Re: Wedged repositories [#6251]

On Tue, Jul 23, 2002 at 04:47:41PM -0400, Keith Bostic wrote:
>
> This really, really sounds like a thread of control killed while
> holding a data resource mutex, and recovery not being run.
>
> I'd say it's time to check to see if that can happen in the code.
>

I don't think svn does any signal handling at all. This means
that the "local" repository access method will pretty much *always*
leave a mutex held if the client is interrupted with ^C.

We could add a SIGINT handler, but SIGKILL would still leave the
repository hung. It seems to me that we need to get the
auto-recover code working correctly in order to fix the SIGKILL
case. That would also fix the SIGINT case, I think, so then
we wouldn't really need the SIGINT handler at all.

Does this make sense? i.e., should we just focus on getting
the call-sequence right for auto-recovery?

> Regards,
> --keith

Highest Regards,

--ben

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Tue Jul 23 23:35:33 2002