A resumed run waits for its old slot while another run holds it #336
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
What happens
Run 158 (vLLM) was stopped so a deploy could go through. The Stocks rerun (run 159, Free) was then admitted into slot 0, the slot run 158 had recorded. Once resumed, run 158 logged
slot 0 is held by run …; its server waits for the slotevery 30 s, and its apply sat atreadywith the holdengine_down. Since #313, runs on different backends go at once, so the holder can run for hours.Fix
When a resumed run's recorded slot is held,
reattachgives it a free slot, as admission would.osf.run.server_startedrecords the new slot, so the run's preview port and compose project move with it.Done when
A test resumes a run whose slot is held: its server starts in a free slot and the run row records it. On production, run 158 resumes and its apply runs alongside the Stocks run.
Shipped in #337 (
bee3ab2). On production, run 158 was resumed while the Stocks run held its recorded slot 0. It started in slot 1 (opencode on :8801) without waiting, and its apply runs as attempt 3 beside the Stocks run. The testtests/engine/test_resume_slot.pyfails without the fix.