Mint.HTTP2 (Mint v1.3.0) View Source
Processless HTTP client with support for HTTP/2.
This module provides a data structure that represents an HTTP/2 connection to
a given server. The connection is represented as an opaque struct %Mint.HTTP2{}
.
The connection is a data structure and is not backed by a process, and all the
connection handling happens in the process that creates the struct.
This module and data structure work exactly like the ones described in the Mint.HTTP
module, with the exception that Mint.HTTP2
specifically deals with HTTP/2 while
Mint.HTTP
deals seamlessly with HTTP/1.1 and HTTP/2. For more information on
how to use the data structure and client architecture, see Mint.HTTP
.
HTTP/2 streams and requests
HTTP/2 introduces the concept of streams. A stream is an isolated conversation
between the client and the server. Each stream is unique and identified by a unique
stream ID, which means that there's no order when data comes on different streams
since they can be identified uniquely. A stream closely corresponds to a request, so
in this documentation and client we will mostly refer to streams as "requests".
We mentioned data on streams can come in arbitrary order, and streams are requests,
so the practical effect of this is that performing request A and then request B
does not mean that the response to request A will come before the response to request B.
This is why we identify each request with a unique reference returned by request/5
.
See request/5
for more information.
Closed connection
In HTTP/2, the connection can either be open, closed, or only closed for writing.
When a connection is closed for writing, the client cannot send requests or stream
body chunks, but it can still read data that the server might be sending. When the
connection gets closed on the writing side, a :server_closed_connection
error is
returned. {:error, request_ref, error}
is returned for requests that haven't been
processed by the server, with the reason of error
being :unprocessed
.
These requests are safe to retry.
HTTP/2 settings
HTTP/2 supports settings negotiation between servers and clients. The server advertises its settings to the client and the client advertises its settings to the server. A peer (server or client) has to acknowledge the settings advertised by the other peer before those settings come into action (that's why it's called a negotiation).
A first settings negotiation happens right when the connection starts. Servers and clients can renegotiate settings at any time during the life of the connection.
Mint users don't need to care about settings acknowledgements directly since they're
handled transparently by stream/2
.
To retrieve the server settings, you can use get_server_setting/2
. Doing so is often
useful to be able to tune your requests based on the server settings.
To communicate client settings to the server, use put_settings/2
or pass them when
starting up a connection with connect/4
. Note that the server needs to acknowledge
the settings sent through put_setting/2
before those settings come into effect. The
server ack is processed transparently by stream/2
, but this means that if you change
a setting through put_settings/2
and try to retrieve the value of that setting right
after with get_client_setting/2
, you'll likely get the old value of that setting. Once
the server acknowledges the new settings, the updated value will be returned by
get_client_setting/2
.
Server push
HTTP/2 supports server push, which
is a way for a server to send a response to a client without the client needing to make
the corresponding request. The server sends a :push_promise
response to a normal request:
this creates a new request reference. Then, the server sends normal responses for the newly
created request reference.
Let's see an example. We will ask the server for "/index.html"
and the server will
send us a push promise for "/style.css"
.
{:ok, conn} = Mint.HTTP2.connect(:https, "example.com", 443)
{:ok, conn, request_ref} = Mint.HTTP2.request(conn, "GET", "/index.html", _headers = [], _body = "")
next_message =
receive do
msg -> msg
end
{:ok, conn, responses} = Mint.HTTP2.stream(conn, next_message)
[
{:push_promise, ^request_ref, promised_request_ref, promised_headers},
{:status, ^request_ref, 200},
{:headers, ^request_ref, []},
{:data, ^request_ref, "<html>..."},
{:done, ^request_ref}
] = responses
promised_headers
#=> [{":method", "GET"}, {":path", "/style.css"}]
As you can see in the example above, when the server sends a push promise then a
:push_promise
response is returned as a response to a request. The :push_promise
response contains a promised_request_ref
and some promised_headers
. The
promised_request_ref
is the new request ref that pushed responses will be tagged with.
promised_headers
are headers that tell the client what request the promised response
will respond to. The idea is that the server tells the client a request the client will
want to make and then preemptively sends a response for that request. Promised headers
will always include :method
, :path
, and :authority
.
next_message =
receive do
msg -> msg
end
{:ok, conn, responses} = Mint.HTTP2.stream(conn, next_message)
[
{:status, ^promised_request_ref, 200},
{:headers, ^promised_request_ref, []},
{:data, ^promised_request_ref, "body { ... }"},
{:done, ^promised_request_ref}
]
The response to a promised request is like a response to any normal request.
