Should we reserve 7 slaves on codfw just for dumps? That seems like a waste of resources? However, dumps should run in case of a datacenter failover. Should we assume there are still some slaves running on the primary datacenter? Should we assume we can create dumps on codfw? And serve from there?
This is to help decide slave weights for codfw.
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Invalid | None | T125673 Switch over from Eqiad to Codfw as primary datacentre | |||
| Resolved | • jcrespo | T124670 Figure out and document the datacenter switchover process | |||
| Resolved | Krinkle | T124671 Switchover of the application servers to codfw | |||
| Resolved | • jcrespo | T124697 Stress-test mediawiki application servers at codfw (specially to figure out db weights configuration) and basic buffer warming | |||
| Resolved | • jcrespo | T125386 Clarify how mysql dumps will be architectured during codfw failover |
We could continue to generate them out of eqiad and serve them as downloadable from codfw, as long as it's certain that the eqiad dbs will be in sync. We've generated them out of a secondary dc in the past. See also T118154 (determine hardware needs for dumps in eqiad and codfw).
From my conversation with Ariel, this seems to have some some blockers. I would like to take dumps out of the codfw test switchover to have db overcapacity during it (and even maybe have fewer dedicated databases for it when it happens, later on). After all, codfw will only be active for a few hours, so I do not think it makes sense to prepare for it before May.
That doesn't mean the ticket mentioned by Ariel should not be worked on, provisioning is important.
Reopen if you disagree with my assessment.
I agree, dumps are not planned to be part of the upcoming test switchover in any case.