benchmark: make http profiling easier

[bnoordhuis]

The new http benchmarks are set up in a way that effectively makes profiling impossible because the interesting code runs in a child process.

That is, if you prof / strace / etc. it, all you see is the master process that spends 99% of the time sleeping in a syscall.

[isaacs]

If you specify all the args, then it'll run the actual test. Like this:

node benchmark/http/simple.js type=bytes length=1024 c=500

If you don't specify all the args, then it'll run with the combinations listed in the script.

[bnoordhuis]

Maybe for the other tests but benchmark/http/simple.js spawns benchmark/http_simple.js in a child process.

I hacked around it like (see below) but it's not very convenient.

diff --git a/benchmark/http/simple.js b/benchmark/http/simple.js
index 04a2a29..ed8545f 100644
--- a/benchmark/http/simple.js
+++ b/benchmark/http/simple.js
@@ -12,7 +12,8 @@ function main(conf) {
   process.env.PORT = PORT;
   var spawn = require('child_process').spawn;
   var simple = require('path').resolve(__dirname, '../http_simple.js');
-  var server = spawn(process.execPath, [simple]);
+  //var server = spawn(process.execPath, [simple]);
+  var server = spawn('perf', ['record', '-i', '-g', '-e', 'cycles:u', process.execPath, simple]);
   setTimeout(function() {
     var path = '/' + conf.type + '/' + conf.length; //+ '/' + conf.chunks;
     var args = ['-r', 5000, '-t', 8, '-c', conf.c];

[isaacs]

Ohh... right. It should probably not spawn http_simple in a child proc, and instead just require() it, like the benchmark/http/cluster.js workers do.