155181 - Stalled browser during critical period when opening pages (100% CPU)

Status: VERIFIED WORKSFORME

(Core :: Layout, defect, P2)

Description

[Jacob Vogelstein]

From Bugzilla Helper:
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.0.0) Gecko/20020530
BuildID:    2002053012

This bug appears consistently when I use the tabbed browser interface.  It is
manifested when i go to a new page or open a link in a new tab -- the tab
initially displays "loading" with the circular-arrows and things work ok, but
shortly after displaying the new page's title, the arrows stop moving, the
animation in the mozilla graphic at the top right corner of the screen halts,
and the browser becomes unresponsive to mouse-clicks or anything else until the
new page finishes loading.  Other mozilla windows are also frozen during this
time.  The status line usually (but not always) says "Loading data from <website>"

Reproducible: Always
Steps to Reproduce:
1.Open any page in mozilla
2.Open a new tab
3.Type in a new website in the address line or click on a bookmark
4.Attempt to switch between tabs or other mozilla windows

Actual Results:  The browser (and mail client) are unresponsive during some
period of time between when the page begins to load and finishes loading

Expected Results:  Switched smoothly between tabs and windows and loaded the new
page in the background

I am using the Pinball theme.

Comment 1

[Sander]

*** Bug 155182 has been marked as a duplicate of this bug. ***

Comment 2

[Sander]

possible dupe of bug 148033 ?

Confirming for now. I never saw this with RC3, but after I switched to the trunk
this occasionally has been occuring to me. I've been futilly trying to narrow
down what's causing it; My current suspicions include a large cookperm.txt
(140kb for me) or something (else) image loading / cache related. Particularly
since the problem only seems to occur when first visiting a page. The chrome
becomes completely unresponsive for up to two seconds while loading something in
a new tab; for example, hovering over bookmarks in the personal bookmark doesn't
result in an underline, pages won't scroll, etc.
This behaviour seems to have worsened between 2002061108 and 2002062109 (win98)
(clean installs, no themes running).

I don't think tabbed browsing is the right component for this, but since I don't
know what _is_ the right component, I'll leave it

Comment 3

[paxunix]

I've been seeing this exact behavior too on trunk 2002062908 (Win98), and it's
been like that for at least the last 3? weeks.  It is especially noticeable when
the page being loaded is long.  I see it reliably here:

http://www.kuro5hin.org/story/2002/6/27/63448/0050

and indeed it's the same for any of the stories with many comments on Kuro5hin
and Slashdot.  When it happens, in the back of my mind I'm always thinking that
there are a lot of comments on this particular page and it has always turned out
to be true.  I never notice any difference on subsequent loads of the same page.
 Interestingly, when the chrome has stalled, clicking on other tabs will change
the window's title bar, even though it will be a few seconds before the content
area is updated.  My cookperm.txt is small--182 bytes, and on all the pages I
can recall seeing this behavior on, there haven't been many images.

Comment 4

[Sander]

That page does indeed see this behaviour even when reloading. Some weird things
to notice: Mozilla also stalls when loading the view source window of that page.
And on first load of the kuro5hin, the chrome seemed to respond in intervals -
not doing anything for half a second, reacting haltingly to the past events, and
once more stalling - maybe four or five times in succession.
Could it be the heavy tables causing this? But why then would the same thing
occur on view source?

Comment 5

[jag (Peter Annema)]

So is this really "in tabs"? Based on the original report's 4th step and later
comments I'm going to move this to layout for now, though it might also be parser.

Comment 6

[Jacob Vogelstein]

Searching for "100% CPU" on Bugzilla shows lots of page-specific problems. 
Personally, I have this problem on almost every page that loads. It's not
noticible all the time because some pages load more quickly than others, but
*EVERY PAGE* boosts CPU up to 100% for some variable amount of time.  It
basically makes the browser unusable for me (because it's so annoying). The
problem is reproducible in both Win2k and Linux.

Comment 7

[Jacob Vogelstein]

Also, Jag - You're right.  It has nothing to do with tabs.  I thought it did at
first because it's more noticible when switching tabs -- you can see the little
arrows chasing each other freeze up and restart as the cpu goes to 100% and the
page loads. Turning on HTTP Pipelining doesn't seem to help.  Problem still
exists (although seems to be improved) in 1.1. I think the problem belongs in
Parser, and I'm running Linux almost full-time now, so I'm changing OS to linux.

Comment 8

[Andrew Schultz]

Windows + Linux = All
no crash, no hang, not critical

reassign

Comment 9

[Kevin McCluskey (gone)]

Patches for 
http://bugzilla.mozilla.org/show_bug.cgi?id=165039
http://bugzilla.mozilla.org/show_bug.cgi?id=157144

Should have helped or fixed this issue on WIN32.

I can not reproduce the problem using 2002103004 trunk build on WinXP.

Jacob: Do you see the problem on recent trunk build?

Comment 10

[Jacob Vogelstein]

Kevin,

Yes, those patches do seem to help on my Win2k box.  Build 2002110808 appears to
function much better than my previous version. But as of 2002110304 on Linux,
the problem still manifests.  One easy way to check (in addition to visiting the
URL in the bug report) is to go to 

http://sports.espn.go.com/nba/scoreboard

and change the "refresh" option to 30 seconds.  Then by browsing in other tabs
or watching the CPU meter or whatever, it is immediately apparent.

Changing OS back to Linux. 

Comment 11

[Carine Held]

Reproduced in 12/23 Linux Trunk. There is no crash, but it does take extremely
long to render the page: 17 seconds. 

Comment 12

[Carine Held]

This just is a very big page with a lot of tables. (There are over 4000 lines of
code). Explorer took even longer to load the page: 25 seconds!

Comment 13

[Jacob Vogelstein]

I have discovered the source of this problem and I think this bug should be
closed.  See http://bugzilla.mozilla.org/show_bug.cgi?id=188006.  Thanks for all
of your help on this issue.

Comment 14

[Boris Zbarsky [:bzbarsky]]

Marking worksforme per comment 13.  Jacob, thanks for digging into this!

Comment 15

[benc]

V. I've been looking at the other bug.