aboutsummaryrefslogtreecommitdiffstats
path: root/doc
diff options
context:
space:
mode:
authorLoïc Hoguin <[email protected]>2016-08-31 17:01:25 +0200
committerLoïc Hoguin <[email protected]>2016-08-31 17:01:25 +0200
commitaa617f23307b38fe3bc2b1295e7bf7bd3ec52d24 (patch)
tree3cf5c1cd17480bd59f0263cfa81d98b56c91edb2 /doc
parentbae10829bafb46e5a1abf4e7eb42fde352d6f0c3 (diff)
downloadcowboy-aa617f23307b38fe3bc2b1295e7bf7bd3ec52d24.tar.gz
cowboy-aa617f23307b38fe3bc2b1295e7bf7bd3ec52d24.tar.bz2
cowboy-aa617f23307b38fe3bc2b1295e7bf7bd3ec52d24.zip
Update the guide chapter for responses
Diffstat (limited to 'doc')
-rw-r--r--doc/src/guide/resp.asciidoc363
1 files changed, 240 insertions, 123 deletions
diff --git a/doc/src/guide/resp.asciidoc b/doc/src/guide/resp.asciidoc
index 1ffdfbd..b8ba7d5 100644
--- a/doc/src/guide/resp.asciidoc
+++ b/doc/src/guide/resp.asciidoc
@@ -1,201 +1,318 @@
[[resp]]
== Sending a response
-The Req object also allows you to send a response.
+The response must be sent using the Req object.
-You can only send one response. Any other attempt will
-trigger a crash. The response may be sent in one go or
-with its body streamed by chunks of arbitrary size.
+Cowboy provides two different ways of sending responses:
+either directly or by streaming the body. Response headers
+and body may be set in advance. The response is sent as
+soon as one of the reply or stream reply function is
+called.
-You can also set headers or the response body in advance
-and Cowboy will use them when you finally do reply.
+Cowboy also provides a simplified interface for sending
+files. It can also send only specific parts of a file.
+
+While only one response is allowed for every request,
+HTTP/2 introduced a mechanism that allows the server
+to push additional resources related to the response.
+This chapter also describes how this feature works in
+Cowboy.
=== Reply
-You can send a reply with no particular headers or body.
-Cowboy will make sure to send the mandatory headers with
-the response.
+Cowboy provides three functions for sending the entire reply,
+depending on whether you need to set headers and body. In all
+cases, Cowboy will add any headers required by the protocol
+(for example the date header will always be sent).
+
+When you need to set only the status code,
+use `cowboy_req:reply/2`:
[source,erlang]
-Req2 = cowboy_req:reply(200, Req).
+Req = cowboy_req:reply(200, Req0).
-You can define headers to be sent with the response. Note
-that header names must be lowercase. Again, Cowboy will
-make sure to send the mandatory headers with the response.
+When you need to set response headers at the same time,
+use `cowboy_req:reply/3`:
[source,erlang]
----
-Req2 = cowboy_req:reply(303, [
- {<<"location">>, <<"http://ninenines.eu">>}
-], Req).
+Req = cowboy_req:reply(303, #{
+ <<"location">> => <<"http://ninenines.eu">>
+}, Req0).
----
-You can override headers that Cowboy would send otherwise.
-Any header set by the user will be used over the ones set
-by Cowboy. For example, you can advertise yourself as a
-different server.
+Note that the header name must always be a lowercase
+binary.
+
+When you also need to set the response body,
+use `cowboy_req:reply/4`:
[source,erlang]
----
-Req2 = cowboy_req:reply(200, [
- {<<"server">>, <<"yaws">>}
-], Req).
+Req = cowboy_req:reply(200, #{
+ <<"content-type">> => <<"text/plain">>
+}, "Hello world!", Req0).
----
-We also saw earlier how to force close the connection by
-overriding the connection header.
+You should always set the content-type header when the
+response has a body. There is however no need to set
+the content-length header; Cowboy does it automatically.
