Bug 173505

Summary: [MSE] Seeking or entering fullscreen can cause extreme CPU usage
Product: WebKit Reporter: Jer Noble <jer.noble>
Component: New BugsAssignee: Jer Noble <jer.noble>
Status: RESOLVED FIXED    
Severity: Normal CC: commit-queue, eric.carlson, thorton, webkit-bug-importer
Priority: P2 Keywords: InRadar
Version: WebKit Nightly Build   
Hardware: Unspecified   
OS: Unspecified   
Bug Depends on: 173515    
Bug Blocks:    
Attachments:
Description Flags
Patch
none
Patch for landing none

Jer Noble
Reported 2017-06-16 19:06:49 PDT
[MSE] Seeking or entering fullscreen can cause extreme CPU usage
Attachments
Patch (2.67 KB, patch)
2017-06-16 19:11 PDT, Jer Noble 		no flags
DetailsFormatted DiffDiff
Patch for landing (5.01 KB, patch)
2017-06-17 23:36 PDT, Jer Noble 		no flags
DetailsFormatted DiffDiff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.
Jer Noble
Comment 1 2017-06-16 19:11:40 PDT
<rdar://problem/32623694>
Jer Noble
Comment 2 2017-06-16 19:11:45 PDT
Created attachment 313175 [details] Patch
Tim Horton
Comment 3 2017-06-16 19:13:41 PDT
Comment on attachment 313175 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=313175&action=review > Source/WebCore/ChangeLog:11 > + performance regression in the normal playback path, where WebKit will effective effective append?
Jer Noble
Comment 4 2017-06-16 19:20:15 PDT
Committed r218438: <http://trac.webkit.org/changeset/218438>
WebKit Commit Bot
Comment 5 2017-06-17 12:47:43 PDT
Re-opened since this is blocked by bug 173515
Jer Noble
Comment 6 2017-06-17 23:00:26 PDT
I found the underlying issue behind the test failures, and it has nothing to do with this patch, except in that this patch causes it to happen much more frequently. When a client specifies a timestamp offset, the SourceBuffer code stores the current presentation time as a float (because a MediaTime + float = float). But the sample itself store it as a MediaTime. Generally speaking, we use the sample's presentation time as the key to our storage, but in one instance (the decodeQueue) we are using the local value of presentation time (the float). This leads to a mismatch later when it comes time to remove those samples from the decode queue, so the samples are left in the decode queue, which causes a decode error, which breaks this test. (BTW, I'm pretty sure that this is the cause of some existing flakiness of some of these tests.)
Jer Noble
Comment 7 2017-06-17 23:36:32 PDT
Created attachment 313224 [details] Patch for landing
WebKit Commit Bot
Comment 8 2017-06-18 08:55:59 PDT
Comment on attachment 313224 [details] Patch for landing Clearing flags on attachment: 313224 Committed r218463: <http://trac.webkit.org/changeset/218463>
WebKit Commit Bot
Comment 9 2017-06-18 08:56:01 PDT
All reviewed patches have been landed. Closing bug.
Note You need to log in before you can comment on or make changes to this bug.