176886 - stack signatures from talkback on linux are "missing"

Status: RESOLVED FIXED

(Core Graveyard :: Talkback Client, defect)

Description

[Andrew Schultz]

looking at the topcrashers at the URL, there are 79 crashes on linux from three
different stack signatures:
MISSING-SIGNATURE (67), 0x0000000b (11) and 0x00000004 (1)
the 0x0000000b stack is only a single frame and so it's as useless as
"MISSING-SIGNATURE"

I don't know if this is a client-side or server-side problem, but it is quite
bad.  Those "MISSING-SIGNATURE" crashes are supposed to contain useful info to
help diagnose crashes, but are completely useless (except for comments/URL). 
Submitting a linux crash bug without a debug build becomes an exercise in futility.

Comment 1

[Jan Carpenter]

MISSING-SIGNATURE happens mostly on Linux, but also happens a little on Mac, and
Windows.  

Don't know if it's significant, but I looked at the detailed "Machine State -
Overview" on several bbid's and they all had the same Trigger Reason:

Trigger Reason:   SIGSEGV: Segmentation Fault: (signal 11)

Also, the last "Machine State - Environment" of them all was /mozilla-bin

Comment 2

[Jan Carpenter]

Attachment: Attaching today's crash info as an example

Comment 3

[Andrew Schultz]

My brain is unable to process the following pieces of data:

bug 184045 comment 3   (linux talkback works)

http://ftp.mozilla.org/pub/data/crash-data/Trunk-topcrashers.html  (linux
talkback doesn't work)

Comment 4

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Is there variation between kernel versions or libc versions or something like that?

Comment 5

[Andrew Schultz]

Attachment: list of OSs with MISSING-SIGNATURE

taken from
http://ftp.mozilla.org/pub/data/crash-data/seamonkey-crash-analysis-detailed.txt


looks like it doesn't discriminate

Comment 6

[Bradley Baetz (:bbaetz)]

Is there a particular set of buildids which starts this? Did anyone change the
talkback stuff in the comm tree?

Comment 7

[Shiva Thirumazhusai]

Missing Signature has been an issue for a long long time(from the Introduction
of Talkback usage). It used to be ""(NULL values in the stack signature and
Stack Trace). Six months back we started replacing ""(NULL values)in the Stack
Signature with "MISSING SIGNATURES"(String) to group and investigate the problem. 

  We haven't done any new changes to Talkback system recently.

Comment 8

[Bradley Baetz (:bbaetz)]

This is probably bug 148453, where profile manager is grabbing the signal to do
lock cleanup. Did init order change, or something, to make that bug suddenly
trigger?

Comment 9

[Tom Mraz]

This should get higher priority as it seems to me that the recent Linux
nightlies (since 1.2) are much much more crash prone than it was in the 1.2
times. I'm sending at least 1-2 crashes a day, sometimes more. But the talkback
data on Linux are almost useless. I think this way the crashes can't be properly
investigated and fixed.

Comment 10

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

FWIW, my archives of talkback data show that the last time we got good data from
Linux builds seems to be the 2002-10-10 14:00 build.

Comment 11

[Andrew Schultz]

checkin for bug 171343 would then appear to be the culprit.

Comment 12

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: a possible workaround

Based on the guess of what caused this, this might work around the problem and
let us get talkback data again.

Comment 13

[Brian Ryner (not reading)]

Bug 171343 fixed a problem on OS X (at the time) where all 3 arguments from the
signal handler weren't getting passed to the next signal handler in the chain
(talkback's), which interferes with getting the stack trace.

We've since turned off symlink-based locking on OS X, so we could probably just
back out bug 171343, too.  I don't think the original issue affects any
platforms other than OS X.

Comment 14

[Shiva Thirumazhusai]

-> [e-mail]

Comment 15

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

It might be nice to have the code available conditionally, though.  If you
disagree, I'm happy to go with backing out, though.

Comment 16

[Brian Ryner (not reading)]

I just don't see a point in making the code debug-only though.  It doesn't buy
us anything.  I'd prefer to go with making the file more readable and backing
out 171343.

Comment 17

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Attachment: backout patch

This patch just does a simple backout.

Comment 18

[Asa Dotzler [:asa]]

Comment on attachment 112742 [details] [diff] [review]
backout patch

a=asa (on behalf of drivers) for checkin to 1.3beta.

Comment 19

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Comment on attachment 112742 [details] [diff] [review]
backout patch

I checked this in to the trunk, 2003-01-27 17:41 PST.  Leaving bug open until
we see whether it fixes the problem.

Comment 20

[Tom Mraz]

It seems to me that in the last two days the talkback server doesn't work at all
so I can't send my recent talkback data.

Comment 21

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

Marking fixed.  Today's talkback report shows two numeric signatures with stacks
showing the problem is bug 184363 and also a single Linux instance of bug
190001.  See the Linux column near the left of
http://ftp.mozilla.org/pub/data/crash-data/Trunk-topcrashers.html .