Re: Concurrent MERGE

Josh Berkus <josh@agliodbs.com> writes:
> Yes; it's a major project.  Our detector works pretty well for deadlocks
> which are 2-process locks or even several processes all locking against
> the same first process. However, triangular and quadralateral deadlocks
> (which I've seen more than once) it completely cannot handle,

Hm?  Please explain what you're talking about.

> and some
> types of activity which can cause deadlocks (like autovacuum or DDL
> activity) also seem to be outside its purview.

There's some known issues with deadlocks involving LWLocks as well as
regular locks, which I agree aren't fixable without some significant
rework.  But I don't believe there's anything fundamentally wrong with
the deadlock detector --- the real problem there is stretching LWLocks
beyond their design intention, namely to be used only for situations
where deadlock is impossible.

> Also, even where the deadlock detector does its job, it's still the most
> expensive type of serialization failure:

Well, that's certainly true --- you don't want deadlock recovery to be
part of any high-performance path.

> 3. deadlocks can, and do, result in cancelling several transactions
> instead of just one; there is no "winner" which is allowed to complete.

Not sure I believe this either; one deadlock kills one transaction.
If you lose multiple transactions I think you had multiple deadlocks.
        regards, tom lane