How exactly does ElasticSearch handle Split Brain scenarios?
I've asked this question on the Solr ML the other day and this is the
finding: http://search-lucene.com/m/3L3ht1xuETB1
The summary is, I think, as follows:
the cluster partition with insufficient number of ZooKeepers (less than
what the quorum is), the partition will not accept adds/updates/deletes
the cluster partition with enough ZooKeepers to form the quorum will
continue to take updates IFF it has at least one copy of each shard from
the original cluster
I think searches will be served by 2), but not 1) and there is now an issue
open in Solr's JIRA to have 1) return search results, too, but with a flag
indicating that results may not be complete or could be stale.
How exactly does ES handle cluster partitioning?
I've asked a concrete Q about how Rivers are handled in Split Brain
situation and it looks like they are/were not handled well, meaning that
each partition of the cluster would end up with its copy of the River,
which I assume means there could be duplicate in the index once things are
fixed and cluster partitions are united in a single cluster: https://groups.google.com/d/msg/elasticsearch/5ndIc0qokDw/QdoJvHkO004J
How exactly does Elasticsearch handle Split Brain scenarios?
I've asked this question on the Solr ML the other day and this is the
finding: http://search-lucene.com/m/3L3ht1xuETB1
The summary is, I think, as follows:
the cluster partition with insufficient number of ZooKeepers (less than
what the quorum is), the partition will not accept adds/updates/deletes
the cluster partition with enough ZooKeepers to form the quorum will
continue to take updates IFF it has at least one copy of each shard from
the original cluster
I think searches will be served by 2), but not 1) and there is now an
issue open in Solr's JIRA to have 1) return search results, too, but with a
flag indicating that results may not be complete or could be stale.
How exactly does ES handle cluster partitioning?
I've asked a concrete Q about how Rivers are handled in Split Brain
situation and it looks like they are/were not handled well, meaning that
each partition of the cluster would end up with its copy of the River,
which I assume means there could be duplicate in the index once things are
fixed and cluster partitions are united in a single cluster: https://groups.google.com/d/msg/elasticsearch/5ndIc0qokDw/QdoJvHkO004J
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.