Mint v0.4.0 Mint.HTTP View Source

Processless HTTP connection data structure and functions.

Single interface for Mint.HTTP1 and Mint.HTTP2 with support for version negotiation and proxies.

Usage

To establish a connection with a given server, use connect/4. This will return an opaque data structure that represents the connection to the server. To send a request, you can use request/5. Sending a request does not take care of the response to that request, instead we use Mint.stream/2 to process the response, which we will look at in just a bit. The connection is a wrapper around a TCP (:gen_tcp module) or SSL (:ssl module) socket that is set in active mode (with active: :once). This means that TCP/SSL messages will be delivered to the process that started the connection.

The process that owns the connection is responsible for receiving the messages (for example, a GenServer is responsible for defining handle_info/2). However, Mint.HTTP makes it easy to identify TCP/SSL messages that are coming from the connection with the server with the stream/2 function. This function takes the connection and a term and returns :unknown if the term is not a TCP/SSL message belonging to the connection. If the term is a message for the connection, then a response and a new connection are returned. It's important to store the new returned connection data structure over the old one since the connection is an immutable data structure.

Let's see an example of a common workflow of connecting to a server, sending a request, and processing the response. We start by using connect/3 to connect to a server.

{:ok, conn} = Mint.HTTP.connect(:http, "httpbin.org", 80)

conn is a data structure that represents the connection.

To send a request, we use request/5.

{:ok, conn, request_ref} = Mint.HTTP.request(conn, "GET", "/", [], nil)

As you can see, sending a request returns a new updated conn struct and a request_ref. The updated connection struct is returned because the connection is an immutable structure keeping the connection state, so every action we do on it must return a new, possibly updated, connection that we're responsible for storing over the old one. request_ref is a unique reference that can be used to identify which request a given response belongs to.

Now that we sent our request, we're responsible for receiving the messages that the TCP/SSL socket will send to our process. For example, in a GenServer we would do that with a handle_info/2 callback. In our case, we're going to use a simple receive. Mint.HTTP provides a way to tell if a message comes from the socket wrapped by our connection or not: the stream/2 function. If the message we pass to it is not destined for our connection, this function returns :unknown. Otherwise, it returns an updated connection and one or more responses.

receive do
  message ->
    case Mint.HTTP.stream(conn, message) do
      :unknown -> handle_normal_message(message)
      {:ok, conn, responses} -> handle_responses(conn, responses)
    end
end

responses is a list of possible responses. The most common responses are:

  • {:status, request_ref, status_code} for the status code
  • {:headers, request_ref, headers} for the response headers
  • {:data, request_ref, binary} for pieces of the response body
  • {:done, request_ref} for the end of the response

As you can see, all responses have the unique request reference as the second element of the tuple, so that we know which request the response belongs to. See Mint.Types.response/0 for the full list of responses returned by Mint.HTTP.stream/2.

Architecture

A processless architecture like the one here requires a few modifications to how we use this HTTP client. Usually, you will want to create this data structure in a process that acts as connection manager. Sometimes, you might want to have a single process responsible for multiple connections, either to just one host or multiple hosts. For more discussion on architectures based off of this HTTP client, see the Architecture page in the docs.

SSL certificates

When using SSL, you can pass in your own CA certificate store or use one provided by Mint. Mint doesn't ship with the certificate store itself, but it has an optional dependency on CAStore, which provides an up-to-date certificate store. If you don't want to use your own certificate store, just add :castore to your dependencies.

Mode

By default Mint operates in active mode meaning that the process that started the connection receives socket messages. Mint also supports passive mode, where no messages are sent to the process and the process needs to fetch data out of the socket manually. The mode can be controlled at connection time through the :mode option in connect/4 or changed dynamically through set_mode/2. Passive mode is generally only recommended for special use cases.

Link to this section Summary

Functions

Closes the given connection.

Creates a new connection to a given server.

Changes the controlling process of the given connection to new_pid.

Deletes a value in the private store.

