cowboy_stream - Stream handlers


The module cowboy_stream defines a callback interface and a protocol for handling HTTP streams.

An HTTP request and its associated response is called a stream. A connection may have many streams. In HTTP/1.1 they are executed sequentially, while in HTTP/2 they are executed concurrently.

Cowboy calls the stream handler for nearly all events related to a stream. Exceptions vary depending on the protocol.

Extra care must be taken when implementing stream handlers to ensure compatibility. While some modification of the events and commands is allowed, it is generally not a good idea to completely omit them.


Stream handlers must implement the following interface:

init(StreamID, Req, Opts) -> {Commands, State}
data(StreamID, IsFin, Data, State) -> {Commands, State}
info(StreamID, Info, State) -> {Commands, State}
terminate(StreamID, Reason, State) -> any()
early_error(StreamID, Reason, PartialReq, Resp, Opts) -> Resp

StreamID   :: cowboy_stream:streamid()
Req        :: cowboy_req:req()
Opts       :: cowboy:opts()
Commands   :: cowboy_stream:commands()
State      :: any()
IsFin      :: cowboy_stream:fin()
Data       :: binary()
Info       :: any()
Reason     :: cowboy_stream:reason()
PartialReq  - cowboy_req:req(), except all fields are optional
Resp       :: cowboy_stream:resp_command()

HTTP/1.1 will initialize a stream only when the request-line and all headers have been received. When errors occur before that point Cowboy will call the callback early_error/5 with a partial request, the error reason and the response Cowboy intends to send. All other events go throuh the stream handler using the normal callbacks.

HTTP/2 will initialize the stream when the HEADERS block has been fully received and decoded. Any protocol error occuring before that will not result in a response being sent and will therefore not go through the stream handler. In addition Cowboy may terminate streams without sending an HTTP response back.

The stream is initialized by calling init/3. All streams that are initialized will eventually be terminated by calling terminate/3.

When Cowboy receives data for the stream it will call data/4. The data given is the request body after any transfer decoding has been applied.

When Cowboy receives a message addressed to a stream, or when Cowboy needs to inform the stream handler that an internal event has occurred, it will call info/3.


Stream handlers can return a list of commands to be executed from the init/3, data/4 and info/3 callbacks. In addition, the early_error/5 callback must return a response command.

The following commands are defined:


Send an informational response to the client.

{inform, cowboy:http_status(), cowboy:http_headers()}

Any number of informational responses may be sent, but only until the final response is sent.


Send a response to the client.

{response, cowboy:http_status(), cowboy:http_headers(),

No more data can be sent after this command.


Initiate a response to the client.

{headers, cowboy:http_status(), cowboy:http_headers()}

This initiates a response to the client. The stream will end when a data command with the fin flag or a trailer command is returned.


Send data to the client.

{data, fin(), iodata()}


Send response trailers to the client.

{trailers, cowboy:http_headers()}


Push a resource to the client.

{push, Method, Scheme, Host, inet:port_number(),
    Path, Qs, cowboy:http_headers()}

Method = Scheme = Host = Path = Qs = binary()

The command will be ignored if the protocol does not provide any server push mechanism.


{flow, pos_integer()}

Request more data to be read from the request body. The exact behavior depends on the protocol.


Inform Cowboy that a process was spawned and should be supervised.

{spawn, pid(), timeout()}


Send an error response if no response was sent previously.

{error_response, cowboy:http_status(), cowboy:http_headers(), iodata()}


Switch to a different protocol.

{switch_protocol, cowboy:http_headers(), module(), state()}

Contains the headers that will be sent in the 101 response, along with the module implementing the protocol we are switching to and its initial state.

Note that the 101 informational response will not be sent after a final response.


Stop the stream.


While no more data can be sent after the fin flag was set, the stream is still tracked by Cowboy until it is stopped by the handler.

The behavior when stopping a stream for which no response has been sent will vary depending on the protocol. The stream will end successfully as far as the client is concerned.

To indicate that an error occurred, either use error_response before stopping, or use internal_error.


Stop the stream with an error.

{internal_error, Reason, HumanReadable}

Reason        = any()
HumanReadable = atom()

This command should be used when the stream cannot continue because of an internal error. An error_response command may be sent before that to advertise to the client why the stream is dropped.

Predefined events

Cowboy will forward all messages sent to the stream to the info/3 callback. To send a message to a stream, send a message to the connection process with the form {{Pid, StreamID}, Msg}. The connection process will then forward Msg to the stream handlers.

Cowboy will also forward the exit signals for the processes that the stream spawned.


A process spawned by this stream has exited.

{'EXIT', pid(), any()}

This is the raw exit message without any modification.


Same as the inform command.

Sent when the request process reads the body and an expect: 100-continue header was present in the request, or when the request process sends an informational response on its own.


Same as the response command.

Usually sent when the request process replies to the client. May also be sent by Cowboy internally.


Same as the headers command.

Sent when the request process starts replying to the client.


Same as the data command.

Sent when the request process streams data to the client.


Same as the trailers command.

Sent when the request process sends the trailer field values to the client.


Same as the push command.

Sent when the request process pushes a resource to the client.


Same as the switch_protocol command.

Sent when switching to the HTTP/2 or Websocket protocol.


The following function should be called by modules implementing stream handlers to execute the next stream handler in the list:



commands() :: [Command]

See the list of commands for details.


fin() :: fin | nofin

Used in commands and events to indicate that this is the end of the stream.


req() :: #{
    method  => binary(),               %% case sensitive
    version => cowboy:http_version() | atom(),
    scheme  => binary(),               %% lowercase; case insensitive
    host    => binary(),               %% lowercase; case insensitive
    port    => inet:port_number(),
    path    => binary(),               %% case sensitive
    qs      => binary(),               %% case sensitive
    headers => cowboy:http_headers(),
    peer    => {inet:ip_address(), inet:port_number()}

Partial request information received when an early error is detected.


reason() :: normal | switch_protocol
    | {internal_error, timeout | {error | exit | throw, any()}, HumanReadable}
    | {socket_error, closed | atom(), HumanReadable}
    | {stream_error, Error, HumanReadable}
    | {connection_error, Error, HumanReadable}
    | {stop, cow_http2:frame(), HumanReadable}

Error         = atom()
HumanReadable = atom()

Reason for the stream termination.


resp_command() :: {response, cowboy:http_status(),
    cowboy:http_headers(), cowboy_req:resp_body()}

See the response command for details.


streamid() :: any()

The identifier for this stream.

The identifier is unique over the connection process. It is possible to form a unique identifier node-wide and cluster-wide by wrapping it in a {self(), StreamID} tuple.


  • 2.2: The trailers command was introduced.
  • 2.0: Module introduced.

See also

cowboy(7), cowboy_http(3), cowboy_http2(3)

Cowboy 2.5 Function Reference


Version select

Like my work? Donate!

Donate to Loïc Hoguin because his work on Cowboy, Ranch, Gun and is fantastic:

Recurring payment options are also available via GitHub Sponsors. These funds are used to cover the recurring expenses like food, dedicated servers or domain names.