Re: [meta-oe][PATCH] protobuf-c: fix race condition

[Martin Jansa <martin.jansa@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 9533 bytes --]

On Wed, Oct 23, 2019 at 08:50:10AM -0700, akuster808 wrote:
> 
> 
> On 10/22/19 2:37 PM, Martin Jansa wrote:
> > On Fri, Oct 18, 2019 at 09:06:54PM +0200, Martin Jansa wrote:
> >> One more data point is that it might be fixed completely with 1.3.2 version
> >> used in zeus/master and failing less often then without this patch in 1.3.1
> >> used in thud/warrior.
> >>
> >> I was rebuilding it 300 times in thud/warrior/zeus and got 5-6 failures in
> >> thud/warrior but none in zeus.
> >>
> >> I've started another for loop to rebuilt it 1000 time with zeus and if it
> >> works then I'll retest warrior and thud with 1.3.2 upgrade backported to
> >> them - if it works then we can either backport whole upgrade or check the
> >> 1.3.1 1.3.2 diff to find out what else we need to backport from there.
> > 1000 times in zeus = 0 failures
> > 100 times in thud with 1.3.2 backported ~ 10 failures
> > 1000 times in thud with 1.3.2 as well as protobuf-native 3.9.2 = 0 failures
> >
> > The protobuf-c upgrade isn't need to be backported to thud, there aren't
> > many changes:
> > https://github.com/protobuf-c/protobuf-c/compare/269771b4b45d3aba04e59569f53600003db8d9ff...1390409f4ee4e26d0635310995b516eb702c3f9e
> >
> > protobuf on the other hand was upgraded from 3.6.1 to 3.9.2 in zeus and
> > the diff is big (1028 commits):
> > https://github.com/protocolbuffers/protobuf/compare/48cb18e5c419ddd23d9badcfe4e9df7bde1979b2...52b2447247f535663ac1c292e088b4b27d2910ef
> >
> > quick grep in git log didn't reveal what might be causing the race
> > condition, upgrading whole protobuf in warrior and thud isn't an option
> > as well.
> >
> > Anyone interested in digging a bit deeper?
> >
> > I don't want to block this change, because it seems to completely fix it
> > for zeus/master and partially improve it also for warrior and thud.
> >
> > We don't use protobuf much, so for our builds I think I'll just backport
> > whole protobuf upgrade to thud once protobuf-c fix is in thud.
> 
> So is https://patchwork.openembedded.org/patch/165905/ still a go for
> warrior?

I think it is and for thud as well. Once there is a fix for protobuf as
well, it should be in separate follow-up change anyway.

with protobuf-c fix included I got 28 failures in 1000 rebuilds
without protobuf-c fix I got 5 failures in 308 rebuilds (still running)

so the rate seems similar if not the same (but there might be other
aspects like load on the build machine from other builds), before I was
seeing lower frequency of failures with this included.

