Re: Server inconsistent state Core Reload issue
: If you look at my email the container that is running SOLR got the request : params (http access logs provided in first email) but when it goes through : the SOLR app/code on the container (probably through request filters or : dispatchers..I don't know exactly) its getting lost, which is what I am : trying to understand. I want to understand under what situations this mat : happen. i can't think of any situation where anything in solr would do this ... i do however note that in the stacktraces you provided from your threaddump, there is a *lot* more forwarding and dispatching going on then i have ever seen in a typical jetty or tomcat type setup -- aprently because of how you have glassfish running you've got the usual stuff prior to the SolrDispatchFilter (org.apache.catalina.*) but you've also got things i have never seen before... com.sun.enterprise.web.WebPipeline com.sun.enterprise.web.PESessionLockingStandardPipeline com.sun.enterprise.web.connector.grizzly.* ...i would bet money that whatever is fausing problems with your request paramaters is happening somewhere in there. It would be trivial to test: just add some logging of the full request params in SolrDispatchFilter and see what it's getting from our servlet container when the doFilter method is called. FWIW, some sporadic googling has lead to a handful of indications that you are not alone in encountering problems with glassfish sporadically swallowing request params (or POST bodies) so that applications running in glassfish don't get them... http://jira.icesoft.org/browse/ICE-5298 http://www.coderanch.com/t/601685/glassfish/Body-Request-occasionally-lost -Hoss
Server inconsistent state Core Reload issue
We are using Solr 3.6.2 with a single core setup on a glassfish server,
every 4-5 hours the server gradually gets into a some kind of a
inconsistent state and stops accepting any queries giving back cached
results. Even the core reload fails giving the following. Has anybody
experienced such behavior ? Can anybody help me understand why this might
happen ?
http://searchserver:80/solr/admin/cores?action=RELOADcore=core1
response
lst name=responseHeader
int name=status0/int
int name=QTime9/int
/lst
lst name=status
lst name=core1
str name=namecore1/str
str name=instanceDir/data/solr/core1-home//str
str name=dataDir/data/solr/core/core1-data//str
date name=startTime2013-05-01T19:16:31.32Z/date
long name=uptime137850/long
lst name=index
int name=numDocs21479/int
int name=maxDoc25170/int
long name=version1367184551418/long
int name=segmentCount4/int
bool name=currenttrue/bool
bool name=hasDeletionstrue/bool
str
name=directoryorg.apache.lucene.store.MMapDirectory:org.apache.lucene.store.MMapDirectory@/data/solr/core/core1-data/index
lockFactory=org.apache.lucene.store.NativeFSLockFactory@71d9673/str
date name=lastModified2013-05-01T19:15:04Z/date
/lst
/lst
/lst
/response
During the inconsistent state any queries being issued to the server loose
the query parameters. We can see the proper queries in the container's http
access logs but solr somehow solr doesn't get the query params at all. Also
note that content length on the container's access logs is always 68935,
which implies its always giving the same docs irrespective of the query.
If we restart the server everything is back to normal and the same queries
run properly.
