Advanced migration options

You can automate your migrations and modify the MigPlan and MigrationController custom resources in order to perform large-scale migrations and to improve performance.

Terminology

Table 1. MTC terminology
TermDefinition

Source cluster

Cluster from which the applications are migrated.

Destination cluster[1]

Cluster to which the applications are migrated.

Replication repository

Object storage used for copying images, volumes, and Kubernetes objects during indirect migration or for Kubernetes objects during direct volume migration or direct image migration.

The replication repository must be accessible to all clusters.

Host cluster

Cluster on which the migration-controller pod and the web console are running. The host cluster is usually the destination cluster but this is not required.

The host cluster does not require an exposed registry route for direct image migration.

Remote cluster

A remote cluster is usually the source cluster but this is not required.

A remote cluster requires a Secret custom resource that contains the migration-controller service account token.

A remote cluster requires an exposed secure registry route for direct image migration.

Indirect migration

Images, volumes, and Kubernetes objects are copied from the source cluster to the replication repository and then from the replication repository to the destination cluster.

Direct volume migration

Persistent volumes are copied directly from the source cluster to the destination cluster.

Direct image migration

Images are copied directly from the source cluster to the destination cluster.

Stage migration

Data is copied to the destination cluster without stopping the application.

Running a stage migration multiple times reduces the duration of the cutover migration.

Cutover migration

The application is stopped on the source cluster and its resources are migrated to the destination cluster.

State migration

Application state is migrated by copying specific persistent volume claims to the destination cluster.

Rollback migration

Rollback migration rolls back a completed migration.

1 Called the target cluster in the MTC web console.

Migrating applications by using the command line

You can migrate applications with the MTC API by using the command line interface (CLI) in order to automate the migration.

Migration prerequisites

  • You must be logged in as a user with cluster-admin privileges on all clusters.

Direct image migration

  • You must ensure that the secure OpenShift image registry of the source cluster is exposed.

  • You must create a route to the exposed registry.

Direct volume migration

  • If your clusters use proxies, you must configure an Stunnel TCP proxy.

Clusters

  • The source cluster must be upgraded to the latest MTC z-stream release.

  • The MTC version must be the same on all clusters.

Network

  • The clusters have unrestricted network access to each other and to the replication repository.

  • If you copy the persistent volumes with move, the clusters must have unrestricted network access to the remote volumes.

  • You must enable the following ports on an OKD 4 cluster:

    • 6443 (API server)

    • 443 (routes)

    • 53 (DNS)

  • You must enable port 443 on the replication repository if you are using TLS.

Persistent volumes (PVs)

  • The PVs must be valid.

  • The PVs must be bound to persistent volume claims.

  • If you use snapshots to copy the PVs, the following additional prerequisites apply:

    • The cloud provider must support snapshots.

    • The PVs must have the same cloud provider.

    • The PVs must be located in the same geographic region.

    • The PVs must have the same storage class.

Creating a registry route for direct image migration

For direct image migration, you must create a route to the exposed OpenShift image registry on all remote clusters.

Prerequisites

  • The OpenShift image registry must be exposed to external traffic on all remote clusters.

    The OKD 4 registry is exposed by default.

Procedure

  • To create a route to an OKD 4 registry, run the following command:

    1. $ oc create route passthrough --service=image-registry -n openshift-image-registry

Proxy configuration

For OKD 4.1 and earlier versions, you must configure proxies in the MigrationController custom resource (CR) manifest after you install the Migration Toolkit for Containers Operator because these versions do not support a cluster-wide proxy object.

For OKD 4.2 to 4.14, the Migration Toolkit for Containers (MTC) inherits the cluster-wide proxy settings. You can change the proxy parameters if you want to override the cluster-wide proxy settings.

Direct volume migration

Direct Volume Migration (DVM) was introduced in MTC 1.4.2. DVM supports only one proxy. The source cluster cannot access the route of the target cluster if the target cluster is also behind a proxy.

If you want to perform a DVM from a source cluster behind a proxy, you must configure a TCP proxy that works at the transport layer and forwards the SSL connections transparently without decrypting and re-encrypting them with their own SSL certificates. A Stunnel proxy is an example of such a proxy.

TCP proxy setup for DVM

You can set up a direct connection between the source and the target cluster through a TCP proxy and configure the stunnel_tcp_proxy variable in the MigrationController CR to use the proxy:

  1. apiVersion: migration.openshift.io/v1alpha1
  2. kind: MigrationController
  3. metadata:
  4.   name: migration-controller
  5.   namespace: openshift-migration
  6. spec:
  7.   [...]
  8.   stunnel_tcp_proxy: http://username:password@ip:port

Direct volume migration (DVM) supports only basic authentication for the proxy. Moreover, DVM works only from behind proxies that can tunnel a TCP connection transparently. HTTP/HTTPS proxies in man-in-the-middle mode do not work. The existing cluster-wide proxies might not support this behavior. As a result, the proxy settings for DVM are intentionally kept different from the usual proxy configuration in MTC.

