Re: 2-phase commit

[ sorry to repost this, but I didn't receive my mail back... Anything
wrong with the mailserver ? ]

I am involved in a project of open-source, PostgreSQL-backed,
buzzword-compliant replication/high availability software that would
act as an SQL « one-to-many » gateway (but still in the design phase
--- this is *not* an announcement :-). Of course, the topic of 2-phase
commit is important to us ; we currently plan to record all write
commands issued during the transaction in an auxiliary table (some
sort of higher-level WAL). First commit phase would then consist in
closing this record and ensuring it can be REDOne in the case of a
crash (UNDO would be done by just rolling back the current
transaction). But this is quite complicated and may require to
serialize all accesses (both read and write) to a given database so as
to guarantee that REDO will yield the very same result.

I understand it would certainly be better and more profitable for
the community if I could help implement 2-phase commit inside
PostgreSQL. But I am not much of a PostgreSQL hacker yet. What do you
think ?

--
<< Tout n'y est pas parfait, mais on y honore certainement les jardiniers >>

Dominique Quatravaux <dom(at)kilimandjaro(dot)dyndns(dot)org>