Diff Two Runs
Compare two workflow runs: inputs, options, WDL source, and per-task changes.
-
Full Comparison
Inputs, options, source, and task-level differences
-
Cache-aware
Recovers real metrics of call-cached tasks from their source runs
-
JSON Output
Machine-readable diff for scripting
Quick Start
Alias: pumbaa wf compare
Flags
| Flag | Description |
|---|---|
--json |
Output the diff as JSON |
--no-cache-resolve |
Do not fetch cache sources to recover real metrics of cache-hit tasks |
Output
The diff is organized in four sections:
| Section | What it shows |
|---|---|
| Inputs | Added (+), removed (-), and changed (~) input keys |
| Options | Same, for workflow options |
| Source | Whether the submitted WDL source changed (with line counts) |
| Tasks | Tasks only in one run, plus status, docker image, shard count, attempts, and duration changes |
Duration changes come with a verdict (e.g. 2.3× slower), making regressions easy to spot.
Call caching
When a task was a call-cache hit, its wall-clock time says nothing about real performance. By default, Pumbaa follows the cache back to the source run and compares the original metrics. Use --no-cache-resolve to skip this (faster, fewer API calls).
Examples
From the dashboard
The same comparison is available interactively: press C on two workflows in the Dashboard.
Use Cases
-
Performance Regressions
Why did this run take twice as long?
-
Failure Triage
What changed between the run that worked and the one that failed?
-
Image Updates
Confirm which tasks picked up a new docker image