Such change is advertised by Solr creators as something that reduces dependency on ZK and is advised to be done to migrate to solr9: https://solr.apache.org/guide/solr/latest/upgrade-notes/major-changes-in-solr-9.html
We are using solr 8.11.1 and wanted to use solr urls instead of zk. After such change - http calls increased by 40-60% due to additional calls to get clusterState - that was previously retrieved from ZK
In fact, calls to getClusterState that are done internally by the httpclient are heavier in terms of performance after doing such a change as there are additional HTTP calls.
Code in question from CloudHttp2SolrClient class:
if (builder.zkHosts != null) {
this.stateProvider = new ZkClientClusterStateProvider(builder.zkHosts, builder.zkChroot);
} else if (builder.solrUrls != null && !builder.solrUrls.isEmpty()) {
try {
this.stateProvider = new Http2ClusterStateProvider(builder.solrUrls, builder.httpClient);
} catch (Exception e) {
throw new RuntimeException("Couldn't initialize a HttpClusterStateProvider (is/are the "
+ "Solr server(s), " + builder.solrUrls + ", down?)", e);
}
}
Had anyone configured this client in a way that performance is not decreased after such a change? Maybe there is a way to not ask for cluster state so many times?
