[ https://issues.apache.org/jira/browse/SOLR-12415?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16493880#comment-16493880 ]
Shawn Heisey commented on SOLR-12415: ------------------------------------- I see these as possible solutions in situations where the base URLs do not include a core/collection name: * Change the zombie server check query to a benign action on one of the admin handlers that responds quickly. ** This would only be guaranteed to work on versions where the chosen handler is implicitly added, which might mean possible compatibility issues with older server versions. ** A different admin handler might be needed for checks on URLs that *do* include the core/collection name. ** The Javadoc for the client would need to declare which handlers are used for zombie checks, what server version added those handlers implicitly, and indicate that there must be explicit config in older versions. * Move 'setDefaultCollection' from CloudSolrClient to SolrClient, fix any problems that causes, and require setDefaultCollection for LBHttpSolrClient to work properly. * Have LBHttpSolrClient make a CoreAdmin call to get a list of valid cores and choose one for the zombie server check - but only if setDefaultCollection was not used. If CloudSolrClient relies on its internal LBHttpSolrClient to re-enable zombie servers, this might affect CloudSolrClient too. I suspect that the cloud client relies more on info from zookeeper. I was a little surprised to learn that LBHttpSolrClient assumes all servers are good until a request fails. I would have expected alive checks to happen before then. > Solr Loadbalancer client LBHttpSolrClient not working as expected, if a Solr > node goes down, it is unable to detect when it become live again due to 404 > error > -------------------------------------------------------------------------------------------------------------------------------------------------------------- > > Key: SOLR-12415 > URL: https://issues.apache.org/jira/browse/SOLR-12415 > Project: Solr > Issue Type: Bug > Security Level: Public(Default Security Level. Issues are Public) > Components: SolrJ > Affects Versions: 7.2.1, 7.3.1, 7.4 > Environment: Solr 7.2.1 > 2 servers - master and slave. > Reporter: Grzegorz Lebek > Priority: Critical > > *Context* > When LBHttpSolrClient has been constructed using *base root urls*, and when > a slave goes down, and then back again, the client is unable to mark it as > alive again due to 404 error. > Logs below: > {code:java} > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 >> "GET > /solr/select?q=%3A&rows=0&sort=docid+asc&distrib=false&wt=javabin&version=2 > HTTP/1.1[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 >> > "User-Agent: Solr[org.apache.solr.client.solrj.impl.HttpSolrClient] > 1.0[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 >> "Host: > localhost:8984[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 >> > "Connection: Keep-Alive[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 >> "[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "HTTP/1.1 > 404 Not Found[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "Cache-Control: must-revalidate,no-cache,no-store[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "Content-Type: text/html;charset=iso-8859-1[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "Content-Length: 243[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "[\r][\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "<html>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "<head>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "<meta > http-equiv="Content-Type" content="text/html;charset=utf-8"/>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "<title>Error 404 Not Found</title>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "</head>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "<body><h2>HTTP ERROR 404</h2>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "<p>Problem > accessing /solr/select. Reason:[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << "<pre> Not > Found</pre></p>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "</body>[\n]" > DEBUG [aliveCheckExecutor-1-thread-1] [wire] http-outgoing-83 << > "</html>[\n]"{code} > *Analysis* > when using only *base root urls* in a LBHttpSolrClient we need to pass a > "*collection*" paramter when sending a request. It works fine except that in > a method > {code:java} > private void checkAZombieServer(ServerWrapper zombieServer){code} > it tries to query a solr without the collection parameter, to check if the > server is alive. This causes a html content (apparently dashboard) to be > returned, and as a result it will move to the exception clause in the method > therefore even if the server is back it will never be marked as alive again. > I debugged this and if we pass a collection name there as a second param it > will respond in a right manner. > Suggestion is either to somehow pass the collection name or to change the way > zombie servers are pinged. > *Steps to reproduce* > Run 2 servers - master and slave. Create client using base urls. Index, test > search etc. > Turn off slave server and after couple of seconds turn it on again. > -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org