TPEPerformance/Meetings/2016-08-29

« previous week | index | next week »

OKR confidence

1. O: Make Firefox more stable [current crash rates]
   1. KR: 1.2 nightly browser crashes per 100 ADI (per-month median)
      • Current: 2.46 -> 3.21 -> 2.38
   2. KR: 0.3 nightly startup crashes per 100 ADI (per-month median)
      • Current: 0.81 -> 0.82 -> 0.72
   3. KR: 2.0 nightly content crashes per 100 ADI (per-month median)
      • Current: 14.38 -> 8.70 -> 8.25
2. O: Make crashes easier to find & fix
   1. KR: 90% new crashes have associated bugs opened and tracked
      • Current: no metric
3. O: Improve Google document performance on Firefox
   1. O: Understand Google document performance issues on Firefox
      1. KR: Reduced the problem space to a finite areas that we can improve
         • Current: GC slowness, Inline script blocking content, Vsync/IPC, Redundent tasks

Health Checks

1. Test code coverage (minus crash/no return code path)
   • N/A
2. Team mental health
   • Good
3. ADI / DAU / MAU
   • N/A

Previous Week

This report is automatically generated by https://github.com/kanru/standups_weekly

kanru

• Previous
  • [x] StructuredClone/BufferList memory overhead
  • [x] Crash rate data
  • [x] PBrowser::SendDestroy crash bug
  • [x] Predict script running time and run rendering before run script, ask support
• Finished 8-22 nightly crash triaging
• bug 1296079 RESOLVED::FIXED Disable dump() calls in dom/browser-element
  • Review bug 1296079
• bug 1264642 RESOLVED::FIXED Reduce the contiguous address space needed for StructuredClone serialization
  • Land bug 1264642 again
• bug 1296003 RESOLVED::FIXED [asan] Intermittent test_screenshot.py Content.test_viewport_after_scroll | IOError: Process killed because the connection to Marionette server is lost. Check gecko.log for errors (Reason: Connection timed out after 360s) due to OOM
  • Bug 1296003 found how to switch to smaller ec2 instance
• bug 1297030 RESOLVED::FIXED AnnotateLSBRelease undefined when compile with --disable-sandbox
  • Landed bug 1297030
• bug 1297314 RESOLVED::FIXED Crash in mozilla::ipc::FatalError | mozilla::dom::PContentParent::Write from ContentParent::RecvKeywordToURI()
  • Review patch for bug 1297314
• bug 1296756 NEW:: Record number of open file descriptors in crash report
  • Bug 1296756 requested review
  • Looking at bug 1296756
  • Work on bug 1296756
• bug 1051567 NEW:: [e10s] Crash in [@ mozilla::ipc::MessageChannel::OnChannelErrorFromLink]
  • Looked at bug 1051567
• bug 1297276 RESOLVED::FIXED Rename mfbt/unused.h to mfbt/Unused.h for consistency
  • Landed
• bug 1297981 NEW:: Delete BufferList::FlattenBytes and Pickle::FlattenBytes
  • Requested review, got r+
• bug 1298325 NEW:: Hang in js::CancelOffThreadIonCompile from js::gc::GCRuntime::minorGC
  • Filed bug 1298325
• bug 1269036 RESOLVED::FIXED crash in mozalloc_abort | NS_DebugBreak | mozilla::ipc::LogicError | mozilla::dom::PBrowser::Transition from ABORT: __delete__()d actor
  • Landed
• bug 1295103 RESOLVED::FIXED Use MOZ_MUST_USE in OriginAttributes
  • Address review comments for bug 1295103
  • Landed
• bug 1297326 RESOLVED::FIXED Crash in libc-2.19.so@0x35067 from GLXVsyncSource::GLXDisplay::RunVsync
  • Filed bug 1297326
• bug 1296125 RESOLVED::FIXED position.timestamp is microseconds in native location providers
  • Review patch for bug 1296125

