Context

After shipping a candidate version of MediaWiki 1.43 to staging and creating a number of new 1.43 wikis we saw a number of errors (T408103, T408101, T408789) that we attributed to these 1.43 pods accessing 1.39 databases or 1.39 pods accessing 1.43 databases.

While we could just ignore these errors this is a high risk direction because we:

1. don't know if we're going to leave some persistent breakages in some subtle way
2. we don't know how long we might be running both versions; in the past running db updates has taken a long time or hit snags

Goal

We should prevent this happening by introducing routing work to the right version in:

• frontend
• backend
• standalone job pods