Upgrading Elasticsearch

This is the process for in-place upgrades of Elasticsearch (ES), meaning ones that can use the existing Search feed and index data.

Required

This upgrade is for Elasticsearch 6.8.6 and higher only, with a Search feed created on Swarm 11.
For migrating from Elasticsearch 2.3.3 or 5.6.12, see Migrating from Older Elasticsearch.

Upgrading Elasticsearch by script

On each node in your Elasticsearch cluster, follow this process and run the files from your Swarm download bundle:

Query the running Elasticsearch cluster, before upgrading, for the list of nodes.
```
curl -i http://ELASTICSEARCH:9200/_cat/nodes
```
Note which node is starred. That is the Elasticsearch master node which you should upgrade last to avoid problems electing a new one.
Backup your existing elasticsearch.yml, so that you have a record of any customizations you may have made.
If you have customized path.data, create a symbolic (soft) link: symlink /var/lib/elasticsearch. If you cannot, perform the upgrade manually, as described below.
When upgrading Elasticsearch 6, start by installing the latest Swarm Search, which is the caringo-elasticsearch-search RPM.
```
yum install caringo-elasticsearch-search-VERSION.noarch.rpm
```
Tip

If you inadvertently install the Elasticsearch 7 RPM, it will fail with the error: "ES_PATH_CONF must be set to the configuration path chown: cannot access '/etc/elasticsearch/elasticsearch.keystore': No such file or directory". Simply install the caringo-elasticsearch-search-7.0.0 RPM now and proceed.
Run the script that installs and configures the upgrade.
The script detects that Elasticsearch 6 is installed and that discovery.zen.unicast.hosts is configured, so it runs as with --upgrade instead of configuring a new cluster.
```
/usr/share/caringo-elasticsearch-search/bin/configure_elasticsearch_with_swarm_search.py
```
Compare your backup file to the newly created elasticsearch.yml and add back any customizations you need, such as network.host (which defaults to _site_, meaning all interfaces).
Verify all nodes are accounted for, that all shards are assigned, and that the status is green.
```
curl -i 'http://ELASTICSEARCH:9200/_cat/health?v'
```

If Elasticsearch 7 is already installed, the script updates the configuration files and restarts the service for you.

Troubleshooting

If the Elasticsearch service fails and journalctl -u elasticsearch shows that access is denied (BootstrapException/AccessDeniedException), change its permissions:

chown elasticsearch /etc/elasticsearch

Important

If the cluster loses master during the upgrade process and does not recover, that’s fine, just hit Ctrl-C once, whenever the upgrade script is stuck in retrying status checks and proceed to the next node after the script finishes. Review /etc/elasticsearch/elasticsearch.yml and tail/var/log/elasticsearch/<cluster-name>.log for configuration errors.

Once all nodes have started on Elasticsearch 7.5.2, they should be able to elect the master and recover. Health status should go yellow then eventually green. You will have to re-enable shard allocation, otherwise /_cat/health?v will stop at 50% with health status yellow.

Upgrading Elasticsearch manually

If you should need to upgrade manually, these are the steps that the script automates:

It fixes /etc/sysconfig/elasticsearch to be the correct ES6 version (it's also the same as ES7).
It increases the systemd timeout in /etc/systemd/system/elasticsearch.service.d/override.conf (see github.com/elastic/elasticsearch/issues/60140)
After refreshing the config files for Elasticsearch 6, it prompts to continue with the yum upgrade to 7.5.2.
It disables shard allocation and does a POST synced-flush for safer rolling upgrades.
Important

Disabling shard allocation or sync-flush can fail to contact the node, but do not proceed to upgrade the next node until the cluster health is green again.
It uninstalls the Prometheus Exporter plugin if it exists.
It shells out to yum to install the Elasticsearch 7 RPM in the current directory or from artifacts.elastic.co, if unavailable.

We expect Internet access by default for the upgrade. In case it is not available, ensure that the Elasticsearch RPM is in the current directory.
It updates elasticsearch.yml for version 7 compatibility, including discovery.initial_master_nodes instead of discovery.zen.unicast.hosts, and jvm.options.
It starts the upgraded Elasticsearch 7 and waits for it to be ready.
If the cluster health is green or yellow, it re-enables shard allocation and prompts you to repeat these two steps on the next node.