754567 - Web page will not open and FireFox starts consuming RAM space at a relatively fast rate when a link clicked on has a video on that page.

Status: RESOLVED WONTFIX

(Core Graveyard :: Plug-ins, defect)

Description

[[e-mail]]

Attachment: 3 pngs 2 of Task Manager Process window, one of the website attempting to open.

User Agent: Mozilla/5.0 (Windows NT 6.1; rv:12.0) Gecko/20100101 Firefox/12.0
Build ID: 20120420145725

Steps to reproduce:

When clicking on a link that has a video embedded on the new page, quite often Firefox starts grabbing lots of RAM space (~350 MB and up), CPU (Pentium4D) utilization jumps up to 66% to 95% and browser becomes unresponsive and screen turns white. OS in use is Windows 7 32 bit, updated to May 10, 2012. This also happened in FF ver11 but not as often. This issue also happens on Windows 7 64 bit (different machine, Core i5). When this problem occurs it does NOT stop Windows 7 32 or 64 bit. The issue also does NOT stop Windows XP Pro SP3 but its a lot harder to shut FireFox down when the browser is tying up CPU resources under XP.

Also please note that on each VM image or CPU I run FireFox on, I've downloaded a fresh image from a Mozilla site or a mirror. Its not the same download on multiple boxes. This all began with FireFox version 11. 


Actual results:

Eventually if you are patient, sometimes the issue will resolve itself. Most often though it gets to a point where it just hangs and stops. Usually you can kill the process in Task Manager and things clear up. When FF reopens it offers to restore the previous session (just say NO!). If you choose to go back to the previous session, the issue starts all over again. If you choose a new session, things are fine unless you attempt to return to the same website via a link.  The website chosen seems to make the difference, however I haven't been able to ID what besides having a video window makes a difference.  I also haven't been able to pin down what type of video is the issue.

As the issue progresses, if you watch the Process window, you can see the RAM count slowly oscillate between a lower amount of RAM to a new higher amount. Like its stepping up 3 steps and then dropping back a step or two, to just take three more steps up in the RAM count. Over and over. Like its trying to clear out garbage RAM but not succeeding in keeping up.  The amount of RAM fluctuation is around 50 to 130 KB. 

Overall the system physical RAM usage slowly climbs from 500+ MB to 1500 MB and up in the Performance window. File-swapping is turned on. Drive activity obviously increases. There are no other applications running. 
 
In the attachment, there are 3 PNG files. 2 of the Process window. If you note firefox.exe is grabbing RAM, the two shots are approximately a minute apart. CPU utilization was running at 65 to 70%. This was from the PentiumD equipped machine running Windows 7 32 bit. (4GB RAM) 

This problem does not seem to happen when running recent versions of FireFox under Ubuntu 11.1 or Linux Mint 12 on the same machines. (Dual boot and/or VirtualBox VMs)

I have a 6 Megabit/sec DSL connection for my Internet service at home. The connection at work is a ComCast (cable provider) connection, I have no idea of the speed but it is slower than my home connection. This issue in FireFox occurs at both locations but I've only had access to my laptop (Core i5, Windows 7 64 bit, 4GB RAM) at work. 

Websites this has occurred on include: rarely for cnn.com, yahoo.com (the news site), hulu.com and a host of random less popular sites. Usually these sites have lots of advertising, video and pictures. 

Netflix Silverlight plug-in seems to not work as well in FireFox as it does in IE9 (or IE10 on Windows 8). This likely does not have anything to do with the issue but it is video.



Expected results:

I would expect that on clicking the link it would open the linked page and then render it. 

I would hate to give-up FireFox for Chrome.

Comment 1

[[e-mail]]

All of this behavior is definitely related to specific websites. It is reproducible in that it happens in FireFox but doesn't happen in IE on the same site. Since it happens on four different computers running three versions of Windows OS but NOT under Linux, makes me think it is related to a specific function inside FireFox. The only correlation to hardware I can offer is that the slower computers have the issue more often.

Comment 2

[Jason Smith [:jsmith]]

Anthony - Can we get someone from QA assigned to investigate this?

Comment 3

[u279076]

Miles, can you please verify the same behaviour is evident in the latest nightly (nightly.mozilla.org) and any of the previous releases of Firefox?

Comment 4

[[e-mail]]

I installed the nightly image about 10 minutes ago. 645PM CDT Tuesday Evening
Well I get something different now. 
When I click on the link in the "remembered" typed in URLs (drop box below URL window) sometimes I get a message: "400 Bad Request 
(newline) Request Header or Cookie too large (newline) nginx"  

CNN.com just flat out doesn't make an appearance. Behavior of FF seems to have settled down. 

