bad/no support for keepalive connections

[zdila]

trafficstars

Testing with ab -n 100 -c 1 -k -X localhost:8080 http://localhost/ is more than 10x slower than without -k argument. Also report shows errors.

[greim]

Thanks for finding this. It might be a couple of issues. Upgrading the code to use http.request() (which reuses client objects) speeds things up considerably. But it still reports failed requests. Continuing investigating...

[greim]

And for me at least it was throwing "too many listeners" errors, since it was adding listeners to the same socket each time when connection was reused in keepalive mode. This aspect should hopefully be easy to fix.

[zdila]

Hi greim. After last update the hoxy doesn't work at all and the browser timeouts.

[greim]

Which version of node are you on? The fix for the performance issues has dependencies on the client connection pooling in 0.4.x.

[zdila]

Default Ubuntu Natty: martin@bono:~$ node -v v0.2.6

Please add version check on startup :-)

[zdila]

with v0.4.7 it works but ab shows may failed connections (if used with -k):

ab -n 100 -c 1 -k -X localhost:8080 http://localhost/

• in this case 50 are failed. With -v 2 argument you can see mixed headers and message. BTW my tested page is standard Apache "It works!" page.

[greim]

Thanks. Incidentally I'm also doing my tests on an Apache "It works!" page. The failed requests are some weird low-level stream issue that I'm still trying to understand. Will reply here if/when I find a fix.

[greim]

Version check added.

[greim]

I realize this is old, but can anyone comment if this is still an issue? v0.x branch is still around if fixes are badly needed.