I have a pretty basic setup: 2 physical servers, 1 data node on each
(master on another server). I create an index w/ 1 shard, 1 replica. So one
server is the primary and the other is the replica. Queries against the
primary server might take 50ms. But the query against the replica will be
100ms. Very consistent. It's happening in out QA and load environments.
Seems like queries against the replica is always much slower. We're running
0.19.12. Any ideas? Does the replica need to communicate w/ the primary
during queries and this is the extra overhead? I thought the replica was
pretty independent and had all the data it needed to respond to queries.
I don't think should be happening, no. I assume the hardware is identical?
What do other ES and server metrics/stats look like? Any differences
among them may hint at the source of the problem. See my signature if you
don't have any ES-specific monitoring in place. What happens if you shut
down the faster server?
On Monday, March 4, 2013 10:41:17 PM UTC-5, Tinou Bao wrote:
Hi,
I have a pretty basic setup: 2 physical servers, 1 data node on each
(master on another server). I create an index w/ 1 shard, 1 replica. So one
server is the primary and the other is the replica. Queries against the
primary server might take 50ms. But the query against the replica will be
100ms. Very consistent. It's happening in out QA and load environments.
Seems like queries against the replica is always much slower. We're running
0.19.12. Any ideas? Does the replica need to communicate w/ the primary
during queries and this is the extra overhead? I thought the replica was
pretty independent and had all the data it needed to respond to queries.
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.