System metrics

Metrics() *SystemMetricsHandle

Selects installation-wide metrics without I/O. The selectors below return *MetricHandle. Call Latest or History to read retained observations. There is no system-wide Snapshot method.

Example

Using an initialized client, read the last completed collector pass:

measurement, err := client.System().Metrics().CollectorCompletedTimestamp().Latest(ctx)
if err != nil {
  return err
}
if measurement != nil {
  fmt.Println(time.Unix(int64(measurement.Value), 0))
}

Worker series

All are gauges without attributes. Values include workers owned by the system, streams, and consumer groups.

Selector	Wire name	Code	Unit	Value
`UnclaimedWorkers()`	`sqlstreams.worker.state.unclaimed_workers`	`SQL0067`	`{worker}`	Workers with no live instance and a nonzero target.
`OldestUnclaimedAge()`	`sqlstreams.worker.state.oldest_unclaimed_age`	`SQL0068`	`ms`	Largest elapsed time since instance expiry among unclaimed workers. `0` if no retained expired instance establishes that age.
`FailingWorkers()`	`sqlstreams.worker.state.failing_workers`	`SQL0069`	`{worker}`	Workers with a live instance on a nonzero consecutive failure streak.

Suspended workers are excluded from unclaimed counts. See worker liveness for the alert contract.

Schedule series

All are gauges without attributes.

Selector	Wire name	Code	Unit	Value
`OverdueSchedules()`	`sqlstreams.schedule.state.overdue`	`SQL0070`	`{found}`	Unsuspended schedules more than 10 minutes past their next scheduled production time.
`OldestDueAge()`	`sqlstreams.schedule.state.oldest_due_age`	`SQL0071`	`ms`	Largest elapsed time past next production among unsuspended schedules. `0` if none are due.
`SuspendedSchedules()`	`sqlstreams.schedule.state.suspended`	`SQL0072`	`{found}`	Suspended schedules, excluded from overdue counts.

Alert series

All are gauges. Counts of active and resolved alerts have no attributes. They describe retained alert heads, not lifetime totals.

Selector	Wire name	Code	Unit	Value
`ActiveAlerts()`	`sqlstreams.alert.state.active_alerts`	`SQL0073`	`{alert}`	Active retained alert heads.
`ResolvedAlerts()`	`sqlstreams.alert.state.resolved_alerts`	`SQL0074`	`{alert}`	Resolved retained alert heads.
`CheckStreamsEvaluated(alert string)`	`sqlstreams.alert.check.streams_evaluated`	`SQL0075`	`{stream}`	Streams evaluated in the check’s latest recorded pass.
`CheckStreamsFailed(alert string)`	`sqlstreams.alert.check.streams_failed`	`SQL0076`	`{stream}`	Streams with evaluation or recording errors, or insufficient evidence.
`CheckPublishedAlerts(alert string)`	`sqlstreams.alert.check.published_alerts`	`SQL0077`	`{alert}`	Active reports written in that pass, including reminders.
`CheckResolvedAlerts(alert string)`	`sqlstreams.alert.check.resolved_alerts`	`SQL0078`	`{alert}`	Alerts changed to resolved in that pass.

The four check selectors require the alert attribute. Supported checks use partition_count, compaction_read_cost, or worker_liveness. Their values are per-pass counts, not cumulative counters.

Collector completion

Selector	Wire name	Code	Kind	Unit
`CollectorCompletedTimestamp()`	`sqlstreams.metrics.collector.completed_timestamp`	`SQL0100`	Gauge	`s`

The value is a Unix timestamp in seconds, without attributes. It advances only after every source read and measurement write in a collector pass succeeds. A partial pass can update other measurements without advancing this value.

The collector runs every 30s by default, with jitter. Alert-check series are recorded by their own checks. See collection and retention for shared timing and freshness rules.

CLI

sqlstreams metric latest sqlstreams.metrics.collector.completed_timestamp

Reads the retained completion series. An absent series exits with status 1. Use metric history for earlier observations and explain for definitions.

Read metric measurements — Returns, errors, and series identity.
Register the system — Collector and alert configuration.
Collector progress alert — Detect missing completed collections.