I am porting several stored procedures from Oracle to Postgres. In the Oracle code, if an exception is thrown within a stored procedure, the exception is caught and details are written to a database table using an autonomous transaction (as the main transaction is rolled back).
As far as I can see from the documentation, Postgres doesn't support autonomous transaction (although there is talk about it at https://wiki.postgresql.org/wiki/Autonomous_subtransactions - is this something that is being discussed for a future release?).
The Postgres functions that I'm writing are batch processes that will be invoked via a scheduler (either cron or pgAgent).
Ideally I'd like to record the exceptions in a database table. If this isn't possible then recording in a log fie would be acceptable, but I'd like to keep this separate from the main postgres log.
Alternatives that I've come up with (none of them very satisfactory):
use 'raise' to record in postgres log
put the error recording in the client code (as invoked by scheduler) - use BEGIN TRANSACTION to start a new transaction
use COPY to output to a file
Can anyone suggest something that would meet my requirements above?
It's hacky, and, I haven't tried it in a few years. Setup a foreign table that resides in the same database. When you write to the foreign table, it will be using a 'loopback' connection, and that transaction will be able to commit because it is a separate connection.
To be fair, I haven't actually done this since the days of dblink, I *believe* it should work with fdw though.