Happens also when I click on any link on the CNN page. This message doesn't come up all the time just about 20% of the time. Also the message URL replaces the www.cnn.com URL in the dropdown box listing. 

BTW Netflix runs great and fills the frame nicely in the Nightly version.  

I've tried ABCNews and Yahoo.com and other sites and not had a problem yet!

The RAM issue: If its a memory leak, its now smaller, only 4K additional RAM at a time, about every second or so. RAM count started at around 164,000K and is now up to 184,000K and still climbing. Its doing that with one tab opened on my Yahoo email page and this tab for bugzilla. This is while I'm only running Nightly and nothing else on the PentiumD 32 bit machine. 

I'm sorry but I'm too chicken to put the nightly on my Win 7 64 bit laptop. Gotta a project that I cannot risk! What I will say is that video playback seems to snap quicker BUT that might because I'm not running any other major applications.

Comment 5

[Ioana (away)]

This issue has been tested on Windows 7 32-bit, Windows 7 64-bit, and on Windows XP. You can see the results of the investigations here https://mozqa.etherpad.mozilla.org/bug-754567.

The memory and CPU usage growths reproduced consistently on Windows 7 32-bit, but not on the other OSs. Firefox did not displayed the white screen, hang or crash even once.

The issue was investigated on Firefox 12.0 as well as on the latest Nightly.

I haven't had the time to find the regression range for this issue, but I will try to get it tomorrow.

Comment 6

[[e-mail]]

I do not have access to mozqa.etherpad.mozilla.org/bug-754567.

From my own experience it happens in version 11 as well.

Comment 7

[u279076]

Sorry Miles, the mozqa subdomain is internal. I've moved the results to the public domain: https://etherpad.mozilla.org/bug-754567

Comment 8

[[e-mail]]

Thanks for moving the "results" Although its not obvious, at least to me, that the test performed have anything to do with the underlying issue. Opening a bunch of tabs with video in each tab is NOT how I got the issue.

Open a single tab, click or type in the URL of a webpage with a funky video window and the browser goes off to La-La land. It does not happen all the time. Assuming FireFox is not totally hosed, you can recover by closing FireFox and restarting it.

What's annoying is that once it has failed to open the webpage, it will continue failing to open until (I'm assuming here) the cached page data has timed out and gets flushed.  Even though I reported the issue happens with cnn.com it doesn't always fail on that page. Other pages have fewer issues. Some of that might be because the CNN page gets re-written frequently. Its constantly changing throughout the day  or it might be with a specific timing issue.

An avenue I would suggest looking down might be interaction with timing. Since different computers seem to have the issue to differing degrees, is it possible that timings on the get commands issued by the browser, the network server response time and the ISP service are playing a role here?

If you tell me where to look I can copy whatever configuration file I need to at least give you the browser settings I have.

(I'll put all of this in comments section on Bugzilla which seems to be the only thing I have access to.)

Comment 9

[u279076]

Ioana, can you please retest to address Miles' concerns in comment 8?

Comment 10

[Ioana (away)]

Anthony, Miles, if you read the whole etherpad, you will notice that we tested multiple sites with just a single tab from the beginning. Everything went ok in those cases on Windows XP and Windows 7 64-bit. On Windows 7 32-bit, the CPU and memory usage went up fast and stayed there, especially when opening yahoo.com videos. 

Since that is the only thing that reproduced, we tested with multiple tabs to stress the browser a little, hoping the whole issue will reproduce (CPU and memory usage increase + page not loading, hangs). It didn't happen.

As for the timing issue, we tested on various computers here from the start. Anthony, perhaps there will be a bigger difference if someone over there can test on your computers.

Comment 11

[Ioana (away)]

Miles, since you are the only one who reproduced the hang, could you please try to find a regression range for this bug?

You can find more details about how to do that here http://harthur.github.com/mozregression/

Comment 12

[[e-mail]]

I'm getting a Bugzilla email or two a day. Nobody seems to be able to reproduce it. BUT right now on this laptop (Windows 7 64 bit) the RAM in use by Firefox (32 bit) is slowly increasing at 4Kilobytes every 2 seconds. When I started the email the RAM in use was 244,000KB, its now up to 250,000KB. I have 3 windows open. One for FF, one for Media Player I'm listening to music and an update download for Renases a 32bit embedded processer IDE.

Now the Renases download is finished. I would expect the RAM to decrease, but no, its now at 251,000 KB and still climbing.

If you open FireFox on the standard Firefox-Google web page as home. It will sit there and idle with RAM count fluttering from 182,000 KB up to 205,000
and back. On a STATIC page.

