Hi everyone! This is a minor heads-up about my most recent set of
pushes, culminating with:
7192183 nasty depot testing should be part of core testing
This project revamps the 'nasty' mode of pkg.depotd, and adds test cases
which do nasty testing. ('nasty' mode is one in which the pkg depot
daemon does various antagonist things to the client, including corrupting
responses, hanging up, sending the wrong file, etc).
There is now a small amount of nasty testing which runs as part of the
core testsuite. This is done mostly to smoke test it.
Developers need to know a couple of key things:
- The test suite has three test classes, all in cli/t_pkg_nasty.py.
There is one test in each class. Each test will, by default, "do its
thing" in a loop 10 times.
- You can control how many loop iterations these tests run by setting an
environment variable. Here is, for example, how you'd have the loop do
1000 iterations:
DEBUG=TRUE NASTY_ITERS=1000 python tests/run.py \
-a /tmp/failures -j 8 -o t_pkg_nasty
1000 iterations (3000 total iterations, since there are 3 test classes)
will likely take a couple of hours to run. Make sure to use -j. And at
the end you should see just the same couple of test cases pass.
- In order to be able to debug 'nasty problems' it's helpful to set
DEBUG=TRUE and also to archive failures somewhere with the -a option.
Otherwise these things can be really hard to reproduce. Depot logs
have been enhanced to give you hints about what nasty things the depot
has done.
- If you want to run massive amounts of testing, it's helpful to create
subclasses of e.g. TestNastyInstall (TestNastyInstall2, TestNastyInstall3,
...). These will then run in parallel if you use -j. You can reasonably
do 10000 iterations overnight this way (for example using 5 classes x 2000
iterations).
I've previously spoken to Dave V. about how to set this up in Hudson, and
Dave, I hope you can now do that.
Lastly, to the best of my knowledge, the bugs which make the pkg client
break under nasty mode are all fixed by my putback. We'll find out if
that's true :)
-dp
--
Daniel Price, Solaris Kernel Engineering
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
