Skip to content

Latest commit

 

History

History
134 lines (106 loc) · 4.99 KB

README.md

File metadata and controls

134 lines (106 loc) · 4.99 KB

pglogical_ticker

Periodically churn an in-replication table using a Postgres background worker to get replication time from standpoint of pglogical provider.

How to use pglogical_ticker

Installation

The functionality of this requires postgres 9.5+ and a working install of pglogical.

DEB available on official PGDG repository as postgresql-${PGSQL_VERSION}-pglogical-ticker. See installation instruction on https://wiki.postgresql.org/wiki/Apt

Or to build from source:

make
make install
make installcheck # run regression suite

Although not strictly required, to get access to the configuration settings of pglogical_ticker and to auto-launch the ticker on server restart or a soft crash, add pglogical_ticker to shared_preload_libraries in your postgresql.conf file:

shared_preload_libraries = 'pglogical,pglogical_ticker' #... and whatever others you already may have

Once installed, simply run this on the provider and all subscribers:

CREATE EXTENSION pglogical_ticker;

Deploy ticker tables

Deploy the ticker tables. Run this command on the provider only, which will use pglogical.replicate_ddl_command to send to subscriber.

SELECT pglogical_ticker.deploy_ticker_tables();

This will add a table for each replication_set.

For cascading replication, you can add existing tables to another replication set, that belonging to your 2nd tier subscriber. You pass that set_name to the deploy function like so:

SELECT pglogical_ticker.deploy_ticker_tables('my_cascaded_set_name');

Add the ticker tables to replication:

SELECT pglogical_ticker.add_ticker_tables_to_replication();

Again, to add ticker tables to a cascaded replication set:

SELECT pglogical_ticker.add_ticker_tables_to_replication('my_cascaded_set_name');

For any more custom needs than this, you can freely add ticker tables to replication sets as you choose to manually.

You can manually try the tick() function if you so choose:

SELECT pglogical_ticker.tick();

Configuration

This is only supported if you have added pglogical_ticker in shared_preload_libraries as noted above.

  • pglogical_ticker.database: The database in which to launch the ticker (we currently have no need to support multiple databases, but may add that feature at a later time). The ticker will only auto-launch on restart if this setting is configured.
  • pglogical_ticker.naptime: How frequently the ticker ticks - default 10 seconds
  • pglogical_ticker.restart_time: How many seconds before the ticker auto-restarts, default 10. This is also how long it will take to re-launch after a soft crash, for instance. Set this to -1 to disable. Be aware that you cannot use this setting to prevent an already-launched ticker from restarting. Only a server restart will take this new value into account for the ticker backend and prevent it from ever restarting, if that is your desired behavior.

Launching the ticker

As of version 1.4, the ticker will automatically launch upon server load if you have pglogical_ticker in shared_preload_libraries.

Otherwise, this function will launch the ticker, only if there is not already one running:

SELECT pglogical_ticker.launch();

/**
It is better to use the following function instead, which automatically checks if
the system should have a ticker based on tables existing in replication.
(this assumes you don't want a replication stream open with no tables).
**/
SELECT pglogical_ticker.launch_if_repset_tables();

The background worker launched either by this function or upon server load will run the function pglogical_ticker.tick() every n seconds according to pglogical_ticker.naptime.

Be sure to use caution in monitoring deployment and running of these background worker processes.

To view all ticker tables at once, there are functions you can run to view tables on both provider and subscriber:

Tables by replication set (provider):

SELECT * FROM pglogical_ticker.all_repset_tickers(); 

Tables by pglogical subscription (subscriber):

SELECT * FROM pglogical_ticker.all_subscription_tickers(); 

For Developers

Help is always wanted to review and improve the BackgroundWorker module. It is directly based on worker_spi from Postgres' test suite.

It could be improved to use a different username. I would also like it to be written so as to prevent any possibility of launching more than one worker, which currently is only done through the exposed function in the docs launch().

As of 1.4, I'm also interested in allowing a clean shutdown with exit code 0, as well as (if safe enough) searching databases for the ticker as opposed to having to configure pglogical_ticker.database.

The SQL files are maintained separately to make version control much easier to see. Make changes in these folders and then run pglogical_ticker-sql-maker.sh to build the extension SQL files. This script will need modification with any new release to properly build new extension files based on any new changes.