CloudNativePG Helm Charts

The mode parameter determines how the Helm chart initializes the CloudNativePG cluster:

• standalone (default): Creates a new CNPG cluster or updates an existing one.
• replica: Creates a replica cluster from an existing CNPG cluster.
• recovery: Creates a cluster from a backup, an object store, or via pg_basebackup.

If lag is caused by high transaction volume, consider these strategies:

1. Batching: Break large transactions into smaller ones or use COPY instead of multiple INSERT statements.
2. Row Filtering: Use ALTER PUBLICATION to only replicate necessary data or add table-level filters.
3. Disable Triggers: Temporarily disable triggers on the subscriber for performance-critical periods (ensure you re-enable them).

Example: Row Filtering

-- Only replicate specific operations
ALTER PUBLICATION publication_name SET (publish = 'insert, update, delete');

-- Add table-level filtering
ALTER PUBLICATION publication_name ADD TABLE table_name WHERE (condition);

Permission Issues

Ensure the subscription owner has the necessary rights on the subscriber.

1. Check Owner: SELECT usename, usesuper FROM pg_user WHERE usename = current_user;
2. Grant Privileges:

   GRANT ALL PRIVILEGES ON ALL TABLES IN SCHEMA public TO subscription_user;
   GRANT ALL PRIVILEGES ON ALL SEQUENCES IN SCHEMA public TO subscription_user;

Data Type Mismatches

If the schema between publisher and subscriber differs, replication may fail.

1. Verify Schema: Compare information_schema.columns on both the publisher and subscriber.
2. Fix Schema: Alter the subscriber table to match the publisher.

   ALTER TABLE table_name ALTER COLUMN column_name TYPE new_type;

Receipt lag indicates the subscriber is not receiving WAL messages quickly enough.

Diagnosis:

• Use ping to check latency between the subscriber and the PUBLISHER-HOSTNAME.
• Use nc -zv to check connectivity to the publisher on port 5432.

Resolution:

1. Optimize Network Placement: Place clusters in the same region or availability zone.
2. Tune PostgreSQL Parameters: Adjust the subscriber's configuration to be more aggressive with status updates.

# In the subscriber's postgresql configuration
postgresql:
  parameters:
    wal_sender_timeout: '60s'
    wal_receiver_status_interval: '10s'

Once the console pod is deployed, you can access its shell using kubectl exec to run commands directly within the cluster environment.

kubectl --namespace <namespace> exec --stdin --tty statefulset/<cluster-name>-console -- bash

Use the plugin-barman-cloud Helm chart to install the CNPG-I Barman Cloud Plugin.

Refer to the Barman Cloud Plugin Chart documentation for installation instructions and advanced configuration settings.

To remove the operator, use helm uninstall.

CRITICAL WARNING: Uninstalling the chart does not remove the Custom Resource Definitions (CRDs). If you manually delete the CRDs, it will trigger a cascade deletion of every Cluster resource and all associated PostgreSQL data in their PVCs. This action is irreversible.

helm uninstall cnpg --namespace cnpg-system

Initial Sync Errors (Missing Tables)

If tables exist on the publisher but not the subscriber:

1. Export Schema: pg_dump -h PUBLISHER-HOST -U postgres -s -t table_name database_name
2. Import Schema: psql -h SUBSCRIBER-HOST -U postgres -d database_name < schema_dump.sql

Connection/Timeout Issues

If replication is failing due to network latency or connectivity:

1. Test Connectivity: Run psql -h PUBLISHER-HOST -U postgres -d database_name -c "SELECT 1;" from the subscriber pod.
2. Increase Timeouts: Adjust the subscription configuration parameters.

# In subscription configuration
spec:
  parameters:
    application_name: "my_subscription"
    synchronous_commit: "off"

The Cluster chart can deploy a dedicated console pod via a StatefulSet to execute long-running database tasks (like CREATE INDEX) that are resistant to network interruptions.

To enable this feature, set the following Helm value: cluster.console.enabled=true.

Important Considerations:

• No PDB: There is no PodDisruption Budget for the console StatefulSet. Node maintenance may evict the pod and kill your session.
• Ephemeral Tools: The pod has root access, allowing you to use apt install for additional tools. However, while the /root home folder is persisted, any installed tools will be lost if the pod restarts.
• Startup Delay: Utilities are installed during pod startup; it may take a few seconds after a restart before they are available.

After installation, verify that the deployment is running and the rollout has completed using kubectl.

kubectl -n cnpg-system get deploy
kubectl -n cnpg-system rollout status deploy/cnpg-cloudnative-pg