[meta-virtualization] [PATCH 3/3] docker: update to v19.03.2-rc1

[Stefan Agner]

On 2019-08-30 15:24, Bruce Ashfield wrote:
> On Fri, Aug 30, 2019 at 9:03 AM Stefan Agner <stefan at agner.ch> wrote:
>>
>> On 2019-08-30 14:42, Bruce Ashfield wrote:
>> > On Fri, Aug 30, 2019 at 7:36 AM Stefan Agner <stefan at agner.ch> wrote:
>> >>
>> >> From: Stefan Agner <stefan.agner at toradex.com>
>> >>
>> >> Use the docker/engine fork of moby which provides a stabilized version
>> >> of moby in its 19.03 branch.
>> >
>> > I've mentioned before, this recipe really is going to stay on moby,
>> > not a curated version of moby or anything like that.
>>
>> Hm, I guess you refer to:
>> https://lists.yoctoproject.org/pipermail/meta-virtualization/2019-March/004068.html
>>
>> Generally I see your argumentation, and I think it is good to try to
>> build a recipe which tracks Dockers upstream projects.
>>
>> However, I really don't think its a good idea to build from moby
>> directly since it has no proper release process.
> 
> That's actually what the point of that recipe is, it's tracking -dev completely.
> 

Ok.

>>
>> IMHO, moby is more -dev branch rather than anything a OE recipe should
>> rely on. I tried first finding what matches best the Docker version
>> "19.03.2-rc1" in Moby, but since this is really a development branch its
>> not possible to pick a matching (or close to matching) git commit.
>>
>> docker/engine is really just a fork with a proper release process. We
>> get cli and libnetwork also from the Docker company repos...
> 
> Yes, but there's docker-ce for folks that want to follow all the
> docker repos. I don't actually think we need the cli/libnetwork in
> that recipe anymore, but I haven't had
> time to experiment.
> 

Hm, yeah I was wondering about that too. But should be looked at
properly I guess.

>>
>> The fact that this recipe has the name "docker" and follows the docker
>> versioning, I really think it is more reasonable to track
>> https://github.com/docker/engine.
> 
> Yah, that is a problem. I need to just get on with it and rename the
> recipe moby. I said I was going to do it in the spring, but perhaps
> I'll just do it this weekend instead.
> 
> The problem is, a lot of things (and layers I don't know about)
> currently depend on "docker" not "virtual/docker" as I mentioned in my
> various other email, so I really need to create a 3rd recipe that
> follows the docker engine .. but it really is just a placeholder,
> since it isn't different than docker-ce on any significant level.  It
> probably leaves me as the only 'moby' user at that point, but that's
> ok too :D
> 
> i.e. I'd end up with "moby", "docker" and "docker-ce".
> 
> I could do that for the fall release and then clean things up in the spring.
> 
> Comments ? How does that sound ?
> 

Yes I really think the naming would resolve this confusion nicely. I
agree splitting up to moby, docker and docker-ce seems sensible.

>>
>> Btw, the reason I started investigate was that I see quite often errors
>> like this with the current version:
>> docker: Error response from daemon: OCI runtime create failed:
>> container_linux.go:345: starting container process caused
>> "process_linux.go:299: copying bootstrap data to pipe caused \"write
>> init-p: broken pipe\"": unknown.
> 
> Hmmm. I've never seen that here and I tend to stress things pretty
> heavily in the systems that I build. I assume you aren't seeing it
> with docker-ce ?

I was not able to find a command (combination) which makes it always
reproducible, so it seems to be a race condition of some sort. I haven't
tested with docker-ce. I think I mostly saw it on our lower end devices
(Cortex-A7). 

Given we are dealing with a pre-release I thought its not worth further
investigate but just move to the release and then keep an eye on it.

--
Stefan

> 
> Bruce
> 
>>
>> --
>> Stefan
>>
>>
>>
>> >
>> > I already have some queued updated to docker here, so I'll take care
>> > of updating this recipe and stack your other patches on top.
>> >
>> > Bruce
>> >
>> >>
>> >> Bump all git hashes to what is used in Docker CE v19.03.2-rc1.
>> >> The SRCREVs are selected through an audit of moby and docker-ce to
>> >> synchronize the build points for the various components.
>> >>
>> >> Signed-off-by: Stefan Agner <stefan.agner at toradex.com>
>> >> ---
>> >>  recipes-containers/docker/docker_git.bb | 8 ++++----
>> >>  1 file changed, 4 insertions(+), 4 deletions(-)
>> >>
>> >> diff --git a/recipes-containers/docker/docker_git.bb b/recipes-containers/docker/docker_git.bb
>> >> index e45f87e..3892c65 100644
>> >> --- a/recipes-containers/docker/docker_git.bb
>> >> +++ b/recipes-containers/docker/docker_git.bb
>> >> @@ -35,11 +35,11 @@ DESCRIPTION = "Linux container runtime \
>> >>  #     to a docker.inc recipe
>> >>
>> >>  # moby commit matches the docker-ce swarmkit bump on the 18.09 branch
>> >> -SRCREV_moby = "344b093258fcb2195fa393081e5224a6c766c798"
>> >> +SRCREV_moby = "ed20165a37b40ff1cfbe55e218344c5e89f30ee2"
>> >>  SRCREV_libnetwork = "5ac07abef4eee176423fdc1b870d435258e2d381"
>> >> -SRCREV_cli = "2f1931f9eb2d6bac2efd48d94739f2e9919d4d7d"
>> >> +SRCREV_cli = "578ab52ece34b084621954e0b543003410cdbe84"
>> >>  SRC_URI = "\
>> >> -       git://github.com/moby/moby.git;nobranch=1;name=moby \
>> >> +       git://github.com/docker/engine.git;branch=19.03;name=moby \
>> >>         git://github.com/docker/libnetwork.git;branch=master;name=libnetwork;destsuffix=git/libnetwork \
>> >>         git://github.com/docker/cli;branch=19.03;name=cli;destsuffix=git/cli \
>> >>         file://docker.init \
>> >> @@ -57,7 +57,7 @@ GO_IMPORT = "import"
>> >>
>> >>  S = "${WORKDIR}/git"
>> >>
>> >> -DOCKER_VERSION = "19.03.0-rc3"
>> >> +DOCKER_VERSION = "19.03.2-rc1"
>> >>  PV = "${DOCKER_VERSION}+git${SRCREV_moby}"
>> >>
>> >>  PACKAGES =+ "${PN}-contrib"
>> >> --
>> >> 2.20.1
>> >>
>> >> --
>> >> _______________________________________________
>> >> meta-virtualization mailing list
>> >> meta-virtualization at yoctoproject.org
>> >> https://lists.yoctoproject.org/listinfo/meta-virtualization