As I reported originally it started with VERSION 11 of FireFox. I have no clue what subversion, build, whatever. It did not do it in Version 10. It does it on all three of the computers I've tested it on, a Windows 7 32 bit OS on a PentiumD system, on a Windows 7 64 bit laptop with a Corei5 and on a Windows XP Pro SP3 operating system image running on a Pentium 4 system.

I loaded the Aurora load from Thursday on the PentiumD system this weekend and it did not glitch or trash the web sites with video.  In fact cnn.com runs fine now. 

I will NOT trash my system jumping around willy-nilly testing versions looking for a build number I'll never find or be able to ID. Certainly not on this laptop. This laptop is my job, improperly working laptop, no job.

If you detect irritation in my text, yes it is there. One thing about tech support the Mozilla Bugzilla gang needs to understand, it does not help to feel like you've been passed around behind the scenes. I understand ya'll are doing what you do for love of the game. 

I do not know what to call the ram count going up other than a memory leak.  I've experienced the same thing professionally at a previous employment from other programmers (JAVA to be exact). The size was similar and the behavior was very much the same. An unidentifiable memory leak that continually grew until the system crashed (XP Pro). Is it possible somebody's code is collecting debug info?

The only behavior change I've made is that I don't leave FireFox running any more. It gets shutdown frequently and restarted now. I don't trust it.

Comment 13

[u279076]

I appreciate your frustration, Miles. Unfortunately, you are the only person on this bug report who can reproduce this issue so far. I'm willing to help guide you through the processes of determining a regression range using the tool Ioana pointed out in comment 11, but it sounds like you aren't willing to do that.

Finding a regression range will help us narrow down this down to a very small set of code changes which could be causing the issue.

We can't take this bug report any further unless you are willing to debug this with our assistance. Unfortunately, this means I have no option but to resolve this report as INCOMPLETE. I really hope you change your mind as it would serve to benefit many other users who may also be experiencing this issue but remain silent. 

Please reopen if you change your mind.

Comment 14

[[e-mail]]

Attachment: Screen grab of laptop including taskbar, listing plugins for FF v13

There is also a Yontoo extension added to the FF v13 install. at the time of the issue, Firefox and Outlook are the only two applications running.

Comment 15

[[e-mail]]

When I went to add comments to the attachment I added, it dumped what I had already typed in. This is a really **** tool.

Again I was on edition.cnn.com
I right clicked (Open on a New Tab) an embedded link : http://edition.cnn.com/2012/06/07/world/meast/iraq-execution/index.html?hpt=hp_t3 in the headline "Saddam Hussein Aide executed in Iraq"
It popped up an error page listing error 400 wording exactly as before.

the url in the url text box was as listed above.

If you click back to the original tab and click the link again, it pops open fine. It seems to be the first click on the page that does it. It only happens 10% of the time but its irritating when it does.

I am running version 13 of FF on my laptop (Core i5, now 8GB of RAM in it). 

This is my PRODUCTION machine I will not screw with it. I can screwup my slower PentiumD machine but its a home machine. I do not work at home. So whatever instructions you send me need to be written clearly and I will try to home in on a regression analysis. But since I already know it started with version 11 of FF. That tells me what I need to know. I'd search the branches just pre- and just post- of version 11 becoming the trunk. Do a diff search and you'll be in the ball-park. As close as you'll get.

I have no access to the FireFox trunk and do not wish for it. But I'd look at the code that translates the embedded URL into a HTTP GET command. It seems as if cnn.com's server is getting **** from FireFox on the first click processing. Maybe the buffer that stores the right-click function is trash? Or maybe not initialized yet but something thinks it is? 

I'm not going to tell ya'll how to program but something isn't right and it started with FF version 11.

Comment 16

[[e-mail]]

Oh BTW FireFox is up to 280 Megabytes of RAM space and still climbing.

Comment 17

[u279076]

Quite simply, the first step is to start trying your own steps with progressively earlier Firefox releases until you find a release where you can't reproduce this anymore.

For example:
* Try Firefox 13
* If it reproduces, try Firefox 12
* If it reproduces, try Firefox 11
* If it reproduces, try Firefox 10
* If it reproduces, try Firefox 9
and so on...

Comment 18

[[e-mail]]

Where does one actually download OLD versions of FireFox? Ordinary Users (as far as FireFox is concerned I am an ordinary user) don't have easy access to old versions of anything Mozilla and I'd say that's a good thing.

Since I thought FireFox was mature I changed my install pattern to automatic. The only version I currently have on my Windows PentiumD system (32 bit) is the current version 13. Version 12 seems to have disappeared last week? All traces of other versions are gone. 