Cheers,

> >> On Fri, Oct 18, 2019 at 11:07 AM Khem Raj <raj.khem@gmail.com> wrote:
> >>
> >>> Yes I have seen same build failures on master even after this fix on
> >>> machine with more cpus so the fix while seems to address the problem is not
> >>> fixing is completely when I get a chance I will look into the logic myself
> >>> too but I think we have to delve into it a bit More
> >>>
> >>> On Fri, Oct 18, 2019 at 3:41 AM Martin Jansa <martin.jansa@gmail.com>
> >>> wrote:
> >>>
> >>>> Hmm I spoke too soon, at least rebuilding protobuf-c-native with this
> >>>> change backported to thud still failed once with:
> >>>>
> >>>> ../git/t/generated-code2/cxx-generate-packed-data.cc:28:17: error: ‘foo’
> >>>> is
> >>>> not a namespace-name
> >>>>  using namespace foo;
> >>>>                  ^~~
> >>>> ../git/t/generated-code2/cxx-generate-packed-data.cc:28:20: error:
> >>>> expected
> >>>> namespace-name before ‘;’ token
> >>>>  using namespace foo;
> >>>>                     ^
> >>>> ../git/t/generated-code2/cxx-generate-packed-data.cc:31:36: error:
> >>>> ‘TestEnumSmall’ does not name a type
> >>>>  #define TEST_ENUM_SMALL_TYPE_NAME  TestEnumSmall
> >>>>                                     ^
> >>>> ../git/t/generated-code2/common-test-arrays.h:44:1: note: in expansion of
> >>>> macro ‘TEST_ENUM_SMALL_TYPE_NAME’
> >>>>  TEST_ENUM_SMALL_TYPE_NAME enum_small_0[] = { TEST_ENUM_SMALL(VALUE) };
> >>>>  ^~~~~~~~~~~~~~~~~~~~~~~~~
> >>>> ../git/t/generated-code2/cxx-generate-packed-data.cc:31:36: error:
> >>>> ‘TestEnumSmall’ does not name a type
> >>>>  #define TEST_ENUM_SMALL_TYPE_NAME  TestEnumSmall
> >>>>                                     ^
> >>>> ../git/t/generated-code2/common-test-arrays.h:45:1: note: in expansion of
> >>>> macro ‘TEST_ENUM_SMALL_TYPE_NAME’
> >>>>  TEST_ENUM_SMALL_TYPE_NAME enum_small_1[] = { TEST_ENUM_SMALL(OTHER_VALUE)
> >>>> };
> >>>>  ^~~~~~~~~~~~~~~~~~~~~~~~~
> >>>> ../git/t/generated-code2/cxx-generate-packed-data.cc:31:36: error:
> >>>> ‘TestEnumSmall’ does not name a type
> >>>>  #define TEST_ENUM_SMALL_TYPE_NAME  TestEnumSmall
> >>>>                                     ^
> >>>> ../git/t/generated-code2/common-test-arrays.h:47:1: note: in expansion of
> >>>> macro ‘TEST_ENUM_SMALL_TYPE_NAME’
> >>>>  TEST_ENUM_SMALL_TYPE_NAME enum_small_random[] =
> >>>>
> >>>> {T(0),T(1),T(1),T(0),T(0),T(1),T(1),T(1),T(0),T(0),T(0),T(0),T(0),T(1),T(1),T(1),T(1),T(1),T(1),T(0),T(1),T(1),T(0),T(1),T(1),T(0)
> >>>> };
> >>>>  ^~~~~~~~~~~~~~~~~~~~~~~~~
> >>>> ...
> >>>> so it probably doesn't fix the issue completely.
> >>>>
> >>>>
> >>>> On Thu, Oct 17, 2019 at 10:44 PM Martin Jansa <martin.jansa@gmail.com>
> >>>> wrote:
> >>>>
> >>>>> On Sun, Sep 29, 2019 at 05:29:56PM +0800, Chen Qi wrote:
> >>>>>> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
> >>>>>> ---
> >>>>>>  .../0001-avoid-race-condition.patch           | 36
> >>>> +++++++++++++++++++
> >>>>>>  .../protobuf/protobuf-c_1.3.2.bb              |  1 +
> >>>>>>  2 files changed, 37 insertions(+)
> >>>>>>  create mode 100644
> >>>> meta-oe/recipes-devtools/protobuf/protobuf-c/0001-avoid-race-condition.patch
> >>>>>> diff --git
> >>>> a/meta-oe/recipes-devtools/protobuf/protobuf-c/0001-avoid-race-condition.patch
> >>>> b/meta-oe/recipes-devtools/protobuf/protobuf-c/0001-avoid-race-condition.patch
> >>>>>> new file mode 100644
> >>>>>> index 000000000..4fc7703d8
