For example, https://treeherder.mozilla.org/#/jobs?repo=try&revision=dbc2be21a53a963f6c1112fc5ded13e46dc7fd90.

This is likely related to bug 1589214.

Let's take for example https://treeherder.mozilla.org/#/jobs?repo=try&tier=1%2C2%2C3&revision=7fcabdcda58dea2bbc86de743c97208107eb3d1d.

The try_task_config.json is good: https://hg.mozilla.org/try/rev/7fcabdcda58dea2bbc86de743c97208107eb3d1d.

The set of tasks generated by the Gecko Decision Task is https://firefoxci.taskcluster-artifacts.net/UsDzfMR2T-2wzQT5uNZ3Yw/0/public/target-tasks.json.

The build is probably triggered because it is a dependency of the node debugger tests (bug 1589214), even if the node debugger tests then don't end up getting scheduled (as you can see on Treeherder or in the Taskcluster group view).

So probably we are correctly optimizing out the node debugger tests task, but not optimizing out its dependencies.

who can work on this?

We are working on this issue right now, most likely we are going to remove the two change-sets that are the culprits here:
https://hg.mozilla.org/mozilla-central/rev/2cbc4579e85afc67a37e3273ef723bd88189a683
https://hg.mozilla.org/mozilla-central/rev/50a04d79f020f4f9d760cc30702ded35f07fc726

I understand this is important, and I suspect that reverting these patches is a reasonable way forward, but I'd like to talk to Jason Laster, whom I just CCed, a bit about strategy first, if that's ok. I'll post more details here in the next few hours.

The "right" way to fix this bug is actually characterized in bug 1579162, I think. That said, doing the expedient thing proposed by Andi may well be the best thing to do at the moment. More in a bit...

I think it's fine to revert the first of those two commits; people are not using the results of those builds much yet.

The second of the two is devtools based, and I can't speak for those.

The two commits have been reverted:
https://hg.mozilla.org/mozilla-central/rev/507ec3d5394a
https://hg.mozilla.org/mozilla-central/rev/4ca097f53643

Now m-c should be clean and every patch that is going to be based or re-based on m-c should no longer encounter the PGO build issue.

you can probably close this bug then :)