Issue metadata
Sign in to add a comment
|
VR DON canceled test very flaky |
||||||||||||||||||||||||||||||
Issue descriptionThe WebXrVrTransitionTest#testPresentationPromiseRejectedIfDonCanceled tests started being extremely flaky as of ~2 weeks ago. A bisect points to https://chromium-review.googlesource.com/c/1320191 as the culprit. Looking at the post-failure screenshot and logcat, it appears the issue is that we're getting a renderer crash caused by this DCHECK https://cs.chromium.org/chromium/src/third_party/blink/renderer/platform/graphics/gpu/drawing_buffer.cc?q=is_hidden+file:drawing_buffer.cc&sq=package:chromium&dr=C&l=329. Unfortunately, I'm unable to get an actual stack trace (I have yet to find a way of getting stacks for renderer crashes), but judging by the logcat output, this is occurring as Chrome is resumed after the DON flow (VR entry screens) are canceled. danakj@, any easy fixes you see before I revert this?
,
Nov 26
,
Nov 26
Oh, can you symbolize that stack trace? Do the bots not do that?
,
Nov 26
I don't think so :( I have yet to find a way to symbolize renderer process crashes, locally or on the bots. The bots and the local stack tool work for browser crashes, but seem to ignore renderer crashes.
,
Nov 26
WebGL is_hidden_ comes from the page visibility, aka RenderViewImpl::OnPageWasShown/Hidden(). DrawingBuffer::PrepareTransferableResourceInternal() comes from a couple places but is most /likely/ coming from TextureLayer::Update here. Which runs when the compositor is visible, which is controlled from RenderWidget's visibility aka RenderWidget::OnWasShown/Hidden(). Having 2 different signals for this is a known problem... I'm filing a bug for that specifically.
,
Nov 26
Based on my investigation on 908582, I'd say this has always been a problem just not for the main frame of the webpage, and you could comment out that DCHECK with a TODO over to 908582 if it's causing you trouble. Otherwise I will try to fix this holistically.
,
Nov 30
,
Nov 30
FWIW There will be a lot more work to get visibility more consistent, likely some timing changed for whatever other reason and its landing luckily on the less flaky side right now.
,
Nov 30
We may as well dupe over there though, if we're not going to disable the DCHECK in the meantime. |
|||||||||||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||||||||||
Comment 1 by danakj@chromium.org
, Nov 26