Performance Tab
The Performance tab of the Workflow Detail page breaks down where time is spent across jobs and individual steps, helping you identify the highest-impact optimization targets.
Job Duration
A horizontal bar chart showing each job's average duration (purple) vs its p95 duration (blue), displayed in minutes. The gap between average and p95 is the “tail latency” — a large gap means some runs of that job are dramatically slower than usual.
| Column / Series | Definition |
|---|---|
| Avg | Mean duration of that job across all loaded runs. |
| p95 | 95th-percentile duration — 95% of runs of that job finish within this time. A rising p95 is a stronger signal of regression than a rising average. |
| Gap (avg → p95) | Large gaps indicate non-deterministic jobs: slow on some runs, fast on others. Common causes: cache misses, test flakiness, or shared infrastructure contention. |
Job Composition per Run
A stacked bar chart showing the last 20 runs on the X-axis and total run duration on the Y-axis. Each bar is segmented by job, colour-coded consistently. This reveals which job dominates total run time and whether that share is growing.
Slowest Steps
A ranked table of the top 10 individual step names by average runtime, aggregated across all jobs and loaded runs. Each row shows: step name, job context, run count, average, p95, max durations, and success %.
| Column | What it means |
|---|---|
| RUNS | Number of runs in which this step executed (denominator for averages). |
| AVG | Mean step duration in seconds across all runs. |
| P95 | 95th-percentile step duration. Use this to size timeouts and SLOs. |
| MAX | Worst observed duration for this step — a ceiling for worst-case pipeline time. |
| SUCCESS % | Percentage of step executions that completed successfully. Low values indicate a flaky step. |