Gets a private value from the connection.

Checks whether the connection is open.

Returns the number of open requests.

Assigns a new private key and value in the connection.

Receives data from the socket in a blocking way.

Sends a request to the connected server.

Changes the mode of the underlying socket.

Streams the next batch of responses from the given message.

Streams a chunk of the request body on the connection or signals the end of the body.

Link to this section Types

Link to this section Functions

Link to this function

close(conn) View Source
close(t()) :: {:ok, t()}

Closes the given connection.

This function closes the socket wrapped by the given connection. Once the socket is closed, the connection goes into the "closed" state and open?/1 returns false. You can throw away a closed connection.

Closing a connection does not guarantee that data that is in flight gets delivered to the server.

Always returns {:ok, conn} where conn is the updated connection.

Examples

{:ok, conn} = Mint.HTTP.close(conn)
Link to this function

connect(scheme, hostname, port, opts \\ []) View Source
connect(Mint.Types.scheme(), String.t(), :inet.port_number(), keyword()) ::
  {:ok, t()} | {:error, Mint.Types.error()}

Creates a new connection to a given server.

Creates a new connection struct and establishes the connection to the given server, identified by the given host and port combination. Both HTTP and HTTPS are supported by passing respectively :http and :https as the scheme.

The connection struct wraps a socket, which is created once the connection is established inside this function. If HTTP is used, then the created socket is a TCP socket and the :gen_tcp module is used to create that socket. If HTTPS is used, then the created socket is an SSL socket and the :ssl module is used to create that socket. The socket is created in active mode (with active: :once), which is why it is important to know the type of the socket: messages from the socket (of type t:socket_message/0 will be delivered directly to the process that creates the connection and tagged appropriately by the socket module (see the :gen_tcp and :ssl modules). See stream/2 for more information on the messages and how to process them and on the socket mode.

Options

  • :transport_opts - (keyword) options to be given to the transport being used. These options will be merged with some default options that cannot be overridden. For more details, refer to the "Transport options" section below.

  • :mode - (:active or :passive) whether to set the socket to active or passive mode. See the "Mode" section in the module documentation and set_mode/2.

  • :protocols - (list of atoms) a list of protocols to try when connecting to the server. The possible values in the list are :http1 for HTTP/1 and HTTP/1.1 and :http2 for HTTP/2. If only one protocol is present in the list, then the connection will be forced to use that protocol. If both :http1 and :http2 are present in the list, then Mint will negotiate the protocol. See the section "Protocol negotiation" below for more information. Defaults to [:http1, :http2].

The following options are HTTP/1-specific and will force the connection to be an HTTP/1 connection.

  • :proxy - a {scheme, hostname, port, opts} tuple that identifies a proxy to connect to. See the "Proxying" section below for more information.

The following options are HTTP/2-specific and will only be used on HTTP/2 connections.

  • :client_settings - (keyword) a list of client HTTP/2 settings to send to the server. See Mint.HTTP2.put_settings/2 for more information. This is only used in HTTP/2 connections.

Protocol negotiation

If both :http1 and :http2 are present in the list passed in the :protocols option, the protocol negotiation happens in the following way:

  • If the scheme used to connect to the server is :http, then HTTP/1 or HTTP/1.1 is used.

  • If the scheme is :https, then ALPN negotiation is used to determine the right protocol. This means that the server will decide whether to use HTTP/1 or HTTP/2. If the server doesn't support protocol negotiation, we will fall back to HTTP/1. If the server negotiates a protocol that we don't know how to handle, {:error, {:bad_alpn_protocol, protocol}} is returned.

Proxying

You can set up proxying through the :proxy option, which is a tuple {scheme, hostname, port, opts} that identifies the proxy to connect to. Once a proxied connection is returned, the proxy is transparent to you and you can use the connection like a normal HTTP/1 connection.

If the scheme is :http, we will connect to the host in the most compatible way, supporting older proxy servers. Data will be sent in clear text.

