1813396 - Can't access Google (intermittent)

Status: RESOLVED WORKSFORME

(Core :: Networking, defect, P2)

Description

[John]

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/109.0

Steps to reproduce:

a) entered a query in the search box
b) went to translate.google.com

Actual results:

Occasionally Firefox shows an hourglass in the tab as if it's trying to connect, and after a while it just gives up, leaving me on the same page as before. If I go to Chrome or Edge or Opera while it's trying, I connect instantly. It only seems to affect Google-related pages, and it only happens once every couple of days. I can't detect any pattern to it. Is there any other information I could collect to help diagnose this?

Expected results:

a) search results displayed
b) Google Translate page displayed

Comment 1

[BugBot [:suhaib / :marco]]

The Bugbug bot thinks this bug should belong to the 'Core::Networking' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Comment 2

[edgul]

Thanks for the report John,

You might try to use http logging to get some more information out of the logs. You can submit them to [e-mail] and we will have a look.

You might also try using the profiler in "networking" mode and attach the output, but catching the bug might be tricky.

Please let us know what you find.

Comment 3

[John]

I will try, but it is happening infrequently so I'll basically just have to turn on logging and leave it on for several days in the hope that it happens again... :(

Comment 4

[John]

I managed to capture an event using about:logging -- searching for "java sort text with numbers numerically" from the Google search bar.

The file is close to 700MB for only a few minutes of capture, and even compressed it is 38MB, so I can't attach it.
You can download it from here:
https://drive.google.com/file/d/1Y6Kf5_hSE47auls0bxCaOCxPzl-MjERz

Comment 5

[Kershaw Chang [:kershaw]]

Hi Reporter,

Thanks for the log. It indicates that the problem might be inside our http3 code, but the log doesn't have that information unfortunately.
Could you try to get the log again by setting MOZ_LOG to timestamp,rotate:200,nsHttp:5,neqo_http3::*:5,neqo_transport::*:5?

Thanks!

Comment 6

[John]

Gladly, but where and how do I set this?
You'll have to wait a while before I can manage to capture another example, too...

Comment 7

[John]

OK, I figured out how to do it, and set MOZ-LOG as requested.
Another event just happened, so I went to about:logging, but both the "Start logging" and "Stop logging" buttons were grayed out...
So I still have nothing to report!

Comment 8

[Kershaw Chang [:kershaw]]

(In reply to John from comment #7)

OK, I figured out how to do it, and set MOZ-LOG as requested.
Another event just happened, so I went to about:logging, but both the "Start logging" and "Stop logging" buttons were grayed out...
So I still have nothing to report!

I think you might want to keep the log always running (remember to add rotate:200) and stop it when the event happens. This is the only way to log what happens.
I'll set the ni flag to indicate that we are still waiting for more information. Thanks.

Comment 9

[John]

Here's another log file: https://drive.google.com/file/d/1YJGFGPHjVHkQysXQGwpscgT47f2PBV5L
Firefox 110 has suddenly made it much easier to manage logging!

Comment 10

[Greg Hess]

Leaving in priority review for Kershaw to review logs.

Comment 11

[Kershaw Chang [:kershaw]]

Hi Reporter,

My apologies that I didn't have a chance to look at the log. The provided link in comment #9 doesn't work anymore.
If you can reproduce this, may I ask you to capture a log again?

Thanks.

Comment 12

[John]

Will do. I have again enabled logging as requested earlier, and will upload another log the next time I see it happen.

Comment 13

[John]

That didn"t take me long... :)

It's here: https://drive.google.com/file/d/1TacHtiY9A-XTVedSuHE_zxrUhY-AngwP

Comment 14

[John]

NB: Please let me know when you've looked at it so I can delete it.

Comment 15

[edgul]

I've downloaded to our team's temporary log storage. Someone will review it soon.

Comment 16

[Kershaw Chang [:kershaw]]

Thanks for the report. It shows that the udp socket just receives nothing after some point. This means we might have some issues in our udp socket code, but the log doesn't contain that information.
I think the first step here would be adding some logs in nsUdpSocket to see why we don't receive any packets.

Comment 17

[Valentin Gosu [:valentin]]

Hi John,

Could you capture another log and also a wireshark dump while you're reproducing the issue?
https://firefox-source-docs.mozilla.org/networking/submitting_networking_bugs.html#wireshark-dump

Thanks!

Comment 18

[John]

Here is a zip file contain a log and a pcap file resulting from an attempt to use Google Translate:
https://drive.google.com/file/d/1Tdr78K3bbf0NEEdYWz6MY7ANtDEo1Kdc

HTH

Comment 19

[John]

Does this mean you need something more than a log and wireshark dump (which I provided 6 months ago, see above)?
If so, what?

Comment 20

[Kershaw Chang [:kershaw]]

(In reply to John from comment #19)

Does this mean you need something more than a log and wireshark dump (which I provided 6 months ago, see above)?
If so, what?

No, I think that's enough. However, I still can't figure out the root cause from the log. The log just shows that the server stops responding.

May I ask you to try one more thing: when this issue happens, does it work if you do a shift reload (Ctrl + Shift + R)?
Thanks.

Comment 21

[John]

I'll try, but it may take time -- the bug is intermittent.

Comment 22

[John]

Just got a failure clicking on a link to a search result from Google, which eventually stayed on the Google results page, and then I remembered to try shift reload. But of course it just reloaded the Google page, and the search link then worked. I need to remember to do a shift-reload during the failing request, not after... :(

Comment 23

[Manuel Bucher [:manuel]]

The eventually staying on the Google results page is likely a channel failure (timeout?) that unfortunately currently doesn't generate an error page: Bug 1842429. When improving error reporting, debugging cases like this would become easier.

Comment 24

[John]

I just tried to open 4 new tabs from a Google results page, and they all got stuck -- and then ctrl-shift-R successfully loaded each of them.

Comment 25

[John]

And just to contradict this: I tried to follow a link from a Google results page, got stuck, pressed ctrl-shift-R, and ended up still looking at the Google results page.

Comment 26

[Valentin Gosu [:valentin]]

It's possible this was fixed by bug 1876045.
John, could you check f it still happens for you?

Comment 27

[John]

Since it's intermittent, it's hard to check, but I'll report if it happens again.

Comment 28

[John]

Seems OK now.

Comment 30

[John]

Sorry, I spoke too soon.
Just now using FF 125.0.1, a Google query stalled. After waiting about 20 seconds, pressing Ctrl-Shift-R loaded the requested page immediately.

Comment 31

[John]

As of the last week or so, this has started happening again quite frequently. I am now on FF version 129.0.1.

Comment 32

[Valentin Gosu [:valentin]]

Thank you, John.
Things might have changed since the last time you submitted logs.
Could you go to about:logging?output=profiler&preset=networking, click on the Enable stack traces for log messages checkbox, and click start logging?

Thanks!

Comment 33

[Sunil Mayya]

Hello John,
Did you get any chance to collect the logs?
Thanks

Comment 34

[John]

Ah, sorry, I just got this by email -- I must have missed the previous message. I'll get onto it.

Comment 35

[John]

I have had logging on since I last wrote, but the problem has not recurred. I am now on v132.0.1, so I presume whatever it was has been fixed between v129 and 132.

Comment 36

[Valentin Gosu [:valentin]]

Thank you, John! Please let us know if it starts reoccurring.