When switching over redis traffic for xenon from fluorine to mwlog1001 in T123728 (https://gerrit.wikimedia.org/r/#/c/337568/) I noticed not all mediawiki traffic moved right away after scap sync-file to mwlog. A rolling-restart of hhvm has fixed the issue, though I believe all traffic should have migrated, we'd need understand why this hasn't happened.
Description
Description
Related Objects
Related Objects
Event Timeline
Comment Actions
May've had to do with long-running processes (I don't recall if HHVM excluded jobs run over HTTP at the time).
Or may've had to do with T181833.
Closing for now, as it seems too old to investigate at this point.