Previous Entry | Next Entry

Release managers just made the call that Firefox 3.0 will release with a known bug which brings Linux systems to their knees. Firefox's system-killing performance bug

The second-class Linux support policies of the Mozilla foundation continue... I imagine the reasoning went something like it has always gone in the past: "the platform that matters most is the one where we, the Foundation, get the most market penetration."

Webkit, here I come.

Update 2008-05-23: There are now two proposed workable solutions.

Comments

( 57 comments — Leave a comment )
Page 1 of 2
<<[1] [2] >>
(Anonymous)
May. 20th, 2008 08:07 pm (UTC)
most people will get a patched version anyway
Most users will be getting their updates from the distributions, not Mozilla directly. And I've talked with most of them and most of them are waiting for a patch before shipping/updating. So don't worry, most normal people won't be hit by this one. Very few of our users actually get their builds from mozilla.org directly.

And, also, the Mozilla community (and the corporation!) invested a huge amount in Linux this time around. Firefox 3 is the browser that Linux users always wanted. Our native theme code is new and feels great. (In fact, that's where the "native" theme code from WebKit comes from!) It's a first class platform for us, hands down.

Christopher Blizzard
jasondclinton
May. 20th, 2008 08:30 pm (UTC)
Re: most people will get a patched version anyway

The problem is that no one knows what the solution is. The patches that are proposed on the bug are not the solution: the problem is calling fsync. Reducing the number of times that fsync is called by 50% doesn't change that fact that calling fsync at all brings the system to its knees. I repeat: there is no patch for distributions to apply. And--even more disturbingly--it appears that no one has any idea how SQLite actually works. In short: Mozilla has spent little time investigating this problem even though it makes an entire platform completely unusable (after a sufficient quantity of profile data has accumulated).

It's going to be PR disaster if it's released in this state. Linux users will swear off Firefox for years if they experience this kind of system-killing performance. The browser market is heating up; Mozilla can't afford to take that chance.

Re: most people will get a patched version anyway - (Anonymous) - May. 20th, 2008 09:44 pm (UTC) - Expand
Re: most people will get a patched version anyway - (Anonymous) - May. 22nd, 2008 04:57 pm (UTC) - Expand
Re: most people will get a patched version anyway - (Anonymous) - May. 20th, 2008 10:05 pm (UTC) - Expand
Re: most people will get a patched version anyway - (Anonymous) - May. 22nd, 2008 11:10 am (UTC) - Expand
Re: most people will get a patched version anyway - (Anonymous) - May. 23rd, 2008 08:18 am (UTC) - Expand
(Anonymous)
May. 20th, 2008 08:11 pm (UTC)
Missing words
"We would like linux distros to take this patch for their builds if at all possible for 3.0."

While still serious, I would assume most people's builds will be provided by their distro's.
(Anonymous)
May. 20th, 2008 08:15 pm (UTC)
webkit
do you think that apple would care more?
(Anonymous)
May. 20th, 2008 08:46 pm (UTC)
patch to be taken by Linux distributors, and in first FF3 update
The patch in that bug will be in 3.0.1 (if it's not in 3.0 -- we've just said that we won't respin specifically for it, but if there's a respin done we will take it), and we're making sure that Linux distributors take the patch if their users are likely to be on ext3 or other filesystems that effectively convert an fsync into a full-filesystem sync. We believe that the bad Linux jbd behaviour will affect few users of FF3.0, given that it's going to be patched in the packages that the vast majority of our Linux users use. (Spinning and testing RC2 would only get the fix into the hands of those remaining users a couple of weeks sooner, versus shipping FF3 and then getting 3.0.1 out on schedule.)

We've invested a ton in improving the Linux experience in Firefox and Gecko, including a bunch of work on native theme code that I think you might find in another layout engine now :), and I think that shows in the product. We take our cross-platform commitment quite seriously, and I think it stands up very well against that of any other web technology stack.

