Re: Two weeks to feature freeze

>>>>> "Bruce" == Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:

Bruce> Tom Lane wrote:
>> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes: > The question
>> was whether 2PC is useful. The question wasn't if an >
>> unreliable 2PC was useful.
>>
>> My question is whether there is such a thing as reliable 2PC.
>> I sure don't see how you build that.

Bruce> Other databases use 2PC --- are you saying none of them are
Bruce> reliable?

And they use them for both federated read/write (what you refer to as
distributed access through dblink) and for clustered configurations.

I'm not sure if I understand Tom's beef - I think he is concerned
about what happens if a subordinate does not respond to a prepare
message. I would assume that the co-ordinator would not let the commit
go through until it has received confirmations from every
subordinate. The application's commit will remain blocked against the
co-ordinator when this takes place.

That said, I agree that 2PC (and variants) is rather heavy weight when
in widely distributed configurations.

(Although I guess in practice, many people use Presumed Abort and not
vanilla 2PC as PA results in fewer log flushes for read-only
transactions.)

--
Pip-pip
Sailesh
http://www.cs.berkeley.edu/~sailesh