My replicas are actually on different machines so they do come up. The problem I found is that since they can't get the leader they just come up but is not part of the cluster. I can still do local search with distrib=false. They do not retry to get the leader so I have to restarted them after the leader has started in order to get them back into the cluster.
Bill On Thu, Nov 8, 2012 at 4:02 PM, Markus Jelsma <markus.jel...@openindex.io>wrote: > Hi - i think you're seeing: > https://issues.apache.org/jira/browse/SOLR-3993 > > > -----Original message----- > > From:Bill Au <bill.w...@gmail.com> > > Sent: Thu 08-Nov-2012 21:16 > > To: solr-user@lucene.apache.org > > Subject: best practice for restarting the entire SolrCloud cluster > > > > I have a simple SolrCloud cluster with 4 Solr instances and 1 shard. I > can > > start and stop individual Solr instances without any problem. But not > when > > I have to shutdown all the Solr instances at the same time. > > > > After shutting down all the Solr instances, the first instance that > starts > > up wait for all the replicas: > > > > INFO: Waiting until we see more replicas up: total=4 found=3 > > timeoutin=169243 > > > > In the meantime, any additional Solr instances that start up while the > > first one is waiting can't get the leader from zookeeper: > > > > SEVERE: Error getting leader from zk > > org.apache.solr.common.SolrException: Could not get leader props > > > > When the first Solr instance see all the replicas, it becomes the leader: > > > > INFO: Enough replicas found to continue. > > INFO: I may be the new leader - try and sync > > > > But it fails to sync with the instances that had failed to get the leader > > before: > > > > WARNING: PeerSync: core=collection1 url=http://host2:8983/solr exception > > talking to http://host2:8983/solr/collection1/, failed > > org.apache.solr.client.solrj.SolrServerException: Timeout occured while > > waiting response from server at: http://host2:8983/solr/collection1 > > > > So I ended up with one for more replicas down after the restart. I had > to > > figure out which replica is down and restart them. > > > > What I also discovered is that if I start the first Solr instance and > wait > > until it returns after the leaderVoteWait of 3 minutes, the rest of the > > Solr instance can be started without any problem since by then they can > get > > the leader from zookeeper. > > > > Is there a better way to restart an entire SolrCloud cluster? > > > > Bill > > >