Mike Shaver
(Anonymous)
May. 20th, 2008 08:50 pm (UTC)
Question
I didnt understand why you say that there will be no patch, and the bug report says there is one.
What am i missing here?
jasondclinton
May. 20th, 2008 08:55 pm (UTC)
Re: Question
I posted that there would be no patch at the exact same time that the bug changed to reflect acceptance of one of the proposed patches. However, I still believe, as I have tested it, that it is not the right solution.
(Anonymous)
May. 20th, 2008 08:54 pm (UTC)
Reply to topic
I like WebKit as well, but you don't have a WebKit browser for GNOME/Ubuntu.
Midori doesn't count at all, as it is similar to an near-alpha release.

Another thing that bugs me is that I don't know exactly what WebKit renders or not. Seriously.
I googled a month ago and got this in a forum: http://www.neowin.net/forum/index.php?s=093d1cd056e12f6688d7ca24fa7ba7ed&showtopic=628125

Reall bad.

=(
(Anonymous)
May. 22nd, 2008 04:06 am (UTC)
http://live.gnome.org/Epiphany/WebKit
Re: Reply to topic - (Anonymous) - May. 24th, 2008 01:10 am (UTC) - Expand
Re: Reply to topic - (Anonymous) - May. 28th, 2008 12:26 pm (UTC) - Expand
rtcm
May. 20th, 2008 09:15 pm (UTC)
The bug is in the kernel
... which doesn't implement fsync(2) as it should. A PR disaster for Linus?
jasondclinton
May. 20th, 2008 09:17 pm (UTC)
Re: The bug is in the kernel
You know, ext2 has been around since 1994 and this is the first time using fsync this way has come up. Are you really going to point fingers?
Re: The bug is in the kernel - jldugger - May. 20th, 2008 10:13 pm (UTC) - Expand
Re: The bug is in the kernel - (Anonymous) - May. 20th, 2008 10:17 pm (UTC) - Expand
Re: The bug is in the kernel - zlynx - May. 25th, 2008 01:28 am (UTC) - Expand
(Anonymous)
May. 20th, 2008 09:19 pm (UTC)
Is the bug really that severe? I've been running all of the beta series on Linux, haven't encountered anything like what is being described... ext3 and high system load seem to be mentioned often, both of which describe my source-built system...
jasondclinton
May. 20th, 2008 09:22 pm (UTC)
Seeing the performance degrade requires the accumulation of about two months worth of browsing history in the new sqlite data format.
(no subject) - (Anonymous) - May. 22nd, 2008 12:18 am (UTC) - Expand
(Anonymous)
May. 20th, 2008 09:22 pm (UTC)
how to reproduce?
The question for me is more: how do I reproduce this behaviour? I have just pushed my load average to 25 and don't see long freezes. Short ones (1-2 seconds max.) are to be expected from a heavily loaded system. I am using ext3 and my places.sqlite is 23 MB. What am I doing wrong?
(Anonymous)
May. 20th, 2008 10:14 pm (UTC)
good
.. to know. i have to reboot my desktop system, which i never halt, twice a week. good to know the problem is firefox. i will switch to epiphany asap. anyway, my main OS right now is not Linux anymore. I switched to OSX and i'm glad of it. I'm really tired about gnome and it's kind of development. And i'm on linux since 1998, i am a sys admin. my desktop should always work because without it i can't bring money to my wallet. i'll be back on linux when i'll be able to mount again nfs shares in user space through gnome interface, when i will be able to install fonts again and when the develpers will stop to talk about mono, silverlight and which is the best gnome game to include. Start to develop something like Book Fonts, ICC profile manager, etcc.. Start to develop something PROFESIONAL.

cheers
np237
May. 21st, 2008 08:59 am (UTC)
Re: good
I wonder whether epiphany will be affected as well. Even though it stores history and bookmarks in its own XML format, xulrunner automatically stores its shit in sqlite format.

If it is not used, we could easily patch the epiphany startup code so that these files are removed, though.
Re: good - dysynderesis - May. 22nd, 2008 02:43 am (UTC) - Expand
(Deleted comment)
(Deleted comment)
(Anonymous)
May. 20th, 2008 10:52 pm (UTC)
Because with a database you can do a lot of things you can't do without.
I'm using Firefox 3 beta 5 and then rc1 since the beginning of April and I'm really happy of it. The new library/history/bookmarks manager, the awesomebar, they're really good. I was becoming very disaffected about firefox on linux but this release is wonderful.
(Anonymous)
May. 21st, 2008 05:42 am (UTC)
please no...
This is exactly the reason I switched from KDE to Gnome. KDE (and Konqueror) did excessive fsyncs, which killed the performance on laptop drives.

