>From: Aurelien Degremont - Stagiaire <degremont@xxxxxxxxxxx>
>Dean Roehrich a écrit :
>> These are often blocked, but interruptible. Otherwise they're stuck until t
>he
>> HSM application is restarted so it can continue processing events.
>
>Yes, i saw this, no other solutions ?
>
>There's an issue with the DM_EVENT_UNMOUNT event.
>If the HSM application is unavailable, you try to unmount a managed
>filesystem, the filesystem will start to be unmount, the event msg will
>be sent, but not received, so the unmount call will be blocked.
>Unfortunately, the HSM application will never be able to re-register for
>the UNMOUNT event for this filesystem, because this filesystem is no
>longer available as a "loaded and dmapi-compliant" filesystem (we cannot
>get the filesystem handle anymore), but it's also not unloaded. The
>system is quite broken for this FS and i see no solution to solve the
>situation...
>
>Am i clear ? :)
Yes.
There's a comment header for dm_send_unmount_event() that talks about some
issues with this being a synchronous event. The spec still says it
is synchronous:
http://www.opengroup.org/onlinepubs/9657099/toc.htm
This looks like another case where the spec can cause a headache.
Dean
|