if ansible-bender is killed the stale pid file can block future uses

[coreyoconnor]

Suppose the following:

1. ansible-bender starts
2. acquires db access; creates the pid file
3. host computer dies; eg: power failure
4. ansible-bender list-builds is invoked

expectation is that list-builds would work as expected. However, if we run with --debug we'll note repeated output of:

ab is running as PID 12947
ab is running as PID 12947
ab is running as PID 12947

Which, is not correct. The referenced ansible-bender process has died and the current ansible-bender process will wait forever. See the while loop here:

• ansible-bender/ansible_bender/db.py

  Line 89 in ba7eb21

  logger.info("ab is running as PID %s", pid)

Proposal:

• check if a process with that pid exists
• if no process with that pid exists for a period of time then delete the pid file and continue

As a separate feature:

• have an option to timeout and failure if db cannot be acquired within specified time. Not really required: The calling process could handle the timeout .

[TomasTomecek]

check if a process with that pid exists
if no process with that pid exists for a period of time then delete the pid file and continue

That's a good idea! Since you investigated so thoroughly, would you be interested in fixing it as well? :)

[coreyoconnor]

check if a process with that pid exists
if no process with that pid exists for a period of time then delete the pid file and continue

That's a good idea! Since you investigated so thoroughly, would you be interested in fixing it as well? :)

Thanks for confirming my understanding :)

I'm not much of a python programmer, but this should be within bounds. Unfortunately, my queue is long. :\ I'll submit a patch when I can. Still, anyone should feel free to fix

[areese]

@coreyoconnor I found a workaround for this. I hit the same issue, and added a log statement to show where the lock file is stored, and ended up removing it manually.

rm -f ~/.cache/ab/ab.pid