Re: [HACKERS] Transaction logging

>
> Your getting me excited about 6.5. Is there a projected release date for
> 6.5? Is there any information on transaction logging? Is there anything I
> can do to help? I am curious about these items because they will make my
> life much easier in the upcoming months as I migrate my application to
> Postgres. Working around these could be very difficulty or near impossible.

I've spent some time thinking about transaction log. The
first idea was to log queries and in some way. But I had to
drop that approach because there are functions (and users
could have written threir own ones too), that don't return
the same result when the database later get's rolled forward
(e.g. anything handling date's/times). And OTOH an
application could SELECT something from the database that
maybe got created by a sequence, and uses this value then in
another INSERT. But when recovering the database, it isn't
guaranteed that all the data will get the same sequences
again (race conditions in concurrent queries). How should the
transaction log now know that this one constant value in the
query must be substituted by another value to ensure
referential integrity? Absolutely impossible.

So the only way I see is to use some sort of image logging
from inside the heap access methods. Would be much more
tricky to dump and recover.

Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#======================================== jwieck(at)debis(dot)com (Jan Wieck) #