RejectedSoftware Forums

On Sat, 21 Sep 2013 16:13:25 GMT, Dicebot wrote:

On Sat, 21 Sep 2013 15:50:47 GMT, simendsjo wrote:

Some tests using weighttp shows that is doesn't really work very well when running concurrent requests.

..I've actually never had vibe correctly handling many requests in my tests.

weighttp -n10000 -c100 -t2 -k http://localhost:8181/

It can run fine a couple of times, and then it breaks.
Weighttp says `error: connect() failed: Cannot assign requested address (99), while vibe says Handling of connection failed: Error writing data to socket. Remote hung up?`.
If I let it rest for some time, it's possible to re-run weighttp without it failing. Running it right after it fails just makes it keep failing.

Another thing.. weighttp reports this `requests: 10000 total, 10000 started, 10000 done, 35 succeeded, 9965 failed, 0 errored` - Any idea what 9965 failed means?

Another thing is some excessive memory usage. If I keep running the tests, the application uses all of my 8GB without releasing anything back. But this might of course be a bug in my app as I haven't tested a minimal vibe app.

I used to test it up to -c 5000 - GC killed the latency at those rates but it worked with very low fail ration.

Maybe regression :(

I'm still fighting with some really bad regressions introduced a while ago. They don't seem to be real regressions, but rather sleeping bugs that surfaced due to some related changes. Hopefully by tomorrow will be in a sane state again and then I'll finally devote all time to getting the CI server running.

On Sat, 21 Sep 2013 15:50:47 GMT, simendsjo wrote:

..I've actually never had vibe correctly handling many requests in my tests.

I reach about 4000 req/s.

weighttp -n10000 -c100 -t2 -k http://localhost:8181/

It can run fine a couple of times, and then it breaks.
Weighttp says `error: connect() failed: Cannot assign requested address (99), while vibe says Handling of connection failed: Error writing data to socket. Remote hung up?`.

You're running out of sockets.
http://redmine.lighttpd.net/projects/weighttp/wiki#Troubleshooting

On Sat, 28 Sep 2013 20:58:19 GMT, Martin Nowak wrote:

On Sun, 22 Sep 2013 16:51:44 GMT, Dicebot wrote:

On Sun, 22 Sep 2013 12:30:27 GMT, Sönke Ludwig wrote:

I'm still fighting with some really bad regressions introduced a while ago. They don't seem to be real regressions, but rather sleeping bugs that surfaced due to some related changes. Hopefully by tomorrow will be in a sane state again and then I'll finally devote all time to getting the CI server running.

Have you got my e-mail on CI topic?

It would be great if services like https://travis-ci.org offered D support.
But I guess we aren't there yet in terms of stability and ecosystem (packages etc.).

I am going to setup my private server to provide automatic limited CI for all dub registry project. Don't know how long it will take to take care of all details, though.

On 09/29/2013 12:19 PM, Dicebot wrote:

It would be great if services like https://travis-ci.org offered D support.
But I guess we aren't there yet in terms of stability and ecosystem (packages etc.).

BTW, Masahiro is using Travis-CI to test msgpack-d.
https://travis-ci.org/msgpack/msgpack-d
https://github.com/msgpack/msgpack-d/blob/master/.travis.yml
Automatically integrates with github and pull requests.

One nuisance is the huge dmd download (33MB).
The debian/ubuntu packages require multilib which is even bigger.

I am going to setup my private server to provide automatic limited CI for all dub registry project. Don't know how long it will take to take care of all details, though.

That meant we need something like a dub test command.