SOLR Log
--
[#|2013-05-01T15:20:02.031-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=20;_ThreadName=httpSSLWorkerThread-9001-1;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=17 |#]
[#|2013-05-01T15:20:02.034-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=24;_ThreadName=httpSSLWorkerThread-9001-4;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=13 |#]
[#|2013-05-01T15:20:02.055-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=23;_ThreadName=httpSSLWorkerThread-9001-3;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=13 |#]
[#|2013-05-01T15:20:02.081-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=25;_ThreadName=httpSSLWorkerThread-9001-5;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=14 |#]
[#|2013-05-01T15:20:02.106-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=19;_ThreadName=httpSSLWorkerThread-9001-0;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=14 |#]
[#|2013-05-01T15:20:02.136-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=22;_ThreadName=httpSSLWorkerThread-9001-2;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=16 |#]
[#|2013-05-01T15:20:02.161-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=20;_ThreadName=httpSSLWorkerThread-9001-1;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=15 |#]
[#|2013-05-01T15:20:02.185-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=24;_ThreadName=httpSSLWorkerThread-9001-4;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=14 |#]
[#|2013-05-01T15:20:02.209-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=23;_ThreadName=httpSSLWorkerThread-9001-3;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=14 |#]
[#|2013-05-01T15:20:02.241-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=25;_ThreadName=httpSSLWorkerThread-9001-5;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=16 |#]
[#|2013-05-01T15:20:02.266-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=19;_ThreadName=httpSSLWorkerThread-9001-0;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=15 |#]
[#|2013-05-01T15:20:02.288-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=22;_ThreadName=httpSSLWorkerThread-9001-2;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=14 |#]
[#|2013-05-01T15:20:02.291-0400|INFO|sun-appserver2.1.1|org.apache.solr.core.SolrCore|_ThreadID=20;_ThreadName=httpSSLWorkerThread-9001-1;|[core1]
webapp=/solr path=/select params={} hits=21479 status=0 QTime=15 |#]
Container Access Logs
-
xx.xxx.xx.xx 01/May/2013:15:20:02 -0500 GET
Re: Server inconsistent state Core Reload issue
Shawn, I don't believe its the container because we use the same container in another setup that has 6 cores which is serving almost 1.8 Million requests a day without a hitch. If you look at my email the container that is running SOLR got the request params (http access logs provided in first email) but when it goes through the SOLR app/code on the container (probably through request filters or dispatchers..I don't know exactly) its getting lost, which is what I am trying to understand. I want to understand under what situations this mat happen. Having said that this application that uses this problematic SOLR instance retrieves large number of facets results for each of 26 facets for each query and every query is a group query, would that cause any issues with SOLR caches that could lead to the issues like I am facing ??? With regards to the port number, our paranoid security folks wanted me to not reveal our ports so I put it as 80 without thinking :-), I assure use that its not 80. Thanks, Ravi On Wed, May 1, 2013 at 6:03 PM, Shawn Heisey s...@elyograg.org wrote: On 5/1/2013 3:14 PM, Ravi Solr wrote: We are using Solr 3.6.2 with a single core setup on a glassfish server, every 4-5 hours the server gradually gets into a some kind of a inconsistent state and stops accepting any queries giving back cached results. Even the core reload fails giving the following. Has anybody experienced such behavior ? Can anybody help me understand why this might happen ? http://searchserver:80/solr/**admin/cores?action=RELOAD**core=core1http://searchserver:80/solr/admin/cores?action=RELOADcore=core1 response lst name=responseHeader int name=status0/int int name=QTime9/int /lst lst name=status It is dropping the parameters from the /admin/cores request too, so it returns status instead of acting on the RELOAD. This is acting like a servlet container issue more than a Solr issue. It's always possible that it actually is Solr. It's a little unusual to see Solr running on port 80. It's not impossible, just not the normal setup, because exposing Solr directly to the outside world is a very bad idea, so it's a lot safer to have it listen on another port. Is glassfish actually listening on port 80? If it's not, then you probably have something acting as a proxy in front of Solr. If your platform is a UNIX variant or Linux and has a fully functional 'lsof' command, the following will tell you which process is bound to port 80: lsof -nPi | grep :80 Can you try running Solr under the jetty that's included with the Solr download? For Solr 3.6.2, this is a slightly modified Jetty 6. You can't use the Jetty 8 that's included with a newer version of Solr. If port 80 is a requirement, that should be possible as long as it's running as root. Thanks, Shawn
Re: Core Reload issue
As long as your indexed documents have the stop words you will continue to see the stop words in the results. On Fri, May 8, 2009 at 11:24 AM, Sagar Khetkade sagar.khetk...@hotmail.com wrote: From my understanding re-indexing the documents is a different thing. If you have the stop word filter for field type say text then after reloading the core if i type in a query which is stop word only it would get parsed from stop word filter which would eventually will not serach against the index. But in my case i am getting the results having the stop word; so the issue. ~Sagar From: noble.p...@gmail.com Date: Tue, 5 May 2009 10:09:29 +0530 Subject: Re: Core Reload issue To: solr-user@lucene.apache.org If you change the the conf files and if you reindex the documents it must be reflected are you sure you re-indexed? On Tue, May 5, 2009 at 10:00 AM, Sagar Khetkade sagar.khetk...@hotmail.com wrote: Hi, I came across a strange problem while reloading the core in multicore scenario. In the config of one of the core I am making changes in the synonym and stopword files and then reloading the core. The core gets reloaded but the changes in stopword and synonym fiels does not get reflected when I query in. The filters for index and query are the same. I face this problem even if I reindex the documents. But when I restart the servlet container in which the solr is embedded I problem does not resurfaces. My ultimate goal is/was to reflect the changes made in the text files inside the config folder. Is this the expected behaviour or some problem at my side. Could anyone suggest me the possible work around? Thanks in advance! Regards, Sagar Khetkade _ More than messages–check out the rest of the Windows Live™. http://www.microsoft.com/india/windows/windowslive/ -- - Noble Paul | Principal Engineer| AOL | http://aol.com _ Planning the weekend ? Here’s what is happening in your town. http://msn.asklaila.com/events/ -- - Noble Paul | Principal Engineer| AOL | http://aol.com
RE: Core Reload issue
From my understanding re-indexing the documents is a different thing. If you have the stop word filter for field type say text then after reloading the core if i type in a query which is stop word only it would get parsed from stop word filter which would eventually will not serach against the index. But in my case i am getting the results having the stop word; so the issue. ~Sagar From: noble.p...@gmail.com Date: Tue, 5 May 2009 10:09:29 +0530 Subject: Re: Core Reload issue To: solr-user@lucene.apache.org If you change the the conf files and if you reindex the documents it must be reflected are you sure you re-indexed? On Tue, May 5, 2009 at 10:00 AM, Sagar Khetkade sagar.khetk...@hotmail.com wrote: Hi, I came across a strange problem while reloading the core in multicore scenario. In the config of one of the core I am making changes in the synonym and stopword files and then reloading the core. The core gets reloaded but the changes in stopword and synonym fiels does not get reflected when I query in. The filters for index and query are the same. I face this problem even if I reindex the documents. But when I restart the servlet container in which the solr is embedded I problem does not resurfaces. My ultimate goal is/was to reflect the changes made in the text files inside the config folder. Is this the expected behaviour or some problem at my side. Could anyone suggest me the possible work around? Thanks in advance! Regards, Sagar Khetkade _ More than messages–check out the rest of the Windows Live™. http://www.microsoft.com/india/windows/windowslive/ -- - Noble Paul | Principal Engineer| AOL | http://aol.com _ Planning the weekend ? Here’s what is happening in your town. http://msn.asklaila.com/events/
Core Reload issue
Hi, I came across a strange problem while reloading the core in multicore scenario. In the config of one of the core I am making changes in the synonym and stopword files and then reloading the core. The core gets reloaded but the changes in stopword and synonym fiels does not get reflected when I query in. The filters for index and query are the same. I face this problem even if I reindex the documents. But when I restart the servlet container in which the solr is embedded I problem does not resurfaces. My ultimate goal is/was to reflect the changes made in the text files inside the config folder. Is this the expected behaviour or some problem at my side. Could anyone suggest me the possible work around? Thanks in advance! Regards, Sagar Khetkade _ More than messages–check out the rest of the Windows Live™. http://www.microsoft.com/india/windows/windowslive/
Re: Core Reload issue
If you change the the conf files and if you reindex the documents it must be reflected are you sure you re-indexed? On Tue, May 5, 2009 at 10:00 AM, Sagar Khetkade sagar.khetk...@hotmail.com wrote: Hi, I came across a strange problem while reloading the core in multicore scenario. In the config of one of the core I am making changes in the synonym and stopword files and then reloading the core. The core gets reloaded but the changes in stopword and synonym fiels does not get reflected when I query in. The filters for index and query are the same. I face this problem even if I reindex the documents. But when I restart the servlet container in which the solr is embedded I problem does not resurfaces. My ultimate goal is/was to reflect the changes made in the text files inside the config folder. Is this the expected behaviour or some problem at my side. Could anyone suggest me the possible work around? Thanks in advance! Regards, Sagar Khetkade _ More than messages–check out the rest of the Windows Live™. http://www.microsoft.com/india/windows/windowslive/ -- - Noble Paul | Principal Engineer| AOL | http://aol.com