-Finally, you can also send a body with the response. Cowboy
-will automatically set the content-length header if you do.
-We recommend that you set the content-type header so the
-client may know how to read the body.
+The response body and the header values must be either
+a binary or an iolist. An iolist is a list containing
+binaries, characters, strings or other iolists. This
+allows you to build a response from different parts
+without having to do any concatenation:
[source,erlang]
----
-Req2 = cowboy_req:reply(200, [
- {<<"content-type">>, <<"text/plain">>}
-], "Hello world!", Req).
+Title = "Hello world!",
+Body = <<"Hats off!">>,
+Req = cowboy_req:reply(200, #{
+ <<"content-type">> => <<"text/html">>
+}, ["<html><head><title>", Title, "</title></head>",
+ "<body><p>", Body, "</p></body></html>"], Req0).
----
-Here is the same example but sending HTML this time.
+This method of building responses is more efficient than
+concatenating. Behind the scenes, each element of the list
+is simply a pointer, and those pointers are used directly
+when writing to the socket.
-[source,erlang]
-----
-Req2 = cowboy_req:reply(200, [
- {<<"content-type">>, <<"text/html">>}
-], "<html><head>Hello world!</head><body><p>Hats off!</p></body></html>", Req).
-----
+=== Stream reply
-Note that the reply is sent immediately.
+Cowboy provides two functions for initiating a response,
+and an additional function for streaming the response body.
+Cowboy will add any required headers to the response.
-=== Chunked reply
+// @todo For HTTP/1.1 Cowboy should probably not use chunked transfer-encoding if the content-length is set.
-You can also stream the response body. First, you need to
-initiate the reply by sending the response status code.
-Then you can send the body in chunks of arbitrary size.
+When you need to set only the status code,
+use `cowboy_req:stream_reply/2`:
[source,erlang]
-Req2 = cowboy_req:chunked_reply(200, Req),
-cowboy_req:chunk("Hello...", Req2),
-cowboy_req:chunk("chunked...", Req2),
-cowboy_req:chunk("world!!", Req2).
+----
+Req = cowboy_req:stream_reply(200, Req0),
-You should make sure to match on `ok` as an error may be
-returned.
+cowboy_req:stream_body("Hello...", nofin, Req),
+cowboy_req:stream_body("chunked...", nofin, Req),
+cowboy_req:stream_body("world!!", fin, Req).
+----
+
+The second argument to `cowboy_req:stream_body/3` indicates
+whether this data terminates the body. Use `fin` for the
+final flag, and `nofin` otherwise.
-While it is possible to send a chunked response without
-a content-type header, it is still recommended. You can
-set this header or any other just like for normal replies.
+This snippet does not set a content-type header. This is
+not recommended. All responses with a body should have
+a content-type. The header can be set beforehand, or
+using the `cowboy_req:stream_reply/3`:
[source,erlang]
----
-Req2 = cowboy_req:chunked_reply(200, [
- {<<"content-type">>, <<"text/html">>}
-], Req),
-cowboy_req:chunk("<html><head>Hello world!</head>", Req2),
-cowboy_req:chunk("<body><p>Hats off!</p></body></html>", Req2).
+Req = cowboy_req:stream_reply(200, #{
+ <<"content-type">> => <<"text/html">>
+}, Req0),
+
+cowboy_req:stream_body("<html><head>Hello world!</head>", nofin, Req),
+cowboy_req:stream_body("<body><p>Hats off!</p></body></html>", fin, Req).
----
-Note that the reply and each chunk following it are sent
-immediately.
+HTTP provides a few different ways to stream response bodies.
+Cowboy will select the most appropriate one based on the HTTP
+version and the request and response headers.
+
+While not required by any means, it is recommended that you
+set the content-length header in the response if you know it
+in advance. This will ensure that the best response method
+is selected and help clients understand when the response
+is fully received.
+
+// @todo Document trailers here.
