Open Bug 1745402 Opened 3 years ago Updated 3 years ago

Incomplete tab loading

Categories

(Core :: Networking: HTTP, defect, P2)

Firefox 97
Desktop
Windows 10
defect

Tracking

()

Tracking Status
firefox97 --- affected

People

(Reporter: acadkins327, Unassigned)

References

(Blocks 1 open bug)

Details

(Whiteboard: [necko-triaged])

Attachments

(2 files, 3 obsolete files)

Attached image image.png (deleted) —

I am a contest sweeper and load many tabs at one time. Using latest Windows 10, and Internet speed of 300Mbps. For the past month(?) or so if loading more than about 20 tabs several tabs only load partially. I have to refresh the tab to get a complete load. The examples are from a load of about 50 tabs. Estimate about 10-20% do not load completely.

Does the problem still happen if you start Firefox in Safe Mode?
See https://support.mozilla.org/en-US/kb/troubleshoot-firefox-issues-using-safe-mode
If the issue still persists, please test it using a new profile, you can find the steps to do that below:

https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles?redirectlocale=en-US&redirectslug=Managing-profiles#w_starting-the-profile-manager
Thanks.

Flags: needinfo?(acadkins327)

Believe this started with v96. I have refreshed Nightly on a regular basis, especially after each version update, no change after. I have not tried using safe mode. I open the tabs all at once by having a collection of bookmarks in a folder. I have pretty much eliminated the problem by only opening about 25 tabs at a time. I have had Fusion turned on. I may try turning off?

Flags: needinfo?(acadkins327)

Could you please try turning that off and test again?
Thanks.

Flags: needinfo?(acadkins327)

Using Nightly 97.0a1(2021-12-14) there is no option to turn off Fusion in experiments. Today I had about 20% of 36 tabs that only partly opened.
Using Firefox 95.0(64 bit) ALL 36 tabs fully opened with no problem.

Flags: needinfo?(acadkins327)

The severity field is not set for this bug.
:dao, could you have a look please?

For more information, please visit auto_nag documentation.

Flags: needinfo?(dao+bmo)
Component: Tabbed Browser → Networking: HTTP
Flags: needinfo?(dao+bmo)
Product: Firefox → Core

Could you try to record a http log?
Thanks.

Flags: needinfo?(acadkins327)

(In reply to Kershaw Chang [:kershaw] from comment #6)

Could you try to record a http log?
Thanks.

It'd be great if you can also point out the URL of http requests that are not loaded completely. Thanks.

Attached file Google Drive link (obsolete) (deleted) —

Attempted to attach link to log file for 59 tabs. It is 1.6 gig so could not be attached directly. I could send by email if the attached link does not work.

This time many of the tabs failed to load at all. I have concluded that I am simply trying to load too many and things go fine if 20 or less. How many tabs should I be able to load?

97.0a1(2012-12-28) Internet running at 201mbs

Flags: needinfo?(acadkins327)

(In reply to Al Adkins from comment #4)

Using Nightly 97.0a1(2021-12-14) there is no option to turn off Fusion in experiments. Today I had about 20% of 36 tabs that only partly opened.
Using Firefox 95.0(64 bit) ALL 36 tabs fully opened with no problem.

To turn off fission, please go to about:config and set fission.autostart to false.
Could you do this and test again? Thanks.

Flags: needinfo?(acadkins327)

Allowed share for link 7:30am. Will turn off fusion and try again tomorrow.

Flags: needinfo?(acadkins327)
Attached file GOOGLE DRIVE - LOG FILE LINK (obsolete) (deleted) —

Set fusion to false.
Only 3 tabs did obvious partial load and they were variations on the same site:
https://kudosz.com/red-lobster-new-year-bagel-gift-card-basket/
https://kudosz.com/godiva-majestic-grandeur-gourmet-gift-basket/
https://kudosz.com/happy-new-year-buttercream-cookies-gift-basket/
The fourth tab in the chain from this site loaded completely.

When I started entering tabs found that there were others that had to be refreshed before they could be used.

A lot of tabs timed out which I expect your log to show.

Nightly 97.0a1(2021-12-28) Internet checked at 212mbps
System: AMD Ryzen 5 2400G with Radeon Vega Graphics 3.60 GHz; RAM 12.0 GB (10.9 GB usable)

Attached file GOOGLE DRIVE LINK TO VER 95 LOG (obsolete) (deleted) —

Loaded same tabs into 95.0.2 and ALL tabs loaded - no time out tabs. Instead of timing out many did partially load.

