You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@qpid.apache.org by Gordon Sim <gs...@redhat.com> on 2010/11/10 13:22:40 UTC
failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
On 11/08/2010 08:47 AM, Robbie Gemmell wrote:
> Hi all,
>
> 0.8 RC1 can be found for download at:
> http://people.apache.org/~robbie/qpid/0.8/RC1/
>
> Output from running RAT across the 'full release' archive can be found at:
> http://people.apache.org/~robbie/qpid/0.8/0.8rc1_rat_output.txt
>
> *Please* take the time to download and try out this release candidate and
> report your results, such that any additional issues which require
> addressing before the release can proceed are identified as early as
> possible.
The qpid-cpp-0.8.tar.gz doesn't include clustering related test files. I
suspect this is due to conditional inclusion of them in the distribution
list if the relevant dependencies are not available.
Autotools build fails at the compilation stage with:
> /bin/sh ../../libtool --tag=CXX --mode=link g++ -g -O2 -Itestagent_gen -g -O2 -L/usr/lib/openais -L/usr/lib64/openais -L/usr/lib/corosync -L/usr/lib64/corosync -o testagent testagent-testagent.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Parent.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Child.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildCreated.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildDestroyed.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Package.o ../../src/libqmf.la
> libtool: link: g++ -g -O2 -Itestagent_gen -g -O2 -o .libs/testagent testagent-testagent.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Parent.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Child.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildCreated.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildDestroyed.o testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Package.o -L/usr/lib/openais -L/usr/lib64/openais -L/usr/lib/corosync -L/usr/lib64/corosync ../../src/.libs/libqmf.so /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqmfengine.so /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidclient.so /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidcommon.so /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidtypes.so -lboost_program_options -lboost_filesystem -ldl -lrt -lsasl2 -luuid -Wl,-rpath -Wl,/usr/local/lib
> make[3]: *** No rule to make target `run_cluster_tests', needed by `all-am'. Stop.
> make[3]: Leaving directory `/home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/tests'
Cmake build fails on running cmake with:
> CMake Error in src/tests/CMakeLists.txt:
> Cannot find source file "InitialStatusMap". Tried extensions .c .C .c++
> .cc .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp .hxx .in .txx
>
>
> CMake Error in src/tests/CMakeLists.txt:
> Cannot find source file "StoreStatus". Tried extensions .c .C .c++ .cc
> .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp .hxx .in .txx
>
Either we need to regenerate the distribution on a platform with those
dependencies or we need to ensure the relavent files are included
regardless of whether the deps were available at the time of running
make dist. Thoughts?
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
RE: failure due to missing cluster artefacts (was Re: 0.8 RC1 available for download)
Posted by Robbie Gemmell <ro...@gmail.com>.
Thanks Alan.
I wasn't really intending to merge it, just install the dependencies...but
it seems like a fairly straight forward change so I'm ok with leaving it in.
Robbie
> -----Original Message-----
> From: Alan Conway [mailto:aconway@redhat.com]
> Sent: 12 November 2010 18:10
> To: dev@qpid.apache.org
> Cc: Carl Trieloff
> Subject: Re: failure due to missing cluster artefacts (was Re: 0.8 RC1
> available for download)
>
> On 11/11/2010 12:45 PM, Alan Conway wrote:
> > On 11/11/2010 09:34 AM, Carl Trieloff wrote:
> >> On 11/10/2010 01:03 PM, Gordon Sim wrote:
> >>>
> >>> In my mind the ideal is that when creating a dist, the full source
> for all
> >>> possible modules are included regardless of whether the deps for
> those are
> >>> available at the time of creation. That way you ease the burden of
> getting the
> >>> system setup for creating the dist and leverage the existing
> systems of a
> >>> wider group of testers for checking all the different parts.
> >>
> >>
> >
> > I've fixed the cluster packaging problem that started this thread in
> revision
> > 1034008.
