Bruce, list, It appears that the freezes were probably not caused by running out of ifeffit resources. When I overloaded Athena with 50+ files I got a very well behaved warning, and then it finally declared it simply couldn't do what was asked of it and refused to do anything until I removed some groups. At no point did it become unresponsive or take the desktop with it. On a different computer I did manage to provoke a segfault when I was just fiddling around without even attempting to imitate actual use, but it still did not take the desktop with it. No significant error messages were produced. I will look at possible driver issues as Stefan suggested. Can anyone else suggest lines of inquiry? David. On Tue, 16 Feb 2010, David Ehle wrote:
Matt, Stefan,
I'll be responding to both posts sent to list in the order recieved.
Matt,
Thank your for your response. I'm going to run athena from a terminal window today (our users use a launcher) and see if I can provoke the problem and collect some actual worthwhile info on the problem. I'm afraid I've been handicapped by only getting 2nd or 3rd hand reports so far - normally I wouldn't even try to file a bug based on that, but it was considered urgent enough by users and staff that I wanted to at least make sure this wasn't a known problem or something with a trivial fix.
To clear up a little confusion - Athena is being run from the desktop, not via an SSH connection. SSH was only used as a method to kill the Athena process when the desktop was non-responsive.
I suspect your right and the problem is some sort of combination issue.
Stefan,
Thank you for your suggestion. The systems in question has not shown this problem in the past and there is a very strong correlation between Athena running and the "freeze" occuring, and the Athena process being killed and the freeze going away (without restarting X) so the video card is not one of my primary suspects - but I will look and see if there has been a recent change in the xorg version or drivers that might match the timeline of when we started seeing the problem. Thanks for the suggestion!
David.
All,
Hopefully by the end of the day I'll be able to provide some actual information.
Thanks again for your help!
David.
Combining messages sent only to list below-
Stefen Mangold on Feb 16th 2:38am: Dear David Ehle,
I had already some trouble at the beamline with a new computer with sudden crashes of X during "doing nothing". We could track this problem down to a bug in the graphic card driver. We could solve this problem by exchanging the consumer graphic card with a 50 Euro more expensive one.
Regards
Stefan
Matt Newville on Feb 15th 3:14pm: David,
I suspect that you *should* get sensible error messages from Athena about running out of Ifeffit resources well before seeing a crash of X. Not in the sense of "Bruce ought to change Athena to report this", but in the sense of "Bruce already does report this". Generally, Ifeffit/Athena should run out of memory well before a Linux system runs out of memory and swap space. Crashing X over ssh seems really peculiar to me -- that's not an out-of-memory issue. I don't often leave Athena open for hours at a time on Linux, but have never seen such crashes on Ubuntu. My suspicion is that this might be caused by Perl/Tk, X, or the lengthy ssh session.
--Matt
On Mon, 15 Feb 2010, Bruce Ravel wrote:
On Monday 15 February 2010, 03:30:51 pm, David Ehle wrote:
Can you think of anything that may have changed in recent versions of Athena that may have aggravated the problem? It seems like it's happening more often now, and suddenly taking X with it every time. Freezing X makes this a much bigger problem than if we just crash Athena.
I cannot imagine what's causing this. I wouldn't say that anything substantial has changed in Ifeffit+friends.
In the short term, any advice on what is important hardware-wise for running Athena would be greatly appreciated. If we can't find a fast solution for Athena related X crashes we will need to set up a dedicated system to run it so it doesn't take down our control stations.
That's not a horrible idea in any case. You could compile up a ginormous version of Ifeffit, link Athena to it, and run into the memory problem much less often.
B
-- Bruce Ravel ------------------------------------ bravel@bnl.gov
National Institute of Standards and Technology Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973
My homepage: http://xafs.org/BruceRavel EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/