Why use a TCP proxy instead of an HTTP/HTTPS proxy?

You can enable DVM by running Rsync between the source and the target cluster over an OpenShift route. Traffic is encrypted using Stunnel, a TCP proxy. The Stunnel running on the source cluster initiates a TLS connection with the target Stunnel and transfers data over an encrypted channel.

Cluster-wide HTTP/HTTPS proxies in OpenShift are usually configured in man-in-the-middle mode where they negotiate their own TLS session with the outside servers. However, this does not work with Stunnel. Stunnel requires that its TLS session be untouched by the proxy, essentially making the proxy a transparent tunnel which simply forwards the TCP connection as-is. Therefore, you must use a TCP proxy.

Known issue

Migration fails with error Upgrade request required

The migration Controller uses the SPDY protocol to execute commands within remote pods. If the remote cluster is behind a proxy or a firewall that does not support the SPDY protocol, the migration controller fails to execute remote commands. The migration fails with the error message Upgrade request required. Workaround: Use a proxy that supports the SPDY protocol.

In addition to supporting the SPDY protocol, the proxy or firewall also must pass the Upgrade HTTP header to the API server. The client uses this header to open a websocket connection with the API server. If the Upgrade header is blocked by the proxy or firewall, the migration fails with the error message Upgrade request required. Workaround: Ensure that the proxy forwards the Upgrade header.

Tuning network policies for migrations

OpenShift supports restricting traffic to or from pods using NetworkPolicy or EgressFirewalls based on the network plugin used by the cluster. If any of the source namespaces involved in a migration use such mechanisms to restrict network traffic to pods, the restrictions might inadvertently stop traffic to Rsync pods during migration.

Rsync pods running on both the source and the target clusters must connect to each other over an OpenShift Route. Existing NetworkPolicy or EgressNetworkPolicy objects can be configured to automatically exempt Rsync pods from these traffic restrictions.

NetworkPolicy configuration
Egress traffic from Rsync pods

You can use the unique labels of Rsync pods to allow egress traffic to pass from them if the NetworkPolicy configuration in the source or destination namespaces blocks this type of traffic. The following policy allows all egress traffic from Rsync pods in the namespace:

  1. apiVersion: networking.k8s.io/v1
  2. kind: NetworkPolicy
  3. metadata:
  4.   name: allow-all-egress-from-rsync-pods
  5. spec:
  6.   podSelector:
  7.     matchLabels:
  8.       owner: directvolumemigration
  9.       app: directvolumemigration-rsync-transfer
  10.   egress:
  11.   - {}
  12.   policyTypes:
  13.   - Egress
Ingress traffic to Rsync pods
  1. apiVersion: networking.k8s.io/v1
  2. kind: NetworkPolicy
  3. metadata:
  4.   name: allow-all-egress-from-rsync-pods
  5. spec:
  6.   podSelector:
  7.     matchLabels:
  8.       owner: directvolumemigration
  9.       app: directvolumemigration-rsync-transfer
  10.   ingress:
  11.   - {}
  12.   policyTypes:
  13.   - Ingress
EgressNetworkPolicy configuration

The EgressNetworkPolicy object or Egress Firewalls are OpenShift constructs designed to block egress traffic leaving the cluster.

Unlike the NetworkPolicy object, the Egress Firewall works at a project level because it applies to all pods in the namespace. Therefore, the unique labels of Rsync pods do not exempt only Rsync pods from the restrictions. However, you can add the CIDR ranges of the source or target cluster to the Allow rule of the policy so that a direct connection can be setup between two clusters.

Based on which cluster the Egress Firewall is present in, you can add the CIDR range of the other cluster to allow egress traffic between the two:

  1. apiVersion: network.openshift.io/v1
  2. kind: EgressNetworkPolicy
  3. metadata:
  4.   name: test-egress-policy
  5.   namespace: <namespace>
  6. spec:
  7.   egress:
  8.   - to:
  9.       cidrSelector: <cidr_of_source_or_target_cluster>
  10.     type: Deny
Choosing alternate endpoints for data transfer

By default, DVM uses an OKD route as an endpoint to transfer PV data to destination clusters. You can choose another type of supported endpoint, if cluster topologies allow.

For each cluster, you can configure an endpoint by setting the rsync_endpoint_type variable on the appropriate destination cluster in your MigrationController CR:

  1. apiVersion: migration.openshift.io/v1alpha1
  2. kind: MigrationController
  3. metadata:
  4.   name: migration-controller
  5.   namespace: openshift-migration
  6. spec:
  7.   [...]
  8.   rsync_endpoint_type: [NodePort|ClusterIP|Route]
Configuring supplemental groups for Rsync pods

When your PVCs use a shared storage, you can configure the access to that storage by adding supplemental groups to Rsync pod definitions in order for the pods to allow access:

Table 2. Supplementary groups for Rsync pods
VariableTypeDefaultDescription

src_supplemental_groups

string

Not set

Comma-separated list of supplemental groups for source Rsync pods

target_supplemental_groups

string

Not set

Comma-separated list of supplemental groups for target Rsync pods

Example usage

The MigrationController CR can be updated to set values for these supplemental groups:

  1. spec:
  2.   src_supplemental_groups: "1000,2000"
  3.   target_supplemental_groups: "2000,3000"

Configuring proxies

Prerequisites

  • You must be logged in as a user with cluster-admin privileges on all clusters.