> >
> > Note that automake automatically does the right thing with source
> files -
> > they're included in the distro even if the foo_SOURCE= line is inside
> an
> > "if..endif" that does not pass for the current configuration. However
> it doesn't
> > do this for EXTRA_DIST, so it should be moved outside any
> conditionals.
> >
>
> I've merged the fix onto 0.8-release-candidates, r1034496.
>
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Alan Conway <ac...@redhat.com>.
On 11/11/2010 12:45 PM, Alan Conway wrote:
> On 11/11/2010 09:34 AM, Carl Trieloff wrote:
>> On 11/10/2010 01:03 PM, Gordon Sim wrote:
>>>
>>> In my mind the ideal is that when creating a dist, the full source for all
>>> possible modules are included regardless of whether the deps for those are
>>> available at the time of creation. That way you ease the burden of getting the
>>> system setup for creating the dist and leverage the existing systems of a
>>> wider group of testers for checking all the different parts.
>>
>>
>
> I've fixed the cluster packaging problem that started this thread in revision
> 1034008.
>
> Note that automake automatically does the right thing with source files -
> they're included in the distro even if the foo_SOURCE= line is inside an
> "if..endif" that does not pass for the current configuration. However it doesn't
> do this for EXTRA_DIST, so it should be moved outside any conditionals.
>
I've merged the fix onto 0.8-release-candidates, r1034496.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Alan Conway <ac...@redhat.com>.
On 11/11/2010 09:34 AM, Carl Trieloff wrote:
> On 11/10/2010 01:03 PM, Gordon Sim wrote:
>>
>> In my mind the ideal is that when creating a dist, the full source for all
>> possible modules are included regardless of whether the deps for those are
>> available at the time of creation. That way you ease the burden of getting the
>> system setup for creating the dist and leverage the existing systems of a
>> wider group of testers for checking all the different parts.
>
>
I've fixed the cluster packaging problem that started this thread in revision
1034008.
Note that automake automatically does the right thing with source files -
they're included in the distro even if the foo_SOURCE= line is inside an
"if..endif" that does not pass for the current configuration. However it doesn't
do this for EXTRA_DIST, so it should be moved outside any conditionals.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Alan Conway <ac...@redhat.com>.
On 11/11/2010 09:34 AM, Carl Trieloff wrote:
> On 11/10/2010 01:03 PM, Gordon Sim wrote:
>>
>> In my mind the ideal is that when creating a dist, the full source for all
>> possible modules are included regardless of whether the deps for those are
>> available at the time of creation. That way you ease the burden of getting the
>> system setup for creating the dist and leverage the existing systems of a
>> wider group of testers for checking all the different parts.
>
>
Agreed, I think this is a regression caused by me, I'll make sure all files are
distributed regardless. I'll mail the list when its done, Is there a JIRA I
should update when it's done?
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Carl Trieloff <cc...@redhat.com>.
On 11/10/2010 01:03 PM, Gordon Sim wrote:
>
> In my mind the ideal is that when creating a dist, the full source for
> all possible modules are included regardless of whether the deps for
> those are available at the time of creation. That way you ease the
> burden of getting the system setup for creating the dist and leverage
> the existing systems of a wider group of testers for checking all the
> different parts.
That makes sense to me. This would then also imply that we need to get
an ack on each part
that someone has tested it as the release manager does not own the
responsibility for the testing
but rather just recording that fact that all pieces have been validated
before taking it to vote.
Carl.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Gordon Sim <gs...@redhat.com>.
On 11/10/2010 07:53 PM, Robbie Gemmell wrote:
> There was a similar issue noted for 0.6
> (https://issues.apache.org/jira/browse/QPID-2316) but as no one replied when
> I asked people to look at it I just ended up nudging it out to 0.9 and
> ensuring I had the relevant dependencies installed for RC1. It is also one
> of the primary reasons I produced the alpha build a few weeks back, hoping
> we wouldn't end up in the RC stage with me still needing to tinker with
> dependencies to actually produce working release candidates. Oh well ;)
Sorry Robbie, I had meant to do a brief run through that but got swamped!
