Re: Deadlock in Index (where index is B+ tree) Locking with Intention Locks
Date: Fri, 13 Dec 2002 17:48:38 GMT
Message-ID: <H72K51.CMt_at_news.boeing.com>
"Mikito Harakiri" <mikharakiri_at_ywho.com> wrote in message
news:ew4K9.7$cw5.102_at_news.oracle.com...
> "Novice" <6tc1_at_qlink.queensu.ca> wrote in message
> news:b80e4a77.0212120907.7cfb153d_at_posting.google.com...
> > And let's imagine the following scenario:
> > T1 - IX(root) IX(B) IX(D) IX(H) X(P) X(H) X(D) X(B) U(P) U(H) U(D)
> > U(B) U(root)
> > T2 - IX(root) IX(B) IX(D) IX(H) X(O) X(H) X(D) X(B) U(O) U(H) U(D)
> > U(B) U(root)
> > T3 - IX(root) IX(B) IX(D) IX(G) X(N) X(G) X(D) X(B) U(N) U(G) U(D)
> > U(B) U(root)
> >
> > There is no permutation I could find of the above that would result in
> > deadlock.
> >
> > If you can think of a more simple set of transactions that would
> > result in deadlock I would also be interested in that.
>
> Is your protocol deadlock safe? Maybe. But it's not very interesting,
since
> all your transactions are serialized on the root node. You need to unlock
> root node early, and then watch problem to happen!
>
>
Unlocking the root node early would certainly provide problems if block/page
splits propogated up the hierarchy to the root...doesn't seem like a viable
alternative.
Received on Fri Dec 13 2002 - 18:48:38 CET