Contradictory output on shutdown; sticky disk committed even after forceful termination of buildkitd

[haines]

  Sending SIGTERM to buildkitd for graceful shutdown
  Warning: buildkitd did not shutdown gracefully after 30 seconds, forcing shutdown with SIGKILL
  Warning: Disk may be in a bad state after SIGKILL - will prevent sticky disk commit
  Warning: buildkitd was forcefully terminated with SIGKILL
  Creating Blacksmith agent client with port: 1031
  Shutdown buildkitd gracefully
  ...
  All critical cleanup steps completed successfully
  Checking for previous step failures before committing sticky disk
  No previous step failures detected, committing sticky disk after successful cleanup
  Creating Blacksmith agent client with port: 1031
  Successfully committed sticky disk

There is a misleading log message here which says "Shutdown buildkitd gracefully" even if it did not.

However, more concerningly, it doesn't seem like the step failure detection works, and the sticky disk is committed regardless.

I guess the problem is that it relies on saving the job state, but I think that is intended for persisting state between the job execution and the post-job hook; I'm not convinced that that works within the post-job hook.