LibreOffice Live, Better Crash Experience

Today was the big day, LibreOffice went live.  We had been testing it for many months, and it fits so well into our architecture that I was confident it would go pretty well; but one never knows.  Last night before I left I reset everyone back to defaults one more time and then when I arrived today I put it live.  The migration was painless for City employees because it didn't cause any disruption.  The next time they requested the word processor, they were pointed to LibreOffice.  Those people that were already in OpenOffice continued in their session until it was closed.  All launch scripts are in common scripts, so I only had to make about a handful of changes and it was done.

The helper applications that I have described in the past seemed to work well in testing and have been deployed.  Notify-send is used in the lower right hand corner to give users tips and FAQs.  The lower left hand corner is a popup button on a timer that allows them to remove deadlocked instances from the process list.  If you leave it alone, it counts down to zero and then just goes away. We have had a few people not read it, and just click on it and then nothing opens because all processes were halted.  Technique will improve quickly as people figure it out.  Non-intrusive, intrusion.  This is required to allow users on 24 hour shifts to clear out processes when IT Support is not available.

With about 100 documents open and about 25 users typing, top looks great.  A few of these processes are still OpenOffice from earlier this morning.  All OpenOffice sessions will flush out as users close documents.

For the number of users on the network and the amount of work being done, things are very stable in regards to teh GNOME desktop.  But as is the case with all software - sometimes things crash.  I kind of had a "duh" moment yesterday when I realized that I could improve their experience.  The GNOME server of course calls bug-buddy when software crashes, by design it gives users a box that contains information that means nothing to them.  A while back I configured bug-buddy on our Evolution server to automatically grab their backtrace and dump to a flat file.  My logs indicate that a few people each day are dumping mail-notification and avant-window-navigator and the server was just giving them the bug-buddy UI and leaving them on their own. Shame on me.  With the panel gone, the session is worthless.  So I moved the bug-buddy binary out of the way and inserted a custom script that logs the crash to our tracking software, and then simply restarts the application for them.  I kill -11'd a session next to me for testing and it's working great.  The user sees a short blink and notify-send of restart and their critical pieces come back.  I'm logging all crashes to see what else might be crashing, but so far it's quiet. I'm thinking this is affecting maybe 5-6 users a day.  If other components are having problems, I should be able to do something similar and just get them up and running again.  The experimental script is below.  I'm pondering the potential for an endless loop and will monitor and adjust as needed.