If the connection scheme is :https, we will connect to the host with a tunnel through the proxy. Using :https for both the proxy and the connection scheme is not supported, it is recommended to use :https for the end host connection instead of the proxy.

Transport options

The options specified in :transport_opts are passed to the module that implements the socket interface: :gen_tcp when the scheme is :http, and :ssl when the scheme is :https. Please refer to the documentation for those modules, as well as for :inet.setopts/2, for a detailed description of all available options.

The behaviour of some options is modified by Mint, as described below.

A special case is the :timeout option, which is passed to the transport module's connect function to limit the amount of time to wait for the network connection to be established.

Common options for :http and :https:

  • :active - controlled by the :mode option. Cannot be overridden.

  • :mode - set to :binary. Cannot be overriden.

  • :packet - set to :raw. Cannot be overridden.

  • :timeout - connect timeout in milliseconds. Defaults to 30_000 (30 seconds), and may be overridden by the caller. Set to :infinity to disable the connect timeout.

Options for :https only:

  • :alpn_advertised_protocols - managed by Mint. Cannot be overridden.

  • :cacertfile - if :verify is set to :verify_peer (the default) and no CA trust store is specified using the :cacertfile or :cacerts option, Mint will attempt to use the trust store from the CAStore package or raise an exception if this package is not available.

  • :ciphers - defaults to the list returned by :ssl.cipher_suites/0 filtered according to the blocklist in RFC7540 appendix A; May be overridden by the caller. See the "Supporting older cipher suites" section below for some examples.

  • :depth - defaults to 4. May be overridden by the caller.

  • :partial_chain_fun - unless a custom :partial_chain_fun is specified, Mint will enable its own partial chain handler, which accepts server certificate chains containing a certificate that was issued by a CA certificate in the CA trust store, even if that certificate is not last in the chain. This improves interoperability with some servers (for example, with a cross-signed intermediate CA or some misconfigured servers), but is a less strict interpretation of the TLS specification than the Erlang/OTP default behaviour.

  • :reuse_sessions - defaults to true. May be overridden by the caller.

  • :secure_renegotiate - defaults to true. May be overridden by the caller.

  • :server_name_indication - defaults to specified destination hostname. May be overridden by the caller.

  • :verify - defaults to :verify_peer. May be overridden by the caller.

  • :verify_fun - unless a custom :verify_fun is specified, or :verify is set to :verify_none, Mint will enable hostname verification with support for wildcards in the server's 'SubjectAltName' extension, similar to the behaviour implemented in :public_key.pkix_verify_hostname_match_fun(:https) in recent Erlang/OTP releases. This improves compatibility with recently issued wildcard certificates also on older Erlang/OTP releases.

  • :versions - defaults to [:"tlsv1.2"] (TLS v1.2 only). May be overridden by the caller.

Supporting older cipher suites

By default only a small list of modern cipher suites is enabled, in compliance with the HTTP/2 specification. Some servers, in particular HTTP/1 servers, may not support any of these cipher suites, resulting in TLS handshake failures or closed connections.

To select the default cipher suites of Erlang/OTP (including for example AES-CBC), use the following :transport_opts:

# Erlang/OTP 20.3 or later:
transport_opts: [ciphers: :ssl.cipher_suites(:default, :"tlsv1.2")]
# Older versions:
transport_opts: [ciphers: :ssl.cipher_suites()]

Recent Erlang/OTP releases do not enable RSA key exchange by default, due to known weaknesses. If necessary, you can build a cipher list with RSA exchange and use it in :transport_opts:

ciphers =
  :ssl.cipher_suites(:all, :"tlsv1.2")
  |> :ssl.filter_cipher_suites(
    key_exchange: &(&1 == :rsa),
    cipher: &(&1 in [:aes_256_gcm, :aes_128_gcm, :aes_256_cbc, :aes_128_cbc])
  )
  |> :ssl.append_cipher_suites(:ssl.cipher_suites(:default, :"tlsv1.2"))

