Considerations

Deploy Multiple MongoDB Replica Sets

We recommend that you use a single instance of the Kubernetes Operator to deploy and manage your MongoDB replica sets.

To deploy more than 10 MongoDB replica sets in parallel, you can increase the thread count of your Kubernetes Operator instance.

Specify CPU and Memory Resource Requirements

In Kubernetes, each Pod includes parameters that allow you to specify CPU resources and memory resources for each container in the Pod.

To indicate resource bounds, Kubernetes uses the requests and limits parameters, where:

• request indicates a lower bound of a resource.

• limit indicates an upper bound of a resource.

The following sections illustrate how to:

• set CPU and Memory for the Operator Pod.

• set CPU and Memory for MongoDB Pods.

For the Pods hosting Ops Manager, use the default resource limits configurations.

Consider Recommended Practices for Storage

The following recommendations are applicable to MongoDB deployments and Ops Manager Application database deployments managed by the Kubernetes Operator.

• Use the storage class that provides the best performance given your constraints. See scale a deployment to learn more.

• Provision Persistent Volumes that support ReadWriteOnce for your storage class.

• On a worker node level, apply the best practices from MongoDB on Linux.

• Ensure that you use a storage class that supports volume expansion to enable database volume resizing.

Use Static Containers (Public Preview)

Static containers are simpler and more secure than non-static containers. Static containers are immutable at runtime, which means that they don't change from the image used to create the container. In addition:

• While running, static containers don't download binaries or run scripts or other utilities over network connections. Static containers only download runtime configuration files.

• While running, static containers don't modify any file except storage volume mounts.

• You can run security scans on the container images to determine what is actually run as a live container, and the container that runs won't run binaries other than what was defined in the image.

• Static containers don't require that you host the MongoDB binary on either Ops Manager or another HTTPS server, which is especially useful if you have an air-gapped environment.

• You can't run extensive CMD scripts for the static container.

• You can't copy files between static containers using initContainer.

To learn more, see Static Containers (Public Preview).

Co-locate mongos Pods with Your Applications

You can run the lightweight mongos instance on the same node as your apps using MongoDB. The Kubernetes Operator supports standard Kubernetes node affinity and anti-affinity features. Using these features, you can force install the mongos on the same node as your application.

The following abbreviated example shows affinity and multiple availability zones configuration.

The podAffinity key determines whether to install an application on the same Pod, node, or data center as another application.

To specify Pod affinity:

1. Add a label and value in the spec.podSpec.podTemplate.metadata.labels YAML collection to tag the deployment. See spec.podSpec.podTemplate.metadata, and the Kubernetes PodSpec v1 core API.

2. Specify which label the mongos uses in the spec.mongosPodSpec.podAffinity .requiredDuringSchedulingIgnoredDuringExecution.labelSelector YAML collection. The matchExpressions collection defines the label that the Kubernetes Operator uses to identify the Pod for hosting the mongos.

1
apiVersion: mongodb.com/v1
2
kind: MongoDB
3
metadata:
4
  name: my-replica-set
5
spec:
6
  members: 3
7
  version: 4.2.1-ent
8
  service: my-service
9
10
  ...
11
    podTemplate:
12
      spec:
13
        affinity:
14
          podAffinity:
15
            requiredDuringSchedulingIgnoredDuringExecution:
16
              - labelSelector:
17
                  matchExpressions:
18
                    - key: security
19
                      operator: In
20
                      values:
21
                        - S1
22
                topologyKey: failure-domain.beta.kubernetes.io/zone
23
          nodeAffinity:
24
            requiredDuringSchedulingIgnoredDuringExecution:
25
              nodeSelectorTerms:
26
                - matchExpressions:
27
                    - key: kubernetes.io/e2e-az-name
28
                      operator: In
29
                      values:
30
                        - e2e-az1
31
                        - e2e-az2
32
          podAntiAffinity:
33
            requiredDuringSchedulingIgnoredDuringExecution:
34
            - podAffinityTerm:
35
                topologyKey: nodeId

See the full example of multiple availability zones and node affinity configuration in replica-set-affinity.yaml in the Affinity Samples directory.

This directory also contains sample affinity and multiple zones configurations for sharded clusters and standalone MongoDB deployments.

Name Your MongoDB Service with its Purpose

Set the spec.service parameter to a value that identifies this deployment's purpose, as illustrated in the following example.

1
apiVersion: mongodb.com/v1
2
kind: MongoDB
3
metadata:
4
  name: my-replica-set
5
spec:
6
  members: 3
7
  version: "6.0.0-ent"
8
  service: drilling-pumps-geosensors
9
  featureCompatibilityVersion: "4.0"

Use Labels to Differentiate Between Deployments

Use the Pod affinity Kubernetes feature to:

• Separate different MongoDB resources, such as test, staging, and production environments.

• Place Pods on some specific nodes to take advantage of features such as SSD support.

