Dispatching is a complex and vital mechanism. Things can and did go wrong, both in obvious and subtle ways in the past. To prevent some future incidents, we should look at past problems, and come up with ways to verify that they are not reoccurring now and will notice if they do in the future.
List of past issues that we need observability for:
- T233520: Wikidata changes do not get sent to client sites on initial sitelink addition (in some cases), leading to things such as missing page props in page_props table
- T192198: Wikidata doesn't update recentchanges
- Metrics idea: current time minus the time of the most recent timestamp coming from Wikibase in the client wiki's Recent Changes: select rc_timestamp, now() from recentchanges where rc_source='wb' order by rc_timestamp desc limit 1;
- wb_changes table growing very large T248244
- Metrics idea: monitor the number of rows inside wb_changes