tasks retired without delivery for exceeding SCHEDY_MAX_STALENESS.
schedy_tasks_replayed_total	counter	Finished tasks manually re-armed via replay.
schedy_deliveries_inflight	gauge	Deliveries currently executing. Compare against SCHEDY_MAX_CONCURRENT_DELIVERIES to spot saturation.
schedy_delivery_duration_seconds	histogram	Round-trip time of delivery requests.
schedy_task_lateness_seconds	histogram	Delay between a task’s execute_at and the moment it fired.

# HELP schedy_tasks_overdue Pending tasks whose execute_at has already passed.
# TYPE schedy_tasks_overdue gauge
schedy_tasks_overdue 117

# HELP schedy_deliveries_total Delivery requests fired at task targets, by outcome. Retries count individually.
# TYPE schedy_deliveries_total counter
schedy_deliveries_total{result="success"} 1423
schedy_deliveries_total{result="failure"} 17

Counters reset when Schedy restarts, which is what Prometheus expects - rate() and increase() handle the reset for you. The schedy_tasks gauges are read from the store on every scrape, so they never drift from what is actually stored.

Go runtime metrics (go_goroutines, go_memstats_*) are not exported. Schedy emits the exposition format directly rather than depending on the Prometheus client library.

Scrape config

scrape_configs:
  - job_name: schedy
    static_configs:
      - targets: ["schedy:8080"]
    # SCHEDY_API_KEY travels as a plain header (Prometheus 2.50+)
    http_headers:
      X-API-Key:
        values: ["your-secret"]

Leave SCHEDY_API_KEY unset and the endpoint is open, like /healthz.

The two that matter

schedy_tasks_overdue is the signal that Schedy is not keeping up. A healthy instance holds it near zero: tasks fire within a runner tick of their scheduled time. A sustained non-zero value means work is arriving faster than it can be delivered, or a target is slow enough to hold the worker goroutines.

schedy_task_lateness_seconds is the same story as a distribution. Alert on its high quantiles rather than its mean - a p99 of several minutes with a healthy mean is a subset of targets timing out, which the failure counter alone will not tell you.

# Backlog that isn't draining
schedy_tasks_overdue > 0

# Delivery failure ratio over 5 minutes
rate(schedy_deliveries_total{result="failure"}[5m])
  / rate(schedy_deliveries_total[5m])

# p99 lateness
histogram_quantile(0.99, rate(schedy_task_lateness_seconds_bucket[5m]))

# Saturated: every delivery slot busy
schedy_deliveries_inflight >= 50

Time spent queued behind the concurrency cap counts as lateness, so saturation shows up in schedy_task_lateness_seconds rather than hiding behind it. See Catch-up.

The schedy_tasks gauges scan the key index on every scrape. That is one pass over task keys - no task bodies are read - which is fast into the millions, but worth knowing if you scrape aggressively with a very large store.