Re: [VuFind-Tech] search issue causing java problems in rc2?

[Demian K. <dem...@vi...>]

Sounds good.  I'm not aware of current search problems, but if you find issues after loading the latest trunk, please send me some example problem searches and I'll see what I can do to fix them.

- Demian

From: Mitchell, Erik [mailto:mit...@wf...]
Sent: Wednesday, April 21, 2010 2:40 PM
To: Demian Katz
Cc: Till Kinstler; vuf...@li...
Subject: Re: [VuFind-Tech] search issue causing java problems in rc2?

Good point - on the old release we restarted nightly & I did not migrate that process - no harm in putting that in place.

On an unrelated note, In the release we have it appears that uppercase boolean operators throw an error & truncation symbols do not work.  I think I might just go get the current code & see if that resolves a few of the issues.

E


On Wed, Apr 21, 2010 at 2:31 PM, Demian Katz <dem...@vi...<mailto:dem...@vi...>> wrote:
I'm glad the patch helped!

How often do you restart the VuFind process?  At Villanova, we do a restart from a cron job on a nightly basis -- I found that leaving it running continuously would sometimes lead to instability...  but with the nightly restart, it behaves itself very well.

- Demian

From: Mitchell, Erik [mailto:mit...@wf...<mailto:mit...@wf...>]
Sent: Wednesday, April 21, 2010 2:16 PM
To: Till Kinstler
Cc: vuf...@li...<mailto:vuf...@li...>
Subject: Re: [VuFind-Tech] search issue causing java problems in rc2?

Demian - the patch did the trick in getting rid of those errors - Thanks!

Till - thanks for the in-depth response.

This version of Vufind is an RC2 release checked out from svn about 4 weeks ago.  It is on an Amazon EC2 small server - 1 CPU, 1.7 GB ram.  At the moment it is the only application & we have allocated 1gb of ram to SOLR (same as in another env in which we have 4GB of ram on a 2cpu machine).  We are running on a small machine and have been wondering if that is the best approach.

If we have to scale this version up to a larger server we will but since server load is so low most of the time it seems odd that it just begins climbing uncontrollably.  Solr index is 9GB (1.7 M records).  From what was said about UseConcMarkSweepGC, there is no use in us implementing that unless we switch to a box with multiple cpus.

I cannot thank everyone enough for your help.  I will keep an eye on cpu use for the next day or so & move to a larger box if it does not get better.

Erik
On Wed, Apr 21, 2010 at 6:51 AM, Till Kinstler <kin...@gb...<mailto:kin...@gb...>> wrote:
Mitchell, Erik schrieb:

Hi all,

We have been running rc2 for a week now & are getting java garbage collection / out of memory errors periodically.

What hardware are you running VuFind on (32 bit/64 bit? How much RAM? multiple CPUs/CPU cores?)? How big is your Solr index?
The default memory settings for Jetty/Solr in VuFind may not be optimal for your environment. We had occasional "Out of memory" conditions, too, until we switched to the UseConcMarkSweepGC Java garbage collector instead of the UseParallelGC (which is set in vufind.sh), nevertheless we had assigned sufficient RAM to Solr... But it depends on your hardware (the UseConcMarkSweepGC works nicely if you have multiple CPU cores).

I turned on some more logging this morning and the output seems to point to a search that cannot be parsed properly (see below).

The error below doesn't influence garbage collection or memory usage. I think, these are two different problems you see here...

It's perfectly normal, that memory consumption grows as Solr is running. Solr is filling up caches with every query. As long as cache sizes don't hit the maximum heap size settings (or maximum cache sizes set in Solr), memory consumption grows and grows and grows... Before hitting the maximum allowed heap size, Java garbage collection starts to clean up. I guess, garbage collection doesn't work fast enough in your environment (that was the reason why we changed the garbage collector), so Solr tries faster to allocate more memory than garbage collection frees memory...

Now for the other problem:

org.apache.lucene.queryParser.ParseException: Cannot parse 'Archaeological Excavations at Kerma (Soudan): Preliminary report for 1993-1994 and 1994-1995 campaigns. ': Encountered " ":" ": "" at line 1, column 44.

Solr doesn't like that query because it tells Solr to search in (nonexistent and syntactically wrong) index field (Soudan) for the term Preliminary... That's an invalid query and Solr complains about that. But that shouldn't influence memory usage.
VuFind seems to make a mistake parsing the search terms and sends that invalid query to Solr. I guess, method isAdvanced() in sys/Solr.php is to blame. Can you give us some more information what version of VuFind do you use (1.0RC2 or a recent SVN trunk checkout?)? And you may turn on helpful debug output in web/conf/config.ini by setting debug = true in section [System]. Then you get some helpful debug output in the browser, you may to this list...
In the current trunk code that query seems to work, though the line
if (strstr($query, '(') && strstr($query, ')')) in isAdvanced()
falsely classifies it as advanced query. But buildAdvancedQuery() in Solr.php then takes care of the : and puts out a valid Solr query. I guess you are using an older version of VuFind that is buggy at this point... But give us more details and we will try to help you.

Till

--
Till Kinstler
Verbundzentrale des Gemeinsamen Bibliotheksverbundes (VZG)
Platz der Göttinger Sieben 1, D 37073 Göttingen
kin...@gb...<mailto:kin...@gb...>, +49 (0) 551 39-13431, http://www.gbv.de



--
Erik Mitchell, Ph.D.
Assistant Director for Technology Services
Z. Smith Reynolds Library
Wake Forest University
http://erikmitchell.info



--
Erik Mitchell, Ph.D.
Assistant Director for Technology Services
Z. Smith Reynolds Library
Wake Forest University
http://erikmitchell.info