Re: [Qemu-devel] [PATCH RFC] tests: Run qtest cases in parallel

[Gonglei (Arei)]

Hi Fam,


> -----Original Message-----
> From: Qemu-devel
> [mailto:address@hidden On
> Behalf Of Fam Zheng
> Sent: Friday, September 23, 2016 3:58 PM
> To: John Snow
> Cc: address@hidden; address@hidden
> Subject: Re: [Qemu-devel] [PATCH RFC] tests: Run qtest cases in parallel
> 
> On Wed, 09/21 14:24, John Snow wrote:
> >
> >
> > On 08/12/2016 05:19 AM, Fam Zheng wrote:
> > > Previously all test cases in a category, such as check-qtest-y, are
> > > executed in a single long gtester command. This patch separates each
> > > test into its own make target to allow better parallism.
> > >

That's will be great if we can specify a test to run, especially for the 
scenario
which add one use qtest case.

For example: 

 # make check test-crypto-cipher

then only run the tests/ test-crypto-cipher. 

Do you think it makes sense?

Regards,
-Gonglei

> > > Signed-off-by: Fam Zheng <address@hidden>
> > > ---
> > >
> > > This saves 50% of the time "make check takes" compared to on master
> > > (I use -j8). RFC because I'm not sure if the new gcov usage is correct
> > > with the now much higher level of parallism compared to before.
> > > ---
> > >  tests/Makefile.include | 34 ++++++++++++++++++----------------
> > >  1 file changed, 18 insertions(+), 16 deletions(-)
> > >
> > > diff --git a/tests/Makefile.include b/tests/Makefile.include
> > > index 14be491..9bf0326 100644
> > > --- a/tests/Makefile.include
> > > +++ b/tests/Makefile.include
> > > @@ -691,27 +691,29 @@ GCOV_OPTIONS = -n $(if $(V),-f,)
> > >  # gtester tests, possibly with verbose output
> > >
> > >  .PHONY: $(patsubst %, check-qtest-%, $(QTEST_TARGETS))
> > > -$(patsubst %, check-qtest-%, $(QTEST_TARGETS)): check-qtest-%:
> $(check-qtest-y)
> > > +
> > > +qtest-run-%: tests/%
> > >   $(if $(CONFIG_GCOV),@rm -f *.gcda */*.gcda */*/*.gcda
> */*/*/*.gcda,)
> > > - $(call
> quiet-command,QTEST_QEMU_BINARY=$*-softmmu/qemu-system-$* \
> > > + $(call quiet-command,\
> > > +         $(if $(QTEST_TARGET), \
> > > +
>       QTEST_QEMU_BINARY=$(QTEST_TARGET)-softmmu/qemu-system-$(QTE
> ST_TARGET)) \
> > >           QTEST_QEMU_IMG=qemu-img$(EXESUF) \
> > >           MALLOC_PERTURB_=$${MALLOC_PERTURB_:-$$((RANDOM %
> 255 + 1))} \
> > > -         gtester $(GTESTER_OPTIONS) -m=$(SPEED) $(check-qtest-$*-y)
> $(check-qtest-generic-y),"GTESTER $@")
> > > - $(if $(CONFIG_GCOV),@for f in $(gcov-files-$*-y) 
> > > $(gcov-files-generic-y);
> do \
> > > -   echo Gcov report for $$f:;\
> > > -   $(GCOV) $(GCOV_OPTIONS) $$f -o `dirname $$f`; \
> > > - done,)
> > > +         gtester $< $(GTESTER_OPTIONS) -m=$(SPEED),"GTESTER $<")
> > > + $(if $(CONFIG_GCOV), @echo Gcov report for $<:;\
> > > +   $(GCOV) $(GCOV_OPTIONS) $< -o `dirname $<`; \
> > > + )
> > > +
> > > +$(foreach target, $(QTEST_TARGETS), \
> > > + $(eval check-qtest-$(target): QTEST_TARGET := $(target)) \
> > > + $(eval check-qtest-$(target): $(patsubst tests/%, qtest-run-%, \
> > > +                                             $(check-qtest-y) \
> > > +
> $(check-qtest-$(target)-y) \
> > > +
> $(check-qtest-generic-y))) \
> > > +)
> > >
> > >  .PHONY: $(patsubst %, check-%, $(check-unit-y))
> > > -$(patsubst %, check-%, $(check-unit-y)): check-%: %
> > > - $(if $(CONFIG_GCOV),@rm -f *.gcda */*.gcda */*/*.gcda */*/*/*.gcda,)
> > > - $(call quiet-command, \
> > > -         MALLOC_PERTURB_=$${MALLOC_PERTURB_:-$$((RANDOM % 255 +
> 1))} \
> > > -         gtester $(GTESTER_OPTIONS) -m=$(SPEED) $*,"GTESTER $*")
> > > - $(if $(CONFIG_GCOV),@for f in $(gcov-files-$(subst tests/,,$*)-y)
> $(gcov-files-generic-y); do \
> > > -   echo Gcov report for $$f:;\
> > > -   $(GCOV) $(GCOV_OPTIONS) $$f -o `dirname $$f`; \
> > > - done,)
> > > +$(patsubst %, check-%, $(check-unit-y)): check-tests/%: qtest-run-%
> > >
> > >  # gtester tests with XML output
> > >
> > >
> >
> > I can't vouch for gcov either, but:
> 
> Too bad that this seems to be too big a hammer that breaks the way gcov was
> used: 1) the "rm *.gcda" command now runs in each test, so it's harder to
> get an overall coverage report; 2) there is a risk that the parallism may
> corrupt those files. :(
> 
> Fam
> 
> >
> > Tested-by: John Snow <address@hidden>
> >
> > -j1:
> >
> > real        1m51.195s
> > user        1m5.478s
> > sys 0m21.158s
> >
> > -j9:
> >
> > real        0m53.039s
> > user        1m41.754s
> > sys 0m31.150s
> >
> >
> > This seems useful.
> >
> > --js