Disabling server pushes
HTTP/2 exposes a boolean setting for enabling or disabling server pushes with :enable_push
.
You can pass this option when connecting or in put_settings/2
. By default server push
is enabled.
Link to this section Summary
Functions
Cancels an in-flight request.
Same as Mint.HTTP.connect/4
, but forces a HTTP/2 connection.
Gets the value of the given HTTP/2 client setting.
Gets the value of the given HTTP/2 server settings.
Returns the window size of the connection or of a single request.
Pings the server.
Communicates the given client settings to the server.
Link to this section Types
Specs
error_reason() :: term()
An HTTP/2-specific error reason.
The values can be:
:closed
- when you try to make a request or stream a body chunk but the connection is closed.:closed_for_writing
- when you try to make a request or stream a body chunk but the connection is closed for writing. This means you cannot issue any more requests. See the "Closed connection" section in the module documentation for more information.:too_many_concurrent_requests
- when the maximum number of concurrent requests allowed by the server is reached. To find out what this limit is, useget_setting/2
with the:max_concurrent_streams
setting name.{:max_header_list_size_exceeded, size, max_size}
- when the maximum size of the header list is reached.size
is the actual value of the header list size,max_size
is the maximum value allowed. Seeget_setting/2
to retrieve the value of the max size.{:exceeds_window_size, what, window_size}
- when the data you're trying to send exceeds the window size of the connection (ifwhat
is:connection
) or of a request (ifwhat
is:request
).window_size
is the allowed window size. Seeget_window_size/2
.{:stream_not_found, stream_id}
- when the given request is not found.:unknown_request_to_stream
- when you're trying to stream data on an unknown request.:request_is_not_streaming
- when you try to send data (withstream_request_body/3
) on a request that is not open for streaming.:unprocessed
- when a request was closed because it was not processed by the server. When this error is returned, it means that the server hasn't processed the request at all, so it's safe to retry the given request on a different or new connection.{:server_closed_request, error_code}
- when the server closes the request.error_code
is the reason why the request was closed.{:server_closed_connection, reason, debug_data}
- when the server closes the connection gracefully or because of an error. In HTTP/2, this corresponds to aGOAWAY
frame.error
is the reason why the connection was closed.debug_data
is additional debug data.{:frame_size_error, frame}
- when there's an error with the size of a frame.frame
is the frame type, such as:settings
or:window_update
.{:protocol_error, debug_data}
- when there's a protocol error.debug_data
is a string that explains the nature of the error.{:compression_error, debug_data}
- when there's a header compression error.debug_data
is a string that explains the nature of the error.{:flow_control_error, debug_data}
- when there's a flow control error.debug_data
is a string that explains the nature of the error.
Specs
setting() :: {:enable_push, boolean()} | {:max_concurrent_streams, pos_integer()} | {:initial_window_size, 1..2_147_483_647} | {:max_frame_size, 16384..16_777_215} | {:max_header_list_size, :infinity | pos_integer()}
HTTP/2 setting with its value.
This type represents both server settings as well as client settings. To retrieve
server settings use get_server_setting/2
and to retrieve client settings use
get_client_setting/2
. To send client settings to the server, see put_settings/2
.
The supported settings are the following:
:header_table_size
- (integer) corresponds toSETTINGS_HEADER_TABLE_SIZE
.:enable_push
- (boolean) corresponds toSETTINGS_ENABLE_PUSH
. Sets whether push promises are supported. If you don't want to support push promises, useput_settings/2
to tell the server that your client doesn't want push promises.:max_concurrent_streams
- (integer) corresponds toSETTINGS_MAX_CONCURRENT_STREAMS
. Tells what is the maximum number of streams that the peer sending this (client or server) supports. As mentioned in the module documentation, HTTP/2 streams are equivalent to requests, so knowing the maximum number of streams that the server supports can be useful to know how many concurrent requests can be open at any time. Useget_server_setting/2
to find out how many concurrent streams the server supports.:initial_window_size
- (integer smaller than2147483647
) corresponds toSETTINGS_INITIAL_WINDOW_SIZE
. Tells what is the value of the initial HTTP/2 window size for the peer that sends this setting.:max_frame_size
- (integer in the range16384..16777215
) corresponds toSETTINGS_MAX_FRAME_SIZE
. Tells what is the maximum size of an HTTP/2 frame for the peer that sends this setting.:max_header_list_size
- (integer) corresponds toSETTINGS_MAX_HEADER_LIST_SIZE
.
