Python Elasticsearch Client

Official low-level client for Elasticsearch. Its goal is to provide common ground for all Elasticsearch-related code in Python; because of this it tries to be opinion-free and very extendable.


Install the elasticsearch package with pip:

$ python -m pip install elasticsearch

If your application uses async/await in Python you can install with the async extra:

$ python -m pip install elasticsearch[async]

Read more about how to use asyncio with this project.


Language clients are forward compatible; meaning that clients support communicating with greater or equal minor versions of Elasticsearch. Elasticsearch language clients are only backwards compatible with default distributions and without guarantees made.

If you have a need to have multiple versions installed at the same time older versions are also released as elasticsearch2, elasticsearch5 and elasticsearch6.

Example Usage

from datetime import datetime
from elasticsearch import Elasticsearch
es = Elasticsearch()

doc = {
    'author': 'kimchy',
    'text': 'Elasticsearch: cool. bonsai cool.',
res = es.index(index="test-index", id=1, document=doc)

res = es.get(index="test-index", id=1)


res ="test-index", query={"match_all": {}})
print("Got %d Hits:" % res['hits']['total']['value'])
for hit in res['hits']['hits']:
    print("%(timestamp)s %(author)s: %(text)s" % hit["_source"])


This client was designed as very thin wrapper around Elasticsearch’s REST API to allow for maximum flexibility. This means that there are no opinions in this client; it also means that some of the APIs are a little cumbersome to use from Python. We have created some Helpers to help with this issue as well as a more high level library (elasticsearch-dsl) on top of this one to provide a more convenient way of working with Elasticsearch.

Persistent Connections

elasticsearch-py uses persistent connections inside of individual connection pools (one per each configured or sniffed node). Out of the box you can choose between two http protocol implementations. See Transport classes for more information.

The transport layer will create an instance of the selected connection class per node and keep track of the health of individual nodes - if a node becomes unresponsive (throwing exceptions while connecting to it) it’s put on a timeout by the ConnectionPool class and only returned to the circulation after the timeout is over (or when no live nodes are left). By default nodes are randomized before being passed into the pool and round-robin strategy is used for load balancing.

You can customize this behavior by passing parameters to the Connection Layer API (all keyword arguments to the Elasticsearch class will be passed through). If what you want to accomplish is not supported you should be able to create a subclass of the relevant component and pass it in as a parameter to be used instead of the default implementation.

Automatic Retries

If a connection to a node fails due to connection issues (raises ConnectionError) it is considered in faulty state. It will be placed on hold for dead_timeout seconds and the request will be retried on another node. If a connection fails multiple times in a row the timeout will get progressively larger to avoid hitting a node that’s, by all indication, down. If no live connection is available, the connection that has the smallest timeout will be used.

By default retries are not triggered by a timeout (ConnectionTimeout), set retry_on_timeout to True to also retry on timeouts.


The client can be configured to inspect the cluster state to get a list of nodes upon startup, periodically and/or on failure. See Transport parameters for details.

Some example configurations:

from elasticsearch import Elasticsearch

# by default we don't sniff, ever
es = Elasticsearch()

# you can specify to sniff on startup to inspect the cluster and load
# balance across all nodes
es = Elasticsearch(["seed1", "seed2"], sniff_on_start=True)

# you can also sniff periodically and/or after failure:
es = Elasticsearch(["seed1", "seed2"],

Thread safety

The client is thread safe and can be used in a multi threaded environment. Best practice is to create a single global instance of the client and use it throughout your application. If your application is long-running consider turning on Sniffing to make sure the client is up to date on the cluster location.

By default we allow urllib3 to open up to 10 connections to each node, if your application calls for more parallelism, use the maxsize parameter to raise the limit:

# allow up to 25 connections to each node
es = Elasticsearch(["host1", "host2"], maxsize=25)


Since we use persistent connections throughout the client it means that the client doesn’t tolerate fork very well. If your application calls for multiple processes make sure you create a fresh client after call to fork. Note that Python’s multiprocessing module uses fork to create new processes on POSIX systems.

TLS/SSL and Authentication

You can configure the client to use SSL for connecting to your elasticsearch cluster, including certificate verification and HTTP auth:

from elasticsearch import Elasticsearch

# you can use RFC-1738 to specify the url
es = Elasticsearch(['https://user:secret@localhost:443'])

# ... or specify common parameters as kwargs

es = Elasticsearch(
    ['localhost', 'otherhost'],
    http_auth=('user', 'secret'),

# SSL client authentication using client_cert and client_key

from ssl import create_default_context

context = create_default_context(cafile="path/to/cert.pem")
es = Elasticsearch(
    ['localhost', 'otherhost'],
    http_auth=('user', 'secret'),


elasticsearch-py doesn’t ship with default set of root certificates. To have working SSL certificate validation you need to either specify your own as cafile or capath or cadata or install certifi which will be picked up automatically.

