Release notes

Stops retrying timed-out requests by default. Re-enabling the old behavior can be done by providing a retryOnTimeout option when instantiating the client. #2293

Fixes a bug where newly-added nodes that don’t have HTTP information yet should be skipped during sniffing, to prevent an unexpected TypeError. #1230

This is the first 7.x release where patch versions of the client will no longer (intentionally) align with patch versions of Elasticsearch. The latest patch release of the client will always be compatible with the corresponding minor release of Elasticsearch.

Fixes a type declaration bug that caused TypeScript builds to fail. #1927

Adds TypeScript type declarations file to package.json exports #1930

You can find all the API changes here.

Fixes a bug in the bulk helper that would cause onDrop to send back the wrong JSON document or error on a nonexistent document when an error occurred on a bulk HTTP request that contained a delete action.

You can find all the API changes here.

You can find all the API changes here.

You can find all the API changes here.

If you call an API that returns a mapbox content type, the response body will be a buffer.

You can configure the client to only trust certificates that are signed by a specific CA certificate ( CA certificate pinning ) by providing a caFingerprint option. This will verify that the fingerprint of the CA certificate that has signed the certificate of the server matches the supplied value. a caFingerprint option, which will verify the supplied certificate authority fingerprint. You must configure a SHA256 digest.

const { Client } = require('@elastic/elasticsearch')
const client = new Client({
  node: 'https://example.com'
  auth: { ... },
  // the fingerprint (SHA256) of the CA certificate that is used to sign the certificate that the Elasticsearch node presents for TLS.
  caFingerprint: '20:0D:CA:FA:76:...',
  ssl: {
    // might be required if it's a self-signed certificate
    rejectUnauthorized: false
  }
})

Useful if the errored response does not come from Elasticsearch, but a proxy in the middle for example.

In some edge cases the params and options weren’t available in observabilty events, now they are always defined.

If the client is busy running an async operation, the .abort() call might be executed before sending the actual request. In such case, the error was swallowed, now it will always be emitted, either in the request or response event.

You can find all the API changes here.

The client will verify if it’s working with a supported release of Elasticsearch. Elastic language clients are guaranteed to be able to communicate with Elasticsearch or Elastic solutions running on the same major version and greater or equal minor version.

Language clients are forward compatible; meaning that clients support communicating with greater minor versions of Elasticsearch. Elastic language clients are not guaranteed to be backwards compatible.

If you configure the ELASTIC_CLIENT_APIVERSIONING to true the client will send a compatibility header to allow you to use a 7.x client against a 8.x cluster. In this way it will be easier to migrate your code to a newer release of Elasticsearch.

Bearer authentication, useful for service account tokens. Be aware that it does not handle automatic token refresh:

auth: {
  bearer: 'token'
}

The final stats object will let you know how many noop operations happened. Also, a new .stats getter has been added to allow you to read the stats before the operation finishes.

const b = client.helpers.bulk({ ... })
...
console.log(b.stats)

According to our support matrix.

You can find all the API changes here.

The new type definition is more advanced compared to the legacy one. In the legacy type definitions you were expected to configure via generics both request and response bodies. The new type definitions comes with a complete type definition for every Elasticsearch endpoint.

You can see how to use them now here.

In case of Elasticsearch errors, now the error message show more info about the underlying issue, improving the debugging experience.

In case of http errors in HEAD request, the client was swalling the response body. This is now fixed and in case of error you will get the full body response.

According to our support matrix.

You can find all the API changes here.

You can now pass Transport specific options to the helpers as well.

The client returns a thenable object when you are not configuring a callback. Now the thenable offers a .finally method as well.

You can find all the API changes here.

Two new observability events has been introduced: serialization and deserialization. The event order is described in the following graph, in some edge cases, the order is not guaranteed. You can find in test/acceptance/events-order.test.js how the order changes based on the situation.

serialization
  │
  │ (serialization and compression happens between those two events)
  │
  └─▶ request
        │
        │ (actual time spent over the wire)
        │
        └─▶ deserialization
              │
              │ (deserialization and decompression happens between those two events)
              │
              └─▶ response

Adds the x-elastic-client-meta HTTP header which is used by Elastic Cloud and can be disabled with the enableMetaHeader parameter set to false.

