Containers: use serial terminal optinally for the tests #10424

jlausuch · 2020-06-01T15:37:41Z

Using this option we gain 2 things:

Execution time reduced more than half.
Everything is logged to serial_termina.txt

It is possible to disable it by setting VIRTIO_CONSOLE=0.

Related ticket: https://progress.opensuse.org/issues/67531
Verification runs:
- VIRTIO_CONSOLE=0: http://fromm.arch.suse.de/tests/942 (~29 min)
- VIRTIO_CONSOLE=1: http://fromm.arch.suse.de/tests/941 (~12 min)

jlausuch · 2020-06-01T15:39:20Z

@pdostal @punkioudi please be aware of this. It's a big change in how tests are executed, so I don't want to convince everyone to use it, that's why it's optional. It won't affect QAM jobs, but you might be interested as well.

asmorodskyi · 2020-06-01T19:02:19Z

I thought that we have VIRTIO_TERMINAL to make such decisions , not sure why we need one more variable . Anyway if we will decide to keep it after all I would move this variable inside select_serial_terminal function but generally I think VIRTIO_TERMINAL is enough .
@pevik you might be interested in this discussion

pevik · 2020-06-01T19:08:21Z

@jlausuch yes, this functionality has been already implemented with VIRTIO_CONSOLE=0
https://github.com/os-autoinst/os-autoinst-distri-opensuse/blob/master/lib/opensusebasetest.pm#L1078

jlausuch · 2020-06-01T20:52:23Z

I thought that we have VIRTIO_TERMINAL to make such decisions , not sure why we need one more variable . Anyway if we will decide to keep it after all I would move this variable inside select_serial_terminal function but generally I think VIRTIO_TERMINAL is enough .
@pevik you might be interested in this discussion

I guess you mean VIRTIO_CONSOLE, not VIRTIO_TERMINAL. I thought the same and started using VIRTIO_CONSOLE variable, but then I realized that all container jobs have VIRTIO_CONSOLE=1 See this or this. Does it mean that the variable should be =0 in those jobs?
If so, I will replace all calls select_console("root-console"); to $self->select_serial_terminal and force VIRTIO_CONSOLE=0 in the jobs that are not from us. WDYT?

pevik · 2020-06-02T06:12:04Z

Virtio console is enabled by default (VIRTIO_CONSOLE=1), because we realized there is really no reason to have it disabled (see os-autoinst/os-autoinst#1045, merged as os-autoinst/os-autoinst@7226ced). You need to explicitly set VIRTIO_CONSOLE=0 to disable it.

I should document this in https://github.com/os-autoinst/os-autoinst-distri-opensuse/blob/master/lib/opensusebasetest.pm#L1078, because in general docs it's somehow reflected, but VIRTIO_CONSOLE=1 is specific to openSUSE/SLES tests https://github.com/os-autoinst/openQA/blob/master/docs/WritingTests.asciidoc#using-a-serial-terminal.

VIRTIO_CONSOLE=1 is set on machine setup on machines 64bit, aarch64-virtio, ppc64le-virtio. But because it's default on, on machines aarch64 and ppc64le it's also enabled (it's not in defined in https://openqa.suse.de/tests/4295384#settings, but it's on https://openqa.suse.de/tests/4295384/file/vars.json).

VIRTIO_CONSOLE is for qemu backed. svirt backend has SERIAL_CONSOLE, which is not enabled by default. Thus from many machines using svirt is enabled only on s390x-kvm-sle15 (and maybe s390x-kvm-sle12), but not on vmware and hyperv (which still doesn't have this support, see https://progress.opensuse.org/issues/55985).

pevik · 2020-06-02T06:45:31Z

I also improved doc, can you please have a look whether it's really an improvement?

jlausuch · 2020-06-02T06:59:35Z

Virtio console is enabled by default (VIRTIO_CONSOLE=1), because we realized there is really no reason to have it disabled (see os-autoinst/os-autoinst#1045, merged as os-autoinst/os-autoinst@7226ced). You need to explicitly set VIRTIO_CONSOLE=0 to disable it.

I should document this in https://github.com/os-autoinst/os-autoinst-distri-opensuse/blob/master/lib/opensusebasetest.pm#L1078, because in general docs it's somehow reflected, but VIRTIO_CONSOLE=1 is specific to openSUSE/SLES tests https://github.com/os-autoinst/openQA/blob/master/docs/WritingTests.asciidoc#using-a-serial-terminal.

VIRTIO_CONSOLE=1 is set on machine setup on machines 64bit, aarch64-virtio, ppc64le-virtio. But because it's default on, on machines aarch64 and ppc64le it's also enabled (it's not in defined in https://openqa.suse.de/tests/4295384#settings, but it's on https://openqa.suse.de/tests/4295384/file/vars.json).