Procedure

  1. Get the MigrationController CR manifest:

    1. $ oc get migrationcontroller <migration_controller> -n openshift-migration

  2. Update the proxy parameters:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigrationController
    3. metadata:
    4.   name: <migration_controller>
    5.   namespace: openshift-migration
    6. ...
    7. spec:
    8.   stunnel_tcp_proxy: http://<username>:<password>@<ip>:<port> (1)
    9.   noProxy: example.com (2)
    1Stunnel proxy URL for direct volume migration.
    2Comma-separated list of destination domain names, domains, IP addresses, or other network CIDRs to exclude proxying.

    Preface a domain with . to match subdomains only. For example, .y.com matches x.y.com, but not y.com. Use * to bypass proxy for all destinations. If you scale up workers that are not included in the network defined by the networking.machineNetwork[].cidr field from the installation configuration, you must add them to this list to prevent connection issues.

    This field is ignored if neither the httpProxy nor the httpsProxy field is set.

  3. Save the manifest as migration-controller.yaml.

  4. Apply the updated manifest:

    1. $ oc replace -f migration-controller.yaml -n openshift-migration

Migrating an application by using the MTC API

You can migrate an application from the command line by using the Migration Toolkit for Containers (MTC) API.

Procedure

  1. Create a MigCluster CR manifest for the host cluster:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: migration.openshift.io/v1alpha1
    3. kind: MigCluster
    4. metadata:
    5.   name: <host_cluster>
    6.   namespace: openshift-migration
    7. spec:
    8.   isHostCluster: true
    9. EOF

  2. Create a Secret object manifest for each remote cluster:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: v1
    3. kind: Secret
    4. metadata:
    5.   name: <cluster_secret>
    6.   namespace: openshift-config
    7. type: Opaque
    8. data:
    9.   saToken: <sa_token> (1)
    10. EOF
    1Specify the base64-encoded migration-controller service account (SA) token of the remote cluster. You can obtain the token by running the following command:
    1. $ oc sa get-token migration-controller -n openshift-migration | base64 -w 0

  3. Create a MigCluster CR manifest for each remote cluster:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: migration.openshift.io/v1alpha1
    3. kind: MigCluster
    4. metadata:
    5.   name: <remote_cluster> (1)
    6.   namespace: openshift-migration
    7. spec:
    8.   exposedRegistryPath: <exposed_registry_route> (2)
    9.   insecure: false (3)
    10.   isHostCluster: false
    11.   serviceAccountSecretRef:
    12.     name: <remote_cluster_secret> (4)
    13.     namespace: openshift-config
    14.   url: <remote_cluster_url> (5)
    15. EOF
    1Specify the Cluster CR of the remote cluster.
    2Optional: For direct image migration, specify the exposed registry route.
    3SSL verification is enabled if false. CA certificates are not required or checked if true.
    4Specify the Secret object of the remote cluster.
    5Specify the URL of the remote cluster.

  4. Verify that all clusters are in a Ready state:

    1. $ oc describe cluster <cluster>

  5. Create a Secret object manifest for the replication repository:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: v1
    3. kind: Secret
    4. metadata:
    5.   namespace: openshift-config
    6.   name: <migstorage_creds>
    7. type: Opaque
    8. data:
    9.   aws-access-key-id: <key_id_base64> (1)
    10.   aws-secret-access-key: <secret_key_base64> (2)
    11. EOF
    1Specify the key ID in base64 format.
    2Specify the secret key in base64 format.

    AWS credentials are base64-encoded by default. For other storage providers, you must encode your credentials by running the following command with each key:

    1. $ echo -n "<key>" | base64 -w 0 (1)
    1Specify the key ID or the secret key. Both keys must be base64-encoded.

  6. Create a MigStorage CR manifest for the replication repository:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: migration.openshift.io/v1alpha1
    3. kind: MigStorage
    4. metadata:
    5.   name: <migstorage>
    6.   namespace: openshift-migration
    7. spec:
    8.   backupStorageConfig:
    9.     awsBucketName: <bucket> (1)
    10.     credsSecretRef:
    11.       name: <storage_secret> (2)
    12.       namespace: openshift-config
    13.   backupStorageProvider: <storage_provider> (3)
    14.   volumeSnapshotConfig:
    15.     credsSecretRef:
    16.       name: <storage_secret> (4)
    17.       namespace: openshift-config
    18.   volumeSnapshotProvider: <storage_provider> (5)
    19. EOF
    1Specify the bucket name.
    2Specify the Secrets CR of the object storage. You must ensure that the credentials stored in the Secrets CR of the object storage are correct.
    3Specify the storage provider.
    4Optional: If you are copying data by using snapshots, specify the Secrets CR of the object storage. You must ensure that the credentials stored in the Secrets CR of the object storage are correct.
    5Optional: If you are copying data by using snapshots, specify the storage provider.

