956452 - Network monitor is janky on sites with a ton of requests

Status: RESOLVED DUPLICATE of bug 1487906

(DevTools :: Netmonitor, defect, P1)

Description

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

Attachment: screen shot of profile

For example, https://soundcloud.com/stream had 691 requests for me, and once all the requests started finishing and getting responses, we started janking pretty hard. I'm pretty sure this is the drawing of the waterfall graph. Is it possible to do this in a |DevToolsUtils.yieldingEach| loop or move a bunch of processing to a worker?

(Profile was too big to upload, but I have a screenshot.)

Comment 1

[Victor Porof [:vporof][:vp]]

I'm reasonably sure we could.

Comment 2

[Victor Porof [:vporof][:vp]]

Is it still janky for you now with the new theme changes?

Comment 3

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

Attachment: screenshot of profile

(In reply to Victor Porof [:vporof][:vp] from comment #2)
> Is it still janky for you now with the new theme changes?

Yup, although I had to manually clear the cache so that all the avatar requests would happen again on soundcloud; Cmd-Shift-R wasn't cutting it.

Here is a screenshot of the up to date profile.

Comment 4

[Victor Porof [:vporof][:vp]]

Thanks!

Comment 5

[Panos Astithas (he/him) [:past] (please ni?)]

Nick, you know that you can click 'Save' to export the profile and attach it to the bug, right?

Comment 6

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

(In reply to Panos Astithas [:past] from comment #5)
> Nick, you know that you can click 'Save' to export the profile and attach it
> to the bug, right?

Yeah but that only works with small profiles. Bugzilla's file size limit is like 10 or 20 mb and the profiles can easily double or triple that.

Comment 7

[Panos Astithas (he/him) [:past] (please ni?)]

I guess that could be a problem, although I've never created anything larger than a couple of MB compressed myself.

Comment 8

[Jim Blandy :jimb]

I happened to run this with packet logging enabled, and it was logging *tons* of 'substring' requests, containing lots of base64-encoded stuff.

Comment 9

[Jim Blandy :jimb]

Attachment: Log of remote protocol packets fetching many longString actor substrings

Here is the tail end of the packet log. I've cleaned it up a bit by taking out the actual string contents.

Comment 10

[Victor Porof [:vporof][:vp]]

Is it still slow now that bug 1064262 landed?

Comment 11

[Alexandre Poirot [:ochameau]]

Since the orignal report of this bug, the netmonitor has been almost completely rewritten.
* We do throttle UI updates to avoid DOM operations on each new request.
* all request's data that isn't display in the default table is now lazy loaded, so that the RDP overhead is much lower.
* many optimizations have been made to make the table layout faster to render.

I don't have any soundcloud account.
Is it still slow?

Comment 12

[:Harald Kirschner :digitarald]

Main thread impact got a lot better. We have more actionable bugs open, so duping this for the meta bug 1487906.