Specs
settings() :: [setting()]
HTTP/2 settings.
See setting/0
.
Specs
t()
Link to this section Functions
Specs
cancel_request(t(), Mint.Types.request_ref()) :: {:ok, t()} | {:error, t(), Mint.Types.error()}
Cancels an in-flight request.
This function is HTTP/2 specific. It cancels an in-flight request. The server could have
already sent responses for the request you want to cancel: those responses will be parsed
by the connection but not returned to the user. No more responses
to a request will be returned after you call cancel_request/2
on that request.
If there's no error in canceling the request, {:ok, conn}
is returned where conn
is
the updated connection. If there's an error, {:error, conn, reason}
is returned where
conn
is the updated connection and reason
is the error reason.
Examples
{:ok, conn, ref} = Mint.HTTP2.request(conn, "GET", "/", _headers = [])
{:ok, conn} = Mint.HTTP2.cancel_request(conn, ref)
Specs
See Mint.HTTP.close/1
.
Specs
connect( Mint.Types.scheme(), Mint.Types.address(), :inet.port_number(), keyword() ) :: {:ok, t()} | {:error, Mint.Types.error()}
Same as Mint.HTTP.connect/4
, but forces a HTTP/2 connection.
Specs
controlling_process(t(), pid()) :: {:ok, t()} | {:error, Mint.Types.error()}
Specs
Specs
Gets the value of the given HTTP/2 client setting.
This function returns the value of the given HTTP/2 setting that the client
advertised to the server. Client settings can be advertised through put_settings/2
or when starting up a connection.
Client settings have to be acknowledged by the server before coming into effect.
This function is HTTP/2 specific. For more information on HTTP/2 settings, see the related section in the RFC.
See the "HTTP/2 settings" section in the module documentation for more information.
Supported settings
The possible settings that can be retrieved are described in setting/0
.
Any other atom passed as name
will raise an error.
Examples
Mint.HTTP2.get_client_setting(conn, :max_concurrent_streams)
#=> 500
Specs
Specs
Gets the value of the given HTTP/2 server settings.
This function returns the value of the given HTTP/2 setting that the server advertised to the client. This function is HTTP/2 specific. For more information on HTTP/2 settings, see the related section in the RFC.
See the "HTTP/2 settings" section in the module documentation for more information.
Supported settings
The possible settings that can be retrieved are described in setting/0
.
Any other atom passed as name
will raise an error.
Examples
Mint.HTTP2.get_server_setting(conn, :max_concurrent_streams)
#=> 500
Specs
get_socket(t()) :: Mint.Types.socket()
Specs
get_window_size(t(), :connection | {:request, Mint.Types.request_ref()}) :: non_neg_integer()
Returns the window size of the connection or of a single request.
This function is HTTP/2 specific. It returns the window size of
either the connection if connection_or_request
is :connection
or of a single
request if connection_or_request
is {:request, request_ref}
.
Use this function to check the window size of the connection before sending a full request. Also use this function to check the window size of both the connection and of a request if you want to stream body chunks on that request.
For more information on flow control and window sizes in HTTP/2, see the section below.
HTTP/2 flow control
In HTTP/2, flow control is implemented through a
window size. When the client sends data to the server, the window size is decreased
and the server needs to "refill" it on the client side. You don't need to take care of
the refilling of the client window as it happens behind the scenes in stream/2
.
A window size is kept for the entire connection and all requests affect this window size. A window size is also kept per request.
The only thing that affects the window size is the body of a request, regardless of
if it's a full request sent with request/5
or body chunks sent through
stream_request_body/3
. That means that if we make a request with a body that is
five bytes long, like "hello"
, the window size of the connection and the window size
of that particular request will decrease by five bytes.
If we use all the window size before the server refills it, functions like
request/5
will return an error.
Examples
On the connection:
HTTP.get_window_size(conn, :connection)
#=> 65_536
On a single streamed request:
{:ok, conn, request_ref} = HTTP2.request(conn, "GET", "/", [], :stream)
HTTP.get_window_size(conn, {:request_ref, request_ref})
#=> 65_536
{:ok, conn} = HTTP2.stream_request_body(conn, request_ref, "hello")
HTTP.get_window_size(conn, {:request_ref, request_ref})
#=> 65_531
Specs
See Mint.HTTP.open?/1
.
Specs
open_request_count(t()) :: non_neg_integer()
See Mint.HTTP.open_request_count/1
.
