In HELP-83023, it seemed that chunk migrations had some liveness issues due to the _sessionCloneMutex being held while reads were being performed:

• On the oplog
  • _handleWriteHistory
  • getNext
  • next
  • findOneOplogEntry
  • oplog read
• On the image collection
  • fetchPrePostImageOplog
  • findOne

When handling the same for new writes, we specifically drop the lock for the duration of the read: here and here.

Consider if either we could also drop the lock for the reads related to the session catalog, or if otherwise SessionCatalogMigrationSource could be refactored such that it's not necessary in the first place.