Note: In SolrCloud terminology, a leader is also a replica. IOW, you have
two replicas, one of which (and it can vary over time) is elected as leader
for that shard.
The other shards remain capable of indexing even if one shard becomes
unavailable. That is expected - and desired - behavior in a fault-tolerant,
fully-distributed system. Your application can/should make its own decision
as to what it will do if an indexing operation cannot be serviced.
-- Jack Krupansky
-----Original Message-----
From: elmerfudd
Sent: Wednesday, February 12, 2014 7:54 AM
To: solr-user@lucene.apache.org
Subject: Indexing in the case of entire shard failure
We have a system which consists of 2 shards while every shard has a leader
and one replica.
During indexing one of the shards (both leader and replica) was shut down.
We got two types of HTTP requests: rm= "Service Unavailable" and "rm="OK".
From this we’ve got to the conclusion that the shard which was up was going
on indexing. Is this behavior correct? We expected that in the case of the
entire shard fails the system stop indexing completely.
--
View this message in context:
http://lucene.472066.n3.nabble.com/Indexing-in-the-case-of-entire-shard-failure-tp4116881.html
Sent from the Solr - User mailing list archive at Nabble.com.
