Slow down a bit ;)... First, just to cover the bases, you have done a commit, right? The index generation on the UI screen is a bit misleading as replicas in SolrCloud don't necessarily have the same generation, that's normal. The "master/slave" bits are cruft from the older non-cloud days.
So I'm having a bit of trouble understanding what the problem is. On the one hand you say "The shard has the same number of docs on each host", yet "customers searching in one specific collection reported seeing varying results with the same search". OK, what does that mean? Results in different order? Some docs just not there? What I would do is run the queries at the Solr nodes (after committing and when there was no indexing going on) with &distrib=false set. Fire this at _each_ replica for _each_ shard. If you see varying numbers of hits on the _same_ shard, that's one problem. Do note that the ordering can, in some edge cases, be different so you can't assume that just because a doc appears on one shard and doesn't on another it's really not there, it's possible it was sorted a bit lower in the list. It's rare that a customer would notice this though. Best, Erick On Fri, Apr 22, 2016 at 11:48 AM, tedsolr <tsm...@sciquest.com> wrote: > I have a SolrCloud setup with v5.2.1 - just two hosts. A ZK ensemble of 3 > hosts. Just today, customers searching in one specific collection reported > seeing varying results with the same search. I could confirm this by looking > at the logs - same search with different hits by the solr host. In the admin > console I see that the Replication section on one node is highlighted - > drawing your attention to a version and generation difference between > listing of "master" and "slave". > > The shard has the same number of docs on each host, they are just at > different generations. What's the proper way to re-sync? Should I restart > the host with the out of sync collection? Or click the "optimize" button for > the one shard? Or reload the collection? Do I need to delete the replica and > build a new one? > > Thanks! > > > > -- > View this message in context: > http://lucene.472066.n3.nabble.com/Replicas-for-same-shard-not-in-sync-tp4272236.html > Sent from the Solr - User mailing list archive at Nabble.com.
