Stabilize KDE live installation on aarch64 #10474

OleksandrOrlov · 2020-06-08T14:31:49Z

The PR adds the required steps to stabilize Tumbleweed installation using Live CD on aarch64.

Changes introduced:

live_installation test module sporadically failed on aarch64 because when clicking on "Installation" icon on Desktop, the mouse button was hold too long and popup appeared instead of opening the application. The PR changes the click with x11_start_program, which also allows to open the app but without relying on click;
await_install failed because black screen appeared after some time without activity. The PR just adjusts the existing approach with mouse moving from time to time. As shown from 4 verification runs, this is effective and fixes the issue;
Legacy sub handle_livecd_screenlock seems to be not effective with the current product, but it is still left (maybe temporary), just to not broke existing tests that potentially may use it (e.g. on x64). Though, the way how check_screen is used here was changed. As previously it used array in function arguments instead of arrayref, that is required by function definition;
Also, QEMURAM=4096 QEMUCPUS=4 TIMEOUT_SCALE=4 were added to Test Suite settings.

Related ticket: https://progress.opensuse.org/issues/66355
Verification runs:
https://openqa.opensuse.org/tests/1293429
https://openqa.opensuse.org/tests/1293430
https://openqa.opensuse.org/tests/1293431
https://openqa.opensuse.org/tests/1293432

The commit fixes check_screen method call to pass arrayref there instead of actual array. Related ticket: poo#66355

b10n1k

Approved but you have forgot one small thing In L51

tests/installation/live_installation.pm

tests/installation/await_install.pm

rwx788 · 2020-06-09T06:44:34Z

tests/installation/await_install.pm

@@ -173,7 +169,7 @@ sub run {
            send_key 'alt-o';    # ok
            next;
        }
-        if (get_var('LIVECD') and (match_has_tag('screenlock') || match_has_tag('blackscreen'))) {
+        if (get_var('LIVECD') and (check_screen('screenlock') || check_screen('blackscreen'))) {


Just wondering how is it better of multi tag assert screen we did previously? Does it help?

On this line it does not actually matters. But it matters on the previous steps:

Previously 'screenlock' and 'blackscreen' were part of @tags array. Then this array was used in while loop on line 132: my $ret = check_screen \@tags, 30;. This check was designed to define if installation is finished ('rebootnow' needle) or interrupted with 'yast_error', 'yast2_wrong_digest' or other popups. For some reason 'screenlock' and just recently 'blackscreen' were added to this @tags array. Maybe it made sense in the past, but now this leads to the incorrect detecting of the installation state.

I removed these tags from @tags array. And on this line I just changed to check_screen, because obviously it will not work with match_has_tag.

From the code I don't see any difference. As we have this condition in the end of the list. So to me those look as equivalent (except timing), so if you say that helps, we can keep it.

This seems like a broken change to me. In the screenlock or blackscreen case this code isn't reached due to the next; in line 143, is it?

tests/installation/await_install.pm

The commit changes the behavior of the test module to move mouse, cursor while installation is performed with Live CD. That is done to prevent screenlock to be appeared. Change is done in scope of work to stabilize the installation test suite with Live CD installation on aarch64. Related ticket: poo#66355

rwx788

LGTM

b10n1k

thanks man. Approved

okurz

I see problems with this PR:

You are invalidating the test. The idea is not to call a low-level script but test from user perspective. Clicking the icon to start the installer is the correct way.
I am not convinced that just a handful of verification runs can prove stability in this case. And I don't trust that the additional non-efficient check_screen is better than the multi-tag assert screen

I would very much prefer if you revert your change and find a better way.

@Vogtinator you might also want to comment on this.

okurz · 2020-06-10T08:54:16Z

tests/installation/live_installation.pm

@@ -48,7 +48,7 @@ sub run {
        assert_and_click 'live-upgrade';
    }
    else {
-        assert_and_click 'live-installation';
+        x11_start_program('xdg-su -c "/usr/sbin/start-install.sh"', target_match => 'maximize');


this is invalidating the test case as the point is to simulate what the user does without relying on internals.

Yes, this should definitely be reverted. If assert_and_click does not work for the simple case of starting an application, something is really broken.

I don't agree with you. This is another way how user could do that.

@Vogtinator This is not a problem of a product. You'll not be able to reproduce the issue manually, unless you are holding mouse click for a second or more. This is sporadic issue related to connection/vnc. When we'll revert the change we'll have the test sporadically failing and it will be false fails, as the product still works correctly.

You're practically suggesting to replace all assert_and_click by x11_start_program here because of some weird VNC issue here, which I don't think is adequate.

The VNC issue has to be fixed in any case. That a single 0.15s press is suddenly ~7x longer is really broken.

+1, that's exactly what I have proposed to @OleksandrOrlov just now, as on 64bit scenario works just fine.
@OleksandrOrlov WDYT?

This seems reasonable.

but do you need the change at all as you already bumped RAM and CPUs for the test scenario?

Oliver, you will not believe, but I tried to bump RAM and CPU first before introducing any new changes.

Vogtinator · 2020-06-10T09:01:06Z

Yes, I would also prefer a revert here.

On all architectures, except aarch64, start Installer Wizard by clicking on the appropriate icon. On aarch64 use x11_start_program instead due to sporadic issue when mouse is hold too long, so that context pop-up is appeared. The decision to revert the click on icon for other architectures was made during discussion in PR os-autoinst#10474. Related ticket: https://progress.opensuse.org/issues/66355

Use arrayfef as an argument in check_screen

7bbaf9e

The commit fixes check_screen method call to pass arrayref there instead of actual array. Related ticket: poo#66355

b10n1k approved these changes Jun 8, 2020

View reviewed changes

b10n1k reviewed Jun 8, 2020

View reviewed changes

tests/installation/live_installation.pm Outdated Show resolved Hide resolved

rwx788 reviewed Jun 9, 2020

View reviewed changes

tests/installation/await_install.pm Show resolved Hide resolved

rwx788 reviewed Jun 9, 2020

View reviewed changes

OleksandrOrlov force-pushed the 66355_stabilize_kde_live_install_aarch64 branch 3 times, most recently from 181ef5e to 4c16182 Compare June 9, 2020 08:00

b10n1k reviewed Jun 9, 2020

View reviewed changes

tests/installation/await_install.pm Show resolved Hide resolved

OleksandrOrlov force-pushed the 66355_stabilize_kde_live_install_aarch64 branch from 4c16182 to 7e79403 Compare June 9, 2020 08:09

rwx788 reviewed Jun 9, 2020

View reviewed changes

b10n1k approved these changes Jun 9, 2020

View reviewed changes

rwx788 merged commit bba0f8f into os-autoinst:master Jun 9, 2020

okurz reviewed Jun 10, 2020

View reviewed changes

OleksandrOrlov deleted the 66355_stabilize_kde_live_install_aarch64 branch July 10, 2020 10:29

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stabilize KDE live installation on aarch64 #10474

Stabilize KDE live installation on aarch64 #10474

OleksandrOrlov commented Jun 8, 2020 •

edited

b10n1k left a comment •

edited

rwx788 Jun 9, 2020

OleksandrOrlov Jun 9, 2020

rwx788 Jun 9, 2020

Vogtinator Jun 10, 2020

rwx788 left a comment

b10n1k left a comment

okurz left a comment

okurz Jun 10, 2020

Vogtinator Jun 10, 2020

OleksandrOrlov Jun 10, 2020

OleksandrOrlov Jun 10, 2020

Vogtinator Jun 10, 2020

rwx788 Jun 10, 2020

OleksandrOrlov Jun 10, 2020

okurz Jun 10, 2020

OleksandrOrlov Jun 10, 2020

okurz Jun 10, 2020

Vogtinator commented Jun 10, 2020

Stabilize KDE live installation on aarch64 #10474

Stabilize KDE live installation on aarch64 #10474

Conversation

OleksandrOrlov commented Jun 8, 2020 • edited

b10n1k left a comment • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rwx788 left a comment

Choose a reason for hiding this comment

b10n1k left a comment

Choose a reason for hiding this comment

okurz left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Vogtinator commented Jun 10, 2020

OleksandrOrlov commented Jun 8, 2020 •

edited

b10n1k left a comment •

edited