When I started FireFox, the RAM count was at 56 or 57 Megabytes. Its now up to 134.464 Megabytes. The only pages open are this one and the Yahoo email page. When I close the page the RAM drops down to 97 MB. After another 20 secs its down to 93 MB. The next application RAM hog is that little program explorer.exe.

When I shut cnn.com down and leave only static pages in the tabs. This page seems static. You can try wmileswade.com. It has 3 lines of text and a picture of a bulldozer. There is nothing active on it. RAM in use for the process firefox.exe will slowly, painfully climb usually in 4 KB steps up. Obviously the more tabs open, the faster RAM climbs.

Comment 19

[[e-mail]]

As another experiment, I disabled ALL add-ins, extensions and plug-ins, leaving just this page open. Outside of the TaskManager there are no other applications running. RAM in use dropped to 95 MB. After about 60 seconds it started climbing. Its now up to 98.6MB. This is with version 13 release version. In the time it took to write this note it has climbed to 99.4MB

Comment 20

[[e-mail]]

I changed my home page to this page. RAM count is climbing from 39 MB.

I went even further. I changed my home page to about:blank. I went into the services tab of Task Manager and shut every service I could that had no connection to networking. The only applications running are FireFox 13 and the Task Manager.  Restarted FF, the add-ins and extensions are still disabled. The RAM in use count comes up as 26MB and slowly climbs from there with a freaking blank page!

Assuming I can get complete download packages for versions 10 through 12 I'll try your method this weekend but I don't know how I'll be able to refine the "regression" any tighter than that. You're going to have to point me to where I can get the downloads. I've been assuming that the bulk of the code is in Java but truly do not have a clue what you guys are using.  

I am a hardware systems engineer. I design hardware and have it built. This kind of nit-picking is why I do not like working with software. Yes I program in C#, VB.Net, Fortran, MSP430 assembler, Z80 assembler and x86 MASM assembler but I don't have to like it. (Tell Bugzilla FORTRAN is a real name when entered in upper and lower case, geez no respect.) 

What I will tell you is that the RAM creep is very slow when the pages are static. When they have active content, the RAM usage climbs quickly. It looks like garbage collection is not keeping up, but I don't know why it is running when the page isn't doing a damn thing. The render is finished. 

I'm finished screwing with this tonight.

Comment 21

[[e-mail]]

BTW I had another 400 code popup on a cnn.com page. It was the very first click I made on the page. It was a right-click Open in a new Tab. When I went back the the original page (original first tab) and right-clicked on the same URL embedded headline, it opened clean. When I finished the regression stuff I look at the first click stuff again.

The Mozregression is based on nightly builds and dates. I have no clue what date to start with.  When was version 11 first released?  I know that it started with version 11. I patiently waited until 12 came along but it had the same issues. No I wasn't aware of the RAM leak before version 12 but the behavior that prompted my first report was the same as previous. Since it didn't happen often I attributed it to **** Windows 7. I still use the Windows XP Pro box and when it started happening there that's what got my attention. There are now 4 different computers that all have the same issue. The only common features are: they are all DELL's, they all are running some form of Windows and they are all running FireFox as default browser.  (I'm not a fanboi of DELL, I just trust them slightly better than HP.)

3 of the 4 computers also run dual or triple boot with Debian derivatives Ubuntu and LinuxMint. They do not have either problem I've had with the Windows version of FireFox. (Laptop is single boot.)  That's what has me wondering if somebody left debugger hooks turned on in their software module.

Comment 22

[u279076]

All previous versions of Firefox are available on our public FTP server:
ftp://ftp.mozilla.org/pub/firefox/releases/

Considering you first noticed this in Firefox 11, Firefox 10 would probably be a good place to start working back from. Once you've identified the last-good and first-bad releases I can identify the dates to use for Mozregression.

Comment 23

[[e-mail]]

OK Aurora 14.0a1 does not make the RAM climb on this static page. Date 5/26/2012
So at least there is a top end, some where. 

Is there anything different between the nightly builds and the release builds?

Comment 24

[[e-mail]]

Thanks for the URL. I've already downloaded a 10 and an 11. I'll try them out tomorrow afternoon. The Aurora FireFox build is there something different (other than it insists on downloading every version)?

Comment 25

[u279076]

If Aurora 14.0a2 from May 26th does not display this bug for you, it's possible it was fixed recently. Can you try the latest-mozilla-central build from nightly.mozilla.org to see if it reproduces?

Comment 26

[[e-mail]]

I'll try it when I get home.

Comment 27

[BMO Automation]

Resolving as wont fix, plugin support deprecated in Firefox 85.