[PATCH] lib/tdb: new locking hook.

[Rusty Russell]

On Fri, 28 Aug 2009 01:30:42 pm tridge at samba.org wrote:
> Hi Rusty,
> 
>  > This obsoletes the *mark and *nonblock calls, by allowing the user
>  > to do fancy work replacing, recording or faking fcntl locks.  This
>  > should make ctdb's life easier. 
> 
> that is OK, but I'd like to see this integrated and tested in ctdb
> first, as ctdb is the one that uses the mark locks.
> 
> Ronnie, have you looked at this?

Yes, I was going to port all these patches through to ctdb, then use it...
but it's much cleaner if patches only go one way, so I reconsidered.

>  > We could avoid adding YA open function if we didn't want to allow
>  > non-blocking open, or complete tracking of all lock operations.
> 
> I don't think the new open is needed just yet, unless you have a use
> in mind for it. The way ctdb uses the 'mark' locks is not at open
> time, but when the tdb is already attached and ctdb wants a child
> process to handle the locking. 

Yep, I've looked at what ctdb does.  This patch aims for a superset of what
ctdb needs: eg. we could do non-blocking transaction locks and other fancy
stuff.

But hold off on this patch, once the others are in my life is simpler.  I
can make sure it plays well with ctdb (ie. *mark and *nonblock can be
implemented externally, and I'll do that and test it).

I'm excited about the idea of a proxy locking object: if it can't get the
lock non-blocking, it records all the locks and fails.  Then you get a child
to grab those locks, once it's done the proxy locking object "succeeds" on
those lock requests the next time.

Cheers,
Rusty.