5.4. Slony-I Upgrade

5.4.1. Upgrade from 2.1 to 2.2

In Slony-I 2.2, we change the layout of tables sl_log_1 and sl_log_2. This does not prevent an in-place upgrade from version 2.1, however it does require that there be no unreplicated data lingering in sl_log_1 and sl_log_2.

If any data remains unconfirmed, then SLONIK UPDATE FUNCTIONS will fail with the message:

cannot upgrade to new sl_log_N format due to existing unreplicated data

A suggested approach is as follows:

• Use SLONIK LOCK SET to lock all sets, so that no new changes are being injected into the log tables

• Set slon parameter slon_config_cleanup_interval to a very low value (a few seconds) so that the slon cleanup thread will trim data out of sl_log_1 and sl_log_2 immediately

• Restart slon for each node and let it run through a cleanup cycle to empty out sl_log_1 and sl_log_2 on all nodes.

• Verify that sl_log_1 and sl_log_2 are empty on all nodes in the cluster.

• Use SLONIK UPDATE FUNCTIONS against each node to upgrade to version 2.2

• Use SLONIK UNLOCK SET to unlock all sets

5.4.2. Incompatibilities between 2.1 and 2.2

5.4.3. Incompatibilities between 2.0 and 2.1

5.4.4. Incompatibilities between 1.2 and 2.0

5.4.5. Upgrading to Slony-I version 2.1 from version 2.0

Slony-I version 2.0 can be upgraded to version 2.1 using the slonik command SLONIK UPDATE FUNCTIONS.

While this requires temporarily stopping replication, it does not forcibly require an outage for applications that submit updates.

The proper upgrade procedure is thus:

• Stop the slon processes on all nodes. (e.g. - old version of slon)

• Install the new version of Slony-I; software on all nodes (including new versions of the shared functions and libraries) .

• Execute a slonik script containing the command update functions (id = [whatever]); for each node in the cluster. This will alter the structure of some of the Slony-I configuration tables.

  Note: Remember that your slonik upgrade script like all other slonik scripts must contain the proper preamble commands to function.

• Start all slons.

5.4.6. Upgrading to Slony-I version 2.1 from version 1.2 or earlier

The version 2 branch is substantially different from earlier releases, dropping support for versions of PostgreSQL prior to 8.3, as in version 8.3, support for a "session replication role" was added, thereby eliminating the need for system catalog hacks as well as the not-entirely-well-supported xxid data type.

As a result of the replacement of the xxid type with a (native-to-8.3) PostgreSQL transaction XID type, the slonik command SLONIK UPDATE FUNCTIONS is quite inadequate to the process of upgrading earlier versions of Slony-I to version 2.

In version 2.0.2, we have added a new option to SLONIK SUBSCRIBE SET, OMIT COPY, which allows taking an alternative approach to upgrade which amounts to:

• Uninstall old version of Slony-I

  When Slony-I uninstalls itself, catalog corruptions are fixed back up.

• Install Slony-I version 2

• Resubscribe, with OMIT COPY

The following process is suggested to help make the upgrade process as safe as possible, given the above risks.

• Use Section 6.1.11 to generate a slonik script to recreate the replication cluster.

  Be sure to verify the SLONIK ADMIN CONNINFO statements, as the values are pulled are drawn from the PATH configuration, which may not necessarily be suitable for running slonik.

  This step may be done before the application outage.

• Determine what triggers have SLONIK STORE TRIGGER configuration on subscriber nodes.

  Trigger handling has fundamentally changed between Slony-I 1.2 and 2.0.

  Generally speaking, what needs to happen is to query sl_table on each node, and, for any triggers found in sl_table, it is likely to be appropriate to set up a script indicating either ENABLE REPLICA TRIGGER or ENABLE ALWAYS TRIGGER for these triggers.

  This step may be done before the application outage.

• Begin an application outage during which updates should no longer be applied to the database.

• To ensure that applications cease to make changes, it would be appropriate to lock them out via modifications to pg_hba.conf

• Ensure replication is entirely caught up, via examination of the sl_status view, and any application data that may seem appropriate.

• Shut down slon processes.

• Uninstall the old version of Slony-I from the database.

  This involves running a slonik script that runs SLONIK UNINSTALL NODE against each node in the cluster.

• Ensure new Slony-I binaries are in place.

  A convenient way to handle this is to have old and new in different directories alongside two PostgreSQL builds, stop the postmaster, repoint to the new directory, and restart the postmaster.

• Run the script that reconfigures replication as generated earlier.

  This script should probably be split into two portions to be run separately:

  • Firstly, set up nodes, paths, sets, and such

  • At this point, start up slon processes

  • Then, run the portion which runs SLONIK SUBSCRIBE SET

  Splitting the Section 6.1.11 script as described above is left as an exercise for the reader.

• If there were triggers that needed to be activated on subscriber nodes, this is the time to activate them.

• At this point, the cluster should be back up and running, ready to be reconfigured so that applications may access it again.