When using a body that is a stream to client.search(), and calling req.abort(), the callback is called twice. Once for the RequestAbortedError, as expected, and once for a "premature close" error from end-of-stream, used by pump, used by the client. This issue has now been fixed.

You can find all the API changes here.

If you need to pass through an http(s) proxy for connecting to Elasticsearch, the client offers out of the box a handy configuration for helping you with it. Under the hood it uses the hpagent module.

const client = new Client({
  node: 'http://localhost:9200',
  proxy: 'http://localhost:8080'
})

Basic authentication is supported as well:

const client = new Client({
  node: 'http://localhost:9200',
  proxy: 'http://user:pwd@localhost:8080'
})

From now on the scroll search helper will automatically close the scroll on Elasticsearch, by doing so, Elasticsearch will free resources faster.

It might happen that the underlying socket stops working due to an external cause while reading the body. This could lead to an unwanted DeserialzationError. From now, this will be handled as a generic ConnectionError.

7.11 will be the last version of the client that will support Node.js v8, while 7.12 will be the last one that supports Node.js v10. If you are using this versions you will see a DeprecationWaring in your logs. We strongly recommend to upgrade to newer versions of Node.js as usng an EOL version will expose you to securty risks.

Please refer to ela.st/nodejs-support for additional information.

The client code has been refactored to speed up the performances of the child method. Before this pr, creating many children per second would have caused a high memory consumption and a spike in CPU usage. This pr changes the way the client is created by refactoring the code generation, now the clients methods are no longer added to the instance with a for loop but via prototypal inheritance. Thus, the overall performances are way better, now creating a child is ~5 times faster, and it consumes ~70% less memory.

This change should not cause any breaking change unless you were mocking the client methods. In such case you should refactor it, or use elasticsearch-js-mock.

Finally, this change should also fix once and of all the bundlers support.

Some validation errors were thrown synchronously, causing the callback to be called in th same tick. This issue is known as "The release fo Zalgo" (see here).

The maxRetries parameter can be configured on a per requets basis, if set to zero it was defaulting to the client default. Now the client is honoring the request specific configuration.

The Connection requets option types were not accepting null as valid value.

The size parameter was being passed too the scroll request, which was causing an error. Value of maxRetries set to 0 was resulting in no request at all.

If needed, the http agent can be disabled by setting it to false.

const { Client } = require('@elastic/elasticsearch')
const client = new Client({
  node: 'http://localhost:9200'.
  agent: false
})

Before this, you could set a context option in each request, but there was no way of setting it globally. Now you can by configuring the context object in the global configuration, that will be merged with the local one.

const { Client } = require('@elastic/elasticsearch')
const client = new Client({
  node: 'http://localhost:9200'.
  context: { meta: 'data' }
})

If you are using ES Modules, now you can easily import the client!

import { Client } from '@elastic/elasticsearch'

It was possible in plain JavaScript, but not in TypeScript, now you can do it in TypeScript as well.

const { Client } = require('@elastic/elasticsearch')
const client = new Client({
  node: 'http://localhost:9200',
  name: Symbol('unique')
})

Only Record<string, any> was allowed. Now string is allowed as well.

All HEAD request will have the body casted to a boolean value, true in case of a 200 response, false in case of a 404 response. The type definitions were not reflecting this behavior.

import { Client } from '@elastic/elasticsearch'
const client = new Client({
  node: 'http://localhost:9200'
})

const { body } = await client.exist({ index: 'my-index', id: 'my-id' })
console.log(body) // either `true` or `false`

Added the scheduling: lifo option to the default HTTP agent configuration to avoid maximizing the open sockets against Elasticsearch and lowering the risk of encountering socket timeouts. This feature is only available from Node v14.5+, but it should be backported to v10 and v12 (nodejs/node#33278).

This pr introduce two changes which should not impact the surface API:

You can find all the API changes here.