> Is "qpid/qpid/cpp/INSTALL" up to date? I don't see a mention of 'cpg' - is
> that Corosync?
Yes, either openais or as it was renamed, corosync.
> Alternatively does anyone have a handy list package names or
> source download links for dependencies I should make sure I have installed
> (Fedora 12) to fix this and any other related 'X wasn't included because Y
> wasn't installed' issues for the next RC? Talking of which, I think we
> should just live with it for 0.8 but fix it properly on trunk.
corosync and optionally clusterlib-devel (for clustering), xqilla-devel
(for xml exchange), cyrus-sasl-devel (for sasl support), nss-devel (for
ssl support)
On digging a bit deeper however I can see that all the 'real' sources
*are* included in the RC, including the cluster code. So I think it is
just some of the test related artifacts that aren't there...
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
RE: failure due to missing cluster artefacts (was Re: 0.8 RC1 available for download)
Posted by Robbie Gemmell <ro...@gmail.com>.
I agree with Gordon, it would make more sense to me to include all sources
regardless of dependency status on the box. The only time the build box
configuration should matter is for the binary packages really. In this case
I ran the build process, it compiled what it needed to for the C++ tree
without incident, I assumed it was ok...and given that it's actually a
source package why shouldn't I be able to do so?
There was a similar issue noted for 0.6
(https://issues.apache.org/jira/browse/QPID-2316) but as no one replied when
I asked people to look at it I just ended up nudging it out to 0.9 and
ensuring I had the relevant dependencies installed for RC1. It is also one
of the primary reasons I produced the alpha build a few weeks back, hoping
we wouldn't end up in the RC stage with me still needing to tinker with
dependencies to actually produce working release candidates. Oh well ;)
Is "qpid/qpid/cpp/INSTALL" up to date? I don't see a mention of 'cpg' - is
that Corosync? Alternatively does anyone have a handy list package names or
source download links for dependencies I should make sure I have installed
(Fedora 12) to fix this and any other related 'X wasn't included because Y
wasn't installed' issues for the next RC? Talking of which, I think we
should just live with it for 0.8 but fix it properly on trunk.
Thanks,
Robbie
> -----Original Message-----
> From: Gordon Sim [mailto:gsim@redhat.com]
> Sent: 10 November 2010 18:03
> To: dev@qpid.apache.org
> Subject: Re: failure due to missing cluster artefacts (was Re: 0.8 RC1
> available for download)
>
> On 11/10/2010 05:41 PM, Carl Trieloff wrote:
> > On 11/10/2010 11:20 AM, Gordon Sim wrote:
> >>>
> >>> build it on a platform with deps install. that seems less error
> prone.
> >>
> >> Actually I think that ensuring the distribution is 'complete'
> >> regardless of the existence of dependencies on the platform it was
> >> built is less error prone.
> >>
> >> The question in my mind was more on whether we fix this for 0.8 or
> >> make do by regenerating with deps installed.
> >
> >
> > Ok, I'm confused. I had asumed we would want it regenerated on a
> > platform with dependencies there, so they are in the kit. I had also
> > assumed we would want the tests run on it, to make sure it works.
> i.e.
> > also test that the dependencies are correct in the making dist
> process.
>
> Testing the dist is different from creating it. Creating it is
> inevitably one person's job (release manager's). Building and testing
> it
> can (and should) be done more widely.
>
> In my mind the ideal is that when creating a dist, the full source for
> all possible modules are included regardless of whether the deps for
> those are available at the time of creation. That way you ease the
> burden of getting the system setup for creating the dist and leverage
> the existing systems of a wider group of testers for checking all the
> different parts. I think that is also less error prone because the
> configuration of the system on which the dist is created doesn't
> inadvertently cause important pieces to be missed. (e.g. at present we
> include the cman integration sources regardless of whether cman is
> installed when the dist is created).
