You need to read the full thread. This turned out to be a scalability problem in how the indexing routine worked. IOW you could bloat the kernel fd data structure as much as you want, and sooner or later as your repository grows you will hit the limit, because the number of files open was linear to the size of the repository.
I would suspect the patch is in the main trunk now, but others will probably know more specifics. Cory Snavely University of Michigan Library IT Core Services On Thu, 2007-12-20 at 17:04 +1030, Steve Thomas wrote: > We're having a lot of trouble currently with tomcat crashing with a > "Too many open files" error. This is happening roughly twice a day -- > I am restarting tomcat every morning, and usually get a call around > lunch time that it has crashed and need to restart again. > > Restart is quick and fixes the problem, temporarily, but naturally I > wish it didn't happen at all. > > I did some digging and found a thread from this list from last year, > which petered out without apparent resolution, but where Mark Diggory > suggested tinkering with the fd.files-max value in sysctl.conf. > > [ > http://sourceforge.net/mailarchive/message.php?msg_id=E1Grhzs-0000Nj-JN%40mail.sourceforge.net > ] > > Well, I tried that, but it made no difference. So, back to Google, > where I found (searching for "files-nr") that you can list all the > open file handles used by a process, using > > # ls -l /proc/PID/fd/ > > where PID is the process id. > > So using this with the pid for the DSpace tomcat, I found lots of > items like this: > > lr-x------ 1 uals uals 64 Dec 20 16:30 237 > -> /data/dspace/search/_vzb.cfs (deleted) > > This is a symlink to one of the lucene index "overflow" files, which > [in my limited understanding] are dynamically created and deleted as > the index grows. These "deleted" items increase in number over time, > and I imagine DSpace eventually hits the ulimit for open files (1024) > and dies. > > So I think the problem may be due to the lucene indexing not releasing > file descriptors when they are deleted. Certainly, watching the list > over an hour I've seen the number of "deleted" lines rise steadily. I > guess we're noticing this as a problem here because of the very large > amount of editing work we're engaged in currently. Other sites with a > more "sedate" use of DSpace might never run into it. > > Well, that's how it looks to me right now. Nothing I can do about it, > but maybe someone expert in the lucene side of DSpace could look into > it? > > > Cheers. :D > > Stephen Thomas, > Senior Systems Analyst, > University of Adelaide Library > UNIVERSITY OF ADELAIDE SA 5005 AUSTRALIA > Phone: +61 8 830 35190 > Fax: +61 8 830 34369 > Email: [EMAIL PROTECTED] > URL: http://www.adelaide.edu.au/directory/stephen.thomas > CRICOS Provider Number 00123M > > ----------------------------------------------------------- > This email message is intended only for the addressee(s) and contains > information that may be confidential and/or copyright. If you are not > the intended recipient please notify the sender by reply email and > immediately delete this email. Use, disclosure or reproduction of this > email by anyone other than the intended recipient(s) is strictly > prohibited. No representation is made that this email or any > attachments are free of viruses. Virus scanning is recommended and is > the responsibility of the recipient. > > > > ------------------------------------------------------------------------- > SF.Net email is sponsored by: > Check out the new SourceForge.net Marketplace. > It's the best place to buy or sell services > for just about anything Open Source. > http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace > _______________________________________________ DSpace-tech mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/dspace-tech ------------------------------------------------------------------------- SF.Net email is sponsored by: Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace _______________________________________________ DSpace-tech mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dspace-tech