  7. Verify that the MigStorage CR is in a Ready state:

    1. $ oc describe migstorage <migstorage>

  8. Create a MigPlan CR manifest:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: migration.openshift.io/v1alpha1
    3. kind: MigPlan
    4. metadata:
    5.   name: <migplan>
    6.   namespace: openshift-migration
    7. spec:
    8.   destMigClusterRef:
    9.     name: <host_cluster>
    10.     namespace: openshift-migration
    11.   indirectImageMigration: true (1)
    12.   indirectVolumeMigration: true (2)
    13.   migStorageRef:
    14.     name: <migstorage> (3)
    15.     namespace: openshift-migration
    16.   namespaces:
    17.     - <source_namespace_1> (4)
    18.     - <source_namespace_2>
    19.     - <source_namespace_3>:<destination_namespace> (5)
    20.   srcMigClusterRef:
    21.     name: <remote_cluster> (6)
    22.     namespace: openshift-migration
    23. EOF
    1Direct image migration is enabled if false.
    2Direct volume migration is enabled if false.
    3Specify the name of the MigStorage CR instance.
    4Specify one or more source namespaces. By default, the destination namespace has the same name.
    5Specify a destination namespace if it is different from the source namespace.
    6Specify the name of the source cluster MigCluster instance.

  9. Verify that the MigPlan instance is in a Ready state:

    1. $ oc describe migplan <migplan> -n openshift-migration

  10. Create a MigMigration CR manifest to start the migration defined in the MigPlan instance:

    1. $ cat << EOF | oc apply -f -
    2. apiVersion: migration.openshift.io/v1alpha1
    3. kind: MigMigration
    4. metadata:
    5.  name: <migmigration>
    6.  namespace: openshift-migration
    7. spec:
    8.  migPlanRef:
    9.    name: <migplan> (1)
    10.    namespace: openshift-migration
    11.  quiescePods: true (2)
    12.  stage: false (3)
    13.  rollback: false (4)
    14. EOF
    1Specify the MigPlan CR name.
    2The pods on the source cluster are stopped before migration if true.
    3A stage migration, which copies most of the data without stopping the application, is performed if true.
    4A completed migration is rolled back if true.

  11. Verify the migration by watching the MigMigration CR progress:

    1. $ oc watch migmigration <migmigration> -n openshift-migration

    The output resembles the following:

    Example output

    1. Name:         c8b034c0-6567-11eb-9a4f-0bc004db0fbc
    2. Namespace:    openshift-migration
    3. Labels:       migration.openshift.io/migplan-name=django
    4. Annotations:  openshift.io/touch: e99f9083-6567-11eb-8420-0a580a81020c
    5. API Version:  migration.openshift.io/v1alpha1
    6. Kind:         MigMigration
    7. ...
    8. Spec:
    9.  Mig Plan Ref:
    10.    Name:       migplan
    11.    Namespace:  openshift-migration
    12.  Stage:        false
    13. Status:
    14.  Conditions:
    15.    Category:              Advisory
    16.    Last Transition Time:  2021-02-02T15:04:09Z
    17.    Message:               Step: 19/47
    18.    Reason:                InitialBackupCreated
    19.    Status:                True
    20.    Type:                  Running
    21.    Category:              Required
    22.    Last Transition Time:  2021-02-02T15:03:19Z
    23.    Message:               The migration is ready.
    24.    Status:                True
    25.    Type:                  Ready
    26.    Category:              Required
    27.    Durable:               true
    28.    Last Transition Time:  2021-02-02T15:04:05Z
    29.    Message:               The migration registries are healthy.
    30.    Status:                True
    31.    Type:                  RegistriesHealthy
    32.  Itinerary:               Final
    33.  Observed Digest:         7fae9d21f15979c71ddc7dd075cb97061895caac5b936d92fae967019ab616d5
    34.  Phase:                   InitialBackupCreated
    35.  Pipeline:
    36.    Completed:  2021-02-02T15:04:07Z
    37.    Message:    Completed
    38.    Name:       Prepare
    39.    Started:    2021-02-02T15:03:18Z
    40.    Message:    Waiting for initial Velero backup to complete.
    41.    Name:       Backup
    42.    Phase:      InitialBackupCreated
    43.    Progress:
    44.      Backup openshift-migration/c8b034c0-6567-11eb-9a4f-0bc004db0fbc-wpc44: 0 out of estimated total of 0 objects backed up (5s)
    45.    Started:        2021-02-02T15:04:07Z
    46.    Message:        Not started
    47.    Name:           StageBackup
    48.    Message:        Not started
    49.    Name:           StageRestore
    50.    Message:        Not started
    51.    Name:           DirectImage
    52.    Message:        Not started
    53.    Name:           DirectVolume
    54.    Message:        Not started
    55.    Name:           Restore
    56.    Message:        Not started
    57.    Name:           Cleanup
    58.  Start Timestamp:  2021-02-02T15:03:18Z
    59. Events:
    60.  Type    Reason   Age                 From                     Message
    61.  ----    ------   ----                ----                     -------
    62.  Normal  Running  57s                 migmigration_controller  Step: 2/47
    63.  Normal  Running  57s                 migmigration_controller  Step: 3/47
    64.  Normal  Running  57s (x3 over 57s)   migmigration_controller  Step: 4/47
    65.  Normal  Running  54s                 migmigration_controller  Step: 5/47
    66.  Normal  Running  54s                 migmigration_controller  Step: 6/47
    67.  Normal  Running  52s (x2 over 53s)   migmigration_controller  Step: 7/47
    68.  Normal  Running  51s (x2 over 51s)   migmigration_controller  Step: 8/47
    69.  Normal  Ready    50s (x12 over 57s)  migmigration_controller  The migration is ready.
    70.  Normal  Running  50s                 migmigration_controller  Step: 9/47
    71.  Normal  Running  50s                 migmigration_controller  Step: 10/47

