Hi, We've seen this too on one of the test nodes yesterday, it ran on a build of a few days old. The node receiving documents complained it could not forward them to the fifth node and returned a 503. The fifth node itself only logged a NPE and the 503, nothing more, no stack traces.
There was no heavy committing going on (we only use auto commit of a few seconds) and it were only about 500 documents. The first two batches were accepted but then it died. It failed very consistently. Perhaps it accepted the first batches because no document was sent to the shard leader on that node yet. Restarting the node fixed the trouble. -----Original message----- > From:Chris Hostetter <hossman_luc...@fucit.org> > Sent: Fri 07-Sep-2012 23:02 > To: solr-user@lucene.apache.org > Subject: Re: [Solr4 beta] error 503 on commit > > > : I get sometimes (not often): > : SolrException e where e.code() == > : SolrException.ErrorCode.SERVICE_UNAVAILABLE.code > > Are there any errors in your solr server logs? > Are you using the DistributedUpdateProcessor (ie: SolrCloud) ? > > There aren't many places in Solr that will throw a 503 status code, if i > had to guess i would suspect that your problem is you are committing too > often relative to the amount of warming you are doing, and exceeding the > max number of open searchers... > > https://wiki.apache.org/solr/FAQ#What_does_.22exceeded_limit_of_maxWarmingSearchers.3DX.22_mean.3F > > : When I catch this exception, I try to commit again, the call doesn't throw, > : but the docs are not committed. Am I supposed to add docs again before > > Hmmm... that is odd, i think we'd definitely need to see the logs from one > of these errors (and the lines leading up to it) to understand what might > be happening. > > -Hoss >