Examples

{:ok, conn} = Mint.HTTP.connect(:http, "httpbin.org", 80)

Using a proxy:

proxy = {:http, "myproxy.example.com", 80, []}
{:ok, conn} = Mint.HTTP.connect(:https, "httpbin.org", 443, proxy: proxy)

Forcing the connection to be an HTTP/2 connection:

{:ok, conn} = Mint.HTTP.connect(:https, "http2.golang.org", 443, protocols: [:http2])

Enable all default cipher suites of Erlang/OTP (release 20.3 or later):

opts = [transport_opts: [ciphers: :ssl.cipher_suites(:default, :"tlsv1.2")]]
{:ok, conn} = Mint.HTTP.connect(:https, "httpbin.org", 443, opts)
Link to this function

controlling_process(conn, new_pid) View Source (since 0.3.0)
controlling_process(t(), pid()) ::
  {:ok, t()} | {:error, Mint.Types.error()}

Changes the controlling process of the given connection to new_pid.

The controlling process is a concept that comes from the Erlang TCP and SSL implementations. The controlling process of a connection is the process that started the connection and that receives the messages for that connection. You can change the controlling process of a connection through this function.

This function also takes care of "transferring" all the connection messages that are in the mailbox of the current controlling process to the new controlling process.

Remember that the connection is a data structure, so if you change the controlling process it doesn't mean you "transferred" the connection data structure itself to the other process, which you have to do manually (for example by sending the connection data structure to the new controlling process). If you do that, be careful of race conditions and be sure to retrieve the connection in the new controlling process before accepting connection messages in the new controlling process.

Note that changing controlling process means that you still need to be careful about only calling Mint functions from a single process (ideally the new controlling process) because the connection data structure is immutable.

Examples

send(new_pid, {:conn, conn})
{:ok, conn} = Mint.HTTP.controlling_process(conn, new_pid)

# In the "new_pid" process
receive do
  {:conn, conn} ->
    # Will receive connection messages.
end
Link to this function

delete_private(conn, key) View Source
delete_private(t(), atom()) :: t()

Deletes a value in the private store.

Deletes the private value stored under key in the connection. Returns the updated connection.

See also put_private/3 and get_private/3.

Examples

conn = Mint.HTTP.put_private(conn, :client_name, "Mint")

Mint.HTTP.get_private(conn, :client_name)
#=> "Mint"

conn = Mint.HTTP.delete_private(conn, :client_name)
Mint.HTTP.get_private(conn, :client_name)
#=> nil
Link to this function

get_private(conn, key, default \\ nil) View Source
get_private(t(), atom(), term()) :: term()

Gets a private value from the connection.

Retrieves a private value previously set with put_private/3 from the connection. key is the key under which the value to retrieve is stored. default is a default value returned in case there's no value under the given key.

See also put_private/3 and delete_private/2.

Examples

conn = Mint.HTTP.put_private(conn, :client_name, "Mint")

Mint.HTTP.get_private(conn, :client_name)
#=> "Mint"

Mint.HTTP.get_private(conn, :non_existent)
#=> nil
Link to this function

open?(conn, type \\ :read_write) View Source
open?(t(), :read | :write | :read_write) :: boolean()

Checks whether the connection is open.

This function returns true if the connection is open, false otherwise. It should be used to check that a connection is open before sending requests or performing operations that involve talking to the server.

The type argument can be used to tell whether the connection is closed only for reading, only for writing, or for both. In HTTP/1, a closed connection is always closed for both reading and writing. In HTTP/2, the connection can be closed only for writing but not for reading, meaning that you cannot send any more data to the server but you can still receive data from the server. See the "Closed connection" section in the module documentation of Mint.HTTP2.

If a connection is not open for reading and writing, it has become useless and you should get rid of it. If you still need a connection to the server, start a new connection with connect/4.

Examples

{:ok, conn} = Mint.HTTP.connect(:http, "httpbin.org", 80)
Mint.HTTP.open?(conn)
#=> true
Link to this function

open_request_count(conn) View Source
open_request_count(t()) :: non_neg_integer()

Returns the number of open requests.