State migration

You can perform repeatable, state-only migrations by using Migration Toolkit for Containers (MTC) to migrate persistent volume claims (PVCs) that constitute an application’s state. You migrate specified PVCs by excluding other PVCs from the migration plan. You can map the PVCs to ensure that the source and the target PVCs are synchronized. Persistent volume (PV) data is copied to the target cluster. The PV references are not moved, and the application pods continue to run on the source cluster.

State migration is specifically designed to be used in conjunction with external CD mechanisms, such as OpenShift Gitops. You can migrate application manifests using GitOps while migrating the state using MTC.

If you have a CI/CD pipeline, you can migrate stateless components by deploying them on the target cluster. Then you can migrate stateful components by using MTC.

You can perform a state migration between clusters or within the same cluster.

State migration migrates only the components that constitute an application’s state. If you want to migrate an entire namespace, use stage or cutover migration.

Prerequisites

  • The state of the application on the source cluster is persisted in PersistentVolumes provisioned through PersistentVolumeClaims.

  • The manifests of the application are available in a central repository that is accessible from both the source and the target clusters.

Procedure

  1. Migrate persistent volume data from the source to the target cluster.

    You can perform this step as many times as needed. The source application continues running.

  2. Quiesce the source application.

    You can do this by setting the replicas of workload resources to 0, either directly on the source cluster or by updating the manifests in GitHub and re-syncing the Argo CD application.

  3. Clone application manifests to the target cluster.

    You can use Argo CD to clone the application manifests to the target cluster.

  4. Migrate the remaining volume data from the source to the target cluster.

    Migrate any new data created by the application during the state migration process by performing a final data migration.

  5. If the cloned application is in a quiesced state, unquiesce it.

  6. Switch the DNS record to the target cluster to re-direct user traffic to the migrated application.

MTC 1.6 cannot quiesce applications automatically when performing state migration. It can only migrate PV data. Therefore, you must use your CD mechanisms for quiescing or unquiescing applications.

MTC 1.7 introduces explicit Stage and Cutover flows. You can use staging to perform initial data transfers as many times as needed. Then you can perform a cutover, in which the source applications are quiesced automatically.

Additional resources

Migration hooks

You can add up to four migration hooks to a single migration plan, with each hook running at a different phase of the migration. Migration hooks perform tasks such as customizing application quiescence, manually migrating unsupported data types, and updating applications after migration.

A migration hook runs on a source or a target cluster at one of the following migration steps:

  • PreBackup: Before resources are backed up on the source cluster.

  • PostBackup: After resources are backed up on the source cluster.

  • PreRestore: Before resources are restored on the target cluster.

  • PostRestore: After resources are restored on the target cluster.

You can create a hook by creating an Ansible playbook that runs with the default Ansible image or with a custom hook container.

Ansible playbook

The Ansible playbook is mounted on a hook container as a config map. The hook container runs as a job, using the cluster, service account, and namespace specified in the MigPlan custom resource. The job continues to run until it reaches the default limit of 6 retries or a successful completion. This continues even if the initial pod is evicted or killed.

The default Ansible runtime image is registry.redhat.io/rhmtc/openshift-migration-hook-runner-rhel7:1.8. This image is based on the Ansible Runner image and includes python-openshift for Ansible Kubernetes resources and an updated oc binary.

Custom hook container

You can use a custom hook container instead of the default Ansible image.

Writing an Ansible playbook for a migration hook

You can write an Ansible playbook to use as a migration hook. The hook is added to a migration plan by using the MTC web console or by specifying values for the spec.hooks parameters in the MigPlan custom resource (CR) manifest.

