[minutes] WebPerf Group Call 14 Jun 2017 [ was: Re: [06/14/17] webperf group call @ 10AM PST]

Hi,

The record of this week's call is available as:
https://www.w3.org/2017/06/14-webperf-minutes.html#item03

And text version as below --

WebPerf Group Call

14 Jun 2017

Agenda

See also: IRC log

Attendees

Present
igrigorik, shubhie, Nic, Yoav, Nathan, xiaoqian, Charles
Regrets
Chair
igrigorik
Scribe
igrigorik
Contents

Topics
f2f agenda
HR-Time open issues
feedback on PerformanceObserver + Timeline
Server Timing #14
Preload #19
Summary of Action Items
Summary of Resolutions

f2f agenda: 
https://docs.google.com/document/d/1fcWpKZhJdrmBlpYUB2wIx42svo9zqojoJMZg8aY7E7Y/edit#

some suggestions for topics at the bottom of the doc

HR-Time: prompt to unload test

xiaoqian: the test seems to work in FF, but does not yet work in Chrome
... need to try in Edge and Safari

<xiaoqian> https://www.w3.org/2017/04/webperf-test/unload_manual.html
re, worker reference: we'll use whatwg

Next up, feedback on PerformanceObserver + Timeline

https://github.com/w3c/hr-time/issues/32

https://github.com/w3c/hr-time/issues/46

NicJ: are perf observer and timeline buffers different?
... not clear on relationship between, say, RT buffer and performance 
timeline buffer?
... how does it compare to ServerTiming entries

we have some undefined behaviors: we defined new buffering in "global 
performance timeline", but haven't defined how the global buffer 
interacts with RT buffer, etc.

NicJ: scripts like boomerang are not guaranteed to load before end of 
onload
... we can use method queue as a workaround

second issue: dealing with parent iframe is easy, dealing with children 
(same-origin) iframes is harder

scribe: the challenge is you have to register the perf observer before 
onload of the iframe
... there are race conditions that are hard to deal with
... LongTasks has an interesting model where it bubbles up all 
child-iframe events as well
... one idea is to enable perf observer to do the same?
... we'd like to use PerfObserver more, but the iframes make it a 
challenge

yoav: having consistent buffering behavior is good for developers

AI's to followup on:

- define global perf timeline buffer: how does it interact?

- iframes are tricky with PerfObserver, can we do something here?

- debugging: bad thing happened, how can we support this case?

AI: nick to split last question into separate discussion

Server Timing: https://github.com/w3c/server-timing/issues/14

scribe: it's hard to line up RT and ST entries?

shubhie: it's not a unique problem.. e.g. inflight requests, slow frames 
and long tasks
... perhaps there is some generic solutions?

yoav: for dependencies, the idea I explored is a unique ID for each 
entry, so you can create relationships between entries
... similarly, one could use that to link other types
... we could have a generic mechanism?

Nathan: we have a problem where we can't match Fetch requests with 
resource timing

what's the argument for having RT and Server Timing separate?

Charles: perhaps in the future if you have a lot of different 
extensions.. it paves the way for separate APIs that are stitchable

Shubhie: we'll need this mechanism anyway

Charles: one worry with pointer idea is that these might not exist yet..

AI: discuss next week at F2F

NicJ: PerfObserver + async callback
... you pass in buffered: true, there is no signal that the are no 
entries
... https://github.com/w3c/performance-timeline/issues/74

^ would address the case

Preload

https://github.com/w3c/preload/issues/99

yoav: fastly presented their h2 push at HTTP workshop in London.. 
rel=preload came up
... feedback on no-push was not very positive
... h2o implemented opposite signal 'h2-push-only'
... we opened an issue to explore if explicit push signal would be 
better
... not clear on use cases for the opposite signal, asked for more 
details on the bug
... let's wait for the use case
... re, f2f agenda..
... Wenbo (Google) asked about compression API. Many Google properties 
interested in compressing upstream data.
... in theory, not hard to implement for gzip / brotli?

Nathan: we're interested in this as well, e.g. for perf logging -- we 
implement our own hacky compression
... lots of use cases for it
... for FB including gzip library is too much

let's discuss at F2F

NicJ: will try to get numbers for implications of size of library, 
transferred bytes

Yoav: SpeedCurve is hosting an event the day of f2f
... please RSVP



On 2017-06-14 04:26, Ilya Grigorik wrote:
> Hangout:
> https://hangouts.google.com/hangouts/_/chromium.org/webperf-wg [1]
> 
> WIP agenda for this week's call:
> https://docs.google.com/document/d/10dz_7QM5XCNsGeI63R864lF9gFqlqQD37B4q8Q46LMM/edit#heading=h.6w46azw89too
> [2]
> 
> If you have other topics you'd like to discuss, please leave a comment
> on the agenda—the doc is open to everyone.
> 
> Links:
> ------
> [1] https://hangouts.google.com/hangouts/_/chromium.org/webperf-wg
> [2]
> https://docs.google.com/document/d/10dz_7QM5XCNsGeI63R864lF9gFqlqQD37B4q8Q46LMM/edit#heading=h.6w46azw89too

Received on Wednesday, 14 June 2017 18:34:37 UTC