One common issue with distrubuted locker is that if the process die somehow, who/how/when should we release the lock.

A noraml approach will be have a timeout and if the process doesn’t hand back the lock within a period of time, the serivce just thanks that the process is dead.

But it is not always the case: for some long running operation, it could take a while.

One way to fix it is having the process update the lock service every few seconds to renew the lock. But it introduces additional cost for the communication.

With Container

With kubernete and container, now there is another approach:

• progress grab a lock from service and the service remember the spot-container_name as the owner of the lock
• service can talk to Kubernete and check whether the container is still alive.
• if it no longer exists, just release the lock