In HTTP/2, the number of open requests is the number of requests opened by the client
that have not yet received a :done
response. It's important to note that only
requests opened by the client (with request/5
) count towards the number of open
requests, as requests opened from the server with server pushes (see the "Server push"
section in the module documentation) are not considered open requests. We do this because
clients might need to know how many open requests there are because the server limits
the number of concurrent requests the client can open. To know how many requests the client
can open, see get_server_setting/2
with the :max_concurrent_streams
setting.
Specs
ping(t(), <<_::8>>) :: {:ok, t(), Mint.Types.request_ref()} | {:error, t(), Mint.Types.error()}
Pings the server.
This function is specific to HTTP/2 connections. It sends a ping request to
the server conn
is connected to. A {:ok, conn, request_ref}
tuple is returned,
where conn
is the updated connection and request_ref
is a unique reference that
identifies this ping request. The response to a ping request is returned by stream/2
as a {:pong, request_ref}
tuple. If there's an error, this function returns
{:error, conn, reason}
where conn
is the updated connection and reason
is the
error reason.
payload
must be an 8-byte binary with arbitrary content. When the server responds to
a ping request, it will use that same payload. By default, the payload is an 8-byte
binary with all bits set to 0
.
Pinging can be used to measure the latency with the server and to ensure the connection is alive and well.
Examples
{:ok, conn, ref} = Mint.HTTP2.ping(conn)
Specs
Specs
put_settings(t(), settings()) :: {:ok, t()} | {:error, t(), Mint.Types.error()}
Communicates the given client settings to the server.
This function is HTTP/2-specific.
This function takes a connection and a keyword list of HTTP/2 settings and sends
the values of those settings to the server. The settings won't be effective until
the server acknowledges them, which will be handled transparently by stream/2
.
This function returns {:ok, conn}
when sending the settings to the server is
successful, with conn
being the updated connection. If there's an error, this
function returns {:error, conn, reason}
with conn
being the updated connection
and reason
being the reason of the error.
Supported settings
See setting/0
for the supported settings. You can see the meaning
of these settings in the corresponding section in the HTTP/2
RFC.
See the "HTTP/2 settings" section in the module documentation for more information.
Examples
{:ok, conn} = Mint.HTTP2.put_settings(conn, max_frame_size: 100)
Specs
recv(t(), non_neg_integer(), timeout()) :: {:ok, t(), [Mint.Types.response()]} | {:error, t(), Mint.Types.error(), [Mint.Types.response()]}
See Mint.HTTP.recv/3
.
Specs
request( t(), method :: String.t(), path :: String.t(), Mint.Types.headers(), body :: iodata() | nil | :stream ) :: {:ok, t(), Mint.Types.request_ref()} | {:error, t(), Mint.Types.error()}
See Mint.HTTP.request/5
.
In HTTP/2, opening a request means opening a new HTTP/2 stream (see the
module documentation). This means that a request could fail because the
maximum number of concurrent streams allowed by the server has been reached.
In that case, the error reason :too_many_concurrent_requests
is returned.
If you want to avoid incurring in this error, you can retrieve the value of
the maximum number of concurrent streams supported by the server through
get_server_setting/2
(passing in the :max_concurrent_streams
setting name).
Header list size
In HTTP/2, the server can optionally specify a maximum header list size that
the client needs to respect when sending headers. The header list size is calculated
by summing the length (in bytes) of each header name plus value, plus 32 bytes for
each header. Note that pseudo-headers (like :path
or :method
) count towards
this size. If the size is exceeded, an error is returned. To check what the size
is, use get_server_setting/2
.
Request body size
If the request body size will exceed the window size of the HTTP/2 stream created by the
request or the window size of the connection Mint will return a :exceeds_window_size
error.
To ensure you do not exceed the window size it is recommended to stream the request
body by initially passing :stream
as the body and sending the body in chunks using
stream_request_body/3
and using get_window_size/2
to get the window size of the
request and connection.
Specs
set_mode(t(), :active | :passive) :: {:ok, t()} | {:error, Mint.Types.error()}
See Mint.HTTP.set_mode/2
.
Specs
stream(t(), term()) :: {:ok, t(), [Mint.Types.response()]} | {:error, t(), Mint.Types.error(), [Mint.Types.response()]} | :unknown
See Mint.HTTP.stream/2
.
Specs
stream_request_body( t(), Mint.Types.request_ref(), iodata() | :eof | {:eof, trailing_headers :: Mint.Types.headers()} ) :: {:ok, t()} | {:error, t(), Mint.Types.error()}