Re: [kvm-unit-tests PATCH v2 2/2] run_tests: allow run tests in parallel

[Peter Xu <peterx@redhat.com>]

On Wed, Jan 04, 2017 at 04:09:39PM +0100, Radim Krčmář wrote:
> 2017-01-03 18:10+0800, Peter Xu:
> > run_task.sh is getting slow. This patch is trying to make it faster by
> > running the tests concurrently.
> > 
> > We provide a new parameter "-j" for the run_tests.sh, which can be used
> > to specify how many run queues we want for the tests. Default queue
> > length is 1, which is the old behavior.
> > 
> > Quick test on my laptop (4 cores, 2 threads each) shows 3x speed boost:
> > 
> >    |-----------------+-----------|
> >    | command         | time used |
> >    |-----------------+-----------|
> >    | run_test.sh     | 75s       |
> >    | run_test.sh -j8 | 27s       |
> >    |-----------------+-----------|
> > 
> > Suggested-by: Radim Krčmář <rkrcmar@redhat.com>
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> > ---
> >  run_tests.sh           | 12 ++++++++++--
> >  scripts/functions.bash | 15 ++++++++++++++-
> >  scripts/global.bash    | 11 +++++++++++
> >  3 files changed, 35 insertions(+), 3 deletions(-)
> 
> I like this diffstat a lot more, thanks :)
> 
> The script doesn't handle ^C well now (at least), which can be worked
> around with
> 
>   trap exit SIGINT
> 
> but it would be nice to know if receiving signals in `wait` can't be
> fixed.

When I send SIGINT to "run_tests.sh -j8", I see process hang dead. Not
sure whether you see the same thing:

#0  0x00007f7af2e1559a in waitpid () from /lib64/libc.so.6
#1  0x00005613edf8953e in waitchld.isra ()
#2  0x00005613edf8aae5 in wait_for ()
#3  0x00005613edf8b682 in wait_for_any_job ()
#4  0x00005613edfc7e64 in wait_builtin ()
#5  0x00005613edf616ea in execute_builtin.isra ()
#6  0x00005613edf623ee in execute_simple_command ()
#7  0x00005613edf79e77 in execute_command_internal ()
#8  0x00005613edf7b972 in execute_command ()
#9  0x00005613edf62aca in execute_while_or_until ()
#10 0x00005613edf7a156 in execute_command_internal ()
#11 0x00005613edf79d88 in execute_command_internal ()
...

If I change the "wait -n" into "wait" (this will work, but of course
slower since we'll wait for all subprocesses end before we start
another one), problem disappears. Not sure whether that means a "wait
-n" bug.

Anyway, IMHO squashing you suggestion of "trap exit SIGINT" at the
entry of for_each_unittest() is an acceptable solution - it works in
all cases.

Thanks,

-- peterx