Performance/MemShrink/Meetings/2011-07-05

From MozillaWiki
Jump to: navigation, search

Announcements

Bugs

Make sure P1 bugs are moving

Bug list - 16 bugs

  • Most P1 bugs in good shape :)
    • bug 592833 had been waiting for today's aurora merge
    • bug 414946 (jemalloc-on-mac) had been waiting for today's aurora merge, and will likely land today!
  • bug 637931 demoted to P2: we agree it's "long term"
  • bug 651695 demoted to P2

Triage MemShrink bugs that don't have priorities marked

Bug list - 14 bugs

  • bug 668809 - we decided this is a P1 metabug
    • We considered splitting out "use diffbloatdump after doing some normal browsing" into a separate information-gathering bug
  • bug 668871 - turned out to be multiple bugs, one involving firebug and one with some kind of timing issue.
    • Let's file a separate bug for firebug (which might turn out to be a dup of bug 669096)
    • Do compartments map 1:1 with anything that appears in about:bloat? (Fuzzing uses about:bloat)
      • Should map 1:1 xpc::CompartmentPrivate ... which isn't listed in about:bloat
  • bug 669120 - controversial!
    • heap-minimize is a big hammer. clears user-visible things like bfcache. (but perhaps we could exclude bfcache for the most-recently-active tab or something like that)
    • maybe only if firefox is minimized and remains minimized for over a minute

Make sure P2 bugs are moving

Bug list - 44 bugs

  • We ran out of time before getting to the P2 bugs.

Tests

Endurance testing

Discussion about bug 663416

  • QA/Mozmill Test Automation/Endurance Tests
  • http://mozmill-crowd.brasstacks.mozilla.com/#/endurance/reports
  • Currently consists of 6 tests, each of which have at least two checkpoints at which they measure memory use (explicit and resident for Firefox 6+, mapped and allocated for Firefox 4-5)
  • This framework can also be used for ad-hoc tests (such as tests that load a public web site over and over again)
  • We don't have a great idea of which browser features (or which web sites) are important to test in this manner
  • Intending to ramp up tests in Q3, and align with Memshrink and Telemetry
  • Can run tests with additional addons installed
  • Intending on comparing average results between builds from same hardware

Fuzz testing

  • Jesse is playing with a DOM fuzzer mode that detects more types of leaks. (After doing random stuff: navigate to blank, GC+CC+MP, count windows & documents using about:bloat.)
  • Found 4 bugs just looking for extra documents & windows
  • What should I have the fuzzer look for next?
    • nsGenericElement?
      • nsGenericHTMLElement would have the advantage of having less chrome churn
    • Everything that appears in about:bloat?
      • Likely to be noisy
    • Everything that appears in about:memory?
      • Could enumerate the reporters, like about:memory does
      • Some of the things are computed by about memory js, which doesn't provide an API of its own