Hi Bo-Chun Wang,
Thanks for raising this, I hadn't noticed the regression until now. We haven't changed much around compaction (which is lucky) because it made finding this slightly easier. The only change that's been made to compaction was this one.
This change switched over from using a bespoke worker pool, to a generic one which I added to out tools common library. In this change there's a somewhat subtle bug; we were never actually using the constructed worker pool i.e. compaction was always using a single thread.
This wasn't picked up by the compiler because the worker pool itself was still referenced (to begin/block on teardown). I'll get a patch up to fix this issue (it's worth noting that this is a level above SQLite, so this will affect all storage types).