jobs: use low-priority transactions for claim, adopt, cancel/pause #65643
Conversation
These transactions can be slow and long-running and they hold locks. This is unfortunate for UX reasons. Release note: (bug fix): Improved availability of jobs table for reads in large, global clusters by running background tasks at low priority.
|
This change is |
|
I'm not quite sure who to get to review this. While it is in the jobs ecosystem, it's more about knowing the semantics of priority. @andreimatei you looked at these job loops at some point not too long ago, any interest in stamping? My plan is to let this bake on master for a couple of weeks and then to backport. |
|
@miretskiy maybe I can alternatively interest you in giving this a review (even though I owe you several). |
|
I suppose a test could be written to show that the client can read while one of these loops holds a lock and then gets blocked. |
There was a problem hiding this comment.
Reviewed 2 of 2 files at r1.
Reviewable status:complete! 0 of 0 LGTMs obtained (waiting on @miretskiy)
There was a problem hiding this comment.
Reviewed 1 of 2 files at r1.
Reviewable status:
|
TFTR! bors r+ |
|
Build succeeded: |
These transactions can be slow and long-running and they hold locks. This is
unfortunate for UX reasons.
Fixes #65077.
Release note: (bug fix): Improved availability of jobs table for reads in large,
global clusters by running background tasks at low priority.