933061 - Install debug packages on Linux build machines for symbolic stack traces for test crashes

Status: RESOLVED INCOMPLETE

(Infrastructure & Operations Graveyard :: CIDuty, task)

Description

[Mats Palmgren (inactive)]

Would it be possible to install libc/x11/glib/gtk debug packages to
get symbols in the generated stack traces?

For example: the attached stack trace in bug 932600 would be much more
helpful if it had symbols instead of "libc-2.15.so + 0xdc37f"
and "libxcb.so.1.1.0 + 0x1fff3" etc.

Comment 1

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Found in triage.

Comment 2

[(not currently active) Ted Mielczarek]

This won't help, because we use Breakpad to get the stack traces, and right now it only uses Breakpad-text-format .sym files. I had a plan to write a custom minidump_stackwalk that could load native debug symbols directly (code is here: http://hg.mozilla.org/users/tmielczarek_mozilla.com/minidump-stackwalk/file/31175cf72d19 ) but I didn't finish it.

Comment 3

[Mats Palmgren (inactive)]

Can we use addr2line instead?  (it works fine for me when symbolizing ASAN dumps)

Comment 4

[Mats Palmgren (inactive)]

Speaking of ASAN dumps... is there reason we get symbols in some logs:
https://tbpl.mozilla.org/php/getParsedLog.php?id=29972503&tree=Mozilla-Central
but not others?
https://tbpl.mozilla.org/php/getParsedLog.php?id=29117031&tree=Mozilla-Central
(both are 64-bit Ubuntu on m-c running mochitest-1)

Comment 5

[Karl Tomlinson (:karlt)]

This may be useful for stack traces from assertions.

Comment 6

[Chris Cooper [:coop] (he/him)]

Ted: given comment #2, comment #3, and comment #5, is there sufficient value in releng doing this?

Comment 7

[(not currently active) Ted Mielczarek]

No, this wouldn't actually help us with anything in our current testing setup AFAIK. We've never quite figured out a sane way to get useful symbols for system libraries into our test harness crashes.

Comment 8

[Chris Cooper [:coop] (he/him)]

(In reply to Ted Mielczarek [:ted.mielczarek] from comment #7)
> No, this wouldn't actually help us with anything in our current testing
> setup AFAIK. We've never quite figured out a sane way to get useful symbols
> for system libraries into our test harness crashes.

OK, marking as INCOMPLETE until/if we figure out the symbols piece.