(consider) Rebuilding trigram indexes in production database
There is a small risk of data corruption due to the underlying Postgres bug which caused https://gitlab.com/gitlab-com/infrastructure/issues/3850
To be on the safe side we should at least consider rebuilding the trigram indexes in production after the update to 9.6.8 (infrastructure#3851).
To do this without downtime we would have to do create index concurrently, then drop the old index, then rename the new index to the old name.
We might consider doing this in a migration in case user sites have the problem. Probably it's sufficient to treat this as an operational problem and provide the process we follow here to support to use if users report problems.
Also, the risk of data corruption is quite low so perhaps we can just wait to see if there's any actual problems.