uuid_generate_v4() creates a different UUID in each node of a replicated PostgreSQL environment if Pgpool-II runs in native replication mode.
This has been reported for the similar function gen_random_uuid() with Pgpool-II replication in #sys-ops on Slack. From the Pgpool-II manual:
There is no guarantee that any data provided using a context-dependent mechanism (e.g. random number, transaction ID, OID, SERIAL, sequence), will be replicated correctly on multiple backends.
RMB documentation mentions it in the upgrading guide for RMB 25 and in connection with "generateId": https://github.com/folio-org/raml-module-builder/#the-post-tenant-api
- Create an issue for each module that still uses uuid_generate_v4() from uuid-ossp
- Remove uuid_generate_v4() and "CREATE EXTENSION uuid-ossp".
For predefined records use a hard-coded UUID.
For calculated records re-use the UUID of the primary source record.
Or use the md5sum trick: If a string (for example a JSONB::text) is unique calculate md5sum(s)::text::uuid to get a unique UUID.
To prevent any new usage of uuid_generate_v4() RMB and Spring base should add this function that takes precedence of the uuid-ossp function:
This doesn't affect Pgpool-II streaming replication mode, it only affect Pgpool-II native replication mode.