VIRTIO_CONSOLE is for qemu backed. svirt backend has SERIAL_CONSOLE, which is not enabled by default. Thus from many machines using svirt is enabled only on s390x-kvm-sle15 (and maybe s390x-kvm-sle12), but not on vmware and hyperv (which still doesn't have this support, see https://progress.opensuse.org/issues/55985).

Thanks for clarification. I will remove the new variable and update the PR.

punkioudi · 2020-06-02T08:07:51Z

@ggkioulis this might interests you too

cfconrad

LGTM

Using this option we gain 2 things: 1) Execution time reduced more than half. 2) Everything is logged to serial_termina.txt It is possible to disable it by setting VIRTIO_CONSOLE=0.

asmorodskyi · 2020-06-02T11:03:57Z

tests/console/docker_compose.pm

@@ -30,7 +30,8 @@ use strict;
 use warnings;

 sub run {
-    select_console("root-console");
+    my ($self) = @_;


since we excluding it for SLE . does it make sense to touch it ? do we plan also migrate to VIRTIO_CONSOLE=1 in o3 ?

We might include it in the future and all these tests are a common subset of container tests, which I think should be similar the way they are handled.

Is there any reason why not? BTW where are these tests in o3?

A test called extra_tests_textmode_containers in Tumbleweed builds e.g. https://openqa.opensuse.org/tests/1282756
and Leap https://openqa.opensuse.org/tests/1282504#

pevik · 2020-06-02T12:09:38Z

tests/console/docker_compose.pm

@@ -30,7 +30,8 @@ use strict;
 use warnings;

 sub run {
-    select_console("root-console");
+    my ($self) = @_;


Is there any reason why not? BTW where are these tests in o3?

jlausuch · 2020-06-02T14:55:55Z

@Vogtinator @pdostal : Would it be fine for you to merge this PR and if you don't want this feature then set VIRTIO_CONSOLE=0 in QAM and O3 side?

Vogtinator · 2020-06-02T14:58:30Z

@Vogtinator @pdostal : Would it be fine for you to merge this PR and if you don't want this feature then set VIRTIO_CONSOLE=0 in QAM and O3 side?

Please add openSUSE TW and Leap 15.1 verification runs. If those pass, LGTM.

pevik · 2020-06-02T17:20:16Z

@Vogtinator I tested it for openSUSE:

Build20200528 of opensuse-Tumbleweed-DVD.x86_64 http://quasar.suse.cz/tests/5279
Build665.2 of opensuse-15.2-DVD.x86_64 http://quasar.suse.cz/tests/5279

BTW on my host 11 min vs. 33 min => virtio console is 3x faster :).
IMHO virtio console is OK for o3. It has problems on some old SLES versions (SLE12 SP1 maybe, @mimi1vx probably remember which one).

jlausuch · 2020-06-02T20:03:37Z

@Vogtinator I tested it for openSUSE:

Build20200528 of opensuse-Tumbleweed-DVD.x86_64 http://quasar.suse.cz/tests/5279

Build665.2 of opensuse-15.2-DVD.x86_64 http://quasar.suse.cz/tests/5279

BTW on my host 11 min vs. 33 min => virtio console is 3x faster :).
IMHO virtio console is OK for o3. It has problems on some old SLES versions (SLE12 SP1 maybe, @mimi1vx probably remember which one).

Thanks! I was about to do it :)
btw, correct link for Tumbleweed: http://quasar.suse.cz/tests/5280

pevik · 2020-06-03T05:35:15Z

I did it before, because I was myself curious how much the speedup is. Although prefer virtio/serial console anyway (to get text output), but I have to admit for tests with minimal output and long running commands the speedup can be not that significant.

Vogtinator · 2020-06-03T05:45:29Z

@Vogtinator I tested it for openSUSE:

Build20200528 of opensuse-Tumbleweed-DVD.x86_64 http://quasar.suse.cz/tests/5279

Build665.2 of opensuse-15.2-DVD.x86_64 http://quasar.suse.cz/tests/5279

BTW on my host 11 min vs. 33 min => virtio console is 3x faster :).

Great!

pevik · 2020-06-03T06:25:01Z

@mdoucha @mimi1vx @dzedro FYI not sure if this affect some SLE12 QAM tests. Hope you already have VIRTIO_CONSOLE=0 for isos which need to have virtio disabled.

jlausuch · 2020-06-03T06:31:29Z

@mdoucha @mimi1vx @dzedro FYI not sure if this affect some SLE12 QAM tests. Hope you already have VIRTIO_CONSOLE=0 for isos which need to have virtio disabled.

