OpenSearch (search service)

OpenSearch is a distributed RESTful search engine built for the cloud.

See the OpenSearch documentation for more information.

To switch from Elasticsearch, follow the same procedure as for upgrading.

Supported versions 

Grid Dedicated Gen 3 Dedicated Gen 2
  • 2
  • 1.2
  • 1.1
  • 2
  • 1.2
  • 1.2

From version 2, you only specify the major version. The latest compatible minor version and patches are applied automatically.

The following versions aren’t available in the EU-1 and US-1 regions:

  • 2

  • 1.2

Consider region migration if your project is in those regions.


The format exposed in the $PLATFORM_RELATIONSHIPS environment variable:

    "username": null,
    "scheme": "http",
    "service": "opensearch12",
    "fragment": null,
    "ip": "",
    "hostname": "",
    "port": 9200,
    "cluster": "rjify4yjcwxaa-master-7rqtwti",
    "host": "opensearch.internal",
    "rel": "opensearch",
    "path": null,
    "query": [],
    "password": null,
    "type": "opensearch:1.2",
    "public": false,
    "host_mapped": false

Usage example 

1. Configure the service 

To define the service, use the opensearch type:

    type: opensearch:<VERSION>
    disk: 256

Note that if you later change the name, it’s treated as an entirely new service. This removes all data from your service. Always backup your data before changing the service.

2. Add the relationship 

To define the relationship, use the opensearch endpoint :

You can define <SERVICE_NAME> and <RELATIONSHIP_NAME> as you like, but it’s best if they’re distinct.

Example Configuration 

Service definition

    type: opensearch:2
    disk: 256

App configuration
    ossearch: "searchopen:opensearch"


By default, OpenSearch has no authentication. No username or password is required to connect to it.

You may optionally enable HTTP Basic authentication. To do so, include the following in your services.yaml configuration:

    type: opensearch:1.2
    disk: 2048
            enabled: true

That enables mandatory HTTP Basic auth on all requests. The credentials are available in any relationships that point at that service, in the username and password properties.

This functionality is generally not required if OpenSearch isn’t exposed on its own public HTTP route. However, certain applications may require it, or it allows you to safely expose OpenSearch directly to the web. To do so, add a route to routes.yaml that has search:opensearch as its upstream (where search is whatever you named the service in services.yaml). For example:

    type: upstream
    upstream: search:opensearch


OpenSearch offers a number of plugins. To enable them, list them under the configuration.plugins key in your services.yaml file, like so:

    type: "opensearch:1.2"
    disk: 1024
            - analysis-icu
            - mapper-size

In this example you’d have the ICU analysis plugin and the size mapper plugin.

If there is a publicly available plugin you need that isn’t listed here, contact support.

Available plugins 

This is the complete list of plugins that can be enabled:

Plugin Description 1.2
analysis-icu Support ICU Unicode text analysis *
analysis-kuromoji Japanese language support *
analysis-nori Integrates Lucene Nori analysis module into OpenSearch *
analysis-phonetic Phonetic analysis *
analysis-smartcn Smart Chinese Analysis Plugins *
analysis-stempel Stempel Polish Analysis Plugin *
analysis-ukrainian Ukrainian language support *
ingest-attachment Extract file attachments in common formats (such as PPT, XLS, and PDF) *
mapper-annotated-text Adds support for text fields with markup used to inject annotation tokens into the index *
mapper-murmur3 Murmur3 mapper plugin for computing hashes at index-time *
mapper-size Size mapper plugin, enables the _size meta field *
repository-s3 Support for using S3 as a repository for Snapshot/Restore *
transport-nio Support for NIO transport *

Plugin removal 

Removing plugins previously added in your services.yaml file doesn’t automatically uninstall them from your OpenSearch instances. This is deliberate, as removing a plugin may result in data loss or corruption of existing data that relied on that plugin. Removing a plugin usually requires reindexing.

To permanently remove a previously enabled plugin, upgrade the service to create a new instance of OpenSearch and migrate to it. In most cases it isn’t necessary as an unused plugin has no appreciable impact on the server.


The OpenSearch data format sometimes changes between versions in incompatible ways. OpenSearch doesn’t include a data upgrade mechanism as it’s expected that all indexes can be regenerated from stable data if needed. To upgrade (or downgrade) OpenSearch, use a new service from scratch.

There are two ways to do so.


In your services.yaml file, change the version and name of your OpenSearch service. Then update the name in the relationships block.

When you push that to, the old service is deleted and a new one with the new name is created with no data. You can then have your application reindex data as appropriate.

This approach has the downsides of temporarily having an empty OpenSearch instance, which your application may or may not handle gracefully, and needing to rebuild your index afterward. Depending on the size of your data that could take a while.


With a transitional approach, you temporarily have two OpenSearch services. Add a second OpenSearch service with the new version a new name and give it a new relationship in You can optionally run in that configuration for a while to allow your application to populate indexes in the new service as well.

Once you’re ready to switch over, remove the old OpenSearch service and relationship. You may optionally have the new OpenSearch service use the old relationship name if that’s easier for your app to handle. Your application is now using the new OpenSearch service.

This approach has the benefit of never being without a working OpenSearch instance. On the downside, it requires two running OpenSearch servers temporarily, each of which consumes resources and needs adequate disk space. Depending on the size of your data, that may be a lot of disk space.