Log of the #blacklight channel on chat.freenode.net

Using timezone: GMT-05:00
* erikhatcher joins03:01
* cbeer leaves09:42
* cbeer joins10:08
* ndushay joins10:41
* Naomi joins10:52
* ndushay leaves11:00
* Naomi leaves11:09
* ndushay joins11:34
* bess joins12:17
* bess leaves12:21
* bess joins12:23
* bess leaves12:26
* bess joins12:33
* bess leaves12:58
* bess joins13:13
* bess leaves13:23
* bess1 joins13:44
* bess1 leaves14:14
* bess joins14:21
* bess leaves14:50
* erikhatcher leaves15:02
* bess joins15:18
* bess leaves15:55
<jrochkind>now my solrmarc has slowed down to 1 record per second for some reason. okay, time to try something else. 16:07
* erikhatcher joins
<jrochkind>heh, glad erik missed that message talking to myself, or he'd be telling me "why are you still using solrmarc."
I said, fwiw, that my solrmarc is for some reason now down to 1 marc record per second.
<erikhatcher>path-e-tech
<jrochkind>okay, a ruby try. must do. too many things i want to do, not enough time. 16:08
<erikhatcher>jrochkind: i know the feeling well
<jrochkind>Huh, and also it keeps giving me java outfomemory heap excpetions. Which explains the slowness, it was running out of memory. 16:09
But even for a ginormous file, there's no reason it ought to need a lot of memory. Marc files can easily be read one file at a time, streamed from disk, no need to load the whole thing into memory. But apparentlyit maybe does?
s/one file/one record/
I hope ruby-marc is capable of reading a marc file streamed from disk one record at a time, not loading the whole thing into memory? If not, I guess it can be patched. 16:10
<erikhatcher>what runs out of memory? solr-marc? or solr?16:13
yeah, ruby-marc does fine streaming
<jrochkind>It's unclear what is running out of memory. The error is reported in the solrmarc log. But I guess I should go look in the solr log for it too. Maybe that would explain my solr instability issues. 16:17
<erikhatcher>that's why i asked about OOM for Solr
<jrochkind>yeah,I gotcha.
<erikhatcher>Solr has never been unstable, given enough RAM for the task at hand
<jrochkind>What log would I look at in Jetty/Solr for that? jetty/logs/stderr.log?
<erikhatcher>if solrmarc is logging it, i'd guess it's on the solrmarc side of things though
how'd you launch Solr?16:18
by default it logs to console
you have to tweak logging config to have it do otherwise
jetty logs requests, but it won't log the detail solr itself will
<jrochkind>Yep, it's indeed in that log. However jetty is setup, it's throwing things there. Launched solr with jetty. And indeed there is an OOM in there. 16:19
Okay, now to figure out how to seriously bump up solr memory in jetty.
<erikhatcher>java -Xmx2048m -jar start.jar ;)16:20
-server is a good option too
how much RAM do you have?
shouldn't need a lot for indexing
<jrochkind>It's being run inside a jetty, I'm not executing the command line manually. so not sure how to do it. I guess I need to bump up jetty's own ram? 16:21
<MrDys>jrochkind: change it in the jetty init.d script
<jrochkind>MrDys: Sweet, thanks. I'll go see what it's got now, and how much mem on th server. It _shouldn't_ need a lot for indexing, but apparently needs more than it's got.
<erikhatcher>how is jetty launched?16:22
<MrDys>jrochkind: heh. /proc/meminfo is saying that VM only has 1gb of physical ram
<erikhatcher>oh, init.d16:23
<jrochkind>MrDys: You think we need to ask Tony for more? Bah. So many JIRA tickets.
Any idea how much I _ought_ to give it, and how much the VM needs for that?
<MrDys>jrochkind: we're going to need more than a gig if it's going to handle even a minimal amount of users
<jrochkind>You think 2G is enough for now?
MrDys: What should I ask Tony for? I don't care, this is not something I know about. 16:24
<MrDys>I don't know what the performance aspects of most of the ruby/solr stack are
2 would be a start, probably more would be even better
<jrochkind>I can ask Tony for whatever we want. I don't know the performance aspects either. 16:25
You think I shoudl ask for 3?
4?
10?
:)
I have no idea either, but you seem to have some more idea than me, MrDys. Since I didn't even know 1 wasn't enough. :)
<MrDys>I'd ask for 416:26
<jrochkind>Okay, I'll ask for 4, see if we can get it.
<MrDys>not sure how divvied up that machine is
<erikhatcher>i think the default for java without specifying is like 64meg
so even 512m might do the trick16:27
<jrochkind>right now the init.d doesn't seem to specify it all, as far as i can tell. So no surprise that 64m isn't enough. :)
But 1G on the host OS isn't enough even to bump it to 512M.
Of course, we've got TWO cores running in solr, which is the only thing running in jetty. So if 512M is enough... does that mean for two cores, double it to 1G?
<erikhatcher>i'd give 512 a try 16:28
<MrDys>unless those two cores are going to be hit with the same amount of requests, I wouldn't think so
<erikhatcher>you really don't need a lot of RAM for indexing
<MrDys>but I really think that the solrmarc process is just eating the machine
<erikhatcher>on the searching side, for facets and caching, you'll need more, but for fiddling around indexing should be fine
well, run solrmarc on another machine ;)
that's the idea for posting over HTTP anyway
<jrochkind>I don't have another machien to run it on at the moment. But, yeah, ultimately, I understand that's the goal. 16:29
<erikhatcher>run 5 machines, posting in marc all separately, pegging each machine with the mapping crap
<jrochkind>I'll try giving it 512m, we probably have enough OS mem for even that now.
* bess joins16:34
* bess leaves16:37
* cbeer_ joins16:46
* erikhatcher leaves17:32
* bess joins18:10
* [LorD-VipsS] joins18:12
* [LorD-VipsS] leaves18:14
* bess leaves18:26
* bess joins20:14
* cbeer_ leaves20:35
* bess leaves21:06
* cbeer_ joins21:45
* cbeer_ leaves21:47
* erikhatcher joins21:51
* rsinger_ joins22:16
* ndushay leaves22:20
* rsinger leaves22:28
* rduplain joins22:33
* rduplain leaves22:51
* erikhatcher leaves23:09
* ndushay joins00:00
* ndushay leaves00:03
* erikhatcher joins00:24
* erikhatcher leaves00:30

Generated by Sualtam