2012/3/16 Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp>:
> I have a plan to support 'Create index on foreign table' for 9.3. Here
> is my plan.
Very interesting idea, but...
> The index creation is supported for a flat file such as CSV and a remote
> table on a RDB e.g., Postgres using CREATE INDEX.
Why do you limit the target type to those two? How about web
services and non-relational databases? Some web services would
provide id-to-content mapping, and KVSs are obviously accessible by
key. IMHO CREATE INDEX for foreign tables should have general design,
not specific to some kind of FDWs.
> I'd like to build the index physical data file for a flat file using the
> index access method of regular tables (ie btree, hash, gin, gist, and
> spgist) based on the following transformation between the TID and the
> file offset to some data in the file:
>
> block_number = file_offset_to_some_data / BLCKSZ
> offset_number = file_offset_to_some_data % BLCKSZ
Indeed these information would help searching data stored in local
files. But, again, it seems too specific to file-based FDWs. I'd
suggest separating basic general design and implementation by FDWs.
The design you shown here seems indexed-file_fdw to me...
Regards,
--
Shigeru HANADA
