mongodb-enterprise-kubernetes

• The Operator and Database images are now based on ubuntu:16.04
• The Operator now uses a single CustomResourceDefinition MongoDB instead of MongoDbReplicaSet, MongoDbShardedCluster and MongoDbStandalone
  • It's important to follow the upgrade procedure described here to transfer existing MongoDbReplicaSet, MongoDbShardedCluster and MongoDbStandalone resources to the new format

A list of the packages installed, and any security vulnerabilities detected in our build process, are outlined here:

• For the MongoDB Enterprise Operator
  https://quay.io/repository/mongodb/mongodb-enterprise-operator?tab=tags
• And for the MongoDB Enterprise Database
  https://quay.io/repository/mongodb/mongodb-enterprise-database?tab=tags

• The Operator and Database images are now based on debian:stretch-slim which is the latest and up-to-date Docker image for Debian 9.

• Perform Ops Manager clean-up on deletion of MongoDB resource without the use of finalisers.
• Bug fix: race conditions when communicating with Ops Manager.
• Bug fix: ImagePullSecrets being incorrectly initialized in OpenShift.
• Bug fix: unintended fetching of closed projects.
• Bug fix: creation of duplicate organizations.
• Bug fix: reconciliation could fail for the MongoDB resource if some other resources in Ops Manager were in error state

• Improved detailed status field for MongoDB resources
• The Operator watches changes to configuration parameters in project config map and credentials secret and performs rolling upgrade for relevant Kubernetes resources
• Added json structured logging for Automation Agent pods
• Support DNS SRV records for MongoDb access
• Bug fix: avoiding unnecessary reconciliation
• Bug fix: Improved Ops Manager/Cloud Manager state management for deleted resources

Release Notes

• Codebase was refactored to use controller-runtime library. This fixed some problems when Operator could leave resources in inconsistent state and introduced proper reconciliation process.
  • Important: as the removal process was refactored to use finalizer headers it's important to delete mongodb resources before deleting Operator/namespace otherwise the deletion operation can get stuck. More details in the docs
• New status field for all MongoDB Kubernetes resources
• Operator can be configured to watch any single namespace or all namespaces in a cluster (requires cluster role)
• Database logging has been improved: new configuration property logLevel (INFO by default) was added. Automation Agent and mongodb logs are merged in to a single log stream.
• New configuration Operator timeouts was added. It defines waiting time for database pods start while updating MongoDB Kubernetes resources
• Bug fix: Fixed failure detection for mongos

Release Notes

• Database image downloads Automation Agent from OpsManager/Cloud Manager during start up
• Bug fix: Fixed Ops Manager/Cloud Manager errors for duplicated project names

Release Notes

• For Replica Set or Sharded Cluster with configured backup, the Operator can now properly orchestrate resource removal and shutdown backup process.
• Improved Persistent Volume support:
  • It is now possible to mount mongodb data, the mongodb journal and all logs into different volumes. Configure podSpec.persistence.multiple to enable this behaviour.
    Define podSpec.persistence.single to mount all there directories into a single volume. This is the default behavior.
  • A new parameter, labelSelector, defines labels for a selector which controls which volumes are mounted.
  • A default StorageClass will be used if storageClass is not defined in the persistence configuration. Most Kubernetes deployments support dynamic volume provisioning, so volumes should be available by default without needing to configure this field.

Release Notes

• The Operator will not create the CustomResourceDefinition objects any more, they need to be created manually by the administrator. A new yaml file is provided (crds.yaml) that can be applied to create/configure these objects.
• ClusterRoles are not required any more. We have changed the way the Operator watches resources. Until the last release, the Operator would watch for any resource on any Namespace. With 0.3, the operator will watch for resources in the same Namespace that it was created in. To support multiple namespaces, multiple Operators can be installed. This allows isolation of MongoDB deployments.
• We have made permissions changes to how PersistentVolumes are mounted.
• Added configuration to operator to not create SecurityContexts for Pods. This solves an issue with OpenShift which disallows this setting when SecurityContextContraints are in place. If using Helm, the variable managedSecurityContext set to true will tell the operator to not create SecurityContexts for Pods, satisfying the OpenShift requirement.
• A combination of projectName and orgId is now used to configure Ops Manager or Cloud Manager. If a project with this projectName does not exist, it will be created by the Operator.

Known Issues

• In a MongoDB deployment with no PersistentVolume configured, disk metrics will not show up in Ops Manager or Cloud Manager Deployments view.

• WiredTiger Cache Size attribute is now set by the operator.

The WiredTiger Cache Size for individual mongods is now set by the operator, considering the memory limits of the mongod container. The formula used to calculate this attribute can be found in this this document.