A tool for generating repeatable load tests of haproxy configurations.
Four machines, running Linux (centos, Amazon Linux etc.) ideally on the same network segment. In AWS parlance, this would be the same AZ within a region.
- Proxy: the instance running haproxy.
- Server: target for backend connections from haproxy.
- Client: generates HTTP requests into the proxy frontend.
- Logspout: a UDP collector for haproxy log messages.
A local clone of h1load (git clone https://github.com/wtarreau/h1load.git)
A recent version of gnuplot (brew install gnuplot)
A way to provision compute. Putting some terraform in the compute dir would
be good.
Provide the IPs of the four instances:
$ cat instances
10.10.10.1 proxy # ubuntu
10.10.10.2 server # amazonlinux2
10.10.10.3 client # amazonlinux2
10.10.10.4 logspout # amazonlinux2
If you used terraform to provision your compute, this list can be generated
using tf2instances.rb.
# set up each machine with OS settings, download compile tools, etc.
setup syscfg
setup init
# push app configs to each machine
setup appcfg
# run tests, using the "simple" method
# download & process data of the last run
setup science simple
# or as two steps...
setup try simple
setup analyze simple
# <make app config edits>
vi stuff
# push app config changes
setup appcfg
# run tests again
...
pushcfg
run cfgs
run init
run run
# observe action
# get perf file from h1load, gnuplot it
# tweak configs
run cfgs
run run
Sets up each machine to serve a specific role:
server: absorbs load using httpterm client: generates load using h1load proxy: handles load using haproxy logspout: accepts logs from haproxy