EDB CloudNativePG Cluster 1.30.2 release notes v1.30.2

Released: 17 November 2026

This release of EDB Postgres® AI for CloudNativePG™ Cluster includes the following:

Highlights

PostgreSQL 18.7 is the new default image: the default operand image is now 18.7-standard-ubi9, which fixes several CVEs. It applies only to new clusters that don't set an image. To get the fixes, update existing clusters to 18.7, 17.12, 16.16 or 15.20 by changing imageName or the ImageCatalog. Before updating, read the PostgreSQL 18.7 announcement and the release notes it links. (#11833)

PostgreSQL 14 is no longer supported: the PostgreSQL community stopped fixing PostgreSQL 14 on November 12, 2026, and EDB Postgres for Kubernetes supports a major version only as long as PostgreSQL does. If you still run PostgreSQL 14, plan an upgrade to a newer major version. PostgreSQL 15 stops receiving fixes on November 11, 2027. (#11935)

The operator now needs the delete permission on leases. Users who manage the RBAC manually must add it. (#11609)

Enhancements

DescriptionAddresses
Streaming retry for standbys on the WAL archive

PostgreSQL tries streaming only after a restore from the archive fails. If WAL is produced about as fast as it is restored, that may never happen, and the standby stays on the archive, with no WAL receiver and no part in synchronous replication. Now, after 30 seconds of continuous restores, WAL restore fails on purpose for 5 seconds, so PostgreSQL tries streaming. These failures are expected and can appear in the logs.

#11868
Server-side encryption with customer-provided keys (SSE-C)

added sseCustomerKey in s3Credentials, to use SSE-C with S3-compatible object stores. It works with backups, WAL archiving and recovery.

#11711, #11775
Backup and ScheduledBackup resources that use the barmanObjectStoremethod now return an admission warning about the removal of native Barman Cloud support in 1.32, as Cluster resources already did.#11725
extension_control_path and dynamic_library_path no longer list directories that are missing from an extension image, for example share in an image that ships only lib.#11652
Transient network errors from barman-cloud-wal-restore are now logged (needs Barman 3.20.0 or later).#11541
Moved the "Defaulting for ..." operator log lines from info to debug level.

They were printed at every reconciliation and hid other messages.

#11790
kubectl cnp status no longer fails without the pods/exec permission.

It shows not permitted: requires the pods/exec permission instead of the size and configuration sections.

#11687

Security Fixes

DescriptionAddresses
Updated golang.org/x/net to v0.62.0, fixing GO-2026-6617, GO-2026-6612, GO-2026-6611, GO-2026-6610, GO-2026-6603, GO-2026-6860 and GO-2026-6866, and bumped the other golang.org/x modules.#11631, #11867

Changes

DescriptionAddresses
Updated the default PgBouncer image from 1.25.2 to 1.26.1.

Poolers that use the default image roll out the new version when the operator is upgraded.

#11571, #11704
Updated the barman-cloud library to v0.7.0, which skips failed backups when it reads the backup catalog.#11711

Bug Fixes

DescriptionAddresses
Fixed enabling a sidecar plugin on a running cluster.

Pods created before that have no sidecar until they are recreated. They logged Unknown plugin errors, their metrics scrapes failed, and a switchover could wait forever. When the plugin archives WAL, the former primary crash looped until the Pod was deleted. The operator now recreates it.

#11858
Fixed replicas created from an online volume snapshot ignoring the backup_label and tablespace_map stored in the snapshot.

PostgreSQL did crash recovery and did not fetch WAL from the archive.

#11847
Fixed logical decoding stalling while a new replica was bootstrapping, with synchronizeLogicalDecoding enabled.

The replica is now left out of synchronized_standby_slots (or pg_failover_slots.standby_slot_names) until its volume is initialized.

#11802
Fixed status.phaseReason being cleared and set again several times a second while a cluster was being created or changed, for example "Some instances are not yet active".

This loaded the API server and made kubectl get cluster -w noisy.

#11789
Fixed the operator retrying instance status requests that failed with an error response, as if the failure was a network error.

During a switchover the old primary answers with an error while PostgreSQL is stopped, so each reconciliation waited about 1.6 seconds, which slowed the whole switchover. Contributed by @kabbydriver.

#11793
Fixed the WAL count and size metrics counting .partial, .backup and .history files.#9967
Fixed replicas crash looping with recovery aborted because of insufficient parameter settings after a hot standby parameter such as max_connectionswas removed from .spec.postgresql.parameters.

Contributed by @ichibahn.

#11236
Fixed the instance manager waiting forever on a frozen postmaster, one that accepts connections but never answers.

This blocked the instance status, the metrics exporter and the check before an in-tree backup. Local connection attempts to PostgreSQL now time out after 5 seconds.

#11718
Fixed a Cluster recreated over leftover resources (for example by a backup tool) staying in Unable to create required cluster objects with cannot set an ownerRef on a resource you can't delete.

This happened with the OwnerReferencesPermissionEnforcement admission plugin. When restoring over orphan PVCs, the operator now deletes the stale primary Lease, so the new primary does not wait for it to expire.

#11609
Fixed WAL restore reporting the end of the WAL archive on the wrong timeline.

A replica restored from a snapshot could miss valid WAL of the new timeline and replay WAL of the parent timeline instead. Contributed by @ardentperf.

#11133
Fixed a Cluster not being reconciled again when a plugin asked for a retry without a delay.

For example, a Cluster created before the ObjectStore of the Barman Cloud plugin stayed in BootstrapPending. The operator now retries after one second. Contributed by @valeriodelsarto.

#11683
Fixed ConsistentSystemID turning False while PostgreSQL is not running, for example in replica clusters after a demotion, or on fenced instances.

The system ID is now read from disk, so it no longer needs a running PostgreSQL.

#11645
Fixed a Backup that could be started more than once, or moved back to pending after it started.#11617
Fixed kubectl cnp backup for volume snapshot backups: when only one of --immediate-checkpoint and --wait-for-archive was set, the other option used its default instead of the value in .spec.backup.volumeSnapshot.onlineConfiguration.#11733