How to quickly change the number of replicas setting in an Elasticsearch Index

Written by Claudio Kuenzler - 0 comments

Published on July 19th 2021 - Listed in Elasticsearch ELK Monitoring

Depending on the environment where an Elasticsearch cluster is running, the (default) index settings might need to be changed. Especially if the cluster only exists with a one-node setup (which is nothing abnormal) indexes with a replica setting of 1 or more will report unassigned shards.

This can be verified with the check_es_system Elasticsearch monitoring plugin:

root@elasticsearch:~# /usr/lib/nagios/plugins/check_es_system.sh -H localhost -t status
ES SYSTEM WARNING - Elasticsearch Cluster "mycluster" is yellow (1 nodes, 1 data nodes, 3 shards, 0 relocating shards, 0 initializing shards, 3 unassigned shards, 10934 docs)

Unassigned shards and replica settings

The reason can be spotted when listing all the available shards:

root@elasticsearch:~# curl -q -s "http://localhost:9200/_cat/shards"
magento2_product_1_v204 0 p STARTED    3639 6.2mb 127.0.0.1 elasticsearch
magento2_product_1_v204 0 r UNASSIGNED                      
magento2_product_2_v204 0 p STARTED    3639 7.1mb 127.0.0.1 elasticsearch
magento2_product_2_v204 0 r UNASSIGNED                      
magento2_product_3_v204 0 p STARTED    3656 6.2mb 127.0.0.1 elasticsearch
magento2_product_3_v204 0 r UNASSIGNED  

All the unassigned shards have "r" as second flag - which stands for replicated shard. As this is a single node cluster, Elasticsearch does not find any other nodes where to replicate this shard to -> UNASSIGNED.

By looking at the settings of these three indexes, especially the number_of_replicas setting, all indexes have a (default) setting of 1 replica defined:

root@elasticsearch:~# curl -q -s "http://localhost:9200/magento2_product_1_v204/_settings" | jq -r '.magento2_product_1_v204.settings.index.number_of_replicas'
1

root@elasticsearch:~# curl -q -s "http://localhost:9200/magento2_product_2_v204/_settings" | jq -r '.magento2_product_2_v204.settings.index.number_of_replicas'
1

root@elasticsearch:~# curl -q -s "http://localhost:9200/magento2_product_3_v204/_settings" | jq -r '.magento2_product_2_v204.settings.index.number_of_replicas'
1

Changing the number_of_replicas setting

Fortunately this setting can be changed without downtime. With a PUT request the index settings can be updated:

root@elasticsearch:~# curl -X PUT -H "Content-Type: application/json" "http://localhost:9200/magento2_product_1_v204/_settings" -d '{ "number_of_replicas": 0 }'
{"acknowledged":true}

Verification of the current setting value:

root@elasticsearch:~# curl -q -s "http://localhost:9200/magento2_product_1_v204/_settings" | jq -r '.magento2_product_1_v204.settings.index.number_of_replicas'
0

Yes, it worked!

Now looking at the shards again, we can see there are only two unassigned shards left:

root@elasticsearch:~# curl -q -s "http://localhost:9200/_cat/shards"
magento2_product_1_v204 0 p STARTED    3639 6.2mb 127.0.0.1 elasticsearch
magento2_product_2_v204 0 p STARTED    3639 7.1mb 127.0.0.1 elasticsearch
magento2_product_2_v204 0 r UNASSIGNED                      
magento2_product_3_v204 0 p STARTED    3656 6.2mb 127.0.0.1 elasticsearch
magento2_product_3_v204 0 r UNASSIGNED  

The number_of_replicas setting is now also changed on the two remaining indexes:

root@elasticsearch:~# curl -X PUT -H "Content-Type: application/json" "http://localhost:9200/magento2_product_2_v204/_settings" -d '{ "number_of_replicas": 0 }'
{"acknowledged":true}

root@elasticsearch:~# curl -X PUT -H "Content-Type: application/json" "http://localhost:9200/magento2_product_3_v204/_settings" -d '{ "number_of_replicas": 0 }'
{"acknowledged":true}

And voilà, no more unassigned shards are shown:

root@elasticsearch:~# curl -q -s "http://localhost:9200/_cat/shards"
magento2_product_1_v204 0 p STARTED 3639 6.2mb 127.0.0.1 elasticsearch
magento2_product_2_v204 0 p STARTED 3639 7.1mb 127.0.0.1 elasticsearch
magento2_product_3_v204 0 p STARTED 3656 6.2mb 127.0.0.1 elasticsearch

Elasticsearch Status change

Now with all unassigned shards gone, the Elasticsearch status immediately changed to green:

root@elasticsearch:~# /usr/lib/nagios/plugins/check_es_system.sh -H localhost -t status
ES SYSTEM OK - Elasticsearch Cluster "mycluster" is green (1 nodes, 1 data nodes, 3 shards, 10934 docs)|total_nodes=1;;;; data_nodes=1;;;; total_shards=3;;;; relocating_shards=0;;;; initializing_shards=0;;;; unassigned_shards=0;;;; docs=10934;;;;

Add a comment

Comments (newest first)

No comments yet.