Load core process changed between 5.5.3 and 5.3.1
Hi, I'm trying to upgrade my cluster from Solr version 5.3.1. to 5.5.3 and I noticed that the core loading process in 5.5.3 is different from 5.3.1. The number of core loaded in parallel in 5.5.3 are about 5 or 6, when in 5.3.1 all cores were published as state "recovering" all together. This is the new behaviour or something is wrong with my setup? Reload a node with 5.5.3 is slower compared with 5.3.1. - Best regards -- View this message in context: http://lucene.472066.n3.nabble.com/Load-core-process-changed-between-5-5-3-and-5-3-1-tp4306588.html Sent from the Solr - User mailing list archive at Nabble.com.
Re: Load core process changed between 5.5.3 and 5.3.1
see: https://issues.apache.org/jira/browse/SOLR-7280 The problem is that when the number of load threads is unbounded and you have lots of cores, you can get into a state where replicas don't come up because of OOM errors and getting them back up is hard/impossible. Plus an OOM error is scary as the state of your system is questionable. You can adjust the number of threads, see the ref guide for "coreLoadThreads" in the element of solr.xml. This the current ref guide, but it's the same as 5.5: https://cwiki.apache.org/confluence/display/solr/Format+of+solr.xml Best, Erick On Sun, Nov 20, 2016 at 7:46 AM, Yago Riveiro wrote: > Hi, > > I'm trying to upgrade my cluster from Solr version 5.3.1. to 5.5.3 and I > noticed that the core loading process in 5.5.3 is different from 5.3.1. > > The number of core loaded in parallel in 5.5.3 are about 5 or 6, when in > 5.3.1 all cores were published as state "recovering" all together. > > This is the new behaviour or something is wrong with my setup? > > Reload a node with 5.5.3 is slower compared with 5.3.1. > > > > - > Best regards > -- > View this message in context: > http://lucene.472066.n3.nabble.com/Load-core-process-changed-between-5-5-3-and-5-3-1-tp4306588.html > Sent from the Solr - User mailing list archive at Nabble.com.
Re: Load core process changed between 5.5.3 and 5.3.1
Indeed in 5.3.1 the CPU spikes to 80 of load and now the cluster is more stable, slower but more stable. Thanks. \-- /Yago Riveiro  On Nov 20 2016, at 4:31 pm, Erick Erickson wrote: > see: https://issues.apache.org/jira/browse/SOLR-7280 > > The problem is that when the number of load threads is unbounded and you have lots of cores, you can get into a state where replicas don't come up because of OOM errors and getting them back up is hard/impossible. Plus an OOM error is scary as the state of your system is questionable. > > You can adjust the number of threads, see the ref guide for "coreLoadThreads" in the element of solr.xml. This the current ref guide, but it's the same as 5.5: https://cwiki.apache.org/confluence/display/solr/Format+of+solr.xml > > Best, Erick > > On Sun, Nov 20, 2016 at 7:46 AM, Yago Riveiro wrote: > Hi, > > I'm trying to upgrade my cluster from Solr version 5.3.1. to 5.5.3 and I > noticed that the core loading process in 5.5.3 is different from 5.3.1. > > The number of core loaded in parallel in 5.5.3 are about 5 or 6, when in > 5.3.1 all cores were published as state "recovering" all together. > > This is the new behaviour or something is wrong with my setup? > > Reload a node with 5.5.3 is slower compared with 5.3.1. > > > > \- > Best regards > \-- > View this message in context: [http://lucene.472066.n3.nabble.com/Load-core- process-changed- between-5-5-3-and-5-3-1-tp4306588.html](http://lucene.472066.n3.nabble.com /Load-core-process-changed- between-5-5-3-and-5-3-1-tp4306588.html&r=c29sci11c2VyQGx1Y2VuZS5hcGFjaGUub3Jn) > Sent from the Solr - User mailing list archive at Nabble.com.
