If a client sends a lot more pipelined requests than we can handle, then
we need to provide backpressure so that the client knows to back off.
Do this by pausing both the stream and the parser itself when the
responses are not being read by the downstream client.
Backport of 085dd30
Avoid a costly buffer-to-string operation. Instead, allocate a new
buffer, copy the chunk header and data into it and send that.
The speed difference is negligible on small payloads but it really
shines with larger (10+ kB) chunks. benchmark/http/end-vs-write-end
with 64 kB chunks gives 45-50% higher throughput. With 1 MB chunks,
the difference is a staggering 590%.
Of course, YMMV will vary with real workloads and networks but this
commit should have a positive impact on CPU and memory consumption.
Big kudos to Wyatt Preul (@wpreul) for reporting the issue and providing
the initial patch.
Fixes#5941 and #5944.
There was previously up to a second exit delay when exiting node
right after an http request/response, due to the utcDate() function
doing a setTimeout to update the cached date/time.
Fixing this should increase the performance of our http tests.
This reverts commit a40133d10c.
Unfortunately, this breaks socket.io. Even though it's not strictly an
API change, it is too subtle and in too brittle an area of node, to be
done in a stable branch.
Conflicts:
doc/api/http.markdown
Otherwise, writing an empty string causes the whole program to grind to
a halt when piping data into http messages.
This wasn't as much of a problem (though it WAS a bug) in 0.8 and
before, because our hyperactive 'drain' behavior would mean that some
*previous* write() would probably have a pending drain event, and cause
things to start moving again.
This commit adds an optimization to the HTTP client that makes it
possible to:
* Pack the headers and the first chunk of the request body into a
single write().
* Pack the chunk header and the chunk itself into a single write().
Because only one write() system call is issued instead of several,
the chances of data ending up in a single TCP packet are phenomenally
higher: the benchmark with `type=buf size=32` jumps from 50 req/s to
7,500 req/s, a 150-fold increase.
This commit removes the check from e4b716ef that pushes binary encoded
strings into the slow path. The commit log mentions that:
We were assuming that any string can be concatenated safely to
CRLF. However, for hex, base64, or binary encoded writes, this
is not the case, and results in sending the incorrect response.
For hex and base64 strings that's certainly true but binary strings
are 'das Ding an sich': string.length is the same before and after
decoding.
Fixes#5528.
Fixes#3740
In the case of pipelined requests, you can have a situation where
the socket gets destroyed via one req/res object, but then trying
to destroy *another* req/res on the same socket will cause it to
call undefined.destroy(), since it was already removed from that
message.
Add a guard to OutgoingMessage.destroy and IncomingMessage.destroy
to prevent this error.
We were assuming that any string can be concatenated safely to
CRLF. However, for hex, base64, or binary encoded writes, this
is not the case, and results in sending the incorrect response.
An unusual edge case, but certainly a bug.
If an http response has an 'end' handler that throws, then the socket
will never be released back into the pool.
Granted, we do NOT guarantee that throwing will never have adverse
effects on Node internal state. Such a guarantee cannot be reasonably
made in a shared-global mutable-state side-effecty language like
JavaScript. However, in this case, it's a rather trivial patch to
increase our resilience a little bit, so it seems like a win.
There is no semantic change in this case, except that some event
listeners are removed, and the `'free'` event is emitted on nextTick, so
that you can schedule another request which will re-use the same socket.
From the user's point of view, there should be no detectable difference.
Closes#5107
The benefits of the hot-path optimization below start to fall off when
the buffer size gets up near 128KB, because the cost of the copy is more
than the cost of the extra write() call. Switch to the write/end method
at that point.
Heuristics and magic numbers are awful, but slow http responses are
worse.
Fix#4975
This solves the problem of calling `readable.pipe(writable)` after the
readable stream has already emitted 'end', as often is the case when
writing simple HTTP proxies.
The spirit of streams2 is that things will work properly, even if you
don't set them up right away on the first tick.
This approach breaks down, however, because pipe()ing from an ended
readable will just do nothing. No more data will ever arrive, and the
writable will hang open forever never being ended.
However, that does not solve the case of adding a `on('end')` listener
after the stream has received the EOF chunk, if it was the first chunk
received (and thus, length was 0, and 'end' got emitted). So, with
this, we defer the 'end' event emission until the read() function is
called.
Also, in pipe(), if the source has emitted 'end' already, we call the
cleanup/onend function on nextTick. Piping from an already-ended stream
is thus the same as piping from a stream that is in the process of
ending.
Updates many tests that were relying on 'end' coming immediately, even
though they never read() from the req.
Fix#4942
Fix#4948
This adds a check before setting the incoming parser
to null. Under certain circumstances it'll already be set to
null by freeParser().
Otherwise this will cause node to crash as it tries to set
null on something that is already null.
Fix#4948
This adds a check before setting the incoming parser
to null. Under certain circumstances it'll already be set to
null by freeParser().
