We currently use untested handwritten and maybe-run database migration scripts for the main storage database.
Being already a cumbersome solution for the current production setup, it will pupate a real nightmare once we (at last) have proper mirrors here and there.
We need to use better tooling for this, and define a proper migration strategy.
Note that this is also linked to The Journal Migration Problem, aka T1279
- test the migrations
- run the migrations
- avoid mismatches between python and sql
- properly handle our different flavors of database (write db vs. read replica)
Investigate what tooling can be helpful for this.