The Ansible playbook is mounted onto a hook container as a config map. The hook container runs as a job, using the cluster, service account, and namespace specified in the MigPlan CR. The hook container uses a specified service account token so that the tasks do not require authentication before they run in the cluster.

Ansible modules

You can use the Ansible shell module to run oc commands.

Example shell module

  1. - hosts: localhost
  2.   gather_facts: false
  3.   tasks:
  4.   - name: get pod name
  5.     shell: oc get po --all-namespaces

You can use kubernetes.core modules, such as k8s_info, to interact with Kubernetes resources.

Example k8s_facts module

  1. - hosts: localhost
  2.   gather_facts: false
  3.   tasks:
  4.   - name: Get pod
  5.     k8s_info:
  6.       kind: pods
  7.       api: v1
  8.       namespace: openshift-migration
  9.       name: "{{ lookup( 'env', 'HOSTNAME') }}"
  10.     register: pods
  12.   - name: Print pod name
  13.     debug:
  14.       msg: "{{ pods.resources[0].metadata.name }}"

You can use the fail module to produce a non-zero exit status in cases where a non-zero exit status would not normally be produced, ensuring that the success or failure of a hook is detected. Hooks run as jobs and the success or failure status of a hook is based on the exit status of the job container.

Example fail module

  1. - hosts: localhost
  2.   gather_facts: false
  3.   tasks:
  4.   - name: Set a boolean
  5.     set_fact:
  6.       do_fail: true
  8.   - name: "fail"
  9.     fail:
  10.       msg: "Cause a failure"
  11.     when: do_fail

Environment variables

The MigPlan CR name and migration namespaces are passed as environment variables to the hook container. These variables are accessed by using the lookup plugin.

Example environment variables

  1. - hosts: localhost
  2.   gather_facts: false
  3.   tasks:
  4.   - set_fact:
  5.       namespaces: "{{ (lookup( 'env', 'MIGRATION_NAMESPACES')).split(',') }}"
  7.   - debug:
  8.       msg: "{{ item }}"
  9.     with_items: "{{ namespaces }}"
  11.   - debug:
  12.       msg: "{{ lookup( 'env', 'MIGRATION_PLAN_NAME') }}"

Migration plan options

You can exclude, edit, and map components in the MigPlan custom resource (CR).

Excluding resources

You can exclude resources, for example, image streams, persistent volumes (PVs), or subscriptions, from a Migration Toolkit for Containers (MTC) migration plan to reduce the resource load for migration or to migrate images or PVs with a different tool.

By default, the MTC excludes service catalog resources and Operator Lifecycle Manager (OLM) resources from migration. These resources are parts of the service catalog API group and the OLM API group, neither of which is supported for migration at this time.

Procedure

  1. Edit the MigrationController custom resource manifest:

    1. $ oc edit migrationcontroller <migration_controller> -n openshift-migration

  2. Update the spec section by adding parameters to exclude specific resources. For those resources that do not have their own exclusion parameters, add the additional_excluded_resources parameter:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigrationController
    3. metadata:
    4.   name: migration-controller
    5.   namespace: openshift-migration
    6. spec:
    7.   disable_image_migration: true (1)
    8.   disable_pv_migration: true (2)
    9.   additional_excluded_resources: (3)
    10.   - resource1
    11.   - resource2
    12.   ...
    1Add disable_image_migration: true to exclude image streams from the migration. imagestreams is added to the excluded_resources list in main.yml when the MigrationController pod restarts.
    2Add disable_pv_migration: true to exclude PVs from the migration plan. persistentvolumes and persistentvolumeclaims are added to the excluded_resources list in main.yml when the MigrationController pod restarts. Disabling PV migration also disables PV discovery when you create the migration plan.
    3You can add OKD resources that you want to exclude to the additional_excluded_resources list.

  3. Wait two minutes for the MigrationController pod to restart so that the changes are applied.

  4. Verify that the resource is excluded:

    1. $ oc get deployment -n openshift-migration migration-controller -o yaml | grep EXCLUDED_RESOURCES -A1

    The output contains the excluded resources:

    Example output

    1. name: EXCLUDED_RESOURCES
    2. value:
    3. resource1,resource2,imagetags,templateinstances,clusterserviceversions,packagemanifests,subscriptions,servicebrokers,servicebindings,serviceclasses,serviceinstances,serviceplans,imagestreams,persistentvolumes,persistentvolumeclaims

Mapping namespaces

If you map namespaces in the MigPlan custom resource (CR), you must ensure that the namespaces are not duplicated on the source or the destination clusters because the UID and GID ranges of the namespaces are copied during migration.

Two source namespaces mapped to the same destination namespace

  1. spec:
  2.   namespaces:
  3.     - namespace_2
  4.     - namespace_1:namespace_2

If you want the source namespace to be mapped to a namespace of the same name, you do not need to create a mapping. By default, a source namespace and a target namespace have the same name.

