we have 5 shards and each shard with one leader and one replica.
the "3300" connections is only for one JVM. please see the follow analysis
in zabbix.
and we the solrj code as follow:
public synchronized static CloudSolrServer getSolrCloudReadServer() {
if (reviewSolrCloudReadServer == null) {
ModifiableSolrParams params = new ModifiableSolrParams();
params.set(HttpClientUtil.PROP_MAX_CONNECTIONS, 1000);
params.set(HttpClientUtil.PROP_MAX_CONNECTIONS_PER_HOST, 100);
HttpClient client = HttpClientUtil.createClient(params);
LBHttpSolrServer lbServer = new LBHttpSolrServer(client);
lbServer.setConnectionTimeout(ReviewProperties.getCloudConnectionTimeOut());
lbServer.setSoTimeout(ReviewProperties.getCloudSoTimeOut());
reviewSolrCloudReadServer = new
CloudSolrServer(ReviewProperties.getZkHost(),lbServer);
reviewSolrCloudReadServer.setDefaultCollection(ReviewProperties.getZkReviewConnection());
reviewSolrCloudReadServer.setZkClientTimeout(ReviewProperties.getZkClientTimeout());
reviewSolrCloudReadServer.setZkConnectTimeout(ReviewProperties.getZkConnectTimeout());
}
return reviewSolrCloudReadServer;
}
2016-07-13 8:13 GMT+08:00 Shawn Heisey <apa...@elyograg.org>:
> On 7/12/2016 8:30 AM, Kent Mu wrote:
> > We have configed the maxThreads in JBOSS, and the good news is solrcloud
> > now running OK. but I another issue came across. We find the number of
> the
> > HTTP connections is very high, and the number can be around 3300. and
> > solrcloud does no release the connections.
> > I understand that, the solrcloud needs to connect to zookeeper and
> > communication between leader and replica need the connection. but I think
> > the number should not to be so huge.
> > besides, we use the singleton pattern to connect solrcloud in JAVA.
>
> Are you referring to the number of http connections in your SolrJ app,
> or the number of http connections in Solr itself? Hopefully these are
> being run by completely separate JVMs. Where exactly are you looking
> when you see 3300 connections?
>
> The connection to Zookeeper does not use HTTP. It is a TCP connection
> but the protocol is custom. Both Solr and SolrJ will maintain a
> connection to each of the zookeeper hosts that are in the zkHost string
> used when they start.
>
> Thanks,
> Shawn
>
>
