Impacts on instance and database administration¶
Stopping and restarting the instance¶
Using E-Maj doesn’t bring any particular constraint regarding stopping and restarting a PostgreSQL instance.
At instance restart, all E-Maj objects are in the same state as at instance stop: log triggers of tables groups in LOGGING state remains enabled and log tables contain cancel-able updates already recorded.
If a transaction with table updates were not committed at instance stop, it would be rolled back during the recovery phase of the instance start, the application tables updates and the log tables updates being cancelled at the same time.
This rule also applies of course to transactions that execute E-Maj functions, like a tables group start or stop, a rollback, a mark deletion,…
Due to a PostgreSQL constraint, the rollback of application sequences assigned to a tables group is the only operation that is not protected by transactions. That is the reason why application sequences are processed at the very end of the rollback operations. (For the same reason, at set mark time, application sequences are processed at the beginning of the operation.)
In case of an instance stop during an E-Maj rollback execution, it is recommended to rerun this rollback just after the instance restart, to ensure that application sequences and tables remain properly synchronised.
Saving and restoring the database¶
Using E-Maj allows a reduction in the database saves frequency. But E-Maj cannot be considered as a substitute to regular database saves that remain indispensable to keep a full image of databases on an external support.
File level saves and restores¶
When saving or restoring instances at file level, it is essential to save or restore ALL instance files, including those stored on dedicated tablespaces.
After a file level restore, tables groups are in the very same state as at the save time, and the database activity can be restarted without any particular E-Maj operation.
Logical saves and restores of entire database¶
Regarding stopped tables groups (in IDLE state), as log triggers are disabled and the content of related log tables is meaningless, there is no action required to find them in the same state as at save time.
Concerning tables groups in LOGGING state at save time, it is important to be sure that log triggers will only be activated after the application tables rebuild. Otherwise, during the tables rebuild, tables updates would also be recorded in log tables!
When using pg_dump command for saves and psql or pg_restore commands for restores, and processing full databases (schema + data), these tools recreate triggers, E-Maj log triggers among them, after tables have been rebuilt. So there is no specific precaution to take.
On the other hand, in case of data only save or restore (i.e. without schema, using -a or –data-only options), the –disable-triggers must be supplied:
- with pg_dump (or pg_dumpall) with save in plain format (and psql is used to restore),
- with pg_restore command with save in tar or custom format.
Restoring the database structure generates 2 error messages reporting that the _emaj_protection_event_trigger_fnct() function and the emaj_protection_trg event trigger already exist:
... ERROR: function "_emaj_protection_event_trigger_fnct" already exists with same argument types ... ERROR: event trigger "emaj_protection_trg" already exists ...
This message display is normal and does not indicate a defective restore. Indeed, both objects are created with the extension and are then detached from it, so that the trigger can block any attempt of the extension drop. As a result, the pg_dump tool saves them as independent objects. And when restoring, these objects are created twice, first with the emaj extension creation, and then as independent objects, this second attempt generating both error messages.
Logical save and restore of partial database¶
With pg_dump and pg_restore tools, database administrators can perform on a subset of database schemas or tables.
Restoring a subset of application tables and/or log tables generates a heavy risk of data corruption in case of later E-Maj rollback of concerned tables. Indeed, it is impossible to guarantee in this case that application tables, log tables and internal E-Maj tables that contain essential data for rollback, remain coherent.
If it is necessary to perform partial application tables restores, a drop and recreation of all tables groups concerned by the operation must be performed just after.
The same way, it is strongly recommended to NOT restore a partial emaj schema content.
The only case of safe partial restore concerns a full restore of the emaj schema content as well as all tables belonging to all groups that are created in the database.
Beside using pg_restore or psql with files produced by pg_dump, it is possible to efficiently load large amounts of data with the COPY SQL verb or the copy psql meta-command. In both cases, this data loading fires INSERT triggers, among them the E-Maj log trigger. Therefore, there is no constraint to use COPY or copy in E-Maj environment.
With other loading tools, it is important to check that triggers are effectively fired for each row insertion.
Reorganisation of application tables¶
Application tables protected by E-Maj can be reorganised using the SQL CLUSTER command. Whether or not log triggers are enabled, the organisation process has no impact on log tables content.
Reorganisation of E-Maj tables¶
The index corresponding to the primary key of each table from E-Maj schemas (neither log tables nor technical tables) is declared “cluster”.
So using E-Maj may have an operational impact regarding the execution of CLUSTER SQL commands at database level.
When E-Maj is used in continuous mode (with deletion of oldest marks instead of regular tables groups stop and restart), it is recommended to regularly reorganize E-Maj log tables. This reclaims unused disk space following mark deletions.
Using E-Maj with replication¶
Integrated physical replication¶
E-Maj is totally compatible with the use of the different PostgreSQL integrated physical replication modes (WAL archiving and PITR, asynchronous and synchronous Streaming Replication). Indeed, all E-Maj objects hosted in the instance are replicated like all other objects of the instance.
However, because of the way PostgreSQL manages sequences, the sequences’ current values may be a little forward on slave instances than on the master instance. For E-Maj, this may lightly overestimate the number of log rows in general statistics. But there is no consequence on the data integrity.
Integrated logical replication¶
Starting with version 10, PostgreSQL includes logical replication mechanisms. The replication granularity is the table. The publication object used in logical replication is quite close to the E-Maj tables group concept, except that a publication cannot contain sequences.
Replication of application tables managed by E-Maj
An application table that belongs to a tables group can be replicated. The effect of any rollback operation that may occur would be simply replicated on subscriber side, as long as no filter has been applied on replicated SQL verbs types.
Replication of application tables with E-Maj activated on subscriber side
It is possible to include an application table into a tables group, with updates coming from a logical replication flow. But all E-Maj operations (starting/stopping the group, setting marks,…) must of course be executed on the subscriber side. An E-Maj rollback operation can be launched once the replication flow has been stopped (to avoid updates conflicts). But then, tables on both publisher and subscriber sides are not coherent anymore.
Replication of E-Maj log tables
It is technicaly possible to replicate an E-Maj log table (once found a way to get the DDL that creates the log table – using pg_dump for instance). This allows to duplicate or concentrate logs content on another server. But the replicated log table can only be used for log auditing. As log sequences or TRUNCATE verbs are not replicated, these logs cannot be used for other purposes.
Replication of application tables and E-Maj log tables
Application tables and log tables can be simultaneously replicated. But as seen previously, these replicated logs can only be used for auditing purpose. E-Maj rollback operations can only be executed on publisher side.
Other replication solutions¶
Using E-Maj with external replication solutions based on triggers like Slony or Londiste, requires some attention… It is probably advisable to avoid replicating log tables and E-Maj technical tables.