Electrolysis/Meetings/2016-09-29

« previous week | index | next week »

mconley

• bug 1300411 - user reports excessive load spinning when switching tabs with e10s in 48.0.2 release
  • Posted a number of GC logs from a user that I'm in touch with over e-mail. In at least one, the user was OOM'ing.
  • Learning how to use analysis.telemetry.mozilla.org
    • Took a snapshot of pings sent in on September 23rd, and filtered down to clients that should have e10s enabled by default
    • Of these clients, a little over a quarter of them have seen at least one spinner of any length. We'll call these the "spinning clients"
    • Of the spinning clients:
      • ~82% of them saw a spinner of 2s or less ("low" bucket)
      • ~16% of them saw a spinner between 2s and 5s ("mid" bucket)
      • ~16% of them saw a spinner greater than 5s in length ("high" bucket)
    • Note that there is, of course, overlap between the clients in those buckets
      • ~2% of the spinning clients saw spinners in both the "mid" and "high" buckets
      • ~6% of the spinning clients saw spinners in both the "low" and "high" buckets
      • ~7% of the spinning clients saw spinners in both the "low" and "mid" buckets.
  • I think the only way we're going to grind these numbers down is by knowing exactly what the content process is doing so we can prioritize
  • Getting ramped up on bug 1303077
• bug 1241459 - [e10s] A crashed background tab does not give option to submit
  • Have a behaviour spec, and a mostly complete plan on how to proceed
  • If a content process associated with any foreground tab crashes, same as current behaviour
  • If a content process associated with **only** background tabs crashes, show about:tabcrashed for the first of that set that the user selects, and the rest restore on demand.
  • If the user opts in to always sending backlogged crash reports, then if only background tabs crash, then all background tabs go into restore on demand state.

chutten

Helped out :mconley with some analysis stuff. We have a fairly good understanding of how many people are effected, but not how badly those people are effected. For instance: we know how many people are seeing spinners of various lengths... but how frequently are they seeing them?

Beta Release

felipe

• bug 1304164 - Disable russian users from e10s
• bug 1301131 - Disable e10s if the user sees the big tab spinner too much

aklotz

• bug 1304449 - Rework MSAA child id generation for e10s - Main bug that fixes all the perf and stability issues - Still 2 patches needing review
• Re-enabled tests that were disabled when a11y+e10s first landed (these are ones that were also failing on non-e10s mode)
• Starting to look into Windows XP support

shell

• CPOW disabling - investigating timing and extent of what that is
• curious if spinner disabling bug will work for add-ons experiment as well.
  • are we concerned about it beyond e10s populations

Roundtable

• tracking-e10s is no more