WIP

pg_pageprep is an extension that is supposed to help to prepare heap pages for migration to 64bit XID page format.

PostgresPro Enterprise page format reqiures extra 20 bytes per page compared to original PostgreSQL in order to support 64bit transaction IDs. The idea behind this extension is to prepare enough space in pages for new format while database is working on vanilla postgres.

make install USE_PGXS=1

or if postgres binaries are not on PATH

make install USE_PGXS=1 PG_CONFIG=/path/to/pg_config

Add pg_pageprep to shared_preload_libraries parameter in postgresql.conf:

shared_preload_libraries='pg_pageprep'

It is required to restart PostgreSQL cluster in order to apply new configuration. After that perform CREATE EXTENSION pg_pageprep; on each database in cluster.

You can add the following parameters to your postgres config:

• pg_pageprep.database - database name, which starter process will use to get databases list (default 'postgres');
• pg_pageprep.role - user name (default 'postgres');
• pg_pageprep.per_page_delay - delay between consequent page scans in milliseconds (default 100ms);
• pg_pageprep.per_relation_delay - delay b/w relations scans in milliseconds (default 1000ms);
• pg_pageprep.per_attempt_delay - delay b/w attempts to scan next relation in case previous attempt was unsuccessful, e.g. there are all relations are already done (default 60s).

python manager.py -d <database> -U <username> <command>

where <database> is a database used to get the list of all databases in cluster, <username> is a user name on whose behalf the script will work and command is one of the following:

• install - creates extension on each existing database, sets pg_pageprep.databases and pg_pageprep.role parameters to config (ALTER SYSTEM). Note that background workers will start automatically after next cluster restart.
• start - starts background workers right away; note that during its works background worker sets fillfactor to 90% (if needed) for every relation it processes;
• stop - stops background workers;
• status - shows information about current workers activity and relations to be processed;
• restore - stop workers (if any) and restore original fillfactor (this is usually need to be done once after pg_upgrade).

Example:

python manager.py -d postgres -U my_username status

Note: If pg_config isn't on your path you'll need to set PG_CONFIG environment variable: PG_CONFIG=/path/to/pg_config python manager.py -d postgres -U my_username status

Perform on every database in cluster.

CREATE EXTENSION pg_pageprep;
SELECT start_bgworker();

Note that during its works background worker sets fillfactor to 90% (if needed) for every relation it processes. To check todo list use pg_pageprep_todo view:

SELECT * FROM pg_pageprep_todo;

The empty list means that everything is done and database is ready for pg_upgrade. To view current workers status use function get_workers_list():

SELECT * FROM get_workers_list();

 database | status 
----------+--------
 postgres | active
(1 row)

Use following function to stop background worker:

SELECT stop_bgworker();

This only stops worker on current database. After pg_upgrade run the following command to restore fillfactors to original values:

SELECT restore_fillfactors()