MongoDB Enterprise Kubernetes Operator 1.22.0

MongoDB Enterprise Kubernetes Operator 1.22.0

Breaking Changes

• All Resources: The Operator no longer uses the "Reconciling" state. In most of the cases it has been replaced with "Pending" and a proper message

Deprecations

None

Bug Fixes

• MongoDB: Fix support for setting autoTerminateOnDeletion=true for sharded clusters. This setting makes sure that the operator stops and terminates the backup before the cleanup.

New Features

• MongoDB: An Automatic Recovery mechanism has been introduced for MongoDB resources and is turned on by default. If a Custom Resource remains in Pending or Failed state for a longer period of time (controlled by MDB_AUTOMATIC_RECOVERY_BACKOFF_TIME_S environment variable at the Operator Pod spec level, the default is 20 minutes)
  the Automation Config is pushed to the Ops Manager. This helps to prevent a deadlock when an Automation Config can not be pushed because of the StatefulSet not being ready and the StatefulSet being not ready because of a broken Automation Config.
  The behavior can be turned off by setting MDB_AUTOMATIC_RECOVERY_ENABLE environment variable to false.
• MongoDB: MongoDB audit logs can now be routed to Kubernetes pod logs.
  • Ensure MongoDB audit logs are written to /var/log/mongodb-mms-automation/mongodb-audit.log file. Pod monitors this file and tails its content to k8s logs.
  • Use the following example configuration in MongoDB resource to send audit logs to k8s logs:

  spec:
    additionalMongodConfig:
      auditLog:
        destination: file
        format: JSON
        path: /var/log/mongodb-mms-automation/mongodb-audit.log
  

  • Audit log entries are tagged with the "mongodb-audit" key in pod logs. Extract audit log entries with the following example command:

  kubectl logs -c mongodb-enterprise-database replica-set-0 | jq -r 'select(.logType == "mongodb-audit") | .contents'
  

• MongoDBOpsManager: Improved handling of unreachable clusters in AppDB Multi-Cluster resources
  • In the last release, the operator required a healthy connection to the cluster to scale down processes, which could block the reconcile process if there was a full-cluster outage.
  • Now, the operator will still successfully manage the remaining healthy clusters, as long as they have a majority of votes to elect a primary.
  • The associated processes of an unreachable cluster are not automatically removed from the automation config and replica set configuration. These processes will only be removed under the following conditions:
    • The corresponding cluster is deleted from spec.applicationDatabase.clusterSpecList or has zero members specified.
    • When deleted, the operator scales down the replica set by removing processes tied to that cluster one at a time.
• MongoDBOpsManager: Add support for configuring logRotate on the automation-agent for appdb.
• MongoDBOpsManager: systemLog can now be configured to differ from the otherwise default of /var/log/mongodb-mms-automation.