> >>>>>> --- /dev/null
> >>>>>> +++
> >>>> b/meta-oe/recipes-devtools/protobuf/protobuf-c/0001-avoid-race-condition.patch
> >>>>>> @@ -0,0 +1,36 @@
> >>>>>> +From 216e31260b618ec73862f9f5336597f391444dac Mon Sep 17 00:00:00
> >>>> 2001
> >>>>>> +From: Chen Qi <Qi.Chen@windriver.com>
> >>>>>> +Date: Sun, 29 Sep 2019 17:20:42 +0800
> >>>>>> +Subject: [PATCH] avoid race condition
> >>>>>> +
> >>>>>> +It's possible that the cxx-generate-packed-data.cc is compiled
> >>>>>> +while the t/test-full.pb.h is being generated. This will result
> >>>>>> +the following error.
> >>>>>> +
> >>>>>> +  DEBUG:     ./t/test-full.pb.h:4:0: error: unterminated #ifndef
> >>>>>> +  ./t/test-full.pb.h:4:0: error: unterminated #ifndef
> >>>>>> +
> >>>>>> +Add a dependency to avoid such problem.
> >>>>>> +
> >>>>>> +Upstream-Status: Pending
> >>>>> Thanks for this patch, it seems to indeed fix this issue, I've reported
> >>>>> a while ago:
> >>>>>
> >>>>>
> >>>> https://www.mail-archive.com/openembedded-devel@lists.openembedded.org/msg64363.html
> >>>>> I think the upstream would be interested in this fix as well, see:
> >>>>>
> >>>>>
> >>>> https://github.com/protobuf-c/protobuf-c/commit/b0bb56303366e2c072ee38eb5f1f11251b07239c
> >>>>> Cheers,
> >>>>>
> >>>>>> +
> >>>>>> +Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
> >>>>>> +---
> >>>>>> + Makefile.am | 1 +
> >>>>>> + 1 file changed, 1 insertion(+)
> >>>>>> +
> >>>>>> +diff --git a/Makefile.am b/Makefile.am
> >>>>>> +index b0cb065..1608ae0 100644
> >>>>>> +--- a/Makefile.am
> >>>>>> ++++ b/Makefile.am
> >>>>>> +@@ -156,6 +156,7 @@ noinst_PROGRAMS += \
> >>>>>> + t_generated_code2_cxx_generate_packed_data_SOURCES = \
> >>>>>> +     t/generated-code2/cxx-generate-packed-data.cc \
> >>>>>> +     t/test-full.pb.cc
> >>>>>> ++t/generated-code2/cxx-generate-packed-data.cc: t/test-full.pb.h
> >>>>>> + $(t_generated_code2_cxx_generate_packed_data_OBJECTS):
> >>>> t/test-full.pb.h
> >>>>>> + t_generated_code2_cxx_generate_packed_data_CXXFLAGS = \
> >>>>>> +     $(AM_CXXFLAGS) \
> >>>>>> +--
> >>>>>> +2.17.1
> >>>>>> +
> >>>>>> diff --git a/meta-oe/recipes-devtools/protobuf/protobuf-c_1.3.2.bb
> >>>>> b/meta-oe/recipes-devtools/protobuf/protobuf-c_1.3.2.bb
> >>>>>> index 6d1ffc3f4..b92f82dec 100644
> >>>>>> --- a/meta-oe/recipes-devtools/protobuf/protobuf-c_1.3.2.bb
> >>>>>> +++ b/meta-oe/recipes-devtools/protobuf/protobuf-c_1.3.2.bb
> >>>>>> @@ -15,6 +15,7 @@ DEPENDS = "protobuf-native protobuf"
> >>>>>>  SRCREV = "1390409f4ee4e26d0635310995b516eb702c3f9e"
> >>>>>>
> >>>>>>  SRC_URI = "git://github.com/protobuf-c/protobuf-c.git \
> >>>>>> +           file://0001-avoid-race-condition.patch \
> >>>>>>            "
> >>>>>>
> >>>>>>  S = "${WORKDIR}/git"
> >>>>>> --
> >>>>>> 2.17.1
> >>>>>>
> >>>>>> --
> >>>>>> _______________________________________________
> >>>>>> Openembedded-devel mailing list
> >>>>>> Openembedded-devel@lists.openembedded.org
> >>>>>> http://lists.openembedded.org/mailman/listinfo/openembedded-devel
> >>>>> --
> >>>>> Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com
> >>>>>
> >>>> --
> >>>> _______________________________________________
> >>>> Openembedded-devel mailing list
> >>>> Openembedded-devel@lists.openembedded.org
> >>>> http://lists.openembedded.org/mailman/listinfo/openembedded-devel
> >>>>
> >
> 

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 201 bytes --]