Upgrading

Refreshing package sign keys

The key that is used to sign packages may expire from time to time, and needs to be renewed. This is done by the build maintainers, and you may need update your installation to know about the refreshed key.

If you see the following (on Debian or Ubuntu), this affects you:

lenny:~# apt-get update
[...]
Reading package lists... Done
W: GPG error: http://download.opensuse.org  Release: The following signatures were invalid: KEYEXPIRED 1270154736
W: You may want to run apt-get update to correct these problems
lenny:~#

After running the following command, all should work fine again:

lenny:~# apt-key adv --keyserver hkp://wwwkeys.de.pgp.net --recv-keys BD6D129A

If it does not help, it is probably best to contact the build maintainers or the mirrorbrain mailing list.

Upgrading PostgreSQL

General notes

When upgrading PostgreSQL, it is important to look at the version number difference. If the third digit changes, no special procedure is needed (except when the release notes explicitely hint about it).

When the first or second digit change, then a dump-and-reload cycle is usually needed.

For instance, when upgrading from 8.3.5 to 8.3.7 nothing needs to be done. When upgrading from 8.3.7 to 8.4, you need to dump and reload.

You might want to follow the instructions that your vendor provides. If your vendor doesn’t provide an upgrade procedure, be warned that the database needs to be dumped before upgrading PostgreSQL.

See pg_dumpall(1) for how to dump and reload the complete database.

Warning

If you use mod_asn, there is one more caveat. If your vendor’s upgrade procedure automatically saves the previous PostgreSQL binaries in case they are needed later, the procedure might not take into account that the ip4r.so shared object might need to be saved as well. Hence, you might be unable to start the old binaries, when the ip4r shared object has been upgraded already.

Hence, it is recommended that you do a complete dump of the databases before upgrading, and load that after upgrading.

Note

When upgrading to 8.4, ident sameuser is no longer a valid value in pg_hba.conf. Replace it with ident.

Offline upgrade

The following console transcripts should give an idea about upgrading an PostgreSQL installation. It was done on an openSUSE system, but a similar procedure should work on other platforms.

The upgrade in this example is done while the database is taken offline, i.e. you need to plan for a downtime of the server. The cron daemon is stopped so that there are no attempted writes to the database. pg_dumpall is used to save the complete database to a file:

root@doozer ~ # rccron stop
Shutting down CRON daemon                                             done
root@doozer ~ # su - postgres
postgres@doozer:~> pg_dumpall > SAVE
postgres@doozer:~> exit
root@doozer ~ # rcpostgresql stop
Shutting down PostgreSQL server stopped                               done

At this point, you would upgrade the PostgreSQL software.

Next, the data directory is moved away, a new one created:

root@doozer ~ # old /var/lib/pgsql/data
moving /var/lib/pgsql/data to /var/lib/pgsql/data-20090728
root@doozer ~ # rcpostgresql start
Initializing the PostgreSQL database at location /var/lib/pgsql/data  done
Starting PostgreSQL                                                   done
root@doozer ~ #

Now, the authentication setup and the configuration need to be migrated from the old install to the new one:

root@doozer ~ # su - postgres
postgres@doozer:~> cp data/pg_hba.conf data/pg_hba.conf.orig
postgres@doozer:~> cp data/postgresql.conf data/postgresql.conf.orig
postgres@doozer:~> vi -d data-20090728/pg_hba.conf data/pg_hba.conf
postgres@doozer:~> vi -d data-20090728/postgresql.conf data/postgresql.conf

Then, load the dump into the new database:

postgres@doozer:~> psql template1 -f SAVE
[...]

Finally, restart PostgreSQL, Apache and cron:

root@doozer ~ # rcpostgresql restart
Shutting down PostgreSQL server stopped                               done
Starting PostgreSQL                                                   done
root@doozer ~ # rcapache2 reload
Reload httpd2 (graceful restart)                                      done
root@doozer ~ # rccron start
Starting CRON daemon                                                  done

Online upgrade

Using a second PostgreSQL daemon, started temporarily, an online upgrade can be performed as follows.

First, create space for the temporary database:

root@mirrordb ~ # mkdir /space/pgsql-tmp
root@mirrordb ~ # chown postgres:postgres /space/pgsql-tmp

Create the new (temporary) database:

root@mirrordb ~ # su - postgres
postgres@mirrordb:~> initdb /space/pgsql-tmp/data
The files belonging to this database system will be owned by user "postgres".
This user must also own the server process.

