Three (of five) of our maps server nodes have database contents which haven't been replicated since roughly March 2022: graph. Let's restore these servers by depooling and resyncing with the master, one at a time.
The affected servers are:
- maps2005.codfw.wmnet
- maps2006.codfw.wmnet
- maps2008.codfw.wmnet
A script already exists to do the postgres part of the work: resync_replica.sh
We think this fix might last for some time, since the other good servers have maintained up-to-date replicas for as long as we have metrics in grafana (since late June 2022).