cervantes

• Previous
  • [x] Heap in minidump patch
  • [x] GMP shutdown crash permfail
  • [ ] Coverity or MOZ_MUST_USE
  • [x] App verifier
• Running mochitest locally with app verifier enabled
• bug 1285531 NEW:: High number of jemalloc crashes in webgl
  • Helped in debugging bug 1285531
  • Looking at bug 1285531
• bug 1286802 NEW:: Add some heap areas to the minidump
  • Revising the patch for bug 1286802
• bug 1295544 RESOLVED::FIXED Fix typo MOZ_CRASH_REPORTER in ProtocolUtils.cpp
  • Landed bug 1295544
• bug 1279612 NEW:: Near permafailing mda tests on Win7 debug on beta in test_webvtt_disabled.html | application crashed [@ mozalloc_abort(char const * const)]
  • Bug 1279612 - found the root cause and commented about the fix
  • Bug 1279612 - testing a patch on try
  • Investigating the use of nsIAsyncShutdownBlocker for bug 1279612

ting

• Previous
  • [x] Fragmentation
  • [x] ReleaseData
• Checked 0821 nightly crashes
• bug 1297314 RESOLVED::FIXED Crash in mozilla::ipc::FatalError | mozilla::dom::PContentParent::Write from ContentParent::RecvKeywordToURI()
  • Bug 1297314 submitted a patch
  • Bug 1297314 try looks good, about landing
  • Checking bug 1297314 for where could set OptionalInputStreamParams with wrong type
• bug 1293501 NEW:: Add debug annotation/code for the crash in ReleaseData()
  • Bug 1293501 added some more crash annotations, asked for review
• bug 1145613 NEW:: Crash in [@ ReleaseData ] & [@ ReleaseData(void*, unsigned int)]
  • Bug 1145613 added some other similar crashes with different signatures
  • Bug 1145613 checked the crash annotations, am double checking the code around nsHttp::Host
• bug 965936 NEW:: [meta] Virtual address space fragmentation
  • Bug 965936 pushed the work of detours to github
  • Bug 965936 succeed in hooking NtAllocateVirtualMemory
  • Bug 965936 trying to debug why hook on NtAllocateVirtualMemory causes crash

wcpan

• Previous
  • [x] MOZ_MUST_USE in uriloader
  • [x] userContent.css
• bug 1046166 REOPENED:: [e10s] userContent.css does not work
  • Bug 1046166 - force async-load, crashes in DevTools.
• bug 1293212 REOPENED:: Use MOZ_MUST_USE in uriloader/
  • Bug 1293212 - landed.

gregweng

• For the thread delay debugging point make browser too slow to debug; need to print all related things out
• Found JSShell will start off threads but never stop them. Need a new test method for the 500ms syscall delay in GDocs cases
• Found possible bottleneck of off thread compilation in the GDocs case; try to probe it
• Studying the core Ion compiler worker: Ion builder, for adding measuring lines
• Timing Ion builder tasks, but it's too fast to be the bottleneck
• Try to measure GDoc case compilation time; studying Ion code
• Trying to watch the frame and context status when the thread get cancelled for the 500ms delay
• Windows has no perf gap between e10s and non-e10s; try to collect more data and analyse it
• Writing a plain logger to measure the whole compilation time, because the TraceLogger is broken, and it's to costly to fix it for this case

thinker

• Previous
  • [ ] [gfx] reval.js
  • [ ] [gdoc] Categorize need, un-needed tasks
• bug 1271983 NEW:: Animated SVG image is blurred in FF 46
  • Study bug 1271983 and provide a solution.

This Week Priorities

kanru

• Shutdown kill
• Mac run-out-of-fds
• Crash-data from bmugzilla
• Various crash bugs

thinker

• reveal.js
• gdoc needed, undeed tasks on main thread

cervantes

• minidump patches
• coveraity

wcpan

• userContent.css

ting

• ReleaseData
• Fragmantation

gweng

• Log compilaition time

Incoming

• Sep: Some team members plan to take PTOs for the Mid-Autumn Festival
• Sep: Some team members are traveling to MV in Sep 12--Sep16
• Nov: WebSummit/CodeSummit