[Xeicker]

Bug: 574630499

Stack: #5581 → #5582 → #5583 (this). See #5581 for context.

Description

blame and impact are TrustedTasks with no utask split, so they can't run on Swarming, and testcases would wait on them forever (stuck blame_pending, never "complete" for triage). For Swarming jobs they're no longer created (task creation, impact cron, UI redo) and critical_tasks_completed no longer requires impact.

Testing

• Unit tests: each skip is tested with the flag on and the same scenario with the flag off, so the skip is shown to come from Swarming; redo tests also check that no testcase state changes for skipped tasks. Also covers variant creation for untrusted testcases from [Swarming] Bypass host OS/CPU checks and skip TrustedTasks on tworkers #5581.
• E2E: covered by the whole-stack run described in [Swarming] Bypass host OS/CPU checks and skip TrustedTasks on tworkers #5581.

[g-ortuno]

The description only mentions a problem, can you also mention what this PR does to fix the problem?

[g-ortuno]

Ah, I misunderstood the description! I see the solution now.

[IvanBM18]

IIUC we still want impact/blame tasks to be executed for the swarming jobs, but the limit we are facing is that we have logic that makes it so all tasks for a swarming job must pass down trough a swarming bot, or a queue that is only pulled from by bots that route things to swarming bots, e.g. jobs-android-emulator.

So my question here is, What stops us from running this tasks in a regular tworker? Is it the routing logic? or is it because this tasks need to be executed on a specific kind of bot?

If its a routing issue, Can you add TODO comments pointing to a bug that shall make it possible for us to execute impact/blame tasks for swarming jobs?

These bugs are already created under the feature b/469405415

[Xeicker]

That's a good point Ivan. I think I kind of misunderstood the original routing bug.
Added the TODO and created the bug to track that gap

[IvanBM18]

LGTM