670953 - Crash in mozilla::storage::Connection::getFilename when running the memory reporter

Status: RESOLVED DUPLICATE of bug 662989

(Core :: SQLite and Embedded Database Bindings, defect)

Description

[Mounir Lamouri (:mounir)]

Attachment: Backtrace

See attached backtrace.

Happened when clicking on GC + CC button.

Comment 1

[Shawn Wilsher :sdwilsh]

Version information?  I thought this was already fixed a while ago...

Comment 2

[Nicholas Nethercote [inactive]]

Yeah, bug 654573 fixed the crash that was happening after visiting about:permissions.

Bug 662985 is still open -- do you have Test Pilot installed?

And I need to get back to bug 662989 to avoid this once and for all...

Comment 3

[Mounir Lamouri (:mounir)]

(In reply to comment #2)
> Yeah, bug 654573 fixed the crash that was happening after visiting
> about:permissions.
> 
> Bug 662985 is still open -- do you have Test Pilot installed?

No. I got that crash on a dev-only profile so no extension.
And I was using trunk. Probably yesterday's tip.

Comment 4

[Nicholas Nethercote [inactive]]

Huh.  There must be another async connection that isn't being asyncClosed.

Comment 5

[Jonathan Protzenko [:protz]]

I'm now hitting this crash reliably in Thunderbird whereas I'd never seen it before. It came after bumping my mozilla-central revision. Was there any change in the area over the last couple days?

Comment 6

[Jonathan Protzenko [:protz]]

Amendment: the crash seems to happen only when telemetry is enabled.

Comment 7

[Nicholas Nethercote [inactive]]

(In reply to comment #6)
> Amendment: the crash seems to happen only when telemetry is enabled.

Telemetry triggers the bug because it consults the memory reporters, but neither telemetry or the memory reporters are at fault.  An async DB connection somewhere isn't being asyncClosed as it should (see http://blog.mozilla.com/nnethercote/2011/06/09/asynchronous-database-connections-must-be-closed-with-asyncclose/ for details).

I'm about to land the fix for bug 662985 which should stop this crashing.  If it happens, there'll still be a leak, and it should show up in about:memory as an entry like "places.sqlite-LEAKED".  I'll mark this as a dup of 662989.

Comment 8

[Nicholas Nethercote [inactive]]

(In reply to comment #7)
>
> I'm about to land the fix for bug 662985 which should stop this crashing. 
> If it happens, there'll still be a leak

To clarify:  if an async DB connection is GC'd without being asyncClosed first, there'll still be a leak.