Fix sporadic error in t/10-virtio_terminal.t

[cfconrad]

I wasn't able to reproce the problem, but this is what I guess.
We call $console->peak() before we wrote the data into the other side of
the pipe.
Now we use USR2 signal to sync between "sender" and "receiver".
Also calling flush() explicit, so we ensure data is not buffered by
perl.

Ticked: https://progress.opensuse.org/issues/94991
Issue: #1686

[codecov]

Codecov Report

Merging #1870 (9055ea2) into master (ae6eed2) will increase coverage by 0.10%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1870      +/-   ##
==========================================
+ Coverage   67.46%   67.56%   +0.10%     
==========================================
  Files          62       62              
  Lines        6885     6885              
==========================================
+ Hits         4645     4652       +7     
+ Misses       2240     2233       -7     

Impacted Files	Coverage Δ
consoles/sshSerial.pm	100.00% <0.00%> (+24.13%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ae6eed2...9055ea2. Read the comment docs.

[Martchus]

I suppose this could generally work but see the inline comments.

[okurz]

As the problem showed up on OBS and as you already changed the spec file the test is also executed on OBS. That's good. But as that was a sporadic problem I suggest to run much more often. E.g. you could try to run a for-loop within OBS checks temporarily as part of this PR. Or you create a custom OBS branch project checking out from git.

[cfconrad]

Thats what I currently do in here: https://build.opensuse.org/package/show/home:cfconrad:branches:devel:openQA/os-autoinst
And I use this script:

LOOP=0

osc_r() {
    osc r | grep openSUSE_Factory
}

is_building() {
    test $(osc_r | awk '{print $4}' | grep -P 'building|blocked|scheduled|finished' | wc -l) -gt 0
}

is_failed() {
    test $(osc_r | awk '{print $4}' | grep -v succeeded | wc -l) -ne 0
}

osc_rebuild() {
    for i in openSUSE_Factory_PowerPC openSUSE_Factory_ARM openSUSE_Factory; do
        osc rebuild -r $i
    done
    sleep 10;
}

msg() {
    echo "[$LOOP]" $@
}

while true; do
    LOOP=$(( $LOOP + 1 ));
    while is_building; do 
        msg "Wait for builds are ready"
        osc_r
        sleep 1;
    done

    if is_failed; then
        msg "Something failed!!!"
        osc_r
        exit 1 
    fi

    osc_rebuild
done

I'm on loop 16. Any hints to improve this?

[okurz]

I'm on loop 16. Any hints to improve this?

Great! Your approach is better than what I had in mind because you do rebuild the complete OBS job so you might hit different OBS workers with differing performance. Well, to improve you could first try to gather the fail-ratio before your fix so that you know what to test for :) But I guess it's ok

[okurz]

I have reviewed the CI runs and found no unhandled output and comparing to the last successful run of CI tests in master in https://github.com/os-autoinst/os-autoinst/runs/4340445436?check_suite_focus=true I see a runtime of 8s and we still have 8s in your PR