Metrics

Metrics are designed for diagnosis without putting secrets or unbounded request text into labels.

Stream termination

cc_lb_stream_terminations_total{outcome,cause} counts response body outcomes. Use it with request correlation logs to distinguish an upstream error from a client cancellation.

Scheduler metrics

• cclb_scheduler_jobs_total tracks registered job lifecycle events.
• cclb_scheduler_job_duration_seconds records handler duration by job type.
• cclb_scheduler_failures_total counts terminal job failures.
• cclb_scheduler_init_failure reports scheduler initialization state.
• cclb_scheduler_prune_rows_removed_total counts bounded cleanup results.

Job labels are bounded to registered types and lifecycle statuses. Scheduler failure views expose sanitized reasons and hashed summaries rather than prompts or credentials.

Routing and runtime diagnosis

Routing traces record the selected tier, formula version, warning multiplier, and bounded selection reason. Runtime status includes the last apply error and timestamps for operational actions. Combine these views when a resource update is accepted but traffic still follows an earlier view.