Eclipse OMR Architecture, 23 Jun 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OMR Architecture Meeting June 23, 2022

Description

Agenda:
* New PR build features and launch triggers (#6525) [ @babsingh ]

A

Welcome everyone to the june 23rd omar architecture meeting uh today we have uh one topic from uh from bev meat singh, so he'll be talking about uh some of the changes to the pr testing that he's made recently and and some of the launch triggers so bethany. Please take it away when you're ready.

B

Thank you daryl. So today uh I will be discussing about the new features I'm proposing to the omrpr testing framework and just a little background. What is pi testing peer testing refers to the jobs that we launched in a pull request to compile and test the changes which are being proposed in the pr. The command. Syntax is jenkins. Space build space, a platform specification.

B

We currently have 11 platform specifications which are listed on the slide, and some examples of using this command are shown at the bottom of the slide. Jenkins build all will launch a job for each platform. Specifications shown uh jenkins will then p. Linux will launch a job on the linux ppc platform and similarly, the zos example will launch a job on a zos machine.

B

So currently, uh the peer testing has a number of deficiencies. For instance, we cannot run a job inside a container.

B

uh We don't have the ability to choose specific machines where to launch the pr job, and similarly, uh we are unable to specify extra options for configure cmake or extra options for make while compilations and the same applies to specifying extra test options or specifying additional environment variables.

B

So these are the deficiencies that I plan to. I propose to resolve with the new features.

B

So I just a quick review of how the old command looks like and how it differs from the new proposed syntax in the old version, we could only specify the platform specification and the hard-coded settings in the scripts will be used, always in the new syntax.

B

We can provide additional options for each platform specification and brackets, and these should either change the default settings or add more settings to the pr job.

B

So one of them, so a new feature to account for running in a docker container, is the option docker. So that's the keyword. If you specify this keyword in brackets, then the pr drawer will be run inside a docker container and there's a negated version of this keyword, which will say, don't run the sphere drop inside a docker container, and this will this can be used in cases if the default setting is to run inside a docker container.

A

What will happen if you ask to run a in a docker container for a for a platform, but there isn't such a environment. It will available.

B

Implicitly, fail: that's work, that's uh the general rule that I've applied for each new feature. If a user specifies wrong instructions, it should implicitly fail. The job should fail.

B

Okay, it will, for instance, uh in this case the pr draw will say it was unable to find the docker command.

A

Okay, good: the useful message is helpful: cool.

B

And uh on this slide there, I've added a keyword to run a job on a cgr, v1 nerd. uh This was needed for the c group's work which was recently completed.

B

uh Linux systems can have two configurations: it can either be a seeger, v1, node or a sega v2 node, and to have sufficient or complete test coverage for c groups.

B

We had to run on both nodes, so the secret v1 keyword allows us to choose a linux machine with the secret v1 configuration, and this is also a negated version of this keyword and if this keyword gets used, for instance, on z, os, which doesn't have any secret v1 node, the build job should fail, saying that no such node is available, no machine was found and the build will just fill with that error message.

B

uh The next keyword is seeker v2, which is completely identical to see who read one, but it's in this case it's going to find a seeger, v2 linux, machine, configured, linux, machine and the negated version should say: don't run it on a secret v2 machine, and this will be helpful in cases where the default setting is to run the pure drop on the secure v2 node.

B

See the two keywords: secret, v1 and v2 are mutually exclusive, so you can either have a c or b one node or ac or v2 node and in the examples shown on this slide, if you specify both the keywords you hear v1 and see here, b2 there's going to be a conflict, but uh what these scripts are going to do is they're going to accept the last keyword, which, in example, one is secret v2 the job will be run on a secret v2 node and in example, 2.

B

The job will be run on the secret b1 node.

B

The other option would have been to fill saying that book. U was going to be specified simultaneously, but I have chosen the order of the order in which keywords are specified to still accept them and run the jobs. Well,.

A

In general, what semantics did you want with that comma operator in the uh between the parentheses? Is it like an, and so if you had, if it wasn't c group, if it was, I think you're going to get to that eventually in in this presentation, if you had other settings that you wanted to apply, are they all ands? Were they ors or neither.

B

I think other settings don't depend on each other, so they're still commas. At this point. All settings are just separated by a comma.

A

um Okay, maybe I'll wait until you go through the rest of the referral circuit to that again: okay,.

B

B

So in this uh over here, in order to specify additional configure options, the syntax specified in this slide can be used. Configure colon and the options to append will be in could.

B

The examples over here are adding additional c flags and c plus plus flags, and in this case I think the dump debug is a debug flag, which should enable more uh diagnostic output for in the port library. And this may be helpful in cases where we want to better understand failures within the pr job.

A

Sorry does the configure option only apply to cmake, then.

B

Yep, it's cmake specific or I.

A

B

It will also apply to the previous build system previous configure system, which was just a make.

A

Okay, I was just one because I was wondering why don't you just call it cmake instead of configure, because I think we still have.

B

The old system, which is not cmake.

A

B

A

That's good yeah. We want to replicate that, but uh yeah they're.

B

Still around, if that gets removed, then I can just make rename it to cmake or if that's the proposal I can just do it now, and the old one can be removed in the future.

B

A

With you doing it now, yep works for me. I can update appear then. Okay, unless anybody else subjects.

C

A

All right thanks.

B

And the next one is to specify uh more options to the make command, so the syntax is compile colon and similar to the configure previous feature. uh You specify all args in codes.