The database cluster will be initialized with locale en_US.UTF-8.
The default database encoding has accordingly been set to UTF8.
The default text search configuration will be set to "english".

creating directory /space/pgsql-tmp/data ... ok
creating subdirectories ... ok
selecting default max_connections ... 100
selecting default shared_buffers/max_fsm_pages ... 32MB/204800
creating configuration files ... ok
creating template1 database in /space/pgsql-tmp/data/base/1 ... ok
initializing pg_authid ... ok
initializing dependencies ... ok
creating system views ... ok
loading system objects' descriptions ... ok
creating conversions ... ok
creating dictionaries ... ok
setting privileges on built-in objects ... ok
creating information schema ... ok
vacuuming database template1 ... ok
copying template1 to template0 ... ok
copying template1 to postgres ... ok

WARNING: enabling "trust" authentication for local connections
You can change this by editing pg_hba.conf or using the -A option the
next time you run initdb.

Success. You can now start the database server using:

    postgres -D /space/pgsql-tmp/data
or
    pg_ctl -D /space/pgsql-tmp/data -l logfile start

postgres@mirrordb:~>

Copy the configuration and the authentification setup to the temporary database:

postgres@mirrordb:~> cp /space/pgsql/data/postgresql.conf /space/pgsql-tmp/data/
postgres@mirrordb:~> cp /space/pgsql/data/pg_hba.conf /space/pgsql-tmp/data/

Note

The second database server will need RAM — maybe it will be necessary to adjust the shared_buffers setting in postgresql.conf for both daemons, so they don’t try allocate more memory than physically available.

Next, change the port in the temporary postgresql.conf from 5432 to 5433 and start the second PostgreSQL server:

postgres@mirrordb:~> vi /space/pgsql-tmp/data/postgresql.conf
postgres@mirrordb:~> postgres -D /space/pgsql-tmp/data

Note

This assumes that Apache is configured to use a TCP connection to access the database server, not a UNIX domain socket.

Load the dumped data (not forgetting to use the differing port number):

postgres@doozer:~> psql -p 5433 template1 -f SAVE
[...]

Now the Apache server, and possibly other services (/etc/mirrorbrain.conf) need to be changed to the temporary port, and (gracefully) restarted.

Note

Verify that everything works as expected with the temporary database. If it does, stop the primary PostgreSQL server (and verify again that everything still works).

From here on, the next steps are probably obvious. You would proceed as described in the previous section. After upgrading the PostgreSQL install, loading the data, copying/merging postgresql.conf and pg_hba.conf, you would revert the Apache configuration to use port 5432 and reload it.

If everything works, you can stop and remove the temporary database installation.

Version-specific upgrade notes

To 2.18.0:

Due to the modernized HTML of the .mirrorlist page, please check your CSS styling for them (if you have some).

To 2.14.0:

To take advantage of mirror selection by geographical distance (as additional criterion to country, network prefix etc.), the free GeoLite City GeoIP database needs to be used. If you used the simpler database so far, you need to switch to the city edition which contains the required data. The following steps are necessary:

  1. Use the provided geoip-lite-update tool to download it (and keep it updated regularly via cron).
  2. Edit your mod_geoip configuration and change GeoIP.dat to GeoLiteCity.dat.updated
  3. Run mb update -A –all-mirrors to update the mirrors’ GeoIP data (coordinates, country and region), and (if mod_asn is used) autonomous system and prefix.

MirrorBrain will continue to run fine without the extra data. Thus, it is not obligatory to switch to the larger GeoIP database.

From 2.13.x to 2.13.3:

If you actively use Torrents, it may make sense to recreate the hashes. It is not strictly necessary, because the only hash that has been changed is the BitTorrent info hash (btih) that is cached in the database. That hash is used only if requested by <URL>.btih and it is shown on the details page, but it is not used in actual Torrent generation. Thus it is unlikely to matter. Anyhow, it could cause confusion.

You can use mb makehashes with the --force option once to recreate the hashes.

From 2.12.x to 2.13.0:

  • If you created hashes in the past, please edit your /etc/crontab and replace the calls to the former tool metalink-hasher update with a call to mb makehashes. Example:

    # old tool:  metalink-hasher update -t /srv/metalink-hashes/srv/ooo /srv/ooo
    # new tool:  mb makehashes -t /srv/metalink-hashes/srv/ooo /srv/ooo
    
  • If you used the the metalink-hasher with the -b option in the past, check the usage examples that come with mb help makehashes. The option has been rewritten to be easier to use, and it should now be easier to get it to do what you want.

From 2.11.1 to 2.11.2:

The mb vacuum command has been renamed to mb db vacuum. The old command will continue to work for now - but existing cron jobs should be updated; the old command might be depracated later.

Users that happen to use the mirrorprobe with the default timeout of 60 seconds should now run it with -t 60, because the default has been lowered to 20 seconds with release.

From 2.10.3 to 2.11.0:

The MirrorBrainHandleDirectoryIndexLocally directive has been removed. A warning is issued where it is still found in the config. It didn’t really have a function.