Page MenuHomePhabricator

centralauth needs a read-only db flag that allows queries to go to lagged slaves
Closed, ResolvedPublic

Description

Author: afeldman

Description:
There doesn't currently seem to be a way to set centralauth to read-only which results in a site outage when its master needs to be rotated.

When the cluster with centralauth is set to read-only (currently S7), it only applies to wikis also on that cluster. Other wikis will still both want to write to centralauth, and also only send their queries to a caught up slave, which there won't be any of for a time during a master switch.

Normally, when a cluster has $wgReadOnly = true, queries are sent to any slave and replication lag checking is disabled. Setting centralauth to read-only needs this behavior as well.


Version: unspecified
Severity: major

Details

Reference
bz31170
TitleReferenceAuthorSource BranchDest Branch
Update analytics druid netflow loading DAGsrepos/data-engineering/airflow-dags!519joalupdate_analytics_netflow_druidmain
Customize query in GitLab

Event Timeline

bzimport raised the priority of this task from to High.Nov 21 2014, 11:53 PM
bzimport set Reference to bz31170.

Created new "platformeng" keyword, and adding that to this bug. I'll have to coordinate with Tim as to who exactly fixes this bug, but it looks like one we need to address soon.

r106784 adds $wgCentralAuthReadOnly

Just need to look at the lag checking code now

Not sure there is a need for anything else