silent exception handler block in collect_jobs

ValueError is added by David Goetz in a very very old patch (a71164995aaafd626cc16ba59cd01dbb69e33fe0), but I didn't find out what's it prevent from here. But OSError is added reasonable in 46a093f068a158b72479522792509a882d3e47f1.

I think at least we could add a log with traceback here

OSError check was added by https://review.openstack.org/#/c/12378/ , I assume because of os.remove call.

Would it be enough to add something like

- except (ValueError, OSError):
+ except OSError:
+ self.logger.exception('ERROR removing partition directory %s' % job_path)
+ except ValueError:

?

So looking at it again, I think maybe we thought the OSError was for something like ENOPERM from isfile (which just returns False on all os.error), and the ValueError was from int(partition). IIRC the random conversation in IRC that caused me to file this, it was the ValueError case that may have been interesting.

I think we just need tighter exceptions and never skip a path without logging it, maybe it'd be easier to see if we extract it to function and test it real hard?

This code base has changed quite a bit since this bug was filed, but I think we still don't log as Clay initially suggested it. We are now tracking failures for recon[1] , but that's it. Is that enough or should we still add the log message there?

[1] https://github.com/openstack/swift/commit/79ba4a85983641e539b620bd143e62673c98416e

this some how reminds me of the container dir walker fixes that Mahati worked on [1]. The point is we need to iter the dirs and trap errors right w/o blowing up the whole loop and when we find something that we really don't know what to do with emit an error and pointer to the operator how to get rid of it:

"This thing at <path> is not great for me!"

That can be at debug, but warning is probably fine since we think these things are unexpected and not the common case!

1. https://review.openstack.org/#/c/331601/