[jira] [Created] (SOLR-13953) Prometheus exporter in SolrCloud mode limited to 100 nodes

["Alex Jablonski (Jira)" <ji...@apache.org>]

Alex Jablonski created SOLR-13953:
-------------------------------------

             Summary: Prometheus exporter in SolrCloud mode limited to 100 nodes
                 Key: SOLR-13953
                 URL: https://issues.apache.org/jira/browse/SOLR-13953
             Project: Solr
          Issue Type: Bug
      Security Level: Public (Default Security Level. Issues are Public)
            Reporter: Alex Jablonski


When using the Prometheus Exporter in SolrCloud mode against a cluster with more than 100 nodes, only 100 nodes' metrics are collected. For the other nodes, we see "Connection pool shut down" errors show up in logs, and the metrics from those nodes aren't reported.

This seems to be tied to the cache implementation in hostClientCache in SolrCloudScraper. That cache currently has a fixed maximum size of 100. When it approaches that limit begins to evict HttpSolrClients, it closes those clients.

We use the cache to build up a map of base URL to HttpSolrClient. For a >100 node cluster, the cache will successfully return clients for all nodes, sequentially. But once we add the 101st node, the first HttpSolrClient, which the cache still holds a reference to, gets closed. When we then try to get the metrics using all of the HttpSolrClients returned from the cache, the ones that have been closed throw IllegalStateExceptions with message "Connection pool shut down".

 

Original email thread here: [http://mail-archives.apache.org/mod_mbox/lucene-dev/201911.mbox/%3CCAOz296DSV-tt7rWBirBZ%2BP4%3DvT5g29FZrR_2zHrHF084Xq%2Bgyw%40mail.gmail.com%3E]

Github PR here: [https://github.com/apache/lucene-solr/pull/1022]

 



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org