Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


Elasticsearch plugin for Para

Build Status Maven Central Join the chat at

What is this?

Para was designed as a simple and modular back-end framework for object persistence and retrieval. It enables your application to store objects directly to a data store (NoSQL) or any relational database (RDBMS) and it also automatically indexes those objects and makes them searchable.

This plugin allows you to use Elasticsearch as the search engine for Para.


  • Implements the Para Search interface using the official Elasticsearch Java client
  • Supports both Elasticsearch and OpenSearch servers
  • Index sharing and multitenancy support through alias routing and filtering
  • Supports both asynchronous and synchronous document operations
  • Full pagination support for both "search-after" and "from-size" modes
  • Proxy endpoint /v1/_elasticsearch - relays all requests directly to Elasticsearch (disabled by default)
  • Supports AWS Elasticsearch Service with request signing


ES Plugin version Elasticsearch support OpenSearch support
1.40.0 and above 8.x and above (with config flag) 1.0.0 and above (with config flag)
1.39.0 and below up to 7.15.2 1.0.0 and above

After version 1.40.0 a configuration flag is required to destinguish between the two different flavors of Eleasticsearch: = "elasticsearch"
# ==== OR ==== # = "opensearch"

The default option here is elasticsearch.


Read the Docs

Getting started

The plugin is on Maven Central. Here's the Maven snippet to include in your pom.xml:


Alternatively you can download the JAR from the "Releases" tab above put it in a lib folder alongside the server WAR file para-x.y.z.war. Para will look for plugins inside lib and pick up the Elasticsearch plugin.


Here are all the configuration properties for this plugin (these go inside your application.conf):

# ES flavor - elasticsearch or opensearch = "elasticsearch"

# enable this to bypass the DB and read all data straight from ES
para.read_from_index = false = 5 = 0 = "data" = "0-1" = "http" = "localhost" = 9200
# context path of elasticsearch e.g. /es = "" = false = "eu-west-1" = false = 10000
# if login and password are filled then add for each request
# the Authorization header with basic auth = "" = ""

# asynchronous settings = false = 5 = 1000 = 1 = 5000 = 50 = 8 = false

# proxy settings = false = "_elasticsearch" = false

Finally, set the config property: = "ElasticSearch"

This could be a Java system property or part of a application.conf file on the classpath. This tells Para to use the Elasticsearch implementation instead of the default (Lucene).

SSL connection with Elasticsearch

Newer versions of ES are using HTTPS by default and generate a unique SSL certificate on first startup. To have a successful connection, you should first add the generated CA of the ES server to your Java keystore, indicating that you trust that certificate. This is the console command to do that:

echo -n | \
openssl s_client -connect localhost:9200 | \
sed -ne '/-BEGIN CERTIFICATE-/,/-END CERTIFICATE-/p' > ./ca_elasticsearch.cer && \
keytool -import -alias saelk -file ca_elasticsearch.cer -keystore ${JAVA_HOME}/lib/security/cacerts -storepass changeit

Finally, set = "https".

Synchronous versus Asynchronous Indexing

The Elasticsearch plugin supports both synchronous (default) and asynchronous indexing modes. For synchronous indexing, the Elasticsearch plugin will make a single, blocking request through the client and wait for a response. This means each document operation (index, reindex, or delete) invokes a new client request. For certain applications, this can induce heavy load on the Elasticsearch cluster. The advantage of synchronous indexing, however, is the result of the request can be communicated back to the client application. If the setting is set to true, synchronous requests that result in an error will propagate back to the client application with an HTTP error code.

The asynchronous indexing mode uses the Elasticsearch BulkProcessor for batching all requests to the Elasticsearch cluster. If the asynchronous mode is enabled, all document requests will be fed into the BulkProcessor, which will flush the requests to the cluster on occasion. There are several configurable parameters to control the flush frequency based on document count, total document size (MB), and total duration (ms). Since Elasticsearch is designed as a near real-time search engine, the asynchronous mode is highly recommended. Making occasional, larger batches of document requests will help reduce the load on the Elasticsearch cluster.

The asynchronous indexing mode also offers an appealing feature to automatically retry failed indexing requests. If your Elasticsearch cluster is under heavy load, it's possible a request to index new documents may be rejected. With synchronous indexing, the burden falls on the client application to try the indexing request again. The Elasticsearch BulkProcessor, however, offers a useful feature to automatically retry indexing requests with exponential backoff between retries. If the index request fails with a EsRejectedExecutionException, the request will be retried up to times. Even if your use case demands a high degree of confidence with respect to data consistency between your database (DAO) and index (Search), it's still recommended to use asynchronous indexing with retries enabled. If you'd prefer to use asynchronous indexing but have the BulkProcessor flushed upon every invocation of index/unindex/indexAll/unindexAll, simply enabled When this option is enabled, the BulkProcessor's flush method will be called immediately after adding the documents in the request. This option is also useful for writing unit tests where you want ensure the documents flush promptly.

