Feedback platform cache restarted in a loop under its own write log

The feedback platform's cache was restarting repeatedly and never staying up long enough to serve traffic.

Cause. Its on-disk write log had grown to 28.9 MB and was never being compacted, because the compaction threshold sat above the size the log had reached. Replaying a log that size takes roughly 100 seconds, and during a replay the cache answers "still loading" rather than "healthy" — so the health check declared it dead and restarted it, every time, before it could ever finish. Each restart replayed the same log from the beginning.

Fix. Lowered the compaction threshold so the log rewrites itself well before it can outgrow the health-check window. The log fell from 28.9 MB to 308 KB and the service has been up since with no restarts and its full key set intact.