v1.21.0
Release date: Oct 12, 2023
!!! Important "Important changes from previous versions" This release contains a few changes to the default settings of CloudNativePG with the goal to improve general stability and security through predefined values. If you are upgrading from a previous version, please carefully read the "Important Changes" section below, as well as the upgrade documentation.
Features:
-
Volume Snapshot support for backup and recovery: leverage the standard Kubernetes API on Volume Snapshots to take advantage of capabilities like incremental and differential copy for both backup and recovery operations. This first step, covering cold backups from a standby, will continue in 1.22 with support for hot backups using the PostgreSQL API and tablespaces.
-
OLM installation method: introduce support for Operator Lifecycle Manager via OperatorHub.io for the latest patch version of the latest minor release through the stable channel. Many thanks to EDB for donating the bundle of their "EDB Postgres for Kubernetes" operator and adapting it for CloudNativePG.
Important Changes:
- Change the default value of
stopDelay
to 1800 seconds instead of 30 seconds (#2848) - Introduce a new parameter, called
smartShutdownTimeout
, to control the window of time reserved for the smart shutdown of Postgres to complete; the general formula to compute the overall timeout to stop Postgres ismax(stopDelay - smartShutdownTimeout, 30)
(#2848) - Change the default value of
startDelay
to 3600, instead of 30 seconds (#2847) - Replace the livenessProbe initial delay with a more proper Kubernetes startup probe to deal with the start of a Postgres server (#2847)
- Change the default value of
switchoverDelay
to 3600 seconds instead of 40000000 seconds (#2846) - Disable superuser access by default for security (#2905)
- Enable replication slots for HA by default (#2903)
- Stop supporting the
postgresql
label - replaced bycnpg.io/cluster
in 1.18 (#2744)
Security:
- Add a default
seccompProfile
to the operator deployment (#2926)
Enhancements:
- Enable bootstrap of a replica cluster from a consistent set of volume snapshots (#2647)
- Enable full and Point In Time recovery from a consistent set of volume snapshots (#2390)
- Introduce the
cnpg.io/coredumpFilter
annotation to control the content of a core dump generated in the unlikely event of a PostgreSQL crash, by default set to exclude shared memory segments from the dump (#2733) - Allow to configure ephemeral-storage limits for the shared memory and temporary data ephemeral volumes (#2830)
- Validate resource limits and requests through the webhook (#2663)
- Ensure that PostgreSQL's
shared_buffers
are coherent with the pods' allocated memory resources (#2840) - Add
uri
andjdbc-uri
fields in the credential secrets to facilitate developers when connecting their applications to the database (#2186) - Add a new phase
Waiting for the instances to become active
for finer control of a cluster's state waiting for the replicas to be ready (#2612) - Improve detection of Pod rollout conditions through the
podSpec
annotation (#2243) - Add primary timestamp and uptime to the kubectl plugin's
status
command (#2953)
Fixes:
-
Ensure that the primary instance is always recreated first by prioritizing ready PVCs with a primary role (#2544)
-
Honor the
cnpg.io/skipEmptyWalArchiveCheck
annotation during recovery to bypass the check for an empty WAL archive (#2731) -
Prevent a cluster from being stuck when the PostgreSQL server is down but the pod is up on the primary (#2966)
-
Avoid treating the designated primary in a replica cluster as a regular HA replica when replication slots are enabled (#2960)
-
Reconcile services every time the selectors change or when labels/annotations need to be changed (#2918)
-
Defaults to
app
both the owner and database during recovery bootstrap (#2957) -
Avoid write-read concurrency on cached cluster (#2884)
-
Remove empty items, make them unique and sort in the
ResourceName
sections of the generated roles (#2875) -
Ensure that the
ContinuousArchiving
condition is properly set to 'failed' in case of errors (#2625) -
Make the
Backup
resource reconciliation cycle more resilient on interruptions by stopping only if the backup is completed or failed (#2591) -
Reconcile PodMonitor
labels
andannotations
(#2583) -
Fix backup failure due to missing RBAC
resourceNames
on theRole
object (#2956) -
Observability:
- Add TCP port label to default
pg_stat_replication
metric (#2961) - Fix the
pg_wal_stat
default metric for Prometheus (#2569) - Improve the
pg_replication
default metric for Prometheus (#2744 and #2750) - Use
alertInstanceLabelFilter
instead ofalertName
in the provided Grafana dashboard - Enforce
standard_conforming_strings
in metric collection (#2888)
- Add TCP port label to default
Changes:
- Set the default operand image to PostgreSQL 16.0
- Fencing now uses PostgreSQL's fast shutdown instead of smart shutdown to halt an instance (#3051)
- Rename webhooks from kb.io to cnpg.io group (#2851)
- Replace the
cnpg snapshot
command withcnpg backup -m volumeSnapshot
for thekubectl
plugin - Let the
cnpg hibernate
plugin command use theClusterManifestAnnotationName
andPgControldataAnnotationName
annotations on PVCs (#2657) - Add the
cnpg.io/instanceRole
label while deprecating the existingrole
label (#2915)
Technical enhancements:
- Replace
k8s-api-docgen
withgen-crd-api-reference-docs
to automatically build the API reference documentation (#2606)