Meeting 2020-07-02

Round table:

Next on triage duty: Dragana

Valentin

[landed]

1618271 - DNS leaks about browser.dns API

1641222 - Follow CNAME/AliasForm chains

1649401 - Minor TRR code cleanup

1649417 - Don't follow CNAME chains when not necessary

[working on]

1649127 - Make sure we only set AddrHostRecord::mTRRUsed = true when we actually send the request

1649143 - Record telemetry with reason we fell back to Do53

[triage]

PTO for 3 weeks - still working 20% with Tania and urgent issues.

Tania

Continue working on xpidl bindings for structured-headers

Junior

• Land
  • Bug 1633935 - Allow sending onStartRequest via pHttpBackgroundChannel
  • Bug 1648035 - Intermittent Assertion failure: mChannelChild (no channel child in RecvOnStatus), at /builds/worker/checkouts/gecko/netwerk/protocol/http/HttpBackgroundChannelChild.cpp:160
  • Bug 1648712 - Intermittent dom/base/test/chrome/test_bug884693.xhtml | Got expected message count - got 1, expected +0
  • Bug 1578230 - Crash [@ mozilla::net::SpdyConnectTransaction::MapStreamToHttpConnection] with HTTP2 Proxy
  • dup -Bug 1647096 - GeckoProfiler.Markers gtest crashes after changing tests to use socket process
• Ongoing
  • Bug 1649848 - [meta] make test_ext_webRequest_filterResponseData.js stable
  • other regressions
  • cookie and permission via pBg

mayhemer

• investigated 1567863 - Firefox startup slow - window loads quickly, but first network call delayed, works normal after that
  • lot of ~8 seconds delays here and there (I/O, script execution blocking main thread)
• took over and tuning 1605305 - Origin header field is not set to HTTP request in cases where it is required
  • some tests need care
• investigating two sec bugs
• fixed and waiting for field test 1609410 - 421 response code from icloud.com not handled well with a proxy
• fixed 1615121 - Intermittent GECKO(1119) | SUMMARY: ThreadSanitizer: data race /builds/worker/workspace/build/src/netwerk/cache2/CacheEntry.cpp:1436:33 in mozilla::net::CacheEntry::MetaDataReady()
• fixed 1632215 - Intermittent SUMMARY: ThreadSanitizer: data race /builds/worker/checkouts/gecko/netwerk/protocol/http/nsHttpChannel.cpp:8439:7 in mozilla::net::nsHttpChannel::OnDataAvailable(nsIRequest*, nsIInputStream*, unsigned long, unsigned int)
• fixed 1649003 - Log JS stack on more places in nsHttpChannel and Child
• fixed 1648781 - Log timer events
• fixed 1642602 - Some disk cache values in about:cache are incorrect.
• duplicated 1560090 - High refresh rate monitor (>200Hz) slows performance
  • an idle dispatch (with 250ms timeout) on the critical path for all resource loads when filtering extensions are involved
• blog post about preload
• preload for workers (fetch/xhr) seems not be a priority
• plan for better network markers in "Waiting for socket thread" span
  • don't expect anything soonish
• ni, r

Andy

Neqo (3, 52, 53)(-3,+3,-4):

• [merged] Implement more qlog traces: CC and Recovery state
• [merged] client: Make H9 code read all available data on RecvStreamReadable
• [in-review] Add network.http.http3.qlog_dir pref to Firefox
• [approved] qlog: Refactor
• [approved] qlog: more-unique filenames
• QUIC planning mtg: Realization of the scope of work for testing needed for rollout. Prioritization of work vs other projects like Fission. Contacts and coordination with QA. Introduction of Milestone 1.5. Followup mtg early next week.

DoH:

• Wrote a draft doc to start talking about how to interact with OSes rolling out DoH support

Next:

• Write a policy options doc for dealing with ISPs who block canary
• Fix Neqo interop issues
• Confirm DoH canary abuse and take follow-up actions

Kershaw

Landed:

• Bug 1648189 - [socket process] Perma Assertion failure: tqParent, at /builds/worker/checkouts/gecko/netwerk/protocol/http/HttpTransactionParent.cpp:173
• Bug 1650068 - [socket process] reenable profiler tests
• Bug 1648008 - Unable to connect to Google first time using it through DuckDuckGo bang after clearing all history

Ongoing:

• Bug 1527256 - [socket process] make about:networking work
• Bug 1646358 - [socket process] Fix test failures regarding devtool tests
• Bug 1648996 - [socket process] devtools/client/netmonitor/test/browser_net_truncate-post-data.js failed with socket process enabled

Michal

• 1627910 Enable test netwerk/test/unit/test_http3
• 1649410 [socket-process] test_version_fallback() in netwerk/test/unit/test_http3.js fails
• neqo testing

Dragana

Finishing up PUSH for neqo,

Looking into how errors are propagated in necko: there are 3(4) layers and looking which layer should handle which error.

Looking at goodput and crosstraffic test result -> flow control need to be improved.

PR 769 Small changes

PR 770 Small fix 2

Bug 1642559 SPDY_REQUEST_PER_CONN_2 looks broken

Bug 1595269 HTTP/2.0 still displayed even if HTTP/3.0 is used

1648553 Sometimes page loads start hanging in all tabs with http3