Client Usage

Make a Request

Begin by importing the aiohttp module:

import aiohttp

Now, let’s try to get a web-page. For example let’s get GitHub’s public time-line:

async with aiohttp.ClientSession() as session:
    async with session.get('') as resp:
        print(await resp.text())

Now, we have a ClientSession called session and a ClientResponse object called resp. We can get all the information we need from the response. The mandatory parameter of ClientSession.get() coroutine is an HTTP url.

In order to make an HTTP POST request use coroutine:'', data=b'data')

Other HTTP methods are available as well:

session.put('', data=b'data')
session.patch('', data=b'data')


Don’t create a session per request. Most likely you need a session per application which performs all requests altogether.

A session contains a connection pool inside, connection reusage and keep-alives (both are on by default) may speed up total performance.

JSON Request

Any of session’s request methods like request, get, post etc accept json parameter:

async with aiohttp.ClientSession() as session:
    async with{'test': 'object'})

By default session uses python’s standard json module for serialization. But it is possible to use different serializer. ClientSession accepts json_serialize parameter:

import ujson

async with aiohttp.ClientSession(json_serialize=ujson.dumps) as session:
    async with{'test': 'object'})

Passing Parameters In URLs

You often want to send some sort of data in the URL’s query string. If you were constructing the URL by hand, this data would be given as key/value pairs in the URL after a question mark, e.g. Requests allows you to provide these arguments as a dict, using the params keyword argument. As an example, if you wanted to pass key1=value1 and key2=value2 to, you would use the following code:

params = {'key1': 'value1', 'key2': 'value2'}
async with session.get('',
                       params=params) as resp:
    assert str(resp.url) == ''

You can see that the URL has been correctly encoded by printing the URL.

For sending data with multiple values for the same key MultiDict may be used as well.

It is also possible to pass a list of 2 item tuples as parameters, in that case you can specify multiple values for each key:

params = [('key', 'value1'), ('key', 'value2')]
async with session.get('',
                       params=params) as r:
    assert str(r.url) == ''

You can also pass str content as param, but beware – content is not encoded by library. Note that + is not encoded:

async with session.get('',
                       params='key=value+1') as r:
        assert str(r.url) == ''


aiohttp internally performs URL canonization before sending request.

Canonization encodes host part by IDNA codec and applies requoting to path and query parts.

For example URL('путь%30?a=%31') is converted to URL('').

Sometimes canonization is not desirable if server accepts exact representation and does not requote URL itself.

To disable canonization use encoded=True parameter for URL construction:

await session.get(URL('', encoded=True))


Passing params overrides encoded=True, never use both options.

Response Content

We can read the content of the server’s response. Consider the GitHub time-line again:

async with session.get('') as resp:
    print(await resp.text())

will printout something like:


aiohttp will automatically decode the content from the server. You can specify custom encoding for the text() method:

await resp.text(encoding='windows-1251')

Binary Response Content

You can also access the response body as bytes, for non-text requests:


The gzip and deflate transfer-encodings are automatically decoded for you.

You can enable brotli transfer-encodings support, just install brotlipy.

JSON Response Content

There’s also a built-in JSON decoder, in case you’re dealing with JSON data:

async with session.get('') as resp:
    print(await resp.json())

In case that JSON decoding fails, json() will raise an exception. It is possible to specify custom encoding and decoder functions for the json() call.


The methods above reads the whole response body into memory. If you are planning on reading lots of data, consider using the streaming response method documented below.

Streaming Response Content

While methods read(), json() and text() are very convenient you should use them carefully. All these methods load the whole response in memory. For example if you want to download several gigabyte sized files, these methods will load all the data in memory. Instead you can use the content attribute. It is an instance of the aiohttp.StreamReader class. The gzip and deflate transfer-encodings are automatically decoded for you:

async with session.get('') as resp:

In general, however, you should use a pattern like this to save what is being streamed to a file:

with open(filename, 'wb') as fd:
    while True:
        chunk = await
        if not chunk:

It is not possible to use read(), json() and text() after explicit reading from content.


