Control Plane Metrics

Control Plane metrics are especially useful when the problem is not SQL execution itself, but the resource lifecycle: database creation, autostart, auto-stop, and the new writer autoscaler.

Key metrics

cp_up

Shows whether the Control Plane itself is alive.

cp_autostart_total

Counts autostart attempts. It is useful when you want to understand how often sleeping databases are actually woken by the first connection and whether errors are increasing.

cp_cold_start_latency_seconds

Shows cold-start duration. This is one of the most useful metrics when investigating complaints like “the database wakes up too slowly after idle.”

cp_start_attempts_total

Shows the total number of startup attempts and their outcome.

cp_idle_autostop_total

Lets you see how often auto-stop is triggered and whether there are failures on that path.

cp_provisioner_http_requests_total

Helps separate a Control Plane issue from a Provisioner-call issue.

cp_tokens_invalid_total

Useful when investigating worker registration problems and related lifecycle failures.

cp_lease_expired_total

Highlights situations where a lease expires abnormally.

spg99_autoscale_failed_total

One of the most important new signals: it shows at which stage the autoscale handoff failed.

How to read these metrics in practice

Complaints about a slow first connect

Look at these together:

• cp_autostart_total
• cp_cold_start_latency_seconds
• cp_start_attempts_total

This quickly shows whether there is a real cold-start problem or whether the application simply does not account for normal serverless delay.

The database takes a long time to reach ready

Useful signals here are:

• cp_start_attempts_total
• cp_provisioner_http_requests_total
• the overall database state in Console

A profile handoff is stuck or failed

Look at:

• spg99_autoscale_failed_total
• scale_state in the describe API
• Gateway metrics for freeze/drain/pinned behavior
• candidate writer readiness

The database does not go back to sleep

Look at:

• cp_idle_autostop_total
• Gateway lease metrics and active connections
• whether the database is stuck in COOLDOWN or another autoscale state

Practical conclusion

If Gateway answers the question “what the client sees at the entry point,” then Control Plane metrics answer best the question “what is happening to the database lifecycle and the writer handoff inside the platform.”