We could use a replica database to read the latest seen state (chd_seen) from the wb_changes_dispatch table.
In order to prevent double dispatching, the dispatcher bumping chd_seen would need to keep it locked until the master change in question replicated to all replicas.
This especially makes sense when implemented with T162556: Consider only updating wb_changes_dispatch after a successful run which removes writing to the wb_changes_dispatch table right before the actual dispatching starts.
| Subject | Repo | Branch | Lines +/- | |
|---|---|---|---|---|
| Do not update chd_seen twice and use replica for reading | mediawiki/extensions/Wikibase | master | +110 -194 |
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Invalid | None | T108944 [Epic] Improve change dispatching | |||
| Resolved | Ladsgroup | T151681 DispatchChanges: Avoid long-lasting connections to the master DB | |||
| Resolved | Ladsgroup | T162557 Use replica for reading the last dispatch position (chd_seen) |
Change 353478 had a related patch set uploaded (by Ladsgroup; owner: Amir Sarabadani):
[mediawiki/extensions/Wikibase@master] Do not update chd_seen twice and use replica for reading
Change 353478 merged by jenkins-bot:
[mediawiki/extensions/Wikibase@master] Do not update chd_seen twice and use replica for reading