Here's the feedback on replacing VACUUM FULL with VACUUM REWRITE:
a) To date, I have yet to hear a single person bring up an actual
real-life use-case where VACUUM FULL was desireable and REWRITE would
not be. Lots of people have said something hypothetical, but nobody has
come forward with a "I have this database X and several times Y
happened, and only FULL would work ...". This makes me think that there
very likey are no actual use cases where we need to preserve FULL.
b) Several people have strongly pushed for a phased removal of FULL over
more than one PG version, with a warning message about depreciation.
c) Vivek had some points about required implementation:
"However, there still must be a way to compact the tables that is mvcc
safe. From what I have read and recall, cluster is not. Thus, the vacuum
rewrite would be a mandatory feature (or cluster could be made mvcc safe)."
Is Vivek correct about this? News to me ...
PostgreSQL Experts Inc.
pgsql-hackers by date
|Next:||From: Kevin Grittner||Date: 2009-09-16 18:20:27|
|Subject: Re: Feedback on getting rid of VACUUM FULL|
|Previous:||From: Robert Haas||Date: 2009-09-16 17:54:33|
|Subject: Re: WIP: generalized index constraints|