Re: Add visibility map information to pg_freespace.

Hello,

> I'm looking into this patch as a reviewer.

I'd appreciate your time to review.

I've had some suggestions so far,

- I should be cautious in changing existing interface.

You're right. It was somehow gone out of my mind. It might be
better to provide a separate function from the compatibility
view despite the loss of the pertinence to stay in this
extension. However, it is too small to be a standalone
extension.

On the other hand the newly-added-column-to-the-tail could be
said to be harmless for the most cases considering the usage of
this extension, I suppose.

- Historical note is needed in pg_freespace doc.

Agreed, I'll provide documents not only for freespace, but for
other modules I'll touch in this patch later.

- How about pageinspect?

I proposed a simple representation format as a basis for
discussion. Nevertheless, the VM pages has no more structure
than a simple bit string. Given the VM info in pg_freespacemap,
I've come in doubt of the necessity of vm_page_contnets() for
the reason besides the orthogonality in the this extension's
interface (which paid no attention before:-).

- How about pgstattuple?

It could even be said to be meaningful to add the number of
not-all-visible pages or the ratio of it in the total pages..

| postgres=# select * from pgstattuple('t');
| -[ RECORD 1 ]----------------+---------
| table_len | 88711168
| tuple_count | 600001
| tuple_len | 26400044
| tuple_percent | 29.76
| dead_tuple_count | 399999
| dead_tuple_len | 17599956
| dead_tuple_percent | 19.84
| free_space | 33607960
| free_percent | 37.88
+ not_all_visible_page_percent | 23.54

# This column name looks too long, though.

In addition, the discussion above about the stability of the
interface is also applicable to this.

Any suggestions?

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center