Problems

1. Currently, a lock is removed after a certain threshold of time, so automatically remove dead locks. But if a session is taking longer than the timeout, it will also remove that active lock
2. Locking is relatively slow under heavy multithreaded work

Solution for 1

• The timeout problem could be fixed if the active lock periodically (less than timeout duration) rewrites its lock with a newer timestamp, so no other process will remove it. The difficulty is to how this should work, having an extra thread or process doing the refresh work is an option

Solution for 1 and 2

Using a different locking mechanism could fix both problems at once

• os.O_EXLOCK, but not available on linux
• fcntl, but not available on windows
• portalocker
• Aplies to fcntl and portalocker: http://0pointer.de/blog/projects/locking.html (fcntl locks are process bound and cannot be used with multithreaded environments)