Deadlock during initial sync on mirror

Issue Summary

When farm vet detects lot of content inconsistencies during mirror startup and tries to sync them, suddenly syncing process stops and mirror never moves to SYNCHRONIZED status.

Steps to Reproduce

1. Add a project with at least 500-600 repositories on a single mirror (stand-alone or farm)
2. Wait for sync to finish.
3. Shut down the mirror and keep it down while users are pushing new commits/making ref changes on these repositories. Wait till there is at least 1 ref change in at least 100 (ideally 500-600) repositories.
4. Start mirror with -Dexecutor.max.threads=4

Expected Results

Mirror moves to SYNCHRONIZED status

Actual Results

Mirror is stuck in METADATA_SYNCHRONIZED status. Attached thread-dump shows all 4 threads in "threadpool" are in WAITING status forever.

Workaround

Avoid keeping mirror down when there is lot of push activity happening on primary server. Mirror and primary should be down at same time OR mirror should be down when there is none or very minimal user activity on primary.