Not stable response time of elastic cluster

Elastic Stack Elasticsearch
1

60 shards 3 nodes elastic cluster, response time is pretty volatile from 20ms to 300ms, how could I make response time shorter and more stable?

2

What are you doing? What is your hardware like?
What is running on the machine?

3

normal query，hardware is 8core 16G 3nodes， machine is only running es as a search engine.

4

es stores many questions and answers, input a queston, and search for a similar question

5

Like the following?

GET /index/_search

Or a bit more complex than a match_all?

SSD drives? Or spinning disks?

What is the output of:

GET /
GET /_cat/nodes?v
GET /_cat/health?v
GET /_cat/indices?v

What is the typical response (the full one please) of a "slow query"? And what was the request?

Also have a look at this great blog post about Slow Queries. That might help.

6

curl -X GET "http://localhost:8080/_cat/nodes?v"

ip           heap.percent ram.percent cpu load_1m load_5m load_15m node.role master name
10.30.252.56           28          40   8    1.99    3.28     3.72 mdi       -      node-56
10.30.252.55           51          55  17    4.91    6.51     7.49 mdi       *      node-55
10.30.252.54           51          54  16    4.08    6.71     7.40 mdi       -      node-54

curl -X GET "http://localhost:8080/_cat/health?v"

epoch      timestamp cluster      status node.total node.data shards pri relo init unassign pending_tasks max_task_wait_time active_shards_percent
1563276185 19:23:05  es yellow          3         3    750 250    0    0       10             0                  -                 98.7%

curl -X GET "http://localhost:8080/_cat/indices?v"

health status index                 uuid                   pri rep docs.count docs.deleted store.size pri.store.size
green  open   all_with_routing_36_0 -eZAYbBcRqKa6WnjTItJLg  60   2     210232            0    561.2mb          187mb
green  open   all_36_1              tsVykN1GQYe7tScV93ZA0A  60   2        473            0      4.1mb          1.3mb
green  open   all_36_0              F_1uZJ02SzKScZw9wQQMIw  60   2     210232            0    734.3mb        244.7mb
yellow open   tritalkdoctest0627    cDH2mEicQNGuVkJ-prgxjg   5   3        119            0      1.1mb          390kb
yellow open   tritalkdoctest0715    QgUYx6BKTvCdIJc1TiOQqg   5   3        119            0      1.1mb        393.6kb
green  open   all_126_1             tHkz0rr-SO2EMKPp8ffL8g  60   2        677            0      4.9mb          1.6mb

curl -X GET "http://localhost:8080/"

{
  "name" : "node-54",
  "cluster_name" : "es",
  "cluster_uuid" : "5pxKPv4dTsajsfYabzx_2Q",
  "version" : {
    "number" : "6.3.0",
    "build_flavor" : "default",
    "build_type" : "tar",
    "build_hash" : "424e937",
    "build_date" : "2018-06-11T23:38:03.357887Z",
    "build_snapshot" : false,
    "lucene_version" : "7.3.1",
    "minimum_wire_compatibility_version" : "5.6.0",
    "minimum_index_compatibility_version" : "5.0.0"
  },
  "tagline" : "You Know, for Search"
}
7

spinning disk, query like GET /index/_search

8

60 shards is too much IMO.
1 shard is enough here.

spinning disk

Switch to SSD drives.

Also pasting again one of my questions:

What is the typical response (the full one please) of a "slow query"? And what was the request?

Please read this about how to format.

Also note that "6.3.0" has been built more than a year ago.
Lot of improvements have been done in the meantime. Could you upgrade to 7.2.0?

1 Like
9

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.