Re: Sequence Access Method WIP

Hi,

I rewrote the patch with different API along the lines of what was
discussed.

The API now consists of following functions:
sequence_alloc - allocating range of new values
The function receives the sequence relation, current value, number of
requested values amdata and relevant sequence options like min/max and
returns new amdata, new current value, number of values allocated and
also if it needs wal write (that should be returned if amdata has
changed plus other reasons the AM might have to force the wal update).

sequence_setval - notification that setval is happening
This function gets sequence relation, previous value and new value plus
the amdata and returns amdata (I can imagine some complex sequence AMs
will want to throw error that setval can't be done on them).

sequence_request_update/sequence_update - used for background processing
Basically AM can call the sequence_request_update and backend will then
call the sequence_update method of an AM with current amdata and will
write the updated amdata to disk

sequence_seqparams - function to process/validate the standard sequence
options like start position, min/max, increment by etc by the AM, it's
called in addition to the standard processing

sequence_reloptions - this is the only thing that remained unchanged
from previous patch, it's meant to pass custom options to the AM

Only the alloc and reloptions methods are required (and implemented by
the local AM).

The caching, xlog writing, updating the page, etc is handled by backend,
the AM does not see the tuple at all. I decided to not pass even the
struct around and just pass the relevant options because I think if we
want to abstract the storage properly then the AM should not care about
how the pg_sequence looks like at all, even if it means that the
sequence_alloc parameter list is bit long.

For the amdata handling (which is the AM's private data variable) the
API assumes that (Datum) 0 is NULL, this seems to work well for
reloptions so should work here also and it simplifies things a little
compared to passing pointers to pointers around and making sure
everything is allocated, etc.

Sadly the fact that amdata is not fixed size and can be NULL made the
page updates of the sequence relation quite more complex that it used to
be. There are probably some optimizations possible there but I think the
patch is good enough for the review now, so I am adding it to October
commitfest.

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services