On 11/09/2011 04:05 PM, Dimitri Fontaine wrote:
> The more I think about this problem, the more I think that the reason
> why we still don't have declarative partitioning is that it basically
> sucks.
I think that we don't have it because no one has ever dumped the much
larger than might be expected amount of time into pulling all the pieces
together and smoothing out the rough parts. I don't think there's any
design thinking leap needed over what's already been worked out. Just a
lot of work to get all the edge cases right on the simplest possible
thing that is useful.
The path to reach something that could be considered for commit includes
something like this set of things:
1) Add partitioning catalog support
2) Create new syntax for partitioning that writes to the catalog
3) Decide how to represent partition data in memory
4) Route new INSERTed data into the right place
5) Support moving UPDATEd data into a new partition
6) Handle COPY usefully
The last rev of this submitted was still working through (1) here, i.e.
this review from Robert:
http://archives.postgresql.org/message-id/AANLkTikP-1_8B04eyIK0sDf8uA5KMo64o8sorFBZE_CT@mail.gmail.com
And there's a whole pile of issues I don't think have been fully
explored about even the most basic case. How to handle ALTER to these
structures cleanly, locking, etc.. I don't think it's possible to
design such that you skip a large portion of these details; someone
needs to put some number of spend weeks+ getting them all right instead.
--
Greg Smith 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support www.2ndQuadrant.us