I do apologise for wasting anyone's time on this, the PEBKAC (my keyboard and chair unfortunately). When adding the new server to haproxy, I updated the label for the balancer entry to the new server, but left the host name the same, so the server that wasn't using any RAM... wasn't getting any requests.
Again, sorry! Tom On Tue, Apr 21, 2015 at 11:54 AM, Tom Evans <tevans...@googlemail.com> wrote: > We monitor them with munin, so I have charts if attachments are > acceptable? Having said that, they have only been running for a day > with this memory allocation.. > > Describing them, the master consistently has 8GB used for apps, the > 8GB used in cache, whilst the slave consistently only uses ~1.5GB for > apps, 14GB used in cache. > > We are trying to use our SOLR servers to do a lot more facet queries, > previously we were mainly doing searches, and the > SolrPerformanceProblems wiki page mentions that faceting (amongst > others) require a lot of JVM heap, so I'm confused why it is not using > the heap we've allocated on one server, whilst it is on the other > server. Perhaps our master server needs even more heap? > > Also, my infra guy is wondering why I asked him to add more memory to > the slave server, if it is "just" in cache, although I did try to > explain that ideally, I'd have even more in cache - we have about 35GB > of index data. > > Cheers > > Tom > > On Tue, Apr 21, 2015 at 11:25 AM, Markus Jelsma > <markus.jel...@openindex.io> wrote: >> Hi - what do you see if you monitor memory over time? You should see a >> typical saw tooth. >> Markus >> >> -----Original message----- >>> From:Tom Evans <tevans...@googlemail.com> >>> Sent: Tuesday 21st April 2015 12:22 >>> To: solr-user@lucene.apache.org >>> Subject: Confusing SOLR 5 memory usage >>> >>> Hi all >>> >>> I have two SOLR 5 servers, one is the master and one is the slave. >>> They both have 12 cores, fully replicated and giving identical results >>> when querying them. The only difference between configuration on the >>> two servers is that one is set to slave from the other - identical >>> core configs and solr.in.sh. >>> >>> They both run on identical VMs with 16GB of RAM. In solr.in.sh, we are >>> setting the heap size identically: >>> >>> SOLR_JAVA_MEM="-Xms512m -Xmx7168m" >>> >>> The two servers are balanced behind haproxy, and identical numbers and >>> types of queries flow to both servers. Indexing only happens once a >>> day. >>> >>> When viewing the memory usage of the servers, the master server's JVM >>> has 8.8GB RSS, but the slave only has 1.2GB RSS. >>> >>> Can someone hit me with the cluebat please? :) >>> >>> Cheers >>> >>> Tom >>>
