Wasm-opt can take way too long

[sbomer]

I'm seeing this job time out during the "Build product" step, in a way that doesn't get reported to GitHub. On GitHub it looks like the job is still running forever.

Hit in multiple PRs, for example:

• Allow trimming FeatureGuard and FeatureSwitchDefinition attributes #100263
• [mono][wasm] Force interpreter to initialize the pointers #100288

Build Information

Build: https://dev.azure.com/dnceng-public/public/_build/results?buildId=622511&view=logs&jobId=63c2d0c8-fec2-5788-81c8-f3ac95e8841f
Build error leg or test failing: browser-wasm linux Release LibraryTests Build product

Error Message

Fill the error message using step by step known issues guidance.

{
  "ErrorMessage": "Agent was purged, cancelling the pipeline",
  "BuildRetry": false,
  "ExcludeConsoleLog": false
}

Known issue validation

Build: 🔎 https://dev.azure.com/dnceng-public/public/_build/results?buildId=622511
Error message validated: [Agent was purged, cancelling the pipeline]
Result validation: ❌ Known issue did not match with the provided build.
Validation performed at: 4/1/2024 10:15:39 PM UTC

Report

Summary

24-Hour Hit Count	7-Day Hit Count	1-Month Count
0	0	0

[agocke]

@sbomer Looks like this didn't work: ❌ Known issue did not match with the provided build.

I can't find the string in the test log.

[vcsjones]

Based on the logs in one of my PRs: https://helixre107v0xdcypoyl9e7f.blob.core.windows.net/dotnet-runtime-refs-pull-100419-merge-91eb0ec8650642f7b5/WasmTestOnBrowser-System.Runtime.InteropServices.JavaScript.Tests/1/console.cb20659c.log?helixlogtype=result

Maybe fail: Tests timed out. Killing driver service pid is a better ErrorMessage.

[vcsjones]

Actually I think this is a duplicate of #99888. Build analysis correctly picked that one up for my PR.

[sbomer]

This is not about the test failure in "browser-wasm linux Release LibraryTests_Threading" (build analysis identified that as #99888), but about the product build failure in "browser-wasm linux Release LibraryTests":

[lewing]

Given the way it is failing this is probably an infrastructure issue

[riarenas]

https://helix.dot.net/BuildAnalysis/SearchTimeline?error=Agent%20was%20purged,%20cancelling%20the%20pipeline&dateType=Day(s)&dateValue=7&pageNumber=3

Seems to show that this is a runtime specific issue. It seems like instances are consistently happening in the wasm legs (along with catching a few other stray disconnects)

(Maybe you can try to change the error message to Agent was purged, cancelling the pipeline? but I'm not sure if that will be picked up by the known issues infra as it's only being surfaced as a warning)

[sbomer]

Thanks for the suggestion, didn't seem to work. :(

[kg]

I'll note that when I watched the stdout of the build step for one of my PRs, it was building, even after 2 hours. So it seems like the builds for this lane are extremely slow. For comparison, the windows equivalent lane seems to build in around 40 minutes.

[pavelsavara]

Actually I think this is a duplicate of #99888. Build analysis correctly picked that one up for my PR.

#99888 is during test run on helix not related to this.

This is build/agent issue. Probably infra.
AzDo kills the agent and deletes the log file!

Like this

This is before the agent was killed, something is terribly slow

The running step is CPU+memory heavy, maybe the agent is also swapping to disk.

This problem is blocking me

[pavelsavara]

How would it look like if the agent failed with OOM ?
We had similar issues on Helix last year, but Ankit got valid log files about it See #51961.