Migration number 25 causes CPU spike and lock contention on splits tables during metastore startup

[minhyuk-playground]

Summary

Migration number 25 (add-split-size), introduced in #6035, adds a GENERATED ALWAYS AS ... STORED column to the splits table.

In our environment (~1.5M rows in splits table), we observed:

• CPU spike on the PostgreSQL instance
• Lock-related metrics spike
• Connection acquisition timeouts

Adding a GENERATED ALWAYS AS ... STORED column requires an exclusive lock to compute and store values for all existing rows, causing incoming queries to queue up.

Environment

• PostgreSQL metastore with ~1.5M rows in splits table
• Upgrading to a version containing migration number 25 (add-split-size)

[guilload]

Yes, all our migrations suck because there are synchronous. Want to open a PR setting up Diesel Guard? That might help. I don't know if we can exclude the past migrations with the current naming scheme though.