Materialize v0.49

v0.49.0

Change the type of the following system catalog replica ID columns from integer to string:
- mz_catalog.mz_cluster_replicas.id
- mz_internal.mz_cluster_replica_statuses.replica_id
- mz_internal.mz_cluster_replica_heartbeats.replica_id
- mz_internal.mz_cluster_replica_metrics.replica_id
- mz_internal.mz_cluster_replica_frontiers.replica_id
This is part of the work to introduce system replicas, which Materialize will use for verification and testing purposes and which will not affect user billing or system limits (#11579). Note that, since mz_catalog is part of Materialize’s stable interface, the change to mz_catalog.mz_cluster_replicas.id is a breaking change. If this change causes you friction, please let us know.
Add the ALTER OWNER command, which updates the owner of an object. This is part of the work to enable Role-based access control (RBAC)(#11579).
Add permission checks based on object ownership. To DROP or ALTER an object, the executing role must now be an owner of that object or a superuser. This is part of the work to enable Role-based access control (RBAC)(#11579).
Apply PRIMARY KEY, UNIQUE, and NOT NULL constraints to tables ingested from PostgreSQL sources.
Rename replica introspection views for consistency, and use the _per_worker name suffix for per-worker introspection views.
Automatically restart failed SSH tunnels to improve the reliability of SSH-tunneled Kafka sources.

Fix a correctness bug in Top K processing for monotonic, append-only sources.
Fix a bug that prevented superusers from altering an object owner if they weren’t a member of the new owner’s role.
Fix a bug that would cause PostgreSQL sources to error when columns are added to upstream tables. Note that dropping columns from upstream tables that Materialize ingests still results in error.