1264871 - Firefox hangs after a day - with high cpu (kernel time), maybe related to VM or RDP

Status: RESOLVED INCOMPLETE

(Firefox :: Untriaged, defect)

Description

[Moritz Franckenstein]

Attachment: Screenshot Sysinternals ProcessExplorer

User Agent: Mozilla/5.0 (Windows NT 5.2; rv:45.0) Gecko/20100101 Firefox/45.0
Build ID: 20160407164938

Steps to reproduce:

I'm running Firefox 45.0.2 with admin rights on Windows Server 2003 SP2 x86 in a VM, accessed by RDP.
I have hundreds of open Tabs, many of them since years. Since they are loaded on demand, that's normally not a problem.
Firefox is running all the time, at night the VM itself is suspended.


Actual results:

Since a year or so, over all Firefox versions since then, after running about a day or two, Firefox starts to hang for some seconds and reacts to mouse clicks very delayed. The hanging times get longer over time and after some more hours it doesn't come back at all. If I exit Firefox before it completely hangs, the session is correctly saved and restored, so I can continue where I was.
I tried many things, like closing Tabs, to see if one of them causes it, but when it started to hang, it never became better again.
The maybe interesting thing is what I see in the Sysinternals process explorer CPU graph: one of my 4 cores is completely loaded (25%) during the hangs, but in red, which means kernel time, I think - but I have no idea what this tells us. During the hangs, there is no I/O. See screenshot.
When I watch it longer, I sometimes notice that the hangs are exactly for 45 seconds and then Firefox continues to process my clicks for 1-2 seconds. Then the next 45 seconds start - maybe some timeout? But it's not always like that, it can be shorter or longer.

How can I find out more, without throwing away all my Tabs and History? I found some plugins showing memory per Tab, but they didn't help.

Could it be related to suspending the VM at night? I am not sure if it ever happened on the same day after a restart.


Expected results:

Firefox should not hang at all, of course :) Multiple cores should be used correctly and a restart should never be necessary to make Firefox be faster or use less memory.

Comment 1

[Paul Pasca[Away. Please needinfo? Paul Oiegas [:pauloiegasSV]]]

Hi Moritz,

Can you please test this with a new clean Firefox profile, maybe even safe mode, to eliminate custom settings as a possible cause (https://goo.gl/PNe90E), and provide the results?

Thanks,
Paul.

Comment 2

[Moritz Franckenstein]

Hi Paul,

thanks for your message. Are you ok with first trying the "Refresh Firefox" feature? I have all my passwords stored in Firefox and don't want to loose my tabs. Or do you have another suggestion which would preserve at least my most important things?

Comment 3

[Moritz Franckenstein]

Ok, here is what I did now:

- backuped %APPDATA%\Mozilla (there is my profile) and %USERPROFILE%\Local Settings\Application Data\Mozilla (is it correct that this is only the cache and nothing important?)
- uninstalled Firefox and maintenance service and deleted their program files folders
- deleted %USERPROFILE%\Local Settings\Application Data\Mozilla
- deleted some dispensable looking subfolders in my profile (including extensions and others)
- realized that I forgot to download the Firefox installer :) - but managed to install it anyway
- it's all still there except the extensions of course, which I will install when needed

As soon as the issue happens again, I will report here and try "Refresh Firefox" as next step.

Comment 4

[Moritz Franckenstein]

It did not happen again since my last post, so it's unclear what it was. I still hope that someone could comment on my screenshot. What could have caused this?
I will close this bug after some more time.

Comment 5

[Ciprian Muresan [:cmuresan], Ecosystem QA]

Moritz, is this issue still reproducible on the latest Firefox (46.0.1) or the latest Nightly (https://nightly.mozilla.org)?
You don't need to delete the old profile in order to test something with a new profile. You can place a Firefox shortcut on your desktop, right click it, go to Properties and under the Shortcut tab, add "-p -no-remote" to the target. It should look something like: <<"C:\Program Files\Mozilla Firefox\firefox.exe" -p -no-remote>>. Click OK and start the browser from the shortcut.
A profile chooser dialog will appear and you can create a new profile without affecting the old one.

Comment 6

[Moritz Franckenstein]

I think I already deleted the backup and with my current profile it didn't happen again. Anyway it was hard to reproduce because it could take days.

So we can close this.

Thanks for info about multiple profiles.

Comment 7

[Moritz Franckenstein]

Attachment: another Screenshot Sysinternals ProcessExplorer - 1 of 3 cores is busy. the red color means kernel time

Comment 8

[Moritz Franckenstein]

After updating to Firefox 47.0 it just happened again. See new attachment.

Comment 9

[Ciprian Muresan [:cmuresan], Ecosystem QA]

Hi Moritz,

A gecko profile would be very helpful in debugging the issue you are experiencing. Could you please follow the instructions found here (https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Profiling_with_the_Built-in_Profiler) and post the link that will be generated in a comment?

Comment 10

[Moritz Franckenstein]

Sounds great! Plugin installed and I will read through the instructions later. But it could take weeks for the issue to appear...

Any additonal advice?

The "Jank" checkbox looks like it is for me.

Comment 11

[Ciprian Muresan [:cmuresan], Ecosystem QA]

The link you will provide will have all the information in it. The "Jank" checkbox is only a filter, but once you upload the profile (by generating a link) anyone who views the link will be able to check it in order to see "jank" only.

Comment 12

[Moritz Franckenstein]

After reading the description, I thought it might reduce load while I'm waiting for the issue...

Comment 13

[Ciprian Muresan [:cmuresan], Ecosystem QA]

I've exchanged a number of emails with Moritz, this comment is just a reminder. I've asked for a performance profile, so we're waiting for Moritz to post one.

Comment 14

[Ciprian Muresan [:cmuresan], Ecosystem QA]

Marking this as Resolved: Incomplete due to the lack of response from the reporter. Moritz, if the issue is still reproducible please post a performance profile.