herronelou
commented
2 weeks ago I'm pretty sure this has affected us too, the daemon seems to spend a non-insignificant amount of time between creating the file handle and finishing the pickling (https://github.com/shotgunsoftware/shotgunEvents/blob/master/src/shotgunEventDaemon.py#L605).
Every time we stop the daemon, it's like there's a 5% chance of hitting this spot and corrupting the file. I think it might be one of the root causes of #63
We had an unexpected power failure and the id file was corrupted on disk. It looks like this has happened before to us. Also, we have a monitoring service that often loads an incomplete id file.
While it's not 100% guaranteed to be atomic, many tools write to a temp file and rename it in place. Some OSes guarantee atomicity and for the rest it should at least be a significant improvement.