The biggest negative impact is that there will be more frames on the Details view.

pevik · 2020-06-03T06:53:56Z

@jlausuch FYI I meant impact on SLE12-SP1 ppc64le "util-linux on sle12/sle12sp1 rejects root login on any console which isnt listed in /etc/securetty , from SP2 is this file ignored" (https://progress.opensuse.org/issues/18980#note-14), it has been fixed long time ago for LTP:#2820, #6352 (https://progress.opensuse.org/issues/18980, https://progress.opensuse.org/issues/18980)

but not sure about current status for the other tests, this old @dzedro complain was closed, so I guess correct setup has been done
https://progress.opensuse.org/issues/45842

jlausuch requested review from asmorodskyi, mloviska and cfconrad June 1, 2020 15:37

This was referenced Jun 2, 2020

doc: Improve virtio/svirt serial terminal doc os-autoinst/openQA#3146

Merged

select_serial_terminal: Update doc (current state) #10427

Merged

cfconrad approved these changes Jun 2, 2020

View reviewed changes

Containers: use serial terminal for the tests

347e95d

Using this option we gain 2 things: 1) Execution time reduced more than half. 2) Everything is logged to serial_termina.txt It is possible to disable it by setting VIRTIO_CONSOLE=0.

asmorodskyi reviewed Jun 2, 2020

View reviewed changes

pevik approved these changes Jun 2, 2020

View reviewed changes

Vogtinator approved these changes Jun 3, 2020

View reviewed changes

asmorodskyi merged commit a3aa6a4 into os-autoinst:master Jun 3, 2020

jlausuch deleted the containers_use_serial branch June 3, 2020 09:16

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Containers: use serial terminal optinally for the tests #10424

Containers: use serial terminal optinally for the tests #10424

jlausuch commented Jun 1, 2020 •

edited

Loading

jlausuch commented Jun 1, 2020

asmorodskyi commented Jun 1, 2020

pevik commented Jun 1, 2020

jlausuch commented Jun 1, 2020 •

edited

Loading

pevik commented Jun 2, 2020

pevik commented Jun 2, 2020

jlausuch commented Jun 2, 2020

punkioudi commented Jun 2, 2020 •

edited

Loading

cfconrad left a comment

asmorodskyi Jun 2, 2020

jlausuch Jun 2, 2020

pevik Jun 2, 2020

jlausuch Jun 2, 2020

pevik Jun 2, 2020

jlausuch commented Jun 2, 2020

Vogtinator commented Jun 2, 2020

pevik commented Jun 2, 2020 •

edited

Loading

jlausuch commented Jun 2, 2020 •

edited

Loading

pevik commented Jun 3, 2020

Vogtinator commented Jun 3, 2020

pevik commented Jun 3, 2020

jlausuch commented Jun 3, 2020

pevik commented Jun 3, 2020 •

edited

Loading

Containers: use serial terminal optinally for the tests #10424

Containers: use serial terminal optinally for the tests #10424

Conversation

jlausuch commented Jun 1, 2020 • edited Loading

jlausuch commented Jun 1, 2020

asmorodskyi commented Jun 1, 2020

pevik commented Jun 1, 2020

jlausuch commented Jun 1, 2020 • edited Loading

pevik commented Jun 2, 2020

pevik commented Jun 2, 2020

jlausuch commented Jun 2, 2020

punkioudi commented Jun 2, 2020 • edited Loading

cfconrad left a comment

Choose a reason for hiding this comment

asmorodskyi Jun 2, 2020

Choose a reason for hiding this comment

jlausuch Jun 2, 2020

Choose a reason for hiding this comment

pevik Jun 2, 2020

Choose a reason for hiding this comment

jlausuch Jun 2, 2020

Choose a reason for hiding this comment

pevik Jun 2, 2020

Choose a reason for hiding this comment

jlausuch commented Jun 2, 2020

Vogtinator commented Jun 2, 2020

pevik commented Jun 2, 2020 • edited Loading

jlausuch commented Jun 2, 2020 • edited Loading

pevik commented Jun 3, 2020

Vogtinator commented Jun 3, 2020

pevik commented Jun 3, 2020

jlausuch commented Jun 3, 2020

pevik commented Jun 3, 2020 • edited Loading

jlausuch commented Jun 1, 2020 •

edited

Loading

jlausuch commented Jun 1, 2020 •

edited

Loading

punkioudi commented Jun 2, 2020 •

edited

Loading

pevik commented Jun 2, 2020 •

edited

Loading

jlausuch commented Jun 2, 2020 •

edited

Loading

pevik commented Jun 3, 2020 •

edited

Loading