>
> However others may have a different view and/or we may wish to be
> pragmatic for 0.8. At present the RC won't build if cpg is installed
> unless you explicitly disable cpg integration.
>
> ---------------------------------------------------------------------
> Apache Qpid - AMQP Messaging Implementation
> Project: http://qpid.apache.org
> Use/Interact: mailto:dev-subscribe@qpid.apache.org
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1
available for download)
Posted by Andrew Stitcher <as...@redhat.com>.
On Wed, 2010-11-10 at 18:03 +0000, Gordon Sim wrote:
> ...
> In my mind the ideal is that when creating a dist, the full source for
> all possible modules are included regardless of whether the deps for
> those are available at the time of creation. That way you ease the
> burden of getting the system setup for creating the dist and leverage
> the existing systems of a wider group of testers for checking all the
> different parts. I think that is also less error prone because the
> configuration of the system on which the dist is created doesn't
> inadvertently cause important pieces to be missed. (e.g. at present we
> include the cman integration sources regardless of whether cman is
> installed when the dist is created).
I entirely agree with this.
I think the confusion is that an Apache release by its definition is a
_source_ release and so should include all the source not just the
source that would build on the release manager's packaging machine.
The problem is essentially because the C++ release process used to (and
may still - I'm not sure) require a build of all executables that have
man pages so that they can be run --help to feed the man page creation
process. Becuase of this the autotools make dist target has got
confusingly tangled up with the executable build process so that the
configure conditionals apply to it.
Andrew
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Gordon Sim <gs...@redhat.com>.
On 11/10/2010 05:41 PM, Carl Trieloff wrote:
> On 11/10/2010 11:20 AM, Gordon Sim wrote:
>>>
>>> build it on a platform with deps install. that seems less error prone.
>>
>> Actually I think that ensuring the distribution is 'complete'
>> regardless of the existence of dependencies on the platform it was
>> built is less error prone.
>>
>> The question in my mind was more on whether we fix this for 0.8 or
>> make do by regenerating with deps installed.
>
>
> Ok, I'm confused. I had asumed we would want it regenerated on a
> platform with dependencies there, so they are in the kit. I had also
> assumed we would want the tests run on it, to make sure it works. i.e.
> also test that the dependencies are correct in the making dist process.
Testing the dist is different from creating it. Creating it is
inevitably one person's job (release manager's). Building and testing it
can (and should) be done more widely.
In my mind the ideal is that when creating a dist, the full source for
all possible modules are included regardless of whether the deps for
those are available at the time of creation. That way you ease the
burden of getting the system setup for creating the dist and leverage
the existing systems of a wider group of testers for checking all the
different parts. I think that is also less error prone because the
configuration of the system on which the dist is created doesn't
inadvertently cause important pieces to be missed. (e.g. at present we
include the cman integration sources regardless of whether cman is
installed when the dist is created).
However others may have a different view and/or we may wish to be
pragmatic for 0.8. At present the RC won't build if cpg is installed
unless you explicitly disable cpg integration.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Carl Trieloff <cc...@redhat.com>.
On 11/10/2010 11:20 AM, Gordon Sim wrote:
>>
>> build it on a platform with deps install. that seems less error prone.
>
> Actually I think that ensuring the distribution is 'complete'
> regardless of the existence of dependencies on the platform it was
> built is less error prone.
>
> The question in my mind was more on whether we fix this for 0.8 or
> make do by regenerating with deps installed.
Ok, I'm confused. I had asumed we would want it regenerated on a
platform with dependencies there, so they are in the kit. I had also
assumed we would want the tests run on it, to make sure it works. i.e.
also test that the dependencies are correct in the making dist process.
Maybe you are saying the same thing -- not sure from your statement.
Carl.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Gordon Sim <gs...@redhat.com>.