Indexing modes

This plugin has two indexing modes: normal and nested. The nested mode was added after v1.28 to protect against a possible mapping explosion which happens when there are lots of objects with lots of different custom properties in them. This overloads the Elasticsearch index metadata and can crash the whole cluster. This indexing mode affects only custom properties in Sysprop objects.

The old "normal" mode is suitable for most Para deployments, with just a few tenants or a single tenant (one app per server). In this mode, Para objects are indexed without modification (all data types are preserved) but this could lead to a mapping explosion.

The nested data structure for these two indexing modes is shown below:

// NORMAL MODE                   // NESTED MODE
{                                {
  "id": "123",                     "id": "123",
  "appid": "para",                 "appid": "para",
  "type": "custom",                "type": "custom",
  "properties": {                  "properties": [
    "key1": "value1",                {"k": "key1",         "v": "value1"},
    "key2": {                        {"k": "key2-subkey1", "v": "subValue1"},
      "subkey1": "subValue1"         {"k": "numericKey3",  "vn": 5}
    },                             ],
    "numericKey3": 5               "_properties": "{\"key1\":\"value1\"}..."
  }                              }

Switching to the new nested indexing mode is done with the configuration property: = true

Another benefit, when using the "nested" mode, is the support for nested queries in query strings. This is a really useful feature which, at the time of writing this, has not yet been implemented in Elasticsearch (issue elastic/elasticsearch#11322). Even better, you can query objects within nested arrays with pinpoint precision, e.g. ?q=properties.nestedArray[2].key:value. A nested query string query is detected if it contains a field with prefix properties.*. Examples of query string queries:

/v1/search?q=term AND properties.owner.age:[* TO 34]
/v1/search? OR properties.owner.pets[1].name=whiskers

Note: Sorting on nested fields works only with numeric data. For example, sorting on a field properties.year will work, but sorting on properties.month won't (applicable only to the "nested" mode).

Calling Elasticsearch through the proxy endpoint

You can directly call the Elasticsearch API through /v1/_elasticsearch. To enable it set = true first. Then you must specify the path parameter corresponds to the Elasticsearch API resource path. This is done for every GET, PUT, POST, PATCH or DELETE request to Elasticsearch. The endpoint accepts request to either /v1/_elasticsearch or /v1/_elasticsearch/{path} where path is a URL-encoded path parameter. Do not add query parameters to the request path with ?, instead, pass them as a parameter map.

GET /v1/_elasticsearch/_search
GET /v1/_elasticsearch/mytype%2f_search
DELETE /v1/_elasticsearch/tweet%2f1

ParaClient example:

Response get = paraClient.invokeGet("_elasticsearch/" + Utils.urlEncode("tweet/_search"), params);

Response post = paraClient.invokePost("_elasticsearch/_count",
               Collections.singletonMap("type", "cat")))));

If the path parameter is omitted, it defaults to _search.

The response object will be transformed to be compatible with Para clients an looks like this:


If you wish to get the raw query response from Elasticsearch, add the parameter getRawResponse=true to the requst path and also URL-encode it:

GET /v1/_elasticsearch/mytype%2f_search%3FgetRawResponse%3Dtrue

Equivalently, the same can be done by adding the query parameter using ParaClient:

MultivaluedHashMap<String, String> params = new MultivaluedHashMap<>();
params.putSingle("getRawRequest", "true");
paraClient.invokeGet("_elasticsearch/" + Utils.urlEncode("mytype/_search"), params);

Note: This endpoint requires authentication and unsigned requests are not allowed. Keep in mind that all requests to Elasticsearch are prefixed with the app identifier. For example if the app id is "app:myapp, then Para will proxy requests to Elasticsearch at http://eshost:9200/myapp/{path}.

Rebuilding indices through the Elasticsearch proxy endpoint

You can rebuild the whole app index from scratch by calling POST /v1/_elasticsearch/reindex. To enable it set = true first. This operation executes ElasticSearchUtils.rebuildIndex() internally, and returns a response indicating the number of reindexed objects and the elapsed time:

   "reindexed": 154,
   "tookMillis": 365

Additionally, you can specify the destination index to reindex into, which must have been created beforehand:

POST /v1/_elasticsearch/reindex?destinationIndex=yourCustomIndex

Shared indices with alias routing

The plugin also supports index sharing, whereby the root app index is shared with other apps which are created with the flag app.isSharingIndex = true. This feature is enabled with = true and it is off by default. When the root index is created with sharing enabled, a special alias is created for it that contains a routing field which sends all documents of a child app to a particular shard, while providing total isolation between apps. This is useful when there are lots of smaller apps with just a few hundred documents each and we want to avoid the overhead of one index per app.

Deprecation notice

  • Support for the ES Transport client has been removed because it is now deprecated and removed in ES 7.0.
  • The previously bundled IndexBasedDAO has been removed because it had lots of issues.


  • Elasticsearch official Java client
  • Para Core


Apache 2.0