http: save roundtrips, convert buffers to strings

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.

2 files changed, 87 insertions, 2 deletions

diff --git a/benchmark/http/client-request-body.js b/benchmark/http/client-request-body.js
new file mode 100644
index 0000000000..7a3468a670
--- /dev/null
+++ b/benchmark/http/client-request-body.js
@@ -0,0 +1,69 @@
+// Measure the time it takes for the HTTP client to send a request body.
+
+var common = require('../common.js');
+var http = require('http');
+
+var bench = common.createBenchmark(main, {
+  dur: [5],
+  type: ['asc', 'utf', 'buf'],
+  bytes: [32, 256, 1024],
+  method: ['write', 'end  '] // two spaces added to line up each row
+});
+
+function main(conf) {
+  var dur = +conf.dur;
+  var len = +conf.bytes;
+
+  var encoding;
+  var chunk;
+  switch (conf.type) {
+    case 'buf':
+      chunk = new Buffer(len);
+      chunk.fill('x');
+      break;
+    case 'utf':
+      encoding = 'utf8';
+      chunk = new Array(len / 2 + 1).join('ü');
+      break;
+    case 'asc':
+      chunk = new Array(len + 1).join('a');
+      break;
+  }
+
+  var nreqs = 0;
+  var options = {
+    headers: { 'Connection': 'keep-alive', 'Transfer-Encoding': 'chunked' },
+    agent: new http.Agent({ maxSockets: 1 }),
+    host: '127.0.0.1',
+    port: common.PORT,
+    path: '/',
+    method: 'POST'
+  };
+
+  var server = http.createServer(function(req, res) {
+    res.end();
+  });
+  server.listen(options.port, options.host, function() {
+    setTimeout(done, dur * 1000);
+    bench.start();
+    pummel();
+  });
+
+  function pummel() {
+    var req = http.request(options, function(res) {
+      nreqs++;
+      pummel();  // Line up next request.
+      res.resume();
+    });
+    if (conf.method === 'write') {
+      req.write(chunk, encoding);
+      req.end();
+    } else {
+      req.end(chunk, encoding);
+    }
+  }
+
+  function done() {
+    bench.end(nreqs);
+  }
+}
diff --git a/lib/http.js b/lib/http.js
index 4a4fe94f3e..bcde516dae 100644
--- a/lib/http.js
+++ b/lib/http.js
@@ -782,12 +782,28 @@ OutgoingMessage.prototype.write = function(chunk, encoding) {
 
   if (chunk.length === 0) return false;
 
+  // TODO(bnoordhuis) Temporary optimization hack, remove in v0.11. We only
+  // want to convert the buffer when we're sending:
+  //
+  //   a) Transfer-Encoding chunks, because it lets us pack the chunk header
+  //      and the chunk into a single write(), or
+  //
+  //   b) the first chunk of a fixed-length request, because it lets us pack
+  //      the request headers and the chunk into a single write().
+  //
+  // Converting to strings is expensive, CPU-wise, but reducing the number
+  // of write() calls more than makes up for that because we're dramatically
+  // reducing the number of TCP roundtrips.
+  if (chunk instanceof Buffer && (this.chunkedEncoding || !this._headerSent)) {
+    chunk = chunk.toString('binary');
+    encoding = 'binary';
+  }
+
   var len, ret;
   if (this.chunkedEncoding) {
     if (typeof(chunk) === 'string' &&
         encoding !== 'hex' &&
-        encoding !== 'base64' &&
-        encoding !== 'binary') {
+        encoding !== 'base64') {
       len = Buffer.byteLength(chunk, encoding);
       chunk = len.toString(16) + CRLF + chunk + CRLF;
       ret = this._send(chunk, encoding);