See class Urllib3HttpConnection for detailed description of the options.

Connecting via Cloud ID

Cloud ID is an easy way to configure your client to work with your Elastic Cloud deployment. Combine the cloud_id with either http_auth or api_key to authenticate with your Elastic Cloud deployment.

Using cloud_id enables TLS verification and HTTP compression by default and sets the port to 443 unless otherwise overwritten via the port parameter or the port value encoded within cloud_id. Using Cloud ID also disables sniffing.

from elasticsearch import Elasticsearch

es = Elasticsearch(
    http_auth=("elastic", "<password>"),

API Key Authentication

You can configure the client to use Elasticsearch’s API Key for connecting to your cluster. Please note this authentication method has been introduced with release of Elasticsearch 6.7.0.

from elasticsearch import Elasticsearch

# you can use the api key tuple
es = Elasticsearch(
    ['node-1', 'node-2', 'node-3'],
    api_key=('id', 'api_key'),

# or you pass the base 64 encoded token
es = Elasticsearch(
    ['node-1', 'node-2', 'node-3'],
    api_key='base64encoded tuple',


elasticsearch-py uses the standard logging library from python to define two loggers: elasticsearch and elasticsearch.trace. elasticsearch is used by the client to log standard activity, depending on the log level. elasticsearch.trace can be used to log requests to the server in the form of curl commands using pretty-printed json that can then be executed from command line. Because it is designed to be shared (for example to demonstrate an issue) it also just uses localhost:9200 as the address instead of the actual address of the host. If the trace logger has not been configured already it is set to propagate=False so it needs to be activated separately.

Type Hints

Starting in elasticsearch-py v7.10.0 the library now ships with type hints and supports basic static type analysis with tools like Mypy and Pyright.

If we write a script that has a type error like using request_timeout with a str argument instead of float and then run Mypy on the script:

from elasticsearch import Elasticsearch

es = Elasticsearch(...)
    request_timeout="5"  # type error!

# $ mypy
# error: Argument "request_timeout" to "search" of "Elasticsearch" has
#                     incompatible type "str"; expected "Union[int, float, None]"
# Found 1 error in 1 file (checked 1 source file)

For now many parameter types for API methods aren’t specific to a type (ie they are of type typing.Any) but in the future they will be tightened for even better static type checking.

Type hints also allow tools like your IDE to check types and provide better auto-complete functionality.


The type hints for API methods like search don’t match the function signature that can be found in the source code. Type hints represent optimal usage of the API methods. Using keyword arguments is highly recommended so all optional parameters and body are keyword-only in type hints.

JetBrains PyCharm will use the warning Unexpected argument to denote that the parameter may be keyword-only.

Environment considerations

When using the client there are several limitations of your environment that could come into play.

When using an HTTP load balancer you cannot use the Sniffing functionality - the cluster would supply the client with IP addresses to directly connect to the cluster, circumventing the load balancer. Depending on your configuration this might be something you don’t want or break completely.


When using capacity-constrained networks (low throughput), it may be handy to enable compression. This is especially useful when doing bulk loads or inserting large documents. This will configure compression.

from elasticsearch import Elasticsearch
es = Elasticsearch(hosts, http_compress=True)

Compression is enabled by default when connecting to Elastic Cloud via cloud_id.


Custom serializers

By default, JSONSerializer is used to encode all outgoing requests. However, you can implement your own custom serializer

from elasticsearch.serializer import JSONSerializer

class SetEncoder(JSONSerializer):
    def default(self, obj):
        if isinstance(obj, set):
            return list(obj)
        if isinstance(obj, Something):
            return 'CustomSomethingRepresentation'
        return JSONSerializer.default(self, obj)

es = Elasticsearch(serializer=SetEncoder())


For a more high level client library with more limited scope, have a look at elasticsearch-dsl - a more pythonic library sitting on top of elasticsearch-py.

elasticsearch-dsl provides a more convenient and idiomatic way to write and manipulate queries by mirroring the terminology and structure of Elasticsearch JSON DSL while exposing the whole range of the DSL from Python either directly using defined classes or a queryset-like expressions.

It also provides an optional persistence layer for working with documents as Python objects in an ORM-like fashion: defining mappings, retrieving and saving documents, wrapping the document data in user-defined classes.


Copyright 2021 Elasticsearch B.V. Licensed under the Apache License, Version 2.0.

Indices and tables