MonitoringFileHandler can miss rotations

[gmacon]

Very occasionally (about once every six months or so) I have a host go out of disk because logrotate has rotated the log file but MonitoringFileHandler fails to notice.

I think I have found the cause of this. Currently, in emit, _query_fd is called twice: once before writing out to the file and again after. If the rotation happens to happen between these two points, the change won't be noticed because the new self._last_stat will already have been computed before the "before" value is stashed before the first call.

I think that the call to _query_fd in the constructor should also be removed, or at least the default value of delay should be changed to True for the MonitoringFileHandler.

I can't think of a clean way to write a test demonstrating this bug; if anyone has a good idea, I'd be happy to hear it.

[vmalloc]

@gmacon thanks for reporting! We're open to PRs fixing this issue, as I'm not sure when we will be able to write a fix for it.