I recently updated Liferea 1.4.14. Suddenly it's very very slow. And no wonder, it also uses SQlite now.
diegocg
May. 21st, 2008 11:54 am (UTC)
I think that this is a (know) ext3 journaling problem. I don't think mozilla is doing anything wrong - I'm using firefox 3 nightlies on Ubuntu and I don't have any problem.

And I don't think it's wrong to prioritize win32 over Linux - arguing that they should have the same priority is like when debian tried to give the same priority to all the architectures. The vast majority of users (x86) got harmed by this policy. Same would happen for Firefox. Just because Mozilla has a sensible policy doesn't means they hate Linux.
(Anonymous)
May. 21st, 2008 12:46 pm (UTC)
> diegocgI - think that this is a (know) ext3 journaling problem. I don't think mozilla is doing anything wrong - I'm using firefox 3 nightlies on Ubuntu and I don't have any problem.

No, it's not, fsync on ext3 will sync the entire buffer in memory, which is bad for performance. That doesn't change the fact that it shouldn't be used like this in the first place though.

http://www.flamingspork.com/projects/libeatmydata/ = solution, a bad one, but it works. I use it for a few other things (pidgin for example does a lot of fsync for absolutely no reasons, cached buddylist icons for example..), I couldn't get it working with firefox though for some reason.
Chris C [blondechris.com]
May. 21st, 2008 01:46 pm (UTC)
DOING. IT. WRONG.
Every time someone acts like a dick about something like this, more people waste time fueling the fire. I imagine some idiot (like one of the Epiphany advocates above, who always know better than the Mozilla Foundation) will end up sending it to Slashdot and then another hour of people's lives goes on the fire.

Not one human being is going to use a 3.0.0 Linux tarball. Quite a few human beings will use a 3.0.0 Windows binary. So the exact reason for punting it is correct; that breaking process for the sake of a binary that nobody will use isn't worth it.

"PR disaster". I wonder whose fault that will be. MoCo's, for following a reasonable release process, or the same attention whores that whine *every time* a new Mozilla release happens?

- Chris
jasondclinton
May. 21st, 2008 02:47 pm (UTC)
Re: DOING. IT. WRONG.

Except that the problem, of course, is that Mozilla has punted the responsibility to distributions. There isn't a known-good patch, yet. Instead of prioritizing finding a solution, the policy is: pass the buck downstream.

The sick irony here is that downstream diverging from upstream is exactly what the entire Iceweasel debacle was and continues to be all about.



Edited at 2008-05-21 02:47 pm (UTC)
Re: DOING. IT. WRONG. - Andrew Conkling [myopenid.com] - May. 21st, 2008 04:26 pm (UTC) - Expand
Re: DOING. IT. WRONG. - davidnielsen.wordpress.com - May. 21st, 2008 07:31 pm (UTC) - Expand
Re: DOING. IT. WRONG. - Chris C [blondechris.com] - May. 21st, 2008 04:27 pm (UTC) - Expand
Re: DOING. IT. WRONG. - (Anonymous) - May. 21st, 2008 07:17 pm (UTC) - Expand
(Anonymous)
May. 22nd, 2008 12:19 am (UTC)
Beta...?
As with a lot of people, I've used Beta 5 and RC 1 on Ubuntu. No patch/reduction applied, as far as I know, and it runs fine. I love it; if anything, it runs faster than FF 2.
(Anonymous)
May. 22nd, 2008 09:55 pm (UTC)
Re: Beta...?
youre wrong. read the bug again. ubuntu already uses the patch. i use the nightlies and while it works perfectly fine in windows and osx its unbearably slow in linux.

and like jason said ppl might move away from ff. i for one did not know why ff is so insanely slow in lin, nor did i care enough to search why.

