Re: [HACKERS] generic LONG VARLENA

[Charset iso-8859-1 unsupported, filtering to ASCII...]
>
> > I am excited about the long data type. This is _the_ way to do long
> > data types. Have any of the commercial databases figured out this way
> > to do it. I can't imagine a better system.
>
> The commercial db's usually make the dba decide on a per column basis
> whether the value is stored inside the table or in an extra space
> (blobspace,lobspace ...). They all have propietary syntax for this.
> (I would probably like it configurabe, whith some reasonable default)
> It is usually available for the text/byte and user defined datatypes.
> In PostgreSQL the array types come to mind.
>
> What I think would be good is, if you could avoid the need for an index on
> the _LARGE_.. table.
> My Idea would be to store an xtid of the first lob page slot in the user
> table,
> and have an xtid pointer to the next lob page slot in it, and so on.
> That way you could avoid indices on the LARGE table.
> SnapshotAny() would also see the correct long, since an updated value would
> get a new xtid anyway. No need to use up an extra oid.

You are getting the data in 8k chunks, so it shouldn't be bad. I think
using ctid is overly complex and makes vacuum fragile on that table.
Better to use the tools we have like indexing and the standard tuple
layout code. Custom solutions like ctid are better off only when we see
seriouis performance problems and can't resolve them any other way.

For example, having an expanded tuple cache will give us great speed
improvements.

--
Bruce Momjian | http://www.op.net/~candle
maillist(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026