On 11/10/2010 02:33 PM, Carl Trieloff wrote:
> On 11/10/2010 07:22 AM, Gordon Sim wrote:
>> Either we need to regenerate the distribution on a platform with those
>> dependencies or we need to ensure the relavent files are included
>> regardless of whether the deps were available at the time of running
>> make dist. Thoughts?
>
> build it on a platform with deps install. that seems less error prone.
Actually I think that ensuring the distribution is 'complete' regardless
of the existence of dependencies on the platform it was built is less
error prone.
The question in my mind was more on whether we fix this for 0.8 or make
do by regenerating with deps installed.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
Re: failure due to missing cluster artefacts (was Re: 0.8 RC1 available
for download)
Posted by Carl Trieloff <cc...@redhat.com>.
On 11/10/2010 07:22 AM, Gordon Sim wrote:
> On 11/08/2010 08:47 AM, Robbie Gemmell wrote:
>> Hi all,
>>
>> 0.8 RC1 can be found for download at:
>> http://people.apache.org/~robbie/qpid/0.8/RC1/
>>
>> Output from running RAT across the 'full release' archive can be
>> found at:
>> http://people.apache.org/~robbie/qpid/0.8/0.8rc1_rat_output.txt
>>
>> *Please* take the time to download and try out this release candidate
>> and
>> report your results, such that any additional issues which require
>> addressing before the release can proceed are identified as early as
>> possible.
>
> The qpid-cpp-0.8.tar.gz doesn't include clustering related test files.
> I suspect this is due to conditional inclusion of them in the
> distribution list if the relevant dependencies are not available.
> Autotools build fails at the compilation stage with:
>
>> /bin/sh ../../libtool --tag=CXX --mode=link g++ -g -O2
>> -Itestagent_gen -g -O2 -L/usr/lib/openais -L/usr/lib64/openais
>> -L/usr/lib/corosync -L/usr/lib64/corosync -o testagent
>> testagent-testagent.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Parent.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Child.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildCreated.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildDestroyed.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Package.o
>> ../../src/libqmf.la
>> libtool: link: g++ -g -O2 -Itestagent_gen -g -O2 -o .libs/testagent
>> testagent-testagent.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Parent.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Child.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildCreated.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-EventChildDestroyed.o
>> testagent_gen/qmf/org/apache/qpid/agent/example/testagent-Package.o
>> -L/usr/lib/openais -L/usr/lib64/openais -L/usr/lib/corosync
>> -L/usr/lib64/corosync ../../src/.libs/libqmf.so
>> /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqmfengine.so
>> /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidclient.so
>> /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidcommon.so
>> /home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/.libs/libqpidtypes.so
>> -lboost_program_options -lboost_filesystem -ldl -lrt -lsasl2 -luuid
>> -Wl,-rpath -Wl,/usr/local/lib
>> make[3]: *** No rule to make target `run_cluster_tests', needed by
>> `all-am'. Stop.
>> make[3]: Leaving directory
>> `/home/gordon/qpid-0.8-rc1/cpp-build-autotools/src/tests'
>
>
> Cmake build fails on running cmake with:
>
>> CMake Error in src/tests/CMakeLists.txt:
>> Cannot find source file "InitialStatusMap". Tried extensions .c .C
>> .c++
>> .cc .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp .hxx .in .txx
>>
>>
>> CMake Error in src/tests/CMakeLists.txt:
>> Cannot find source file "StoreStatus". Tried extensions .c .C .c++
>> .cc
>> .cpp .cxx .m .M .mm .h .hh .h++ .hm .hpp .hxx .in .txx
>>
>
> Either we need to regenerate the distribution on a platform with those
> dependencies or we need to ensure the relavent files are included
> regardless of whether the deps were available at the time of running
> make dist. Thoughts?
>
> ---------------------------------------------------------------------
> Apache Qpid - AMQP Messaging Implementation
> Project: http://qpid.apache.org
> Use/Interact: mailto:dev-subscribe@qpid.apache.org
>
build it on a platform with deps install. that seems less error prone.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org