1907224 - Uncaptured WebGPU error without device target: Not enough memory left to request device

Status: RESOLVED FIXED

(Core :: Graphics: WebGPU, defect, P2)

Description

[Tyson Smith [:tsmith]]

Attachment: testcase.html

I was adjusting out OOM bucketing and came across this issue. It is happening frequently but only on Windows. This goes back to at least 20240514-d1f40cf63952. I have not been able to reproduce it on any of my local machines but the test case is pretty simple.

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

This bug prevents fuzzing from making progress; however, it has low severity. It is important for fuzz blocker bugs to be addressed in a timely manner (see here why?).
:ErichDonGubler, could you consider increasing the severity?

For more information, please visit BugBot documentation.

Comment 2

[Erich Gubler [:ErichDonGubler] (he/him)]

:tsmith, when you encounter this in fuzzing environments, is this an OOM on the GPU side, or the CPU side?

I think we can resolve this specific snippet OOM'ing by enforcing the standard constraint that an adapter can only have a single device returned from it. However, this would merely punt the OOM issue from requesting devices in a loop to requesting adapters in a loop, once the fuzzer attempted it.

:tsmith, we've had some recent work to reclaim resources better in WGPU upstream by :teoxoy. Has the rate of errors changed? I'm not expecting it to, but I'm interested in that data point for ourselves. 🙂

Comment 3

[Erich Gubler [:ErichDonGubler] (he/him)]

Note: Will not promote severity beyond S3, as WebGPU is still confined to Nightly.

Comment 4

[Jim Blandy :jimb]

On Matrix, Mr. Tyson Smith mused:

jimb: 20240716-2ed6b77c66d3 was the last build that was hitting this frequently then it dropped off to only a few a day
oh maybe it stopped completely

Comment 5

[Tyson Smith [:tsmith]]

(In reply to Erich Gubler [:ErichDonGubler] from comment #2)

:tsmith, when you encounter this in fuzzing environments, is this an OOM on the GPU side, or the CPU side?

I wasn't ever able to reproduce locally, sorry.

:tsmith, we've had some recent work to reclaim resources better in WGPU upstream by :teoxoy. Has the rate of errors changed? I'm not expecting it to, but I'm interested in that data point for ourselves. 🙂

m-c 20240716-2ed6b77c66d3 was the last build we saw with a high report rate of this issue, however I do still see some results. I will schedule reduction of a new report.

Comment 6

[Tyson Smith [:tsmith]]

This issue is no longer being reported by fuzzers. The last report is from m-c 20240913-b0716d2d5fef.