ClientResponse object contains request_info property, which contains request fields: url and headers. On raise_for_status structure is copied to ClientResponseError instance.

Custom Headers

If you need to add HTTP headers to a request, pass them in a dict to the headers parameter.

For example, if you want to specify the content-type for the previous example:

import json
url = ''
payload = {'some': 'data'}
headers = {'content-type': 'application/json'}


Custom Cookies

To send your own cookies to the server, you can use the cookies parameter of ClientSession constructor:

url = ''
cookies = {'cookies_are': 'working'}
async with ClientSession(cookies=cookies) as session:
    async with session.get(url) as resp:
        assert await resp.json() == {
           "cookies": {"cookies_are": "working"}}

Note endpoint returns request cookies in JSON-encoded body. To access session cookies see ClientSession.cookie_jar.

More complicated POST requests

Typically, you want to send some form-encoded data – much like an HTML form. To do this, simply pass a dictionary to the data argument. Your dictionary of data will automatically be form-encoded when the request is made:

payload = {'key1': 'value1', 'key2': 'value2'}
async with'',
                        data=payload) as resp:
    print(await resp.text())
  "form": {
    "key2": "value2",
    "key1": "value1"

If you want to send data that is not form-encoded you can do it by passing a str instead of a dict. This data will be posted directly.

For example, the GitHub API v3 accepts JSON-Encoded POST/PATCH data:

import json
url = ''
payload = {'some': 'data'}

async with, data=json.dumps(payload)) as resp:

POST a Multipart-Encoded File

To upload Multipart-encoded files:

url = ''
files = {'file': open('report.xls', 'rb')}

await, data=files)

You can set the filename, content_type explicitly:

url = ''
data = FormData()
               open('report.xls', 'rb'),

await, data=data)

If you pass a file object as data parameter, aiohttp will stream it to the server automatically. Check StreamReader for supported format information.

Streaming uploads

aiohttp supports multiple types of streaming uploads, which allows you to send large files without reading them into memory.

As a simple case, simply provide a file-like object for your body:

with open('massive-body', 'rb') as f:
   await'', data=f)

Or you can use aiohttp.streamer object:

def file_sender(writer, file_name=None):
    with open(file_name, 'rb') as f:
        chunk =**16)
        while chunk:
            yield from writer.write(chunk)
            chunk =**16)

# Then you can use `file_sender` as a data provider:

async with'',
                        data=file_sender(file_name='huge_file')) as resp:
    print(await resp.text())

Also it is possible to use a StreamReader object. Lets say we want to upload a file from another request and calculate the file SHA1 hash:

async def feed_stream(resp, stream):
    h = hashlib.sha256()

    while True:
        chunk = await resp.content.readany()
        if not chunk:

    return h.hexdigest()

resp = session.get('')
stream = StreamReader()
loop.create_task('', data=stream))

file_hash = await feed_stream(resp, stream)

Because the response content attribute is a StreamReader, you can chain get and post requests together:

r = await session.get('')

Uploading pre-compressed data

To upload data that is already compressed before passing it to aiohttp, call the request function with the used compression algorithm name (usually deflate or zlib) as the value of the Content-Encoding header:

async def my_coroutine(session, headers, my_data):
    data = zlib.compress(my_data)
    headers = {'Content-Encoding': 'deflate'}
    async with'',


To tweak or change transport layer of requests you can pass a custom connector to ClientSession and family. For example:

conn = aiohttp.TCPConnector()
session = aiohttp.ClientSession(connector=conn)


You can not re-use custom connector, session object takes ownership of the connector.

See also

Connectors section for more information about different connector types and configuration options.

Limiting connection pool size

To limit amount of simultaneously opened connections you can pass limit parameter to connector:

conn = aiohttp.TCPConnector(limit=30)

The example limits total amount of parallel connections to 30.

The default is 100.

