The search preference documentation on http://www.elasticsearch.org/guide/reference/api/search/preference/ states
that a custom (string) preference is supported. However, it does not
explicitly state how possible failovers are handled in that case. Suppose
that a custom preference value "user-1234" is currently (internally)
associated with shard 1. What happens if shard 1 then goes down? WIll
search queries with preference value "user-1234" also fail from that point
onwards, or will it fail over, i.e., those custom preferences affected will
get re-mapped internally to other shards?
I'm also really interested in the answer for Patrick's question.
Bumping this in hope to get some answer.
Thanks in advance,
Yarin.
On Wednesday, August 28, 2013 2:59:14 PM UTC+3, Patrick Peschlow wrote:
The search preference documentation on Elasticsearch Platform — Find real-time answers at scale | Elastic that a custom (string) preference is supported. However, it does not
explicitly state how possible failovers are handled in that case. Suppose
that a custom preference value "user-1234" is currently (internally)
associated with shard 1. What happens if shard 1 then goes down? WIll
search queries with preference value "user-1234" also fail from that point
onwards, or will it fail over, i.e., those custom preferences affected will
get re-mapped internally to other shards?
when no one replied, I delved into the Elasticsearch source code to find
the answer myself. What I concluded from the code is that - assuming there
are N data nodes available in the cluster - the custom preference value is
transformed and eventually mapped to an index 0 .. N-1 via modulo. This
way, as long as the available nodes do not change, the queries will get
routed to the same node.
However, once a node leaves the cluster or another one joins, the index
resulting from the transformation may change and thus queries using the
same custom preference may get routed to another node. This implies that
there is a failover of the custom preference mechanism, as long as there
is at least one node available
the preferred node may change whenever the set of available data nodes
changes (not necessarily when that preferred node goes down but it can
happen even when any node leaves or joins)
I hope that clarifies the issue for you. And if my understanding of the
code turns out to be wrong, I'll be glad to learn about it
-Patrick
On Sunday, December 15, 2013 2:49:15 PM UTC+1, Yarin Miran wrote:
I'm also really interested in the answer for Patrick's question.
Bumping this in hope to get some answer.
Thanks in advance,
Yarin.
On Wednesday, August 28, 2013 2:59:14 PM UTC+3, Patrick Peschlow wrote:
The search preference documentation on Elasticsearch Platform — Find real-time answers at scale | Elastic that a custom (string) preference is supported. However, it does not
explicitly state how possible failovers are handled in that case. Suppose
that a custom preference value "user-1234" is currently (internally)
associated with shard 1. What happens if shard 1 then goes down? WIll
search queries with preference value "user-1234" also fail from that point
onwards, or will it fail over, i.e., those custom preferences affected will
get re-mapped internally to other shards?
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.
