That is very odd. Have there been any hard commits performed at all? Even if not, there should still be an index directory.
Solr will do a full replication if the replica is too far out of date, but that shouldn't create (I don't think) a new index directory unless it's a misleading message. Is the cluster still receiving updates while the instance is down? "too far out of date" is about 100 documents currently. Are you sure you aren't just seeing a full replication happen? When you say "only replicates new documents" how long are you waiting? If none of this is germane, we need more details on how you're bringing the nodes up and down. Because this shouldn't be happening as you describe. Also, there have been a lot of changes since 4.0, if you have the bandwidth you might try with a current build. Best Erick On Fri, Jan 4, 2013 at 2:02 AM, Sai Gadde <gadde....@gmail.com> wrote: > I have a single collection and shard in my Solr cloud setup with 3 > nodes. zookeeper ensemble running on three different machines. > > When we restart one of the server other than leader in the cloud the index > directory is getting deleted in that Solr instance. Index starts with '0' > documents and the instance only replicates new documents. > > These are the messages from solr admin panel logging. Solr version: 4.0.0 > > 10:48:26WARNINGSolrCoreNew index directory detected: old=null > new=/solr/mycore/data/index/10:48:26WARNINGSolrCore[mycore] Solr index > directory '/solr/mycore/data/index' doesn't exist. Creating new index... > > Any help regarding this issue would be appreciated. > > Thanks > Shyam > gadde....@gmail.com >
