aasmith/httpterm repository overview

⁠docker-httpterm

A docker image for Willy Tarreu's http terminator. See http://git.1wt.eu/web?p=httpterm.git;a=summary⁠

⁠Notes

By default, an instance is started on port 8000. If are you pushing serious traffic, you probably want --net=host and should disable conntrack in iptables. Starting the docker daemon with --ip-masq=false --iptables=false should be sufficient to do this -- if you understand the implications!

⁠File descriptor limits

As httpterm is used in load testing scenarios, you may want to support a high number of connections. Recent versions of docker use a far more restrictive ulimit -n, so you will probably see the following warning:

[WARNING] 061/144051 (7) : [httpterm.main()] Cannot raise FD limit to 300011.
[WARNING] 061/144051 (7) : [httpterm.main()] FD limit (1024) too low for maxconn=300000/maxsock=300011. Please raise 'ulimit-n' to 300011 or more to avoid any trouble.

Try running the container with --ulimit nofile=1000000 or some other high number to alleviate this.

⁠CPU pinning

If you are pinning interrupts⁠, make sure httpterm runs on a dedicated CPU core. --cpuset-cpus 0 or whichever CPU is free from these interrupts (si in top).

⁠Usage

Any args passed to the container are args directly applied to httpterm process. The default is to start the container with -n 300000 -N 300000 -L :8000 which starts a server listening on port 8000 and caps the max connections per listener and the total global max to 300000.

An example of a more complicated run:

docker run -d --ulimit nofile=1000000 --cpuset-cpus 5 --name httpterm -it --net=host -v /tmp/httpterm.cfg:/tmp/httpterm.cfg aasmith/httpterm -f /tmp/httpterm.cfg -V

A list of parameters can be fetched from localhost:8000/?:

The following arguments are supported to override the default objects:
/?s=<size>[kmg] : return <size> bytes (may be kB, MB, GB). Eg: /?s=20k
/?r=<retcode> : present <retcode> as the HTTP return code. Eg: /?r=404
/?c=<cache> : set the return as not cacheable if zero. Eg: /?c=0
/?t=<time> : wait <time> milliseconds before responding. Eg: /?t=500
/?k={0|1} : Enable transfer encoding chunked with 1 byte chunks
/?S={0|1} : Disable/enable use of splice() to send data
/?R={0|1} : Disable/enable sending random data (disables splicing)
Note that those arguments may be cumulated on one line separated by the '&' sign :
GET /?s=20k&c=1&t=700 HTTP/1.0
GET /?r=500&s=0&c=0&t=1000 HTTP/1.0