Incorrect namespace mapping

  1. spec:
  2.   namespaces:
  3.     - namespace_1:namespace_1

Correct namespace reference

  1. spec:
  2.   namespaces:
  3.     - namespace_1

Excluding persistent volume claims

You select persistent volume claims (PVCs) for state migration by excluding the PVCs that you do not want to migrate. You exclude PVCs by setting the spec.persistentVolumes.pvc.selection.action parameter of the MigPlan custom resource (CR) after the persistent volumes (PVs) have been discovered.

Prerequisites

  • MigPlan CR is in a Ready state.

Procedure

  • Add the spec.persistentVolumes.pvc.selection.action parameter to the MigPlan CR and set it to skip:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigPlan
    3. metadata:
    4.   name: <migplan>
    5.   namespace: openshift-migration
    6. spec:
    7. ...
    8.   persistentVolumes:
    9.   - capacity: 10Gi
    10.     name: <pv_name>
    11.     pvc:
    12. ...
    13.     selection:
    14.       action: skip

Mapping persistent volume claims

You can migrate persistent volume (PV) data from the source cluster to persistent volume claims (PVCs) that are already provisioned in the destination cluster in the MigPlan CR by mapping the PVCs. This mapping ensures that the destination PVCs of migrated applications are synchronized with the source PVCs.

You map PVCs by updating the spec.persistentVolumes.pvc.name parameter in the MigPlan custom resource (CR) after the PVs have been discovered.

Prerequisites

  • MigPlan CR is in a Ready state.

Procedure

  • Update the spec.persistentVolumes.pvc.name parameter in the MigPlan CR:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigPlan
    3. metadata:
    4.   name: <migplan>
    5.   namespace: openshift-migration
    6. spec:
    7. ...
    8.   persistentVolumes:
    9.   - capacity: 10Gi
    10.     name: <pv_name>
    11.     pvc:
    12.       name: <source_pvc>:<destination_pvc> (1)
    1Specify the PVC on the source cluster and the PVC on the destination cluster. If the destination PVC does not exist, it will be created. You can use this mapping to change the PVC name during migration.

Editing persistent volume attributes

After you create a MigPlan custom resource (CR), the MigrationController CR discovers the persistent volumes (PVs). The spec.persistentVolumes block and the status.destStorageClasses block are added to the MigPlan CR.

You can edit the values in the spec.persistentVolumes.selection block. If you change values outside the spec.persistentVolumes.selection block, the values are overwritten when the MigPlan CR is reconciled by the MigrationController CR.

The default value for the spec.persistentVolumes.selection.storageClass parameter is determined by the following logic:

  1. If the source cluster PV is Gluster or NFS, the default is either cephfs, for accessMode: ReadWriteMany, or cephrbd, for accessMode: ReadWriteOnce.

  2. If the PV is neither Gluster nor NFS or if cephfs or cephrbd are not available, the default is a storage class for the same provisioner.

  3. If a storage class for the same provisioner is not available, the default is the default storage class of the destination cluster.

You can change the storageClass value to the value of any name parameter in the status.destStorageClasses block of the MigPlan CR.

If the storageClass value is empty, the PV will have no storage class after migration. This option is appropriate if, for example, you want to move the PV to an NFS volume on the destination cluster.

Prerequisites

  • MigPlan CR is in a Ready state.

Procedure

  • Edit the spec.persistentVolumes.selection values in the MigPlan CR:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigPlan
    3. metadata:
    4.   name: <migplan>
    5.   namespace: openshift-migration
    6. spec:
    7.   persistentVolumes:
    8.   - capacity: 10Gi
    9.     name: pvc-095a6559-b27f-11eb-b27f-021bddcaf6e4
    10.     proposedCapacity: 10Gi
    11.     pvc:
    12.       accessModes:
    13.       - ReadWriteMany
    14.       hasReference: true
    15.       name: mysql
    16.       namespace: mysql-persistent
    17.     selection:
    18.       action: <copy> (1)
    19.       copyMethod: <filesystem> (2)
    20.       verify: true (3)
    21.       storageClass: <gp2> (4)
    22.       accessMode: <ReadWriteMany> (5)
    23.     storageClass: cephfs
    1Allowed values are move, copy, and skip. If only one action is supported, the default value is the supported action. If multiple actions are supported, the default value is copy.
    2Allowed values are snapshot and filesystem. Default value is filesystem.
    3The verify parameter is displayed if you select the verification option for file system copy in the MTC web console. You can set it to false.
    4You can change the default value to the value of any name parameter in the status.destStorageClasses block of the MigPlan CR. If no value is specified, the PV will have no storage class after migration.
    5Allowed values are ReadWriteOnce and ReadWriteMany. If this value is not specified, the default is the access mode of the source cluster PVC. You can only edit the access mode in the MigPlan CR. You cannot edit it by using the MTC web console.