1
mongosPodSpec:
2
  podAffinity:
3
    requiredDuringSchedulingIgnoredDuringExecution:
4
      - labelSelector:
5
        matchExpressions:
6
        - key: security
7
          operator: In
8
          values:
9
          - S1
10
        topologyKey: failure-domain.beta.kubernetes.io/zone

Customize the CustomResourceDefinitions that the Kubernetes Operator Watches

You can specify which custom resources you want the Kubernetes Operator to watch. This allows you to install the CustomResourceDefinition for only the resources that you want the Kubernetes Operator to manage.

You must use helm to configure the Kubernetes Operator to watch only the custom resources you specify. Follow the relevant helm installation instructions, but make the following adjustments:

1. Decide which CustomResourceDefinitions you want to install. You can install any number of the following:

   Value	Description
   mongodb	Install the CustomResourceDefinitions for database resources and watch those resources.
   mongodbusers	Install the CustomResourceDefinitions for MongoDB user resources and watch those resources.
   opsmanagers	Install the CustomResourceDefinitions for Ops Manager resources and watch those resources.

2. Install the Helm Chart and specify which CustomResourceDefinitions you want the Kubernetes Operator to watch.

   Separate each custom resource with a comma:

   helm install <deployment-name> mongodb/enterprise-operator \
     --set operator.watchedResources="{mongodb,mongodbusers}" \
        --skip-crds

Ensure Proper Persistence Configuration

The Kubernetes deployments orchestrated by the Kubernetes Operator are stateful. The Kubernetes container uses Persistent Volumes to maintain the cluster state between restarts.

To satisfy the statefulness requirement, the Kubernetes Operator performs the following actions:

• Creates Persistent Volumes for your MongoDB deployment.

• Mounts storage devices to one or more directories called mount points.

• Creates one persistent volume for each MongoDB mount point.

• Sets the default path in each Kubernetes container to /data.

To meet your MongoDB cluster's storage needs, make the following changes in your configuration for each replica set deployed with the Kubernetes Operator:

• Verify that persistent volumes are enabled in spec.persistent. This setting defaults to true.

• Specify a sufficient amount of storage for the Kubernetes Operator to allocate for each of the volumes. The volumes store the data and the logs.

  • To set multiple volumes, each for data, logs, and the oplog, use spec.podSpec.persistence.multiple.data.

  • To set a single volume to store data, logs, and the oplog, use spec.podSpec.persistence.single.

The following abbreviated example shows recommended persistent storage sizes.

1
apiVersion: mongodb.com/v1
2
kind: MongoDB
3
metadata:
4
  name: my-replica-cluster
5
spec:
6
7
  ...
8
  persistent: true
9
10
11
  shardPodSpec:
12
  ...
13
    persistence:
14
      multiple:
15
        data:
16
          storage: "20Gi"
17
        logs:
18
          storage: "4Gi"
19
          storageClass: standard

For a full example of persistent volumes configuration, see replica-set-persistent-volumes.yaml in the Persistent Volumes Samples directory. This directory also contains sample persistent volumes configurations for sharded clusters and standalone deployments.

Set CPU and Memory Utilization Bounds for the Kubernetes Operator Pod

When you deploy MongoDB replica sets with the Kubernetes Operator, the initial reconcilliation process increases CPU usage for the Pod running the Kubernetes Operator. However, when the replica set deployment process completes, the CPU usage by the Kubernetes Operator reduces considerably.

For production deployments, to satisfy deploying up to 50 MongoDB replica sets or sharded clusters in parallel with the Kubernetes Operator, set the CPU and memory resources and limits for the Kubernetes Operator Pod as follows:

• spec.template.spec.containers.resources.requests.cpu to 500m

• spec.template.spec.containers.resources.limits.cpu to 1100m

• spec.template.spec.containers.resources.requests.memory to 200Mi

• spec.template.spec.containers.resources.limits.memory to 1Gi

If you use Helm to deploy resources, define these values in the values.yaml file.

The following abbreviated example shows the configuration with recommended CPU and memory bounds for the Kubernetes Operator Pod in your deployment of 50 replica sets or sharded clusters. If you are deploying fewer than 50 MongoDB clusters, you may use lower numbers in the configuration file for the Kubernetes Operator Pod.

1
apiVersion: apps/v1
2
kind: Deployment
3
metadata:
4
 name: mongodb-enterprise-operator
5
 namespace: mongodb
6
spec:
7
 replicas: 1
8
 selector:
9
  matchLabels:
10
     app.kubernetes.io/component: controller
11
     app.kubernetes.io/name: mongodb-enterprise-operator
12
     app.kubernetes.io/instance: mongodb-enterprise-operator
13
 template:
14
  metadata:
15
   labels:
16
     app.kubernetes.io/component: controller
17
     app.kubernetes.io/name: mongodb-enterprise-operator
18
     app.kubernetes.io/instance: mongodb-enterprise-operator
19
   spec:
20
     serviceAccountName: mongodb-enterprise-operator
21
     securityContext:
22
       runAsNonRoot: true
23
       runAsUser: 2000
24
     containers:
25
     - name: mongodb-enterprise-operator
26
       image: quay.io/mongodb/mongodb-enterprise-operator:1.9.2
27
       imagePullPolicy: Always
28
       args:
29
        - "-watch-resource=mongodb"
30
        - "-watch-resource=opsmanagers"
31
        - "-watch-resource=mongodbusers"
32
       command:
33
        - "/usr/local/bin/mongodb-enterprise-operator"
34
       resources:
35
         limits:
36
           cpu: 1100m
37
           memory: 1Gi
38
         requests:
39
           cpu: 500m
40
           memory: 200Mi

