Repository navigation
Wait for shared-filesystem output of completed Slurm jobs - #1104
jan-janssen with Copilot wants to merge 3 commits into
Conversation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
|
Important Review skippedBot user detected. To trigger a single review, invoke the
|
Co-authored-by: jan-janssen <3854739+jan-janssen@users.noreply.github.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1104 +/- ##
==========================================
- Coverage 94.24% 93.87% -0.37%
==========================================
Files 39 39
Lines 2190 2221 +31
==========================================
+ Hits 2064 2085 +21
- Misses 126 136 +10 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
|
@copilot Can the same analysis be extended to support the flux job scheduler and not only the slurm scheduler? In particular it would be great to use the pysqa package which should already provide an abstraction layer for checking the status of the job. |
Co-authored-by: jan-janssen <3854739+jan-janssen@users.noreply.github.com>
Extended the shared output grace period to Flux using pysqa’s |
Fixes #1102
Problem
When a Slurm job left
squeueand its_o.h5file wasn't yet visible on the shared filesystem, the task was failed with "terminated without producing output", even though the job had completed successfully.Change
sacctfor its state.COMPLETED: don't fail; allow a 60 s grace period for the output file to appear (it is loaded as soon as it does). If it never appears, the task fails after the grace period.Testing
Added unit tests (mocked queue/accounting) for delayed output, failed jobs and grace-period expiry; the targeted file-backend tests pass. A live Slurm reproduction wasn't possible as no Slurm is available in this environment.