There seems to be resource contention problem with Solrj under load. To
reproduce the problem: set up a sample webapp with solrj connect to a HTTP
Solr instance and hammer the webapp with Apache ab (say 10 concurrent
connection with 100 requests). You'll notice that the webapp's servlet
container quickly consumes 100% CPU and stays there unless you restart it. I
can confirm that this happens with both Tomcat and Jetty. Meanwhile, the
server that Solr is deployed on seems to be running fine.
>From this observation, I suspect that Solrj has connection contention
problem. And this seems to be the case if you look at CommonHttpSolrServer.
This class uses MultiThreadedHttpConnectionManager which has
maxConnectionsPerHost set to 2 by default. When the number of thread
increases, this is obviously not enough and leads to connection contention
problem. I quickly solve problem by adding another constructor to
CommonHttpSolrServer that allows setting maxConnectionsPerHost and
maxTotalConnections:
public CommonsHttpSolrServer(int maxConsPerHost, int maxTotalCons, String
solrServerUrl) throws MalformedURLException {
this(solrServerUrl);
this.maxConsPerHost = maxConsPerHost;
this.maxTotalCons = maxTotalCons;
HttpConnectionManagerParams params = new HttpConnectionManagerParams();
params.setDefaultMaxConnectionsPerHost(maxConsPerHost);
params.setMaxTotalConnections(maxTotalCons);
_connectionManager.setParams(params);
}
Hope this information would help others.
--
Regards,
Cuong Hoang
