Subject: Re: X and date change causes hang
To: Steven M. Bellovin <firstname.lastname@example.org>
From: David H.Gutteridge <email@example.com>
Date: 11/22/2004 22:41:13
Here's what I just tried: I removed the root crontab file (no other users have crontabs), I made a date change one year forward, and the results were:
X server before: 0.02 CPU time
X server after: 1.29 CPU time
As soon as a remote xterm session I had opened unfroze, the first time I tried this top reported that cron was indeed running, but with comparatively little CPU time, and it maxed out at 60% of CPU before quickly dropping. It appeared (by eye) that cron had started right after the freeze ended, but my eyes aren't a fair arbiter (I didn't look at it fast enough). What I can say is that cron had 0:03 CPU time before the freeze, and was accumulating at least some of its present total of 0:23 after the freeze ended. I redid the test a second time, and cron remained at 0:23. (The second time X showed a massive number for CPU time, obviously affected by the date change.)
lastcomm shows no processes between the date command I issued and the time the system unfroze. No other processes show any appreciable changes. So it appears X is indeed the culprit.
It's curious that this does not happen when the date is moved backwards, only forwards. I know nothing about X internals, so I have no idea if that's significant or not.
> From: "Steven M. Bellovin" <firstname.lastname@example.org>
> I wonder if this is really a bug at all. Does cron try to catch up?
> An X server is just an application (unlike, say, ping replies or
> console output without X); if it's competing for CPU with other,
> high-priority processes, it's not a surprise that it would freeze.
> There may be some component to X itself -- if it's in a select() loop
> trying to make something move smoothly, it may be running through a lot
> of steps catching up to the current time, even if no draw events are
> pending. But let's try to understand whether or not the CPU is 110%
> busy. What is the accumulated CPU time of X before and after the change?
> What about other processes? Does lastcomm show anything?
> --Steve Bellovin, http://www.research.att.com/~smb