For comparison attached a link to 95 log file.

Thanks for the log.
The log shows that there are lot's off TRR requests are cancelled because of timeout.
Not sure if this is regressed by bug 1737198.

Since you said this doesn't happen on 95.0.2, could you try to use mozregression to help us figure out which revision starts to have this problem?

Thanks.

Flags: needinfo?(acadkins327)

I am flattered you would suggest the mozregression process but that exceeds both my time available and skill level. I will be glad to assist at a more basic level.
Best regards,
Al Adkins

Flags: needinfo?(acadkins327)

(In reply to Kershaw Chang [:kershaw] from comment #15)

Thanks for the log.
The log shows that there are lot's off TRR requests are cancelled because of timeout.

I've looked into some trr requests. It seems that the time spent on doing proxy resolution on main thread is too long. We are working on bug 1745979 to improve this.

2021-12-28 11:21:27.255000 UTC - [Parent 15992: TRR Background]: V/nsHttp TRRServiceChannel ctor [this=1c3941c5000]
2021-12-28 11:21:27.255000 UTC - [Parent 15992: TRR Background]: V/nsHttp TRRServiceChannel::ResolveProxy [this=1c3941c5000]
2021-12-28 11:21:27.255000 UTC - [Parent 15992: TRR Background]: V/nsHttp TRRServiceChannel::ResolveProxy [this=1c3941c5000]
2021-12-28 11:21:28.181000 UTC - [Parent 15992: Main Thread]: V/nsHttp TRRServiceChannel::OnProxyAvailable [this=1c3941c5000 pi=0 status=0 mStatus=0]
2021-12-28 11:21:28.181000 UTC - [Parent 15992: TRR Background]: V/nsHttp TRRServiceChannel::OnProxyAvailable [this=1c3941c5000 pi=0 status=0 mStatus=0]

I think you can try to do things below and see if they help.

  1. Increase the trr time out by opening about:config and setting network.trr.request_timeout_ms to a larger value.
  2. Go to about:config and turn on network.proxy.detect_system_proxy_changes. Note that this feature is still in development, so it may cause some problem.
Flags: needinfo?(acadkins327)

Turned fission back on - changed network.trr from 1500 to 2000 - turned network.proxy to true.
NOTHING timed out and 65 tabs loaded. About 90% loaded fully the balance loaded partially.

Flags: needinfo?(acadkins327)

(In reply to Al Adkins from comment #18)

Turned fission back on - changed network.trr from 1500 to 2000 - turned network.proxy to true.
NOTHING timed out and 65 tabs loaded. About 90% loaded fully the balance loaded partially.

I just realized that proxy resolution for regular http requests is also really slow.

2021-12-28 11:21:28.795000 UTC - [Parent 15992: Main Thread]: V/nsHttp nsHttpChannel::ResolveProxy [this=1c38d02de00]
2021-12-28 11:21:30.228000 UTC - [Parent 15992: Main Thread]: V/nsHttp nsHttpChannel::OnProxyAvailable [this=1c38d02de00 pi=0 status=0 mStatus=0]

It looks like enabling network.proxy.detect_system_proxy_changes helps a bit, but not fix the problem completely.

To help us understand why proxy resolution is so slow, could you try to add proxy:5 to MOZ_LOG and get a log again? Could you make sure that rotate:200 is set as well to avoid producing a huge log file?
Also, do you have any proxy-related extensions installed? Could you go to network settings page and select No Porxy and see if this helps?
Thanks.

Flags: needinfo?(acadkins327)
Attached image Extensions.jpg (deleted) —
Attachment #9256963 - Attachment is obsolete: true
Attachment #9257019 - Attachment is obsolete: true
Attachment #9257023 - Attachment is obsolete: true
Flags: needinfo?(acadkins327)

https://drive.google.com/file/d/1Nw44K0cfns9jPft-F7J8VSSpxlzcR6IV/view?usp=sharing
https://drive.google.com/file/d/1c6UUty5Mf9IOUI7uO0i0ViG_2c8bSz3w/view?usp=sharing

Attached screen shot of current extensions. Both logs are with proxy set to NO. Suspected Ad Blocker might cause a problem but not noticeable difference with or without it.

