Dear list,
I'm still experiencing sporadic occurrences of the "Error opening new
searcher" issue with the Solr statistics core. In our case Tomcat always
shuts down cleanly, and even so, manually deleting the write locks before
Tomcat startup doesn't seem to help. Thinking about this a bit more, it
sounds plausible to me that high load or slow disks could indeed cause a
timeout waiting for a write lock with large Solr cores. Our server has an
SSD, but is indeed busy, and we have nine years of statistics with hundreds
of millions of events.
Some Solr parameter that seems like it should help is writeLockTimeout. The
default in solrconfig.xml is 1000ms (1 sec). I tried setting it to 5000 and
1, neither of which helped. Should I try to bump this setting up to
some obscenely high value? Could there be another factor here, like JVM
heap size or settings? Happy to discuss experiences others are having.
Thanks,
P.S. I've been running Solr 4.10.4 in DSpace 5.8 production for over one
month now (versus 4.10.2 in DSpace pom.xml).
On Thu, Mar 21, 2019 at 11:54 PM Alan Orth wrote:
> Thanks for the response, Mark.
>
> I have started testing Solr 4.10.4 in a few development environments will
> share my notes after some time. Regarding the lock files, I might be
> barking up the wrong tree because it seems like Solr never deletes these: I
> did a quick check by adding a `find /dspace -type f -iname '*.lock'` before
> and after the Tomcat service starts and stops, and the lock files are
> always there. I'm not sure what to make of that...
>
> Regards,
>
> On Tue, Mar 19, 2019 at 6:02 PM Mark H. Wood wrote:
>
>> On Tuesday, March 19, 2019 at 11:35:47 AM UTC-4, Alan Orth wrote:
>>>
>>> Dear list,
>>>
>>> In the last few months we've been having issues with Solr throwing
>>> "Error creating core" errors one out of every two times we start Tomcat.
>>> This results in our statistics from previous years being inaccessible. For
>>> example, from the Solr log yesterday:
>>>
>>> 2019-03-18 12:32:39,799 ERROR org.apache.solr.core.CoreContainer @ Error
>>> creating core [statistics-2018]: Error opening new searcher
>>> ...
>>> Caused by: org.apache.solr.common.SolrException: Error opening new
>>> searcher
>>> at
>>> org.apache.solr.core.SolrCore.openNewSearcher(SolrCore.java:1565)
>>> at org.apache.solr.core.SolrCore.getSearcher(SolrCore.java:1677)
>>> at org.apache.solr.core.SolrCore.(SolrCore.java:845)
>>> ... 31 more
>>> Caused by: org.apache.lucene.store.LockObtainFailedException: Lock
>>> obtain timed out: NativeFSLock@
>>> /dspace/solr/statistics-2018/data/index/write.lock
>>>
>>> Some old Stack Overflow posts recommend setting Solr's address space to
>>> unlimited (ulimit -v unlimited), but the issue still occurs seemingly
>>> randomly for us.
>>>
>>
>>
>> I think that raising ulimit would only help if Tomcat's JVM memory
>> settings add up to more than the OS was configured to give the process.
>> Those recommendations may have been for stand-alone Solr running in its own
>> Jetty instance.
>>
>>
>>
>>> Now I usually just try to restart Tomcat again, or sometimes I shut it
>>> down cleanly, delete all the Solr write locks, and then start Tomcat back
>>> up. It's starting to feel a bit superstitious... maybe I should try to kill
>>> a chicken.
>>>
>>
>>
>>> We are running DSpace 5.8 with Tomcat 7.0.93 on Ubuntu 16.04. We
>>> upgraded from DSpace 5.5 in late 2018 and 2019 was the first year that the
>>> yearly stats-util sharding completed successfully (fixed in DSpace 5.7). I
>>> believe our problems are related to the existence of these shards. It feels
>>> like there is some kind of *race condition* because it only every so
>>> often and it's not always the same core that Solr is refusing to create.
>>> Sometimes it's statistics-2018, statistics-2015, etc.
>>>
>>
>>
>> So, sometimes Solr doesn't remove all of its locks when shut down. Does
>> it log ([DSpace]log/solr*.log) anything interesting at shutdown? does
>> Tomcat?
>>
>>
>>
>>> On this note, is there any reason we are still using Solr 4.10.2 with
>>> DSpace 5.x and 6.x? The Solr project issued two bug fix releases in that
>>> series—4.10.3 and 4.10.4—and there are about fifty bug fixes in those
>>> releases, some of which address memory leaks and shard handling. See the
>>> change logs for 4.10.3¹ and 4.10.4². As an experiment I just bumped the
>>> version to 4.10.3 in my test environment and DSpace starts up... so that's
>>> a good sign. I will do more testing of ingests, indexing, etc and report
>>> back.
>>>
>>>
>>
>> I think the reason is that nobody went looking for newer releases. Your
>> experiment is definitely worth trying, and your results will be interesting.
>>
>> --
>> All messages to this mailing list should adhere to the DuraSpace Code of
>> Conduct: https://duraspace.org/about/policies/code-of-conduct/
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "DSpace Technical S