deps: bump to race-free c/image and c/storage along with test to verify concurrent/parallel builds

[flouthoc]

Commits explained in bullet points.

• Bump c/storage to main/d06b0f
• Bump c/image to main/9a9cd9

New Names API in c/storage and c/image ensures that Name modification operations are race-free.

• Adds a test to verify parallel/concurrent builds via podman don't race against each other.

Read more about problem statement here: containers/buildah#3794 and #13339

[giuseppe]

LGTM

[edsantiago]

/hold

[edsantiago]

This does not test what you think it tests. There is no parallelization here anywhere that I can see.

[edsantiago]

I'm rewriting the test. I will post my changes once I'm done. This must not merge.

[edsantiago]

This is super-dangerous, it runs the risk of hanging forever, but I can find no way around that.

I tried to grok the bug but can't; so I don't know if it's important to build lots-of-different-images or if you want all-the-same. If you want the latter, just change echo hi$i in the Containerfile to echo hi or something like that.
UPDATED: indeed, with just RUN echo hi (no $i) my test fails pretty consistently on main.

@test "podman parallel build should not race" {
    # Start thirty builds in parallel
    local count=30
    local -a pids
    for i in $(seq --format '%02g' 1 $count); do
        local d=$PODMAN_TMPDIR/build-test$i
        mkdir $d
        cat >$d/Containerfile <<EOF
FROM $IMAGE
RUN echo hi
EOF
        $PODMAN build -t i$i $d &>/dev/null &
    done

    # DANGER DANGER DANGER: this has the potential to hang forever,
    # should any individual build fail. I can find no way to add a timeout.
    wait

    # Now delete each image. If an image wasn't built, rmi will fail
    for i in $(seq --format '%02g' 1 $count); do
        run_podman rmi i$i
    done
}

[giuseppe]

    # DANGER DANGER DANGER: this has the potential to hang forever,
    # should any individual build fail. I can find no way to add a timeout.

could we add the timeout on the podman build command itself, e.g. timeout -s KILL 60 podman build ...?

[flouthoc]

@edsantiago @giuseppe Instead of wait could we use a sleep here ? So we don't hang for infinite ?

[edsantiago]

@giuseppe excellent idea, and it seems to work! Thank you! Here's the new code:

@test "podman parallel build should not race" {
    # Start thirty builds in parallel
    local count=30
    local -a pids
    for i in $(seq --format '%02g' 1 $count); do
        local d=$PODMAN_TMPDIR/build-test$i
        mkdir $d
        cat >$d/Containerfile <<EOF
FROM $IMAGE
RUN echo hi
EOF
        timeout --foreground -v --kill=10 60 $PODMAN build -t i$i $d &>/dev/null &
    done

    wait

    # Now delete all built images. If an image wasn't built, rmi will fail
    run_podman rmi $(seq --format 'i%02g' 1 $count)
}

(note the optimization in the rmi)

@flouthoc we do not use sleep if at all possible. It is evil: it forces a constant delay no matter how long the builds take (forcing CI to take longer and longer, which makes the world a worse place). It is also not a guarantee: it is possible for some circumstance to make build take very long one day, and the sleep would finish first, and we would have an impossible-to-debug flake.

[flouthoc]

@edsantiago Thanks a lot for patching the test and i have verified it 👍 it does not fails on my local lets see on the CI.

[edsantiago]

Update: here's a cleaner, less sloppy version of the test. Please use this one, it is less misleading (i.e. there is no reason to create multiple Containerfiles; and the comments are better)

@test "podman parallel build should not race" {
    # Run thirty parallel builds using the same Containerfile
    cat >$PODMAN_TMPDIR/Containerfile <<EOF
FROM $IMAGE
RUN echo hi
EOF

    local count=30
    for i in $(seq --format '%02g' 1 $count); do
        timeout --foreground -v --kill=10 60 \
                $PODMAN build -t i$i $PODMAN_TMPDIR &>/dev/null &
    done

    # Wait for all background builds to complete. Note that this succeeds
    # even if some of the individual builds fail! Our actual test is below.
    wait

    # Now delete all built images. If any image wasn't built, rmi will fail
    # and test will fail.
    run_podman rmi $(seq --format 'i%02g' 1 $count)
}

[openshift-ci]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: edsantiago, flouthoc, giuseppe

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [edsantiago,flouthoc,giuseppe]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[edsantiago]

Oops. Looks like you need skip_if_remote

[edsantiago]

/lgtm

[flouthoc]

FYI @TomSweeneyRedHat