As a follow up to T224591: Migrate contint* hosts to Buster, we need a runbook explaining how to switch the CI masters.
Runbook to be at https://www.mediawiki.org/wiki/Continuous_integration/Data_center_switch
Once written, it should be linked at: https://www.mediawiki.org/wiki/Wikimedia_Release_Engineering_Team/Runbooks
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Declined | None | T256422 switch contint prod server back from contint2001 to contint1001 | |||
| Resolved | LSobanski | T256396 Create a runbook for switching CI master |
Literally extract the information from T224591
The runbook address the issues we had with uid/gid changing during the transfer which then require painful chmod all over the place. The reason is the quickdata copy puppet classes that spawns rsync within a chroot which in turns prevent uid/gid mapping.
Tyler bootstrapped https://www.mediawiki.org/wiki/Continuous_integration/Data_center_switch and I have edited it with the basic skeleton of how to do a switch over. The biggest devils are:
@jijiki We can take this based on:
hieradata/role/common/ci/master.yaml:profile::contacts::role_contacts: ['ServiceOps-Collab']
https://www.mediawiki.org/wiki/Continuous_integration/Data_center_switch exists thanks to Hashar, it's just that it's still in draft status.
https://gerrit.wikimedia.org/r/q/topic:contint2002 is a topic branch that lists all patches that are expected to be needed to switch. currently at 10, minus decom'ing the old host
@hashar do you think there's still value in doing this with the Zuul upgrade coming soon?