Responses never terminate (ab dies with timeout error) (OS X)
joshgoebel opened this issue · comments
Using latest jester and Apache bench and for some reason it thinks the requests timeout. Perhaps we're not closing a socket properly or something? It's 100% reproducible here and only with httpbeast. asynchttpserver does not have the same issue.
nim c -d:release -r tests/example
...
INFO Jester is making jokes at http://0.0.0.0:5000
Starting 1 threads
ab -r -n 1000 -c 1000 http://0.0.0.0:5000/
This is ApacheBench, Version 2.3 <$Revision: 1826891 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking 0.0.0.0 (be patient)
apr_pollset_poll: The timeout specified has expired (70007)
Seems to work fine in browser and with curl. AB is receiving the responses (if I used -v2), but obviously something else is off.
Thoughts?
Pretty sure this is because we don't yet support close=0 or keep-alive=0.
Hrm, interesting. Not sure what the problem could be, I'd need to look into how apache bench creates requests.
if data.sendQueue.len == data.bytesSent:
data.bytesSent = 0
data.sendQueue.setLen(0)
data.data.setLen(0)
# selector.unregister(fd)
# fd.SocketHandle.close()
selector.updateHandle(fd.SocketHandle,
{Event.Read})
Adding the two commented lines fixed it for me... but of course (if this is the proper place to add them) they would need to be dependent on the Close/Keep-Alive/ and HTTP version. You can see how it's done in asynchttpserver if you want a reference.
ab is using HTTP/1.0 (though it doesn't seem to mind the 1.1 being returned) which by default expects keep-alive to be off and the connection to be terminated by the server - which we aren't doing.
@yyyc514 You can use wrk instead of ab.
Although ab is a single thread and old tool, but wrk runs in multithreading.
If you want to use wrk with single thread, simply specify -t 1 with the command line option.
wrk -t 1 http://localhost:8080
Techempower, a famous benchmark, also uses wrk to measure it.
Yeah I've been using wrk, but still think httpbeast isn't doing the correct connection handling here.
Same problem here.
I also ran into this.
I tried this work around by adding req.client.close() to end of requests, but I get this error after some requests:
Error: unhandled exception: ioselectors_epoll.nim(147, 11) `s.fds[fdi].ident == InvalidIdent` Descriptor 118 already registered [AssertionDefect]```