> -----Original Message-----
> From: owner-pgsql-hackers@postgreSQL.org
> [mailto:owner-pgsql-hackers@postgreSQL.org]On Behalf Of Vadim Mikheev
> Sent: Sunday, May 02, 1999 12:23 AM
> To: PostgreSQL Developers List
> Subject: [HACKERS] I'm planning some changes in lmgr...
>
>
> but have no time to do them today and tomorrow -:(.
>
> 1. Add int waitMask to LOCK to speedup checking in LockResolveConflicts:
> if lock requested conflicts with lock requested by any waiter
> (and we haven't any lock on this object) -> sleep
>
> 2. Add int holdLock (or use prio) to PROC to let other know
> what locks we hold on object (described by PROC->waitLock)
> while we're waiting for lock of PROC->token type on
> this object.
>
> I assume that holdLock & token will let us properly
> and efficiently order waiters in LOCK->waitProcs queue
> (if we don't hold any lock on object -> go after
> all waiters with holdLock > 0, etc etc etc).
>
> Comments?
>
First, I agree to check conflicts for ( total - own ) hodling lock of
the target object if transaction has already hold some lock on the
object and when some conflicts are detected,the transaction
should be queued with higher priority than transactions which hold
no lock on the object.
Secondly, if a transaction holds no lock on the object, we should
check conflicts for ( holding + waiting ) lock of the object.
And I have a question as to the priority of queueing.
Does the current definition of priority mean the urgency
of lock ?
It may prevent lock escalation in some cases.
But is it effective to avoid deadlocks ?
It's difficult for me to find such a case.
Thanks.
Hiroshi Inoue
Inoue@tpf.co.jp