Skip to content

Latest commit

 

History

History
80 lines (45 loc) · 3.1 KB

README.md

File metadata and controls

80 lines (45 loc) · 3.1 KB
Raw

pg_kaboom

pg_kaboom logo PGXN version

Where's the kaboom?! There's supposed to be an Earth-shattering kaboom!

This extension serves to crash postgresql in multiple varied and destructive ways.

But why?

Testing of failover can be hard to do from SQL; some things are nice to expose via SQL functions. This is one of those things.

Is this safe?

Hell, no. Under no circumstances should you use this extension on a production cluster; this is purely for testing things out in a development environment.

We require you to set a GUC variable pg_kaboom.disclaimer to a magic value in order for any of these functions to do anything. That said, there are often times where simulating different breakage scenarios are useful. Under no way are we liable for anything you do with this software. This is provided without warranty and complete disclaimer.

This is your final warning! You will lose data!

Installation

$ git clone [email protected]:CrunchyData/pg_kaboom.git
$ cd pg_kaboom
$ make PG_CONFIG=path/to/pg_config && make install PG_CONFIG=path/to/pg_config
$ psql -c 'CREATE EXTENSION pg_kaboom' -U <user> -d <database>

Usage

Once this extension is installed in the database you wish to destroy use, you will just need to run the function pg_kaboom(text) with the given weapon of breakage.

That said, we want to make sure that you are really sure you want to do these destructive operations. You should never install this extension on a production server. And you are required to issue the following per-session statement in order to do anything with this extension:

SET pg_kaboom.disclaimer = 'I can afford to lose this data and server';
SET pg_kaboom.execute = on;  -- required for shell command-based running; additional safety value.  Not all weapons respect this.
SELECT pg_kaboom('segfault');

-- backend segfaults, exeunt

Available Weapons

Currently defined weapons (more to come) are:

  • break-archive :: install a broken archive_command and force a restart

  • fill-log :: allocate all of the space inside the logs directory

  • fill-pgdata :: allocate all of the space inside the $PGDATA directory

  • fill-pgwal :: allocate all of the space inside the $PGDATA/pg_wal directory

  • mem :: allocate some memory

  • restart :: do an immediate restart of the server

  • rm-pgdata :: do a rm -Rf $PGDATA

  • segfault :: cause a segfault in the running backend process

  • signal :: send a SIGKILL to the Postmaster process

  • xact-wrap :: force the database to run an xact-wraparound vacuum

You can also use the following "special" weapons:

  • random :: choose a random weapon

  • null :: don't do anything, just go through the normal flow

Contributions welcome! Let's get creative in testing how PostgreSQL can recover/respond to various systems meddling!

Author

David Christensen [email protected], [email protected]