If you are sending search request at a high rate, this helper might be useful for you. It will use the mutli search API under the hood to batch the requests and improve the overall performances of your application. The result exposes a documents property as well, which allows you to access directly the hits sources.

const { Client } = require('@elastic/elasticsearch')

const client = new Client({ node: 'http://localhost:9200' })
const m = client.helpers.msearch()

// promise style API
m.search(
    { index: 'stackoverflow' },
    { query: { match: { title: 'javascript' } } }
  )
  .then(result => console.log(result.body)) // or result.documents
  .catch(err => console.error(err))

// callback style API
m.search(
  { index: 'stackoverflow' },
  { query: { match: { title: 'ruby' } } },
  (err, result) => {
    if (err) console.error(err)
    console.log(result.body)) // or result.documents
  }
)

If there is a slow producer, the bulk helper might send data with a very large period of time, and if the process crashes for any reason, the data would be lost. This pr introduces a flushInterval option in the bulk helper to avoid this issue. By default, the bulk helper will flush the data automatically every 30 seconds, unless the threshold has been reached before.

const b = client.helpers.bulk({
  flushInterval: 30000
})

The same problem might happen with the multi search helper, where the user is not sending search requests fast enough. A flushInterval options has been added as well, with a default value of 500 milliseconds.

const m = client.helpers.msearch({
  flushInterval: 500
})

From now on, all he search helpers will use the filter_path option automatically when needed to retrieve only the hits source. This change will result in less netwprk traffic and improved deserialization performances.

Before this, the documents key that you can access in any search helper was computed as soon as we got the search result from Elasticsearch. With this change the documents key is now a getter, which makes this process lazy, resulting in better performances and lower memory impact.

The client helpers can’t be used in Node.js < 10 because it needs a custom flag to be able to use them. Given that not every provider allows the user to specify custom Node.js flags, the Helpers has been disabled completely in Node.js < 10.

Now all the user-provided headers names will be lowercased by default, so there will be no conflicts in case of the same header with different casing.

You can find all the API changes here.

From now on, the client comes with an handy collection of helpers to give you a more comfortable experience with some APIs.

The following helpers has been introduced:

What does this mean? It means that you will see less NoLivingConnectionError, which now can only be caused if you set a selector/filter too strict. For improving the debugging experience, the NoLivingConnectionsError error message has been updated.

From now on, it will be possible to abort a request generated with the promise-styl API. If you abort a request generated from a promise, the promise will be rejected with a RequestAbortedError.

const promise = client.search({
  body: {
    query: { match_all: {} }
  }
})

promise
  .then(console.log)
  .catch(console.log)

promise.abort()

Now every API makes better use of the generics and overloading, so you can (or not, by default request/response bodies are Record<string, any>) define the request/response bodies in the generics.

// request and response bodies are generics
client.search(...)
// response body is `SearchResponse` and request body is generic
client.search<SearchResponse>(...)
// request body is `SearchBody` and response body is `SearchResponse`
client.search<SearchResponse, SearchBody>(...)

This should not be a breaking change, as every generics defaults to any. It might happen to some users that the code breaks, but our test didn’t detect any of it, probably because they were not robust enough. However, given the gigantic improvement in the developer experience, we have decided to release this change in the 7.x line.

It can happen in a situation where we are updating the connections list and running sniff, leaving the dead list in a dirty state. Now the ConnectionPool.update cleans up the dead list every time, which makes way more sense given that all the new connections are alive.

It might happen that markDead is called just after a pool update, and in such case, the client was adding the dead list a node that no longer exists, causing unhandled exceptions later.

The client should not retry if it’s sending a stream body, because it should store in memory a copy of the stream to be able to send it again, but since it doesn’t know in advance the size of the stream, it risks to take too much memory. Furthermore, copying everytime the stream is very an expensive operation.

Until now, aborting a request was blocking the HTTP request, but never calling the callback or resolving the promise to notify the user. This is a bug because it could lead to dangerous memory leaks. From now on if the user calls the request.abort() method, the callback style API will be called with a RequestAbortedError, the promise will be rejected with RequestAbortedError as well.

Fixes:

Documentation:

Support for Elasticsearch v7.6.

Fixes:

Documentation:

Support for Elasticsearch v7.5.

Features

Support for Elasticsearch v7.4.

Fixes:

Documentation:

Internals:

Support for Elasticsearch v7.3.

Features:

Fixes:

Internals:

Documentation:

Support for Elasticsearch v7.2

Fixes:

Support for Elasticsearch v7.1

Fixes:

Fixes:

Support for Elasticsearch v7.0

"appendix",role="exclude",id="redirects"] = Deleted pages

The following pages have moved or been deleted.