Re: RE: Estimating Size of Database

Mitesh Shah writes:

> One follow up question. In the example given, it says there are 36
> bytes for each row header and 4 bytes for each pointer to a tuple. I'm
> not sure where these numbers (36 and 4) are coming from. Are they
> standard for *every* table? If my table has more than just two
> integers, for example, will each row header be more than 36 bytes?

More or less. Quoth the source:

typedef struct HeapTupleHeaderData
{
Oid t_oid; /* OID of this tuple -- 4 bytes */

CommandId t_cmin; /* insert CID stamp -- 4 bytes each */
CommandId t_cmax; /* delete CommandId stamp */

TransactionId t_xmin; /* insert XID stamp -- 4 bytes each */
TransactionId t_xmax; /* delete XID stamp */

ItemPointerData t_ctid; /* current TID of this or newer tuple */
int16 t_natts; /* number of attributes */

uint16 t_infomask; /* various infos */

uint8 t_hoff; /* sizeof() tuple header */

/* ^ - 31 bytes - ^ */

bits8 t_bits[MinHeapTupleBitmapSize / 8];
/* bit map of NULLs */

/* MORE DATA FOLLOWS AT END OF STRUCT */
} HeapTupleHeaderData;

Most of the fields are for maintaining information required for
transaction rollback and multi-version concurrency control, in case you
can't quite decode it. ;-)

--
Peter Eisentraut peter_e(at)gmx(dot)net http://yi.org/peter-e/