Release notes

You can find all the API changes here.

Now the client exports the SniffingTransport class.

Fixes a memory leak caused by an error in the upstream dataset of the bulk helper.

The legacy http client was not cleaning up the abort listener, which could cause a memory leak.

Improve the stream body collection and keep alive timeout.

You can find all the API changes here.

Breaking: Yes | Migration effort: Medium

The current TypeScript definitions will be removed from the client, and the new definitions, which contain request and response definitions as well will be shipped by default.

Breaking: Yes | Migration effort: Large

Maintaining both API styles is not a problem per se, but it makes error handling more convoluted due to async stack traces. Moving to a full-promise API will solve this issue.

// callback-style api
client.search({ params }, { options }, (err, result) => {
 console.log(err || result)
})

// promise-style api
client.search({ params }, { options })
  .then(console.log)
  .catch(console.log)

// async-style (sugar syntax on top of promises)
const response = await client.search({ params }, { options })
console.log(response)

If you are already using the promise-style API, this won’t be a breaking change for you.

Breaking: Yes | Migration effort: Small

The old abort API makes sense for callbacks but it’s annoying to use with promises

// callback-style api
const request = client.search({ params }, { options }, (err, result) => {
 console.log(err) // RequestAbortedError
})

request.abort()

// promise-style api
const promise = client.search({ params }, { options })

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

promise.abort()

Node v12 has added the standard AbortController API which is designed to work well with both callbacks and promises.

const ac = new AbortController()
client.search({ params }, { signal: ac.signal })
  .then(console.log)
  .catch(console.log) // RequestAbortedError

ac.abort()

Breaking: Yes | Migration effort: Small

Thanks to the new types we are developing now we know exactly where a parameter should go. The client API leaks HTTP-related notions in many places, and removing them would definitely improve the DX.

This could be a rather big breaking change, so a double solution could be used during the 8.x lifecycle. (accepting body keys without them being wrapped in the body as well as the current solution).

// from
const response = await client.search({
  index: 'test',
  body: {
    query: {
      match_all: {}
    }
  }
})

// to
const response = await client.search({
  index: 'test',
  query: {
    match_all: {}
  }
})

Breaking: Yes | Migration effort: Small to none

The separated transport has been rewritten in TypeScript and has already dropped the callback style API. Given that now is separated, most of the Elasticsearch specific concepts have been removed, and the client will likely need to extend parts of it for reintroducing them. If you weren’t extending the internals of the client, this won’t be a breaking change for you.

Breaking: Yes | Migration effort: Small

The client API leaks HTTP-related notions in many places, and removing them would definitely improve the DX. The client will expose a new request-specific option to still get the full response details.

// from
const response = await client.search({
  index: 'test',
  body: {
    query: {
      match_all: {}
    }
  }
})
console.log(response) // { body: SearchResponse, statusCode: number, headers: object, warnings: array }

// to
const response = await client.search({
  index: 'test',
  query: {
    match_all: {}
  }
})
console.log(response) // SearchResponse

// with a bit of TypeScript and JavaScript magic...
const response = await client.search({
  index: 'test',
  query: {
    match_all: {}
  }
}, {
  meta: true
})
console.log(response) // { body: SearchResponse, statusCode: number, headers: object, warnings: array }

Breaking: Yes | Migration effort: Small to none

Move from the current cluster connection pool to a weight-based implementation. This new implementation offers better performances and runs less code in the background, the old connection pool can still be used. If you weren’t extending the internals of the client, this won’t be a breaking change for you.

Breaking: Yes | Migration effort: Small to none

By default, the HTTP client will no longer be the default Node.js HTTP client, but undici instead. Undici is a brand new HTTP client written from scratch, it offers vastly improved performances and has better support for promises. Furthermore, it offers comprehensive and predictable error handling. The old HTTP client can still be used. If you weren’t extending the internals of the client, this won’t be a breaking change for you.

Breaking: Yes | Migration effort: Medium

Currently, every path or query parameter could be expressed in both snake_case and camelCase. Internally the client will convert everything to snake_case. This was done in an effort to reduce the friction of migrating from the legacy to the new client, but now it no longer makes sense. If you are already using snake_case keys, this won’t be a breaking change for you.

Breaking: Yes | Migration effort: Small

People usually refers to this as tls, furthermore, internally we use the tls API and Node.js refers to it as tls everywhere.

// before
const client = new Client({
  node: 'https://localhost:9200',
  ssl: {
    rejectUnauthorized: false
  }
})

// after
const client = new Client({
  node: 'https://localhost:9200',
  tls: {
    rejectUnauthorized: false
  }
})

Breaking: Yes | Migration effort: Small

Prototype poisoning protection is very useful, but it can cause performances issues with big payloads. In v8 it will be removed, and the documentation will show how to add it back with a custom serializer.

Breaking: Yes | Migration effort: Large

Nowadays the client support the entire Elasticsearch API, and the transport.request method can be used if necessary. The client extensions API have no reason to exist.

client.extend('utility.index', ({ makeRequest }) => {
  return function _index (params, options) {
    // your code
  }
})

client.utility.index(...)

If you weren’t using client extensions, this won’t be a breaking change for you.

Breaking: No | Migration effort: None

The new separated transport is already written in TypeScript, and it makes sense that the client v8 will be fully written in TypeScript as well.

Breaking: Yes | Migration effort: Small

Currently, the client offers a subset of methods of the EventEmitter class, v8 will ship with a diagnostic property which will be a proper event emitter.

// from
client.on('request', console.log)

// to
client.diagnostic.on('request', console.log)

Breaking: Yes | Migration effort: Small

The Cloud configuration does not support ApiKey and Bearer auth, while the auth options does. There is no need to keep the legacy basic auth support in the cloud configuration.

// before
const client = new Client({
  cloud: {
    id: '<cloud-id>',
    username: 'elastic',
    password: 'changeme'
  }
})

// after
const client = new Client({
  cloud: {
    id: '<cloud-id>'
  },
  auth: {
    username: 'elastic',
    password: 'changeme'
  }
})

If you are already passing the basic auth options in the auth configuration, this won’t be a breaking change for you.

Once you call client.close every new request after that will be rejected with a NoLivingConnectionsError. In-flight requests will be executed normally unless an in-flight request requires a retry, in which case it will be rejected.

The v7 client provided snake_cased methods, such as client.delete_by_query. This is no longer supported, now only camelCased method are present. So client.delete_by_query can be accessed with client.deleteByQuery