Ref T13613. Improve the performance of this migration by using a temporary table and an "UPDATE x JOIN y ..." pattern.
- Ran on secure, got exit after a few seconds since the migration is idempotent and changesets already had PHIDs.
- Ran on secure with the continue; commented out, got valid new PHIDs in 53s (from 153s).
- Tried a larger page size (16K), didn't see any improvement.
- From "--trace", client PHID generation seems to be the limiting factor.
From "--trace", client PHID generation seems to be the limiting factor.
I think this isn't quite true. On secure, generating 400K PHIDs in a standalone script takes 7s, so this isn't totally trivial but is only ~13% of remaining runtime. I don't see any obvious wins remaining here, though.