Regressions
Automatic statistical detection of duration spikes, cost anomalies, and queue-time degradation in your workflows.
Gitera continuously compares each workflow's recent behavior with its own baseline and flags meaningful degradation — no manual threshold tuning per workflow.
How detection works
- A 3-day measurement window is compared against a 14-day rolling baseline using p50 and p95 percentiles.
- Warning: ≥ 25% deviation. Critical: ≥ 50% deviation, with minimum absolute floors so tiny workflows do not create noise.
Regression types
| Type | Detects |
|---|---|
| Duration | Workflows that got significantly slower |
| Cost | Runs consuming abnormally more compute minutes |
| Queue time | Runner starvation — jobs waiting much longer than usual |
Lifecycle
Each regression moves through Active → Resolved or Dismissed. Dismissed regressions stay quiet for a 7-day cooldown so the same known-slow pipeline does not keep alerting.
What to do with a regression
- Open it to see the baseline vs current comparison.
- Jump into log search or the workflow detail to find the slow step.
- Resolve it when the pipeline is back to normal — Gitera also closes it automatically when metrics recover.