[ https://issues.apache.org/jira/browse/SOLR-6931?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14269729#comment-14269729 ]
Mark Miller commented on SOLR-6931: ----------------------------------- See SOLR-4509. HttpClient uses a stale connection check to avoid using a bad pooled connection. This check has a race, and we can use a bad connection sometimes. In most of these cases, it is actually safe for us to retry. We can't use the default retry handler because it attempts to detect idempotent updates and Solr allows update type requests via GET requests. If we turn off the idempotent detection, the retry is safe and we can avoid some very problematic problems like 'connection reset' exceptions. On a heavy working SolrCloud cluster, even a rare response like this from a replica can cause a recovery and heavy cluster disruption. > We should do a limited retry when using HttpClient. > --------------------------------------------------- > > Key: SOLR-6931 > URL: https://issues.apache.org/jira/browse/SOLR-6931 > Project: Solr > Issue Type: Bug > Reporter: Mark Miller > Assignee: Mark Miller > Attachments: SOLR-6931.patch > > -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org