Could you try again with setting network.trr.strict_native_fallback to false?
Thanks.

Flags: needinfo?(acadkins327)
Flags: needinfo?(acadkins327)

(In reply to Al Adkins from comment #23)

https://drive.google.com/file/d/1irO2NpgeFHfgQauRkUJ9m9zX3x9eY9Xl/view?usp=sharing
No Proxy with all specified settings.

Thanks for this log. With disabling network.trr.strict_native_fallback, I can see that there is no DNS request timeout.
I still see some requests are failed, but unfortunately the reason is not clear to me. One thing for sure is that everything seems really slow according to the log. Dispatching a http transaction to socket thread takes 2-3 seconds.
Anyway, I can't find anything wrong obviously from the log.

Well, I think mozregression is the best way to find out the problem.
Please reconsider to use this tool to help us. Thank you.

Flags: needinfo?(acadkins327)

Ok - I will take another look. I have not used it before so ther will be a learning curve. Will keep you posted.

Flags: needinfo?(acadkins327)

I loaded mozregression and was surprised at how easy it is to use. It runs a little slow when mass loading the bookmarks but is amazing. I ran out of time today but should be able to get into it tomorrow.

Mozregression revealed no problem - the Bug report should be closed.

Disabled all extensions and refreshed Nightly. There were two problems. Only with the Roboform extension did I ever see the server error. The second problem seemed to be the Nightly Experiment setting WebAPI:WebGPU which contributed to partial loading.

Thank you for your assistance and patience. You are a great teacher and I have learned a lot. In the future will be able to do a much better job of
trouble shooting on my own. I do apologize for taking so much of your very valuable time.

Best regards,
Al Adkins

(In reply to Al Adkins from comment #28)

Mozregression revealed no problem - the Bug report should be closed.

Disabled all extensions and refreshed Nightly. There were two problems. Only with the Roboform extension did I ever see the server error. The second problem seemed to be the Nightly Experiment setting WebAPI:WebGPU which contributed to partial loading.

So, could you check which one cause the partial loading issue? It's the Roboform extension or WebAPI:WebGPU experiment?
Thanks.

Flags: needinfo?(acadkins327)

I turned off "WebAPI:WebGPU" and I am still using Roboform. I have reduced the tab load to less than 30 and no longer have the server error.
I now have several rounds of loading groups of 25 to 30 tabs. The first round of loads are complete and I enter the contest. The second and later rounds may have occasional partial items. Beginning to suspect something in History - cache? I will try clearing History between loads.

When testing with Mozregression I simply loaded the tab groups but did not actually enter any contest.

Flags: needinfo?(acadkins327)

Tomorrow I will disable Roboform and enter contests without it. I have been reluctant because the process will be much slower.

Set the ni flag to indicate that we are waiting for more information.

Flags: needinfo?(acadkins327)

Loaded 27 tabs in Nightly 98(2022-01-11) with all 4 of my extensions including Roboform - 2 partial tabs.
I disabled ALL extensions in Nightly 98(2022-01-11) and loaded 27 tabs - 5 partial tabs.

I loaded 27 tabs in Firefox 96 and ALL tabs loaded with no problem.

Something is wrong in Nightly. I have used Firefox and/or Nightly for many years and will continue to do so. I feel that as a user my only obligation was to tell you that I found a problem. I regret that the testing I did was not effective in revealing when the problem occurred or what the problem is. I will not do any additional testing. I have other obligations and cannot take the time to get involved in testing. I will answer any specific questions regarding my configuration or what I am doing and cooperate as best I can.

Best regards,
Al Adkins

Flags: needinfo?(acadkins327)

When testing with Mozregression I simply loaded the tab groups but did not actually enter any contest.

Did you mean that you can't reproduce this with mozregression?
Could you try to use mozregression again with adding --profile $PATH_TO_YOUR_PROFILE option?
Thanks.

Flags: needinfo?(acadkins327)

I could not reproduce this with mozregression.

Unfortunately I cannot do any further testing.

Flags: needinfo?(acadkins327)
Blocks: doh
Severity: -- → S4
Priority: -- → P2
Whiteboard: [necko-triaged]

The bug has a release status flag that shows some version of Firefox is affected, thus it will be considered confirmed.

Status: UNCONFIRMED → NEW
Ever confirmed: true
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: