QA » openQA Project » openQA Tests

@jorauch just to be 100% clear: You mean current line 685 right? :)

Indeed, I mean the new Line 685

PR merged, waiting for verification on OSD

jorauch wrote:

After the emergency PR has been merged we should be done with this ticket?

After the verification on OSD, yes

Could be next step of this to use select_serial_terminal() (instead of select_console 'root-console'), which is wrapper trying to handle everything. It's adjustable with variables (VIRTIO_CONSOLE=0 to disable virtio console; SERIAL_CONSOLE=0 to disable svirt serial console).

Nobody is forced to use virtio console / serial console, but on the other hand it's much faster and we hear complains about lack of resources (osd workers being under heavy load).

pvorel wrote:

Could be next step of this to use select_serial_terminal() (instead of select_console 'root-console')

the idea was actually the opposite, to use select_console 'whatever-console-you-define' instead of the wrapper. Of course the console "whatever-console-you-define" can exactly be the serial terminal whenever configured. As we can freely create the consoles and activate them within the test code all this should be possible without needing to change any os-autoinst code.

But simple select_console 'root-console' just ignores the fact that some people want virtio console. Having wrapper which just select "proper console" makes tests more portable and avoid if/else code in tests.

When I created select_serial_terminal() I wanted to have smart way to select console. I also thought that speed of openQA is important thus virtio console should be used as much as it can (i.e. when needles can be avoided, which should IMHO be in consoletests, but tests/console/consoletest_setup.pm does needle matching due testing fonts, it's a question if this must be done in every test).
But that was wrong assumption, as there are probably more tests which needs needles (maybe there could be workaround in needles functions to switch to root-console, do the matching and then switch back).

But now I understand it, that select_console 'root-console' is for all tests matching needles and select_serial_terminal() for kernel and any console tests (which don't work with needles).
Assuming that there is now no need to have any wrapper for needles based testing regardless used backend? (will select_console 'root-console' always work on any backend without any other helper function).
Than the state should be documented, so it's clear.

okurz wrote:

I hope you all agree that it makes sense to keep this ticket open when the discussion is not yet resolved? Setting to "Feedback".

+1

@pvorel Sure, I agree with all your points regarding a smart way to select consoles based on needs and preferring the faster serial terminal where possible. I think we are not too far from our goal. Maybe it can be as simple as reworking the function select_serial_terminal() into an according console selection?

I don't understand how you want to achieve it.
The goal for select_serial_terminal() was not to have parameters, but set preferences via variables (thus code is simple).

I'd be for keeping select_console() and select_serial_terminal() as it is + add select_needles_console() (need better way), which would be equivalent of select_console 'user-serial-terminal' (e.g. both ways use wrapper, not just console based testing).
But that might be strange/radical idea.

Maybe this needs to be moved to os-autoinst and also makes sense to have there for the benefit of multiple test distributions.

So that e.g. select_console 'root-serial-terminal' would do what select_serial_terminal() does so far? Same for select_console 'user-serial-terminal' replacing select_user_serial_terminal?

I originally wanted to have it in the library. But it depends on variables which are set only in our testing repository, that's why it's there. IMHO generally there are some parts in our test repo which should be in the library, but are in test repo due the difficulty to make changes in the library. To put it into library, the decision would have to be some general mechanism or at least on variables which aren't specific on our testing repository.

hm, this sounds like we have a conflicting proposal to the os-autoinst design. I suggest you wait for coolo's return and ask for his opinion on the architectural design.

BTW: just a comment: I still think moving back to select_console this would be step back. select_serial_terminal() allows to avoid many if conditions (currently used in 287 tests). It'd be great to have similar functionality in os-autoinst but consoles are based on distros git, thus not possible.

Sure it would be possible to define custom console classes within the test distribution and use that one with select_console. The source location of the files does not matter as we already include from the lib/ folder of the test distribution

This ticket was set to Normal priority but was not updated within the SLO period. Please consider picking up this ticket or just set the ticket to the next lower priority.

This ticket was set to Normal priority but was not updated within the SLO period. Please consider picking up this ticket or just set the ticket to the next lower priority.