=== Preset response headers
-You can define response headers in advance. They will be
-merged into the headers given in the reply call. Headers
-in the reply call override preset response headers which
-override the default Cowboy headers.
+Cowboy provides functions to set response headers without
+immediately sending them. They are stored in the Req object
+and sent as part of the response when a reply function is
+called.
+
+To set response headers:
[source,erlang]
-Req2 = cowboy_req:set_resp_header(<<"allow">>, "GET", Req).
+Req = cowboy_req:set_resp_header(<<"allow">>, "GET", Req0).
+
+Header names must be a lowercase binary.
-You can check if a response header has already been set.
-This will only check the response headers that you set,
-and not the ones Cowboy will add when actually sending
-the reply.
+Do not use this function for setting cookies. Refer to
+the xref:cookies[Cookies] chapter for more information.
+
+To check if a response header has already been set:
[source,erlang]
cowboy_req:has_resp_header(<<"allow">>, Req).
-It will return `true` if the header is defined, and `false`
-otherwise.
+It returns `true` if the header was set, `false` otherwise.
-Finally, you can also delete a preset response header if
-needed. If you do, it will not be sent.
+To delete a response header that was set previously:
[source,erlang]
-Req2 = cowboy_req:delete_resp_header(<<"allow">>, Req).
+Req = cowboy_req:delete_resp_header(<<"allow">>, Req0).
-=== Preset response body
+=== Overriding headers
-You can set the response body in advance. Note that this
-body will be ignored if you then choose to send a chunked
-reply, or if you send a reply with an explicit body.
+As Cowboy provides different ways of setting response
+headers and body, clashes may occur, so it's important
+to understand what happens when a header is set twice.
-[source,erlang]
-Req2 = cowboy_req:set_resp_body("Hello world!", Req).
+Headers come from five different origins:
+
+* Protocol-specific headers (for example HTTP/1.1's connection header)
+* Other required headers (for example the date header)
+* Preset headers
+* Headers given to the reply function
+* Set-cookie headers
+
+Cowboy does not allow overriding protocol-specific headers.
+
+Set-cookie headers will always be appended at the end of
+the list of headers before sending the response.
+
+Headers given to the reply function will always override
+preset headers and required headers. If a header is found
+in two or three of these, then the one in the reply function
+is picked and the others are dropped.
-You can also set a fun that will be called when it is time
-to send the body. There are three different ways of doing
-that.
+Similarly, preset headers will always override required
+headers.
-If you know the length of the body that needs to be sent,
-you should specify it, as it will help clients determine
-the remaining download time and allow them to inform the
-user.
+To illustrate, look at the following snippet. Cowboy by
+default sends the server header with the value "Cowboy".
+We can override it:
[source,erlang]
----
-F = fun (Socket, Transport) ->
- Transport:send(Socket, "Hello world!")
-end,
-Req2 = cowboy_req:set_resp_body_fun(12, F, Req).
+Req = cowboy_req:reply(200, #{
+ <<"server">> => <<"yaws">>
+}, Req0).
----
-If you do not know the length of the body, you should use
-a chunked response body fun instead.
+=== Preset response body
+
+Cowboy provides functions to set the response body without
+immediately sending it. It is stored in the Req object and
+sent when the reply function is called.
+
+To set the response body:
[source,erlang]
-----
-F = fun (SendChunk) ->
- Body = lists:duplicate(random:uniform(1024, $a)),
- SendChunk(Body)
-end,
-Req2 = cowboy_req:set_resp_body_fun(chunked, F, Req).
-----
+Req = cowboy_req:set_resp_body("Hello world!", Req0).
+
+// @todo Yeah we probably should add that function that
+// also sets the content-type at the same time...
-Finally, you can also send data on the socket directly,
-without knowing the length in advance. Cowboy may be
-forced to close the connection at the end of the response
-though depending on the protocol capabilities.
+To check if a response body has already been set:
+
+[source,erlang]
+cowboy_req:has_resp_body(Req).
