Elasticsearch restart performance

Elastic Stack Elasticsearch
1

Hi All,
We are using a one node cluster elasticsearch(version 5.3.0) in a 96RAM 24cores centos7 machine. The cluster holds about 1TB time-based documents in about 30 Indexes. If I restart the cluster using systemctl restart elasticsearch, it takes about 5 minutes to recovery all indexes from red to green. Because the cluster will hold more documents like 10TB in the future , I wonder if there are any methods or settings to speed up the restart process? Following is our index settings, all index settings are the same.

{
      "index" : {
        "search" : {
          "slowlog" : {
            "threshold" : {
              "fetch" : {
                "warn" : "1s",
                "trace" : "200ms",
                "debug" : "500ms",
                "info" : "800ms"
              },
              "query" : {
                "warn" : "20s",
                "trace" : "2s",
                "debug" : "5s",
                "info" : "10s"
              }
            }
          }
        },
        "indexing" : {
          "slowlog" : {
            "threshold" : {
              "index" : {
                "warn" : "10s",
                "trace" : "500ms",
                "debug" : "2s",
                "info" : "5s"
              }
            }
          }
        },
        "number_of_shards" : "1",
        "provided_name" : "7b2c96f57ea1_nic1_4h_2017121016",
        "creation_date" : "1512892957368",
        "number_of_replicas" : "0",
        "uuid" : "9xm9ZANUS_ufWqGx05EmdQ",
        "version" : {
          "created" : "5030099"
        }
      }
    }
  }
2

Having multiple nodes would speed this up.

3

Thank you for the quick reply. Do you mean add another machine or use 2 nodes in one machine?

4

It depends if you are restarting the Elasticsearch service, or the entire host.

Mind you, you will be lucky to get 10TB to an Elasticsearch node, there are limitations in the JVM that will stop you before that density.

5

With that powerful machine, why not try and set up more than one node on that machine! Have a look if you can dockerise your setup!

6

I am restarting the elatsicsearch service by systemctl. So you mean there is a limitation for the size of the node?

7

Actually I have tried this and ask question here about the multiple-nodes archtechture. And the results seems it does not help the performance in my situation, the only good is the gc count and time.

8

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