Open requests are requests that have not yet received a :done response. This function returns the number of open requests for both HTTP/1 and HTTP/2, but for HTTP/2 only client-initiated requests are considered as open requests. See Mint.HTTP2.open_request_count/1 for more information.

Examples

{:ok, conn, _ref} = Mint.HTTP.request(conn, "GET", "/", [])
Mint.HTTP.open_request_count(conn)
#=> 1
Link to this function

put_private(conn, key, value) View Source
put_private(t(), atom(), term()) :: t()

Assigns a new private key and value in the connection.

This storage is meant to be used to associate metadata with the connection and it can be useful when handling multiple connections.

The given key must be an atom, while the given value can be an arbitrary term. The return value of this function is an updated connection.

See also get_private/3 and delete_private/2.

Examples

Let's see an example of putting a value and then getting it:

conn = Mint.HTTP.put_private(conn, :client_name, "Mint")
Mint.HTTP.get_private(conn, :client_name)
#=> "Mint"
Link to this function

recv(conn, byte_count, timeout) View Source (since 0.3.0)
recv(t(), non_neg_integer(), timeout()) ::
  {:ok, t(), [Mint.Types.response()]}
  | {:error, t(), Mint.Types.error(), [Mint.Types.response()]}

Receives data from the socket in a blocking way.

By default Mint operates in active mode, meaning that messages are delivered to the process that started the connection. However, Mint also supports passive mode (see the "Mode" section in the module documentation).

In passive mode, you'll need to manually get bytes out of the socket. You can do that with this function.

byte_count is the number of bytes you want out of the socket. If byte_count is 0, all available bytes will be returned.

timeout is the maximum time to wait before returning an error.

This function will raise an error if the socket is in active mode.

Examples

{:ok, conn, responses} = Mint.HTTP.recv(conn, 0, 5000)
Link to this function

request(conn, method, path, headers, body \\ nil) View Source
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()}

Sends a request to the connected server.

This function sends a new request to the server that conn is connected to. method is a string representing the method for the request, such as "GET" or "POST". path is the path on the host to send the request to. headers is a list of request headers in the form {header_name, header_value} with header_name and header_value being strings. body can have one of three values:

  • nil - no body is sent with the request. This is the default value.

  • iodata - the body to send for the request.

  • :stream - when the value of the body is :stream the request body can be streamed on the connection. See stream_request_body/3. In HTTP/1, you can't open a request if the body of another request is streaming.

If the request is sent correctly, this function returns {:ok, conn, request_ref}. conn is an updated connection that should be stored over the old connection. request_ref is a unique reference that can be used to match on responses for this request that are returned by stream/2. See stream/2 for more information.

If there's an error with sending the request, {:error, conn, reason} is returned. reason is the cause of the error. conn is an updated connection. It's important to store the returned connection over the old connection in case of errors too, because the state of the connection might change when there are errors as well. An error when sending a request does not necessarily mean that the connection is closed. Use open?/1 to verify that the connection is open.

Requests can be pipelined so the full response does not have to received before the next request can be sent. It is up to users to verify that the server supports pipelining and that the request is safe to pipeline.

In HTTP/1, you can't open a request if the body of another request is streaming. See Mint.HTTP1.request/5 for more information.

For a quick discussion on HTTP/2 streams and requests, see the Mint.HTTP2 module and Mint.HTTP2.request/5.

Examples

Mint.HTTP.request(conn, "GET", "/", _headers = [])
Mint.HTTP.request(conn, "POST", "/path", [{"content-type", "application/json"}], "{}")
Link to this function

set_mode(conn, mode) View Source (since 0.3.0)
set_mode(t(), :active | :passive) ::
  {:ok, t()} | {:error, Mint.Types.error()}

Changes the mode of the underlying socket.

To use the connection in active mode, where the process that started the connection receives socket messages, set the mode to :active (see also stream/2). To use the connection in passive mode, where you need to manually receive data from the socket, set the mode to :passive (see also recv/3).