Otherwise this will cause node to crash as it tries to set
null on something that is already null.
This adds the following to HTTP:
* server.setTimeout(msecs, callback)
Sets all new connections to time out after the specified time, at
which point it emits 'timeout' on the server, passing the socket as an
argument.
In this way, timeouts can be handled in one place consistently.
* req.setTimeout(), res.setTimeout()
Essentially an alias to req/res.socket.setTimeout(), but without
having to delve into a "buried" object. Adds a listener on the
req/res object, but not on the socket.
* server.timeout
Number of milliseconds before incoming connections time out.
(Default=1000*60*2, as before.)
Furthermore, if the user sets up their own timeout listener on either
the server, the request, or the response, then the default behavior
(destroying the socket) is suppressed.
Fix#3460
Ability to return just the length of listeners for a given type, using
EventEmitter.listenerCount(emitter, event). This will be a lot cheaper
than creating a copy of the listeners array just to check its length.
Register the 'close' event listener with .once(), not .on().
It doesn't matter in the grand scheme of things because the listener
doesn't keep references to any heavy-weight objects but using .once()
for a oneshot listener is something of a best practice.
This makes it so that `stream.push(chunk)` is the only way to signal the
end of reading, removing the confusing disparity between the
callback-style _read method, and the fact that most real-world streams
do not have a 1:1 corollation between the "please give me data" event,
and the actual arrival of a chunk of data.
It is still possible, of course, to implement a `CallbackReadable` on
top of this. Simply provide a method like this as the callback:
function readCallback(er, chunk) {
if (er)
stream.emit('error', er);
else
stream.push(chunk);
}
However, *only* fs streams actually would behave in this way, so it
makes not a lot of sense to make TCP, TLS, HTTP, and all the rest have
to bend into this uncomfortable paradigm.
This appears to fix#4673. That bug is very hard to reproduce, so it's
hard to tell for certain, but this approach is more correct anyway.
Hat-tip: @dougwilson
Prior to v0.10, Node ignored ECONNRESET errors in many situations.
There *are* valid cases in which ECONNRESET should be ignored as a
normal part of the TCP dance, but in many others, it's a very relevant
signal that must be heeded with care.
Exacerbating this problem, if the OutgoingMessage does not have a
req.connection._handle, it assumes that it is in the process of
connecting, and thus buffers writes up in an array.
The problem happens when you reuse a socket between two requests, and it
is destroyed abruptly in between them. The writes will be buffered,
because the socket has no handle, but it's not ever going to GET a
handle, because it's not connecting, it's destroyed.
The proper fix is to treat ECONNRESET correctly. However, this is a
behavior/semantics change, and cannot land in a stable branch.
Fix#4775
This is similar to commit 2cbf458 but this time for 204 No Content
instead of 304 Not Modified responses.
When the user sends a 204 response with a Transfer-Encoding: chunked
header, suppress sending the zero chunk and force the connection to
close.
Force the connection to close when the response is a 304 Not Modified
and the user has set a "Transfer-Encoding: chunked" header.
RFC 2616 mandates that 304 responses MUST NOT have a body but node.js
used to send out a zero chunk anyway to accommodate clients that don't
have special handling for 304 responses.
It was pointed out that this might confuse reverse proxies to the point
of creating security liabilities, so suppress the zero chunk and force
the connection to close.
Fix the following exception:
http.js:974
this._httpMessage.emit('close');
^
TypeError: Cannot call method 'emit' of null
at Socket.onServerResponseClose (http.js:974:21)
at Socket.EventEmitter.emit (events.js:124:20)
at net.js:421:10
at process._tickCallback (node.js:386:13)
at process._makeCallback (node.js:304:15)
Fixes#4586.
Apply the same optimization to res.end(buf) that is applied to res.end(str).
Speeds up `node benchmark/http_simple_auto -k -c 1 -n 25000 buffer/1`
(non-chunked response body) by about 750x. That's not a typo.
Chunked responses:
$ cat tmp/http-chunked-client.js
// Run `node benchmark/http_simple` in another terminal.
var http = require('http'), url = require('url');
var options = url.parse('http://127.0.0.1:8000/buffer/1/1');
options.agent = new http.Agent({ maxSockets: 1 });
for (var i = 0; i < 25000; ++i) http.get(options);
Before:
$ time out/Release/node tmp/http-chunked-client.js
real 16m40.411s
user 0m9.184s
sys 0m0.604s
After:
$ time out/Release/node tmp/http-chunked-client.js
real 0m5.386s
user 0m2.768s
sys 0m0.728s
That's still a 185x speed-up.
Fixes#4415.
isaacs
Ben Noordhuis
Peter Rust
Nathan Zadoks
koichik
hc
Eugene Girshov
Trevor Norris
Bert Belder
Alexandr Emelin
Ryunosuke SATO
Shigeki Ohtsu