If you explicitly want not to have limits, pass 0. For example:

conn = aiohttp.TCPConnector(limit=0)

To limit amount of simultaneously opened connection to the same endpoint ((host, port, is_ssl) triple) you can pass limit_per_host parameter to connector:

conn = aiohttp.TCPConnector(limit_per_host=30)

The example limits amount of parallel connections to the same to 30.

The default is 0 (no limit on per host bases).

Resolving using custom nameservers

In order to specify the nameservers to when resolving the hostnames, aiodns is required:

from aiohttp.resolver import AsyncResolver

resolver = AsyncResolver(nameservers=["", ""])
conn = aiohttp.TCPConnector(resolver=resolver)

SSL control for TCP sockets

By default aiohttp uses strict checks for HTTPS protocol. Certification checks can be relaxed by setting verify_ssl to False:

r = await session.get('', verify_ssl=False)

If you need to setup custom ssl parameters (use own certification files for example) you can create a ssl.SSLContext instance and pass it into the proper ClientSession method:

sslcontext = ssl.create_default_context(
r = await session.get('', ssl_context=sslcontext)

If you need to verify self-signed certificates, you can do the same thing as the previous example, but add another call to ssl.SSLContext.load_cert_chain() with the key pair:

sslcontext = ssl.create_default_context(
r = await session.get('', ssl_context=sslcontext)

There is explicit errors when ssl verification fails


    await session.get('')
except aiohttp.ClientConnectorSSLError as e:
    assert isinstance(e, ssl.SSLError)


    await session.get('')
except aiohttp.ClientConnectorCertificateError as e:
    assert isinstance(e, ssl.CertificateError)

If you need to skip both ssl related errors


    await session.get('')
except aiohttp.ClientSSLError as e:
    assert isinstance(e, ssl.SSLError)

    await session.get('')
except aiohttp.ClientSSLError as e:
    assert isinstance(e, ssl.CertificateError)

You may also verify certificates via SHA256 fingerprint:

# Attempt to connect to
# with a pin to a bogus certificate:
bad_fingerprint = b'0'*64
exc = None
    r = await session.get('',
except aiohttp.FingerprintMismatch as e:
    exc = e
assert exc is not None
assert exc.expected == bad_fingerprint

# cert's actual fingerprint
assert == b'...'

Note that this is the fingerprint of the DER-encoded certificate. If you have the certificate in PEM format, you can convert it to DER with e.g:

openssl x509 -in crt.pem -inform PEM -outform DER > crt.der


Tip: to convert from a hexadecimal digest to a binary byte-string, you can use binascii.unhexlify().

All verify_ssl, fingerprint and ssl_context could be passed to TCPConnector as defaults, params from ClientSession.get() and others override these defaults.


verify_ssl and ssl_context params are mutually exclusive.

MD5 and SHA1 fingerprints are deprecated but still supported – they are famous as very insecure hash functions.

Unix domain sockets

If your HTTP server uses UNIX domain sockets you can use UnixConnector:

conn = aiohttp.UnixConnector(path='/path/to/socket')
session = aiohttp.ClientSession(connector=conn)

Proxy support

aiohttp supports HTTP/HTTPS proxies. You have to use proxy parameter:

async with aiohttp.ClientSession() as session:
    async with session.get("",
                           proxy="") as resp:

It also supports proxy authorization:

async with aiohttp.ClientSession() as session:
    proxy_auth = aiohttp.BasicAuth('user', 'pass')
    async with session.get("",
                           proxy_auth=proxy_auth) as resp:

Authentication credentials can be passed in proxy URL:


Contrary to the requests library, it won’t read environment variables by default. But you can do so by passing trust_env=True into aiohttp.ClientSession constructor for extracting proxy configuration from HTTP_PROXY or HTTPS_PROXY environment variables (both are case insensitive):

async with aiohttp.ClientSession() as session:
    async with session.get("", trust_env=True) as resp:

Response Status Codes

We can check the response status code:

async with session.get('') as resp:
    assert resp.status == 200

Response Headers

We can view the server’s response ClientResponse.headers using a CIMultiDictProxy:

>>> resp.headers
 'CONTENT-TYPE': 'application/json',
 'DATE': 'Tue, 15 Jul 2014 16:49:51 GMT',
 'SERVER': 'gunicorn/18.0',
 'CONNECTION': 'keep-alive'}

The dictionary is special, though: it’s made just for HTTP headers. According to RFC 7230, HTTP Header names are case-insensitive. It also supports multiple values for the same key as HTTP protocol does.