The mode can also be controlled at connection time by the :mode option passed to connect/4.

Note that if you're switching from active to passive mode, you still might have socket messages in the process mailbox that you need to consume before doing any other operation on the connection.

See the "Mode" section in the module documentation for more information on modes.

Examples

{:ok, conn} = Mint.HTTP.set_mode(conn, :passive)
Link to this function

stream(conn, message) View Source
stream(t(), term()) ::
  {:ok, t(), [Mint.Types.response()]}
  | {:error, t(), Mint.Types.error(), [Mint.Types.response()]}
  | :unknown

Streams the next batch of responses from the given message.

This function processes a "message" which can be any term, but should be a message received by the process that owns the connection. Processing a message means that this function will parse it and check if it's a message that is directed to this connection, that is, a TCP/SSL message received on the connection's socket. If it is, then this function will parse the message, turn it into a list of responses, and possibly take action given the responses. As an example of an action that this function could perform, if the server sends a ping request this function will transparently take care of pinging the server back.

If there's no error, this function returns {:ok, conn, responses} where conn is the updated connection and responses is a list of responses. See the "Responses" section below. If there's an error, {:error, conn, reason, responses} is returned, where conn is the updated connection, reason is the error reason, and responses is a list of responses that were correctly parsed before the error.

If the given message is not from the connection's socket, this function returns :unknown.

Socket mode

Mint sets the socket in active: :once mode. This means that a single socket message at a time is delivered to the process that owns the connection. After a message is delivered, then no other messages are delivered (we say the socket goes in passive mode). When stream/2 is called to process the message that was received, Mint sets the socket back to active: :once. This is good to know in order to understand how the socket is handled by Mint, but in normal usage it just means that you will process one message at a time with stream/2 and not pay too much attention to the socket mode.

Mint also supports passive mode to avoid receiving messages. See the "Mode" section in the module documentation.

Responses

Each possible response returned by this function is a tuple with two or more elements. The first element is always an atom that identifies the kind of response. The second element is a unique reference t:request_ref/0 that identifies the request that the response belongs to. This is the term returned by request/5. After these two elements, there can be response-specific terms as well, documented below.

These are the possible responses that can be returned.

  • {:status, request_ref, status_code} - returned when the server replied with a response status code. The status code is a non-negative integer.

  • {:headers, request_ref, headers} - returned when the server replied with a list of headers. Headers are in the form {header_name, header_value} with header_name and header_value being strings. A single :headers response will come after the :status response. A single :headers response may come after all the :data responses if trailing headers are present.

  • {:data, request_ref, binary} - returned when the server replied with a chunk of response body (as a binary). The request shouldn't be considered done when a piece of body is received because multiple chunks could be received. The request is done when the :done response is returned.

  • {:done, request_ref} - returned when the server signaled the request as done. When this is received, the response body and headers can be considered complete and it can be assumed that no more responses will be received for this request. This means that for example, you can stop holding on to the request ref for this request.

  • {:error, request_ref, reason} - returned when there is an error that only affects the request and not the whole connection. For example, if the server sends bad data on a given request, that request will be closed an an error for that request will be returned among the responses, but the connection will remain alive and well.

  • {:pong, request_ref} - returned when a server replies to a ping request sent by the client. This response type is HTTP/2-specific and will never be returned by an HTTP/1 connection. See Mint.HTTP2.ping/2 for more information.

  • {:push_promise, request_ref, promised_request_ref, headers} - returned when the server sends a server push to the client. This response type is HTTP/2 specific and will never be returned by an HTTP/1 connection. See Mint.HTTP2 for more information on server pushes.

Examples

Let's assume we have a function called receive_next_and_stream/1 that takes a connection and then receives the next message, calls stream/2 with that message as an argument, and then returns the result of stream/2:

defp receive_next_and_stream(conn) do
  receive do
    message -> Mint.HTTP.stream(conn, message)
  end
end

Now, we can see an example of a workflow involving stream/2.

