Upgrade from Elasticsearch OSS to OpenSearch

If you want to upgrade from an existing Elasticsearch OSS cluster to OpenSearch and find the snapshot approach unappealing, you can upgrade your existing nodes from Elasticsearch OSS to OpenSearch.

If your existing cluster runs an older version of Elasticsearch OSS, the first step is to upgrade to version 6.x or 7.x. Elasticsearch OSS supports two types of upgrades: rolling and cluster restart.

• Rolling upgrades let you shut down one node at a time for minimal disruption of service.

  Rolling upgrades work between minor versions (e.g. 6.5 to 6.8) and also support a single path to the next major version (e.g. 6.8 to 7.10.2). Performing these upgrades might require intermediate upgrades to arrive at your desired version and can affect cluster performance as nodes leave and rejoin, but the cluster remains available throughout the process.

• Cluster restart upgrades require you to shut down all nodes, perform the upgrade, and restart the cluster.

  Cluster restart upgrades work between minor versions (e.g. 6.5 to 6.8) and the next major version (for example, 6.x to 7.10.2). Cluster restart upgrades are faster to perform and require fewer intermediate upgrades, but require downtime.

Upgrade paths

Upgrade Elasticsearch OSS

1. Disable shard allocation to prevent Elasticsearch OSS from replicating shards as you shut down nodes:

   PUT _cluster/settings
   {
     "persistent": {
       "cluster.routing.allocation.enable": "primaries"
     }
   }

2. Stop Elasticsearch OSS on one node (rolling upgrade) or all nodes (cluster restart upgrade).

   On Linux distributions that use systemd, use this command:

   sudo systemctl stop elasticsearch.service

   For tarball installations, find the process ID (ps aux) and kill it (kill <pid>).

3. Upgrade the node (rolling) or all nodes (cluster restart).

   The exact command varies by package manager, but likely looks something like this:

   sudo yum install elasticsearch-oss-7.10.2 --enablerepo=elasticsearch

   For tarball installations, extract to a new directory to ensure you do not overwrite your config, data, and logs directories. Ideally, these directories should have their own, independent paths and not be colocated with the Elasticsearch application directory. Then set the ES_PATH_CONF environment variable to the directory that contains elasticsearch.yml (e.g. /etc/elasticesarch/). In elasticsearch.yml, set path.data and path.logs to your data and logs directories (e.g. /var/lib/elasticsearch and /var/log/opensearch).

4. Restart Elasticsearch OSS on the node (rolling) or all nodes (cluster restart).

   On Linux distributions that use systemd, use this command:

   sudo systemctl start elasticsearch.service

   For tarball installations, run ./bin/elasticsearch -d.

5. Wait for the node to rejoin the cluster (rolling) or for the cluster to start (cluster restart). Check the _nodes summary to verify that all nodes are available and running the expected version:

   # Elasticsearch OSS
   curl -XGET 'localhost:9200/_nodes/_all?pretty=true'
   # Open Distro for Elasticsearch with security plugin enabled
   curl -XGET 'https://localhost:9200/_nodes/_all?pretty=true' -u 'admin:admin' -k

   Specifically, check the nodes.<node-id>.version portion of the response. Also check _cat/indices?v for a green status on all indices.

6. (Rolling) Repeat steps 2–5 until all nodes are using the new version.

7. After all nodes are using the new version, re-enable shard allocation:

   PUT _cluster/settings
   {
     "persistent": {
       "cluster.routing.allocation.enable": "all"
     }
   }

8. If you upgraded from 5.x to 6.x, reindex all indices.

9. Repeat all steps as necessary until you arrive at your desired Elasticsearch OSS version.

Upgrade to OpenSearch

1. Disable shard allocation to prevent Elasticsearch OSS from replicating shards as you shut down nodes:

   PUT _cluster/settings
   {
     "persistent": {
       "cluster.routing.allocation.enable": "primaries"
     }
   }

2. Stop Elasticsearch OSS on one node (rolling upgrade) or all nodes (cluster restart upgrade).

   On Linux distributions that use systemd, use this command:

   sudo systemctl stop elasticsearch.service

   For tarball installations, find the process ID (ps aux) and kill it (kill <pid>).

3. Upgrade the node (rolling) or all nodes (cluster restart).

   1. Extract the OpenSearch tarball to a new directory to ensure you do not overwrite your Elasticsearch OSS config, data, and logs directories.

   2. (Optional) Copy or move your Elasticsearch OSS data and logs directories to new paths. For example, you might move /var/lib/elasticsearch to /var/lib/opensearch.

   3. Set the OPENSEARCH_PATH_CONF environment variable to the directory that contains opensearch.yml (e.g. /etc/opensearch).

   4. In opensearch.yml, set path.data and path.logs. You might also want to disable the security plugin for now. opensearch.yml might look something like this:

      path.data: /var/lib/opensearch
      path.logs: /var/log/opensearch
      plugins.security.disabled: true

   5. Port your settings from elasticsearch.yml to opensearch.yml. Most settings use the same names. At a minimum, specify cluster.name, node.name, discovery.seed_hosts, and cluster.initial_master_nodes.

   6. (Optional) Add your certificates to your config directory, add them to opensearch.yml, and initialize the security plugin.

4. Start OpenSearch on the node (rolling) or all nodes (cluster restart).

   For the tarball, run ./bin/opensearch -d.

5. Wait for the OpenSearch node to rejoin the cluster (rolling) or for the cluster to start (cluster restart). Check the _nodes summary to verify that all nodes are available and running the expected version:

   # Security plugin disabled
   curl -XGET 'localhost:9200/_nodes/_all?pretty=true'
   # Security plugin enabled
   curl -XGET -k -u 'admin:admin' 'https://localhost:9200/_nodes/_all?pretty=true'

   Specifically, check the nodes.<node-id>.version portion of the response. Also check _cat/indices?v for a green status on all indices.

6. (Rolling) Repeat steps 2–5 until all nodes are using OpenSearch.

7. After all nodes are using the new version, re-enable shard allocation:

   PUT _cluster/settings
   {
     "persistent": {
       "cluster.routing.allocation.enable": "all"
     }
   }