We encountered performance issues for our use cases when we deployed
RLS to production. We made the decision to back that approach out and
solve the security concerns in application-level code instead.
Role management via migrations is tricky business. They are server-wide
objects, and given our database templating approach for parallelism it
leads to the role having dependent objects in multiple databases that
outlive any one test run.
This leads to a situation where the migrations complete successfully,
but the role remains as an artifact because those other test databases
technically depend on it.
To avoid this, and since we're no longer using this strategy, the prior
migrations that referred to the policy + role have been commented away
with notes pointing to the migrations introduced here, which remove
everything in a safe and idempotent way.
For Reviewers
do the tests pass cleanly for you?
does the application boot cleanly for you?
after doing the above, is the policy absent from \d repo?
after doing the above, is the role absent from \du?