{:ok, conn, request_ref} = Mint.HTTP.request(conn, "GET", "/", _headers = [])

{:ok, conn, responses} = receive_next_and_stream(conn)
responses
#=> [{:status, ^request_ref, 200}]

{:ok, conn, responses} = receive_next_and_stream(conn)
responses
#=> [{:headers, ^request_ref, [{"Content-Type", "application/json"}]},
#=>  {:data, ^request_ref, "{"}]

{:ok, conn, responses} = receive_next_and_stream(conn)
responses
#=> [{:data, ^request_ref, "}"}, {:done, ^request_ref}]
Link to this function

stream_request_body(conn, ref, body) View Source
stream_request_body(
  t(),
  Mint.Types.request_ref(),
  iodata() | :eof | {:eof, trailing_headers :: Mint.Types.headers()}
) :: {:ok, t()} | {:error, t(), Mint.Types.error()}

Streams a chunk of the request body on the connection or signals the end of the body.

If a request is opened (through request/5) with the body as :stream, then the body can be streamed through this function. The function takes a conn, a request_ref returned by request/5 to identify the request to stream the body for, and a chunk of body to stream. The value of chunk can be:

  • iodata - a chunk of iodata is transmitted to the server as part of the body of the request.

  • :eof - signals the end of the streaming of the request body for the given request. Usually the server won't send any reply until this is sent.

  • {:eof, trailing_headers} - sends trailing headers and signals the end of the streaming of the request body for the given request. This behaves the same way as :eof but first sends the trailing headers. See the "Trailing headers" section below.

This function always returns an updated connection to be stored over the old connection.

For information about transfer encoding and content length in HTTP/1, see Mint.HTTP1.stream_request_body/3.

Trailing headers

HTTP trailing headers can be sent after the body of a request. The behaviour is slightly different for HTTP/1 and HTTP/2.

In HTTP/1, trailing headers are only supported if the transfer encoding is set to chunked. See Mint.HTTP1.stream_request_body/3 for more information on chunked transfer encoding.

In HTTP/2, trailing headers behave like normal headers. You don't need to care about the transfer encoding.

The trailer header

As specified in section 4.4 of RFC 7230, in HTTP/1 you need to specify which headers you're going to send as trailing headers using the trailer header. The trailer header applies to both HTTP/1 and HTTP/2. See the examples below for more information.

The te header

As specified in section 4.3 of RFC 7230, the te (or TE) header is used to specify which transfer-encodings the client is willing to accept (besides chunked). Mint supports decoding of trailing headers, but if you want to notify the server that you are accepting trailing headers, use the trailers value in the te header. For example:

Mint.HTTP.request(conn, "GET", "/", [{"te", "trailers"}], "some body")

Note that the te header can also be used to communicate which encodings you support to the server.

Examples

Let's see an example of streaming an empty JSON object ({}) by streaming one curly brace at a time.

headers = [{"content-type", "application/json"}, {"content-length", "2"}]
{:ok, conn, request_ref} = Mint.HTTP.request(conn, "POST", "/", headers, :stream)
{:ok, conn} = Mint.HTTP.stream_request_body(conn, request_ref, "{")
{:ok, conn} = Mint.HTTP.stream_request_body(conn, request_ref, "}")
{:ok, conn} = Mint.HTTP.stream_request_body(conn, request_ref, :eof)

Here's an example of sending trailing headers:

headers = [{"content-type", "application/json"}, {"trailer", "my-trailer, x-expires"}]
{:ok, conn, request_ref} = Mint.HTTP.request(conn, "POST", "/", headers, :stream)

{:ok, conn} = Mint.HTTP.stream_request_body(conn, request_ref, "{}")

trailing_headers = [{"my-trailer", "xxx"}, {"x-expires", "10 days"}]
{:ok, conn} = Mint.HTTP.stream_request_body(conn, request_ref, {:eof, trailing_headers})