In PostgreSQL terminology, an LSN (Log Sequence Number) is a 64-bit integer used to determine a position in WAL (Write ahead log), used to preserve data integrity. Internally in code, it is managed as XLogRecPtr, a simple 64-bit integer. An LSN is represented with two hexadecimal numbers of 8 digits each separated with “/”. For example, when looking on server what is the current position of WAL, you can do something like that:
=# SELECT pg_current_xlog_location(); pg_current_xlog_location -------------------------- 16/3002D50 (1 row)
The first hexadecimal number corresponds to a logical xlog file with 256 segments of 16MB (total of 4096MB), incremented once segments are filled. Segments are represented by the second hexadecimal number, as an offset of the logical xlog file, and can go up to FFFFFFFF (FF000000 up to 9.3, for a maximum of 4080MB for a single logical xlog file). Up to 9.3, all the functions using LSN have been using as a substitute “text” to represent an LSN number, so all the functions using LSN numbers had to transform manual the output into a text before sending back the result to client. 9.4 improves the situation by using a [datatype dedicated to LSN] (http://www.postgresql.org/docs/devel/static/datatype-pg-lsn.html), introduced by this commit and called pg_lsn:
commit 7d03a83f4d0736ba869fa6f93973f7623a27038a Author: Robert Haas <firstname.lastname@example.org> Date: Wed Feb 19 08:35:23 2014 -0500 Add a pg_lsn data type, to represent an LSN. Robert Haas and Michael Paquier
This has been completed later on by another patch switching a couple of system functions to use this datatype. Here are the functions:
Even if this is actually not backward-compatible, it is thought that this should not have much consequence on user applications. Using a dedicated datatype has as well several advantages:
- Internal PostgreSQL code (extensions as well, I saw that many times) does not need to manipulate anymore internally XLogRecPtr to change it into an LSN.
- Validation of LSN format (2 8-digit hexadecimal numbers separated by “/”) is now inside the data type itself.
- External utilities can really take advantage of that.
- Basic operators (=, !=, <, >, <=, >=, -) are included in the data type.
To finish, here are a couple of things doable now when manipulating LSNs, like validation of LSN record format:
=# SELECT 'G/0'::pg_lsn; ERROR: 22P02: invalid input syntax for type pg_lsn: "G/0" LINE 1: SELECT 'G/0'::pg_lsn; ^ LOCATION: pg_lsn_in, pg_lsn.c:41
Or some simple arithmetic operations:
=# SELECT pg_current_xlog_location() = pg_current_xlog_location() as diff; diff ------ t (1 row) =# SELECT pg_current_xlog_location() != pg_current_xlog_location() as diff; diff ------ f (1 row)
The operator “-“ reduces the interest of pg_xlog_location_diff, but it needs to be kept for backward compatibility purposes (actually what this does now is only calling pg_lsn_mi, the function of type pg_lsn called behind operator “-“). Also, when using “-“, the difference in bytes between two LSN positions is calculated:
=# \dfS+ pg_lsn_mi List of functions -[ RECORD 1 ]-------+----------------------------- Schema | pg_catalog Name | pg_lsn_mi Result data type | numeric Argument data types | pg_lsn, pg_lsn Type | normal Security | invoker Volatility | immutable Owner | easter_egg Language | internal Source code | pg_lsn_mi Description | implementation of - operator =# \x Expanded display (expanded) is off. =# SELECT pg_current_xlog_location() - '0/3000000' as diff; diff ------- 12896 (1 row)
This is going to make the code of some tools, like recovery/backup things, much more simplified…