i just dropped it altogether and went back to opera. you could ask why i didnt go back to ff2 and the answer is rather simple as well. im not so much the fan of slow leakware.
Re: Beta...? - (Anonymous) - May. 23rd, 2008 08:16 am (UTC) - Expand
(Anonymous)
May. 22nd, 2008 02:30 am (UTC)
Welll.....
I'm using Firefox 3 RC1 under wine and it works great. Better than native. Is that what you're upset about? I suppose that is lame of mozilla....
jangid
May. 22nd, 2008 04:18 am (UTC)
I think that is OK as long as they are freeing users from platform locking. When more than 80% users will be using Firefox then they can decide which platform to support and not run around MS Windows.
bitobear
May. 22nd, 2008 05:45 am (UTC)
Bad filesystem modes...
I use the firefox 3 RC 5 on my laptop and never see a pause.

Having pursued the bug report I think this is because I always tune my ext3 file systems to use journal_data mode. Any laptop and most desktops should be set this way anyway.

The fsync() calls won't tie up the system trying to flush the whole file system contents just to flush the journal. fsync commits to the journal and you are done.

This isn't a firefox bug, its a linux feature.
(Anonymous)
May. 22nd, 2008 06:33 pm (UTC)
Re: Bad filesystem modes...
Isn't Writeback mode is the fastest mode because it does the least amount of journaling?


(Anonymous)
May. 22nd, 2008 06:58 am (UTC)
Suggestion
Would it be much work to add a db abstraction layer to Foxy?
Then I could use my existing MySql install which has no issues on ext3.

I think as offline web2.0 apps propagate this flexibility will be more in demand, particularly from enterprise users.
vakhitov
May. 22nd, 2008 10:10 am (UTC)
I have a problem on Windows that i think is related. Quite often when I type one or two letters into the not-so-awesomebar, it hangs for some tens of seconds thrashing the HD and not allowing me to do anything in Firefox...
(Anonymous)
May. 22nd, 2008 04:33 pm (UTC)
patched versions?
> Most users will be getting their updates from the distributions,
> not Mozilla directly. And I've talked with most of them and most
> of them are waiting for a patch before shipping/updating

Remember when Debian tried this and Mozilla revoked their use of the Firefox name and logos? How are other distros supposed to apply patches without going afoul of Mozilla's hardcore stance??
(Anonymous)
May. 22nd, 2008 05:19 pm (UTC)
I think this bug has already been released into the wild. Ever since I updated Ubuntu to 8.04, Firefox 3 Beta 5 has been killing my performance. It doesn't always happen, and sometimes only happens after I've had a lot of tabs open for a while (22+ tabs usually, across 2-3 browser windows). I thought it might be some web pages loading up troublesome flash ads, but the performance sometimes doesn't truly return until I close Firefox (or a good number of the tabs). And my file system is running ext3, for the record.

Sticking with Ubuntu and Firefox of course... eagerly awaiting a patch.

Grrrr...

Michael Duffy
grahammacleod.blogspot.com
May. 23rd, 2008 06:55 am (UTC)
I'm happy enough with FF2 for the time being until this is fixed. It's fast enough for me and any difference in speed, in my opinion, would be negligible.
(Anonymous)
May. 23rd, 2008 07:15 pm (UTC)
performance killer
i've been on ff b5 and now rc1 on ubuntu 7.10.

i had massive performance problems on my dated-ish laptop until i deactivated the phishing check

the described delay may nonetheless occur at times. i'm glad i'm not the only one experiencing this ;)
edoloughlin
May. 26th, 2008 11:01 am (UTC)
Temporary workaround using ext2
I moved all the sqlite files in my profile to a temporary ext2 filesystem mounted from a file. It seems to help - no freezes in the last couple of days (just the usual Flash-related crashes). I've posted instructions here for those interested. You'll have to put the pieces back together when there's a fix available but it's just a matter of copying the files back...
Page 1 of 2
<<[1] [2] >>
( 57 comments — Leave a comment )

Profile

color, uphair, smile
jasondclinton
Jason D. Clinton

Latest Month

September 2011
S M T W T F S
    123
45678910
11121314151617
18192021222324
252627282930 

Tags

Powered by LiveJournal.com
Designed by Tiffany Chow