For a full example of CPU and memory utilization resources and limits for the Kubernetes Operator Pod that satisfy parallel deployment of up to 50 MongoDB replica sets, see the mongodb-enterprise.yaml file.

Set CPU and Memory Utilization Bounds for MongoDB Pods

The values for Pods hosting replica sets or sharded clusters map to the requests field for CPU and memory for the created Pod. These values are consistent with considerations stated for MongoDB hosts.

The Kubernetes Operator uses its allocated memory for processing, for the WiredTiger cache, and for storing packages during the deployments.

For production deployments, set the CPU and memory resources and limits for the MongoDB Pod as follows:

• spec.podSpec.podTemplate.spec.containers.resources.requests.cpu to 0.25

• spec.podSpec.podTemplate.spec.containers.resources.limits.cpu to 0.25

• spec.podSpec.podTemplate.spec.containers.resources.requests.memory to 512M

• spec.podSpec.podTemplate.spec.containers.resources.limits.memory to 512M

If you use Helm to deploy resources, define these values in the values.yaml file.

The following abbreviated example shows the configuration with recommended CPU and memory bounds for each Pod hosting a MongoDB replica set member in your deployment.

1
apiVersion: mongodb.com/v1
2
kind: MongoDB
3
metadata:
4
name: my-replica-set
5
spec:
6
  members: 3
7
  version: 4.0.0-ent
8
  service: my-service
9
  ...
10
11
  persistent: true
12
  podSpec:
13
    podTemplate:
14
      spec:
15
        containers:
16
        - name: mongodb-enterprise-database
17
          resources:
18
            limits:
19
              cpu: "0.25"
20
              memory: 512M

For a full example of CPU and memory utilization resources and limits for Pods hosting MongoDB replica set members, see the replica-set-podspec.yaml file in the MongoDB Podspec Samples directory.

This directory also contains sample CPU and memory limits configurations for Pods used for:

• A sharded cluster, in the sharded-cluster-podspec.yaml.

• Standalone MongoDB deployments, in the standalone-podspec.yaml.

Use Multiple Availability Zones

Set the Kubernetes Operator and StatefulSets to distribute all members of one replica set to different nodes to ensure high availability.

The following abbreviated example shows affinity and multiple availability zones configuration.

1
apiVersion: mongodb.com/v1
2
kind: MongoDB
3
metadata:
4
  name: my-replica-set
5
spec:
6
  members: 3
7
  version: 4.2.1-ent
8
  service: my-service
9
  ...
10
    podAntiAffinityTopologyKey: nodeId
11
    podAffinity:
12
      requiredDuringSchedulingIgnoredDuringExecution:
13
      - labelSelector:
14
        matchExpressions:
15
        - key: security
16
          operator: In
17
          values:
18
          - S1
19
        topologyKey: failure-domain.beta.kubernetes.io/zone
20
21
    nodeAffinity:
22
       requiredDuringSchedulingIgnoredDuringExecution:
23
         nodeSelectorTerms:
24
         - matchExpressions:
25
           - key: kubernetes.io/e2e-az-name
26
           operator: In
27
           values:
28
           - e2e-az1
29
           - e2e-az2

In this example, the Kubernetes Operator schedules the Pods deployment to the nodes which have the label kubernetes.io/e2e-az-name in e2e-az1 or e2e-az2 availability zones. Change nodeAffinity to schedule the deployment of Pods to the desired availability zones.

See the full example of multiple availability zones configuration in replica-set-affinity.yaml in the Affinity Samples directory.

This directory also contains sample affinity and multiple zones configurations for sharded clusters and standalone MongoDB deployments.

Increase Thread Count to Run multiple Reconciliation Processes in Parallel

If you plan to deploy more than 10 MongoDB replica sets in parallel, you can configure the Kubernetes Operator to run multiple reconciliation processes in parallel by setting MDB_MAX_CONCURRENT_RECONCILES environment variable in your Kubernetes Operator deployment or or through the operator.maxConcurrentReconciles field in your Helm values.yaml file to configure a higher thread count.

Increasing the thread count of the Kubernetes Operator allows you to vertically scale your Kubernetes Operator deployment to hundreds of MongoDB resources running within your Kubernetes cluster and optimize CPU utilization.

Please monitor Kubernetes API server and Kubernetes Operator resource usage and adjust their respective resource allocation if necessary.