Bug 131358 – [WK2][GTK] Split the 64bit bots in Build and Test

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Sergio Villar Senin]

Currently we have two 64-bit bots, one building debug and the other building release. Apart from that they both run all the tests. We need to have separate bots for building and testing in order to detect build and test breakages easily. With the current setup is quite difficult to know the status of the port.

[Carlos Alberto Lopez Perez]

Created attachment 231005 [details]
Patch

[Carlos Alberto Lopez Perez]

Patch attached

[Carlos Garcia Campos]

Comment on attachment 231005 [details]
Patch

Thanks for the patch!. Why are we bringing back the WK2 thing? There's no longer WebKit1 API in GTK+ port, so I think we can keep the names of the bots without the WK2 part.

[Carlos Garcia Campos]

Comment on attachment 231005 [details]
Patch

I think with the split we are missing the GObject DOM bindings API breaks step. We need to add the step RunGtkWebKitGObjectDOMBindingsAPIBreakTests to BuildAndTestFactory when port is gtk

[Carlos Garcia Campos]

(In reply to comment #4)
> (From update of attachment 231005 [details])
> I think with the split we are missing the GObject DOM bindings API breaks step. We need to add the step RunGtkWebKitGObjectDOMBindingsAPIBreakTests to BuildAndTestFactory when port is gtk

I meant to TestFactory

[Carlos Alberto Lopez Perez]

Created attachment 231060 [details]
Patch

[Carlos Alberto Lopez Perez]

Thanks for the comments. I have addressed them and uploaded a new version of the patch.

[Carlos Garcia Campos]

Comment on attachment 231060 [details]
Patch

ok, let's try. Thanks!

[WebKit Commit Bot]

Comment on attachment 231060 [details]
Patch

Clearing flags on attachment: 231060

Committed r168471: <http://trac.webkit.org/changeset/168471>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Carlos Alberto Lopez Perez]

This broke the master. Bug https://bugs.webkit.org/show_bug.cgi?id=132688 follows the issue

[Carlos Alberto Lopez Perez]

There was another problem with the original path (the builder names were not updated correctly on the scheduler list). Tracked here: https://bugs.webkit.org/show_bug.cgi?id=132699

[Csaba Osztrogonác]

(In reply to comment #0)
> Currently we have two 64-bit bots, one building debug and the other building release. Apart from that they both run all the tests. We need to have separate bots for building and testing in order to detect build and test breakages easily. With the current setup is quite difficult to know the status of the port.

Are you sure if it is efficient to split the builder and the tester for the
debug bot? Uploading gigabytes to the master by builder and then download
by the tester will take long time, maybe an hour per build if your bots
aren't very close to build.webkit.org.

[Carlos Alberto Lopez Perez]

(In reply to comment #13)
> 
> Are you sure if it is efficient to split the builder and the tester for the
> debug bot? Uploading gigabytes to the master by builder and then download
> by the tester will take long time, maybe an hour per build if your bots
> aren't very close to build.webkit.org.

We are planning to workaround this issue by wrapping the zip/unzip commands with a shell script that will copy the built product to the test buildslave directly (both machines are on the same LAN so it should be pretty fast).

And instead of uploading the built product, the script will upload to the buildmaster a fake zip file that contains only a txt with a link to the real build product.

This way we will avoid the problem with the slow upload/download.

Probably it will take some few days of try-error-fix until we fully stabilize this new setup.

Once the deployment is working as expected I plan to send a mail to the ML [e-mail] announcing this.

[Carlos Alberto Lopez Perez]

(In reply to comment #14)
> Once the deployment is working as expected I plan to send a mail to the ML [e-mail] announcing this.

BTW, https://lists.webkit.org/pipermail/webkit-gtk/2014-May/001909.html