Got a question, the last 3 days (Under heavy sales volume) our index's via logstash rotate at GMT midnight.
We are noticing that it is taking +20 minutes for the new indexes to get allocated to a Node which during that time the Cluster is in a RED state
I have already adjusted my "Indexing" queue to
indices.memory.index_buffer_size: 70%
because we are consuming about 200GB a day on this cluster and searches are comparatively light
What should I look for that would influence SHARE slowness in creation besides system load?
BTW, after the new indexes are create I have a total of ~1600 Shards (some indexes are small) maybe I should merge some of them and reduce the shard count instead of increasing # of nodes?
That's odd, new indices should be created very quickly as they have no data.
How many shards are you creating for the new indices, how many replicas?
How big is your exisiting cluster state; how many existing indices, shards and replicas; how many nodes?
Nuts just noticed some of the servers are GMT and some are EDT ick got to fix that
its 9 nodes - each one on its own system ( 8 cors 16GB ram 150GB for data disk )
( at 8GB Heap, 150GB disk space each)
100 index ( 5 shard with 1 replication) ,
SAR
02:12:01 AM CPU %user %nice %system %iowait %steal %idle
02:13:01 AM all 54.01 0.00 6.13 0.74 0.00 39.12
02:14:01 AM all 61.54 0.00 5.98 0.17 0.00 32.32
02:15:01 AM all 55.65 0.00 6.11 0.20 0.00 38.04
02:16:01 AM all 54.54 0.00 5.59 0.28 0.00 39.58
02:17:01 AM all 70.41 0.00 5.55 0.17 0.00 23.86
02:18:01 AM all 56.08 0.00 6.81 1.04 0.00 36.07
02:19:02 AM all 56.93 0.00 6.13 0.27 0.00 36.67
02:20:01 AM all 59.85 0.00 6.16 0.25 0.00 33.75
02:21:01 AM all 58.47 0.00 6.02 0.24 0.00 35.27
Average: all 43.77 0.00 5.61 0.40 0.00 50.22
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.