https://gerrit.wikimedia.org/r/c/mediawiki/core/+/440022 would solve this too. Which db group we should use for dispatching? vslow? @jcrespo your input is very valuable here.

@Ladsgroup check the other existing groups, but unless it has a huge overlap with on of them, we should agree on a new one e.g. "dispatch". For that (or even if we reuse an existing one) we should add more servers to that group, assuming it is considerable load. Do you have stats in QPS or others?

There is nothing that has overlap so a group called dispatch (with db1087 only) I think makes sense. Regarding QPS, I can't give you any numbers right now and it's rather hard but given the graphs I can say there is four concurrent processes all the time spending 53% of their run time selecting from the database.

Let me ask in another way- which percentage, even with a lot of error, of work dispatch creates on s8 (wikidata), if we include jobqueue, webrequests and other traffic.

I would put at least 2 servers for redundancy reasons.

@Ladsgroup perhaps something you want to do or not do as part of the ongoing current efforts around T265198: Don’t require extra cronjobs / maintenance scripts for Wikibase and T256208: Consolidate places that read/write 'wb_changes' table

I suggest we decline this. We should be getting rid of groups not to add more (T263127: Remove groups from db configs).

The issues in change dispatching should be solved in a higher level which is being worked on in Wikibase change dispatching scripts to jobs