man pg_staging (1): Prepare a staging environment from http accessible backups

SYNOPSIS

pg_staging [--version] [-c configuration file] [-t tmpdir] [-d debug] [-v verbose] [-q quiet] <command> <dbname> [<date>]

DESCRIPTION

pg_staging is a tool to prepare and maintain a staging environment from http accessible backups. Its main job (see command restore) is to:

: 1. create target database, named dbname_YYYYMMDD

: 2. add this database to the pgbouncer setup, reload pgbouncer

: 3. fetch a backup file named dbname.date -I.dump

: 4. prepare a filtered catalog for the given dump, depending on configuration

: 5. given sql_path and a pre directory, psql -f *.sql in there

: 6.
pg_restore the backup with the custom catalog to the created database

: 7. if restore_vacuum is true, VACUUM ANALYZE the database

: 8. given sql_path setup and a post directory, psql -f *.sql in there

: 9. switch pgbouncer entry for dbname to target dbname_YYYYMMDD and reload pgbouncer again.

pg_staging is able to do some more, mainly the commands given allow to give fine grain control over what it does rather than only providing the full restore option.

DEPENDENCIES

pg_staging will use the following external tools:

: •
pgbouncer in order to maintain more than one staging database

: •
pg_restore which major version must match target database cluster

: •
scp to upload new pgbouncer.ini configuration files

: •
ssh to run the staging-client.sh on the target host

See next section for how to prepare those components.

INITIAL SETUP

In order for pg_staging to be able to manage any target you give it, the following conditions have to be met:

1. install staging-client.sh on the target host

Currently you have to ssh non interactively (setup a password free ssh key authentication) to the target host. pg_staging will run the following command:

ssh <host> sudo ./staging-client.sh <pgbouncer.xxxx.ini> <pgbouncer_port>

: 2. install and open pgbouncer "trust" connection as maintenance user (dbuser) on the maintenance database (maintdb).
This connection will get used to CREATE DATABASE and DROP DATABASE.

COMMANDS

commands

: This will show available commands and a docstring for each.

main operation

init <dbname>

: Prepare a cluster given a pg_dumpall -g file, see option dumpall_url.

dump <dbname> [<filename>]

: dump given database to file, using pg_dump in custom format. The pgbouncer_port is used and the filename defaults to dbname.YYYYMMDD.dump. The pg_restore directory part is considered for finding the pg_dump binary. This command will prevent you from overwriting an existing dump file, see redump if you do not want this behavior.

redump <dbname> [<filename>]

: dump even when destination filename already exists.

restore <dbname> [<YYYYMMDD>]

: See description section, it explains the details. It may be of importance to recall that restore will clean up its temporary files, including the dump file itself. The clean up happens in case of success and in case of error. This command will source pre and post sql files, as per sql_path config.

drop <dbname> [<YYYYMMDD>]

: DROP DATABASE the given database, or today's one if none given. It won't edit pgbouncer configuration accordingly though, as of version 0.5.

switch <dbname> [<YYYYMMDD>]

: Change the canonical <dbname> entry in pgbouncer to point to given dated instance, default to today's one.

purge <dbname>

: Clean the database section by dropping out the older databases and keeping online only the keep_bases most recent.

vacuumdb <dbname> [<YYYYMMDD>]

: VACUUM ANALYZE given database.

load <dbname> <filename>

: pg_restore given dump file, this allow to skip the auto downloading part of the restore command.

fetch <dbname> [<YYYYMMDD>]

: Only fetch the dump, do not restore it, do not remove it afterwards.

presql <dbname> [<YYYYMMDD>]

: Source the sql_path/pre/*.sql files into the database by means of psql -f, in alphabetical order, without recursive walking into subdirs.

postsql <dbname> [<YYYYMMDD>]

: Source the sql_path/post/*.sql files into the database by means of psql -f, in alphabetical order, without recursive walking into subdirs.

listings

databases

: Show the list of database sections parsed into the .ini file.

backups <dbname> [<YYYYMMDD>]

: Show <dbname> available backups on the http host.

dbsize <dbname> [<YYYMMDD>]

: Show database size of given instance, as returned by SELECT pg_size_pretty(pg_database_size(dbname_YYYYMMDD));

dbsizes --all | --match <pattern> | <dbname>

: Show database sizes of given instances. With --all show sizes of all instances of all configured section, with --match you can reduce the listing to regexp matching section names, with a <dbname> it'll show sizes of all instances of given section.

show <dbname> [<YYYYMMDD>] <setting>

: Show current value of <setting> for given database.

pgbouncer

pgbouncer <dbname>

: Show pgbouncer database listing for given dbname.

pause <dbname> [<YYYMMDD>]

: Issue a pgbouncer pause <dbname> command.

resume <dbname> [<YYYMMDD>]

: Issue a pgbouncer resume <dbname> command.

londiste

londiste <dbname> [<YYYMMDD>]

: Prepare londiste configuration files in TMPDIR, then send them over to the provider hosts in ~pgstating/londiste and start the daemons (pgqadm.py and londiste.py).

remote service management

Note that all actions (start, stop, restart, status) are not available to all services (londiste, ticker, pgbouncer).

start <service> <dbname> [<YYYMMDD>]

: Start remote service for given dbname, where service is one of londiste, ticker, or pgbouncer.

stop <service> <dbname> [<YYYMMDD>]

: Stop remote service for given dbname, where service is one of londiste, ticker, or pgbouncer.

restart <service> <dbname> [<YYYMMDD>]

: Restart remote service for given dbname, where service is one of londiste, ticker, or pgbouncer.

status <service> <dbname> [<YYYMMDD>]

: Show status remote service for given dbname, where service is one of londiste, ticker, or pgbouncer.

experimental and internal

nodata

: Show tables we want to skip loading DATA for, those we are a subscriber of.

catalog

: Show the filtered out catalog we'll give to pg_restore -L.

triggers

: Show the schema-qualified functions used by triggers, in order to be able to follow dependancies when filtering out a schema definition (such as pgq or londiste).

OPTIONS

Usage: pg_staging.py [-c <config_filename>] command dbname <args>

Options:
  -h, --help            show this help message and exit
  --version             show pg_staging version
  -c CONFIG, --config=CONFIG
                        configuration file, defauts to /etc/hm-
                        tools/pg_staging.ini
  -n, --dry-run         simulate operations, don't do them
  -v, --verbose         be verbose and about processing progress
  -q, --quiet           be terse, almost silent
  -d, --debug           provide python stacktrace when error
  -t TMPDIR, --tmpdir=TMPDIR
                        temp dir where to fetch dumps, /tmp

CONSOLE

If you start pg_staging without command, it will open up an interactive console with basic readline support. All previous commands are supported, except for the experimental ones, and the following are added.

config <filename>

: read given filename as the current configuration file for pg_staging.

set <section> <option> <value>

: set given option to given value for current interactive session only.

get <section> <option>

: print current value of given option.

verbose

: switch on and off the verbosity of pg_staging.

INTERNALS

How we use tools. Will get expanded if questions arise.

AUTHOR

pg_staging is written by Dimitri Fontaine <m[blue][email protected]m[][1]>.

NOTES

1.

[email protected]

: mailto:[email protected]