+
+It returns `true` if the body was set and is non-empty,
+`false` otherwise.
+
+// @todo We probably should also have a function that
+// properly removes the response body, including any
+// content-* headers.
+
+The preset response body is only sent if the reply function
+used is `cowboy_req:reply/2` or `cowboy_req:reply/3`.
+
+=== Sending files
+
+Cowboy provides a shortcut for sending files. When
+using `cowboy_req:reply/4`, or when presetting the
+response header, you can give a `sendfile` tuple to
+Cowboy:
+
+[source,erlang]
+{sendfile, Offset, Length, Filename}
+
+Depending on the values for `Offset` or `Length`, the
+entire file may be sent, or just a part of it.
+
+The length is required even for sending the entire file.
+Cowboy sends it in the content-length header.
+
+To send a file while replying:
[source,erlang]
----
-F = fun (Socket, Transport) ->
- Body = lists:duplicate(random:uniform(1024, $a)),
- Transport:send(Socket, Body)
-end,
-Req2 = cowboy_req:set_resp_body_fun(F, Req).
+Req = cowboy_req:reply(200, #{
+ <<"content-type">> => "image/png"
+}, {sendfile, 0, 12345, "path/to/logo.png"}, Req0).
----
-=== Sending files
+// @todo An example of presetting a file would be useful,
+// but let's wait for the function that can set the
+// content-type at the same time.
+
+// @todo What about streaming many files? For example
+// it should be possible to build a tar file on the fly
+// while still using sendfile. Another example could be
+// proper support for multipart byte ranges. Yet another
+// example would be automatic concatenation of CSS or JS
+// files.
+
+=== Push
+
+The HTTP/2 protocol introduced the ability to push resources
+related to the one sent in the response. Cowboy provides two
+functions for that purpose: `cowboy_req:push/3,4`.
+
+Push is only available for HTTP/2. Cowboy will automatically
+ignore push requests if the protocol doesn't support it.
+
+The push function must be called before any of the reply
+functions. Doing otherwise will result in a crash.
+
+To push a resource, you need to provide the same information
+as a client performing a request would. This includes the
+HTTP method, the URI and any necessary request headers.
-You can send files directly from disk without having to
-read them. Cowboy will use the `sendfile` syscall when
-possible, which means that the file is sent to the socket
-directly from the kernel, which is a lot more performant
-than doing it from userland.
+Cowboy by default only requires you to give the path to
+the resource and the request headers. The rest of the URI
+is taken from the current request (excluding the query
+string, set to empty) and the method is GET by default.
-Again, it is recommended to set the size of the file if it
-can be known in advance.
+The following snippet pushes a CSS file that is linked to
+in the response:
[source,erlang]
----
-F = fun (Socket, Transport) ->
- Transport:sendfile(Socket, "priv/styles.css")
-end,
-Req2 = cowboy_req:set_resp_body_fun(FileSize, F, Req).
+cowboy_req:push("/static/style.css", #{
+ <<"accept">> => <<"text/css">>
+}, Req0),
+Req = cowboy_req:reply(200, #{
+ <<"content-type">> => <<"text/html">>
+}, ["<html><head><title>My web page</title>",
+ "<link rel='stylesheet' type='text/css' href='/static/style.css'>",
+ "<body><p>Welcome to Erlang!</p></body></html>"], Req0).
----
-Please see the Ranch guide for more information about
-sending files.
+To override the method, scheme, host, port or query string,
+simply pass in a fourth argument. The following snippet
+uses a different host name:
+
+[source,erlang]
+----
+cowboy_req:push("/static/style.css", #{
+ <<"accept">> => <<"text/css">>
+}, #{host => <<"cdn.example.org">>}, Req),
+----
+
+Pushed resources don't have to be files. As long as the push
+request is cacheable, safe and does not include a body, the
+resource can be pushed.
+
+Under the hood, Cowboy handles pushed requests the same as
+normal requests: a different process is created which will
+ultimately send a response to the client.