Sunday, 1 June 2014

Chapter 6 part 2 - Pages

Pages

The datafiles are organized as array of fixed length elements called pages. The default size is 8k. Pages of a table's datafile are called heap pages to distinguish from the index pages which differs from the former only for the special space present in the page's end. The figure 6.1 shows an index page structure. The special space is small area in the page's end used to track down the index structure. For example a B-tree index stores in the special space the pointers to the leaf pages.




Figure 6.1: Index page
The page starts with a header 24 bytes long followed by the item pointers, usually 4 bytes long. In the page's bottom are stored the actual items, the tuples.

The item pointers is an array of pairs, offset and length, pointing the actual tuples in the page's bottom. The tuples are put in the page's bottom and going backwards to fill up all the available free space.

The header contains the page's generic space management informations as shown in figure 6.2.




Figure 6.2: Page header
  • pd_lsn identifies the xlog record for last page's change. The buffer manager uses the LSN for WAL enforcement. A dirty buffer is not dumped to the disk until the xlog has been flushed at least as far as the page's LSN.
  • pd_checksum stores the page's checksum if enabled, otherwise this field remain unused
  • pd_flags used to store the page's flags
  • pg_lower offset to the start of the free space
  • pg_upper offset to the end of the free space
  • pg_special offset to the start of special space
  • pd_pagesize_version page size and page version packed together in a single field.
  • pg_prune_xid is a hint field to helpt to determine if pruning is useful. It's used only on the heap pages.
The pd_checksum field substitute the pd_tli field present in the page header up to PostgreSQL 9.2 and used to track the xlog record across the timeline id. The page's checksum can be enabled only when the data area is initialised with initdb and cannot be disabled later.


The offset fields, pg_lower, pd_upper and the optional pd_special, are 2 bytes long, this means PostgreSQL can only support pages up to 32KB.


The page version was introduced with PostgreSQL 7.3. For the prior releases the page version is arbitrarily considered 0; PostgreSQL 7.3 and 7.4 had the page version to 1; PostgreSQL 8.0 used the version 2; PostgreSQL 8.1 and 8.2 used version number 3; From PostgreSQL 8.3 the page's version number is 4.

No comments:

Post a Comment