So, we can access the headers using any capitalization we want:

>>> resp.headers['Content-Type']

>>> resp.headers.get('content-type')

All headers converted from binary data using UTF-8 with surrogateescape option. That works fine on most cases but sometimes unconverted data is needed if a server uses nonstandard encoding. While these headers are malformed from RFC 7230 perspective they are may be retrieved by using ClientResponse.raw_headers property:

>>> resp.raw_headers
((b'SERVER', b'nginx'),
 (b'DATE', b'Sat, 09 Jan 2016 20:28:40 GMT'),
 (b'CONTENT-TYPE', b'text/html; charset=utf-8'),
 (b'CONTENT-LENGTH', b'12150'),
 (b'CONNECTION', b'keep-alive'))

Response Cookies

If a response contains some Cookies, you can quickly access them:

url = ''
async with session.get(url) as resp:


Response cookies contain only values, that were in Set-Cookie headers of the last request in redirection chain. To gather cookies between all redirection requests please use aiohttp.ClientSession object.

Response History

If a request was redirected, it is possible to view previous responses using the history attribute:

>>> resp = await session.get('')
>>> resp
<ClientResponse( [200]>
>>> resp.history
(<ClientResponse( [301]>,)

If no redirects occurred or allow_redirects is set to False, history will be an empty sequence.


aiohttp works with client websockets out-of-the-box.

You have to use the aiohttp.ClientSession.ws_connect() coroutine for client websocket connection. It accepts a url as a first parameter and returns ClientWebSocketResponse, with that object you can communicate with websocket server using response’s methods:

session = aiohttp.ClientSession()
async with session.ws_connect('') as ws:

    async for msg in ws:
        if msg.type == aiohttp.WSMsgType.TEXT:
            if == 'close cmd':
                await ws.close()
                await ws.send_str( + '/answer')
        elif msg.type == aiohttp.WSMsgType.CLOSED:
        elif msg.type == aiohttp.WSMsgType.ERROR:

You must use the only websocket task for both reading (e.g. await ws.receive() or async for msg in ws:) and writing but may have multiple writer tasks which can only send data asynchronously (by ws.send_str('data') for example).


By default all IO operations have 5min timeout. The timeout may be overridden by passing timeout parameter into ClientSession.get() and family:

async with session.get('', timeout=60) as r:

None or 0 disables timeout check.

The example wraps a client call in async_timeout.timeout() context manager, adding timeout for both connecting and response body reading procedures:

import async_timeout

with async_timeout.timeout(0.001):
    async with session.get('') as r:
        await r.text()


Timeout is cumulative time, it includes all operations like sending request, redirects, response parsing, consuming response, etc.

Graceful Shutdown

When ClientSession closes at the end of an async with block (or through a direct .close() call), the underlying connection remains open due to asyncio internal details. In practice, the underlying connection will close after a short while. However, if the event loop is stopped before the underlying connection is closed, an ResourceWarning: unclosed transport warning is emitted (when warnings are enabled).

To avoid this situation, a small delay must be added before closing the event loop to allow any open underlying connections to close.

For a ClientSession without SSL, a simple zero-sleep (await asyncio.sleep(0)) will suffice:

async def read_website():
    async with aiohttp.ClientSession() as session:
        async with session.get('') as response:

loop = asyncio.get_event_loop()
# Zero-sleep to allow underlying connections to close

For a ClientSession with SSL, the application must wait a short duration before closing:

# Wait 250 ms for the underlying SSL connections to close

Note that the appropriate amount of time to wait will vary from application to application.

All if this will eventually become obsolete when the asyncio internals are changed so that aiohttp itself can wait on the underlying connection to close. Please follow issue #1925 for the progress on this.