B

The examples- and this will tell make to run on eight cpus or that dash j8, will tell the make to run on eight cpus and dash d. I think, enables debug output for make.

B

Over here in this one, you can specify more options to be appended to the test command and the examples over here uh dash r protest will only run the for test suite and in the second example dash v will, I think, enable verbose output, more verbose, output for the tests, another useful feature to better debug failures and the last one is to specify environment variables.

B

The syntax is env colon and the extra environment variables are in codes, but if there are multiple environment variables, they need to be separated via a comma.

B

And in this case, uh in the example, I'm setting the g test filter environment variable which will allow us to filter either just one specific test or exclude specific tests. So I think the environment variable allows us to either exclude or.

B

Run specific tests so.

B

This is a more complicated example in this one. We are running all in addition to that. We are specifying custom options for the x linux job, which will mean that the x linux job will use the options a provided, whereas all other pr builds will use their default options.

B

And in this example,.

B

Is a derivation of the previous one x linux? Pr job will still use option. This options dash a but in this case all of the p air bills will use options underscore b.

B

Now, if multiple options are provided for apr or if multiple options are specified for a platform specification for in this example, I think x, linux and linux, x, 86-64 are aliases and two sets of options are two different sets of options are provided for the same platform. Specification uh options dash underscore a and options underscore b.

B

In this case, I think the options that are provided at the end will be in first, so options underscore b will be used. Various options underscore a will be ignored and the same analogy will apply to the scenario. If you have n sets of options for specified for a single platform specification.

B

uh Here's another example in this example, we are, I think, enforcing obvious. All new features are being used, so you could read to don't run it on a secret. V2 node run, try to run it on docker, and then we are also specifying uh test options.

B

uh An environment variable is being specified and then we are also passing extra compilation options options, but we are running it on a zos specification. Zero has neither support c groups nor docker, so this job will fail.

B

So it ultimately will depend upon the user to provide the right set of options for the job to succeed.

B

Pr 6525 is where all this work was done, and it also points to other issues from. I think where the features were proposed.

B

So if any feedback needs to be given, it can go in the sphere, and I think with this I conclude my presentation.

A

Okay, that's thanks a lot of the interesting and I think, useful extensions to the uh to the triggers um there. um How? How did you test these various options? Did you just jenkins build like crazy on this pr or yeah.

B

So yeah there will be so most of the testing was done in the sphere and you will see different variations of the new syntax being tested over here.

B

And I think adam also did some testing in a non-pr job to make sure promotion happens correctly.

C

Yeah, the other one I did just run to test. um If you run a docker command when there's no docker, you just get the machine error of docker not found. So we don't really.

B

C

That, ahead of time,.

B

I think it can, but we will need, I think, for that we will need either labels or.

C

Yeah, like you, should run, if you want option docker, you should have a label on all the machines that have docker really, but I don't think we have that at this point.

B

We don't have a label completely at this point, so.

C

B

It's going to implicitly fail, but it can be a future to do to further.

C

Think that's nice.

C

Now, if that machine did happen to have docker, um but it's still on windows, which we don't have a windows container for, I guess we would get the default ubuntu container, which should work on windows machine, but then you're not really getting what you want. So we might want to understand what would happen there.

C

Maybe just some further testing.

B

Yeah but uh question for daryl: do we do we want to support windows containers also.

A

I don't think we want to preclude it.

A

We may not support it right away.

B

If that's the case, I think just by adding a docker file for each supported platform should be sufficient in this case.

C

So right now we're only supporting.

B

C

And it only builds up 120, so there's the script is kind of hard coded for that right now.

C

So if we did have a one windows machine without had docker on it, you would then test inside that container and you'd actually be testing xlnx, not windows, so that might be a problem.

C

So we might want to adjust the code to help prevent it ever for scenarios, something like that.

A

Okay, any other questions for bev neat.

D

uh Hi bethany um just the general question um to run like these triggers. um Is it limited to only committers or can anyone that opens a pr? I guess run these triggers.

B

I think there is a white list which allows non-committers, which gives non-committers permission to run peer bills, and I I don't know, what's the criteria to request for permission.

A

Yeah, I think yeah generally, I mean all well, all committers are on the list um and and generally, if you've done, um you know a couple of pr's and have you know demonstrated that you know there you there's no ill will toward the community. Then um we can. We can add you to the oil list.

A

um So there's no, um it's not a strenuous criteria to be added to that list. It's if it's going to be much more convenient to add you to the list than for you to continue to find someone to launch jobs on your behalf, then um you can be at it. So.

A

So, for example, salmon. I know you've done a few things, so I can certainly add you to the list.

D

All right, thanks yeah, I just was just wondering the process of that, because I know I tried uh running something and it didn't work, so I wasn't sure but yeah I'll reach out thanks.

A

Other questions.

A

If not so, I'm going to spend some time going through the the pr I haven't uh looked at all the all the groovy code in there, um yet I know adam you've approved it, um but I just want to take uh take a pass through that myself.

A

um If anybody's got further comments, please drop them in that pull request.

A

Okay, if there's no further discussion on this, I I think we can. uh We can move on, and uh this is the only uh talk that we had today, the only topic for today. So um I have created the um the issue for the next uh meeting on july. The 7th um not sure if that meeting is going to go ahead or not just because it's uh it's falling in a holiday period and a lot of people will be away so that meeting may not actually happen, but by all means.

A

If you have a topic you want to bring up, please propose it there and if there's nothing else, I think we can adjourn for the day. So thank you very much babneet for the interesting talk and we'll talk again in a couple weeks thanks. Everyone.

B