Converting storage classes in the MTC web console

You can convert the storage class of a persistent volume (PV) by migrating it within the same cluster. To do so, you must create and run a migration plan in the Migration Toolkit for Containers (MTC) web console.

Prerequisites

  • You must be logged in as a user with cluster-admin privileges on the cluster on which MTC is running.

  • You must add the cluster to the MTC web console.

Procedure

  1. In the left-side navigation pane of the OKD web console, click Projects.

  2. In the list of projects, click your project.

    The Project details page opens.

  3. Click the DeploymentConfig name. Note the name of its running pod.

  4. Open the YAML tab of the project. Find the PVs and note the names of their corresponding persistent volume claims (PVCs).

  5. In the MTC web console, click Migration plans.

  6. Click Add migration plan.

  7. Enter the Plan name.

    The migration plan name must contain 3 to 63 lower-case alphanumeric characters (a-z, 0-9) and must not contain spaces or underscores (_).

  8. From the Migration type menu, select Storage class conversion.

  9. From the Source cluster list, select the desired cluster for storage class conversion.

  10. Click Next.

    The Namespaces page opens.

  11. Select the required project.

  12. Click Next.

    The Persistent volumes page opens. The page displays the PVs in the project, all selected by default.

  13. For each PV, select the desired target storage class.

  14. Click Next.

    The wizard validates the new migration plan and shows that it is ready.

  15. Click Close.

    The new plan appears on the Migration plans page.

  16. To start the conversion, click the options menu of the new plan.

    Under Migrations, two options are displayed, Stage and Cutover.

    Cutover migration updates PVC references in the applications.

    Stage migration does not update PVC references in the applications.

  17. Select the desired option.

    Depending on which option you selected, the Stage migration or Cutover migration notification appears.

  18. Click Migrate.

    Depending on which option you selected, the Stage started or Cutover started message appears.

  19. To see the status of the current migration, click the number in the Migrations column.

    The Migrations page opens.

  20. To see more details on the current migration and monitor its progress, select the migration from the Type column.

    The Migration details page opens. When the migration progresses to the DirectVolume step and the status of the step becomes Running Rsync Pods to migrate Persistent Volume data, you can click View details and see the detailed status of the copies.

  21. In the breadcrumb bar, click Stage or Cutover and wait for all steps to complete.

  22. Open the PersistentVolumeClaims tab of the OKD web console.

    You can see new PVCs with the names of the initial PVCs but ending in new, which are using the target storage class.

  23. In the left-side navigation pane, click Pods. See that the pod of your project is running again.

Additional resources

Performing a state migration of Kubernetes objects by using the MTC API

After you migrate all the PV data, you can use the Migration Toolkit for Containers (MTC) API to perform a one-time state migration of Kubernetes objects that constitute an application.

You do this by configuring MigPlan custom resource (CR) fields to provide a list of Kubernetes resources with an additional label selector to further filter those resources, and then performing a migration by creating a MigMigration CR. The MigPlan resource is closed after the migration.

Selecting Kubernetes resources is an API-only feature. You must update the MigPlan CR and create a MigMigration CR for it by using the CLI. The MTC web console does not support migrating Kubernetes objects.

After migration, the closed parameter of the MigPlan CR is set to true. You cannot create another MigMigration CR for this MigPlan CR.

You add Kubernetes objects to the MigPlan CR by using one of the following options:

  • Adding the Kubernetes objects to the includedResources section. When the includedResources field is specified in the MigPlan CR, the plan takes a list of group-kind as input. Only resources present in the list are included in the migration.

  • Adding the optional labelSelector parameter to filter the includedResources in the MigPlan. When this field is specified, only resources matching the label selector are included in the migration. For example, you can filter a list of Secret and ConfigMap resources by using the label app: frontend as a filter.

Procedure

  1. Update the MigPlan CR to include Kubernetes resources and, optionally, to filter the included resources by adding the labelSelector parameter:

    1. To update the MigPlan CR to include Kubernetes resources:

      1. apiVersion: migration.openshift.io/v1alpha1
      2. kind: MigPlan
      3. metadata:
      4.   name: <migplan>
      5.   namespace: openshift-migration
      6. spec:
      7.   includedResources:
      8.   - kind: <kind> (1)
      9.     group: ""
      10.   - kind: <kind>
      11.     group: ""
      1Specify the Kubernetes object, for example, Secret or ConfigMap.

    2. Optional: To filter the included resources by adding the labelSelector parameter:

      1. apiVersion: migration.openshift.io/v1alpha1
      2. kind: MigPlan
      3. metadata:
      4.   name: <migplan>
      5.   namespace: openshift-migration
      6. spec:
      7.   includedResources:
      8.   - kind: <kind> (1)
      9.     group: ""
      10.   - kind: <kind>
      11.     group: ""
      12. ...
      13.   labelSelector:
      14.     matchLabels:
      15.       <label> (2)
      1Specify the Kubernetes object, for example, Secret or ConfigMap.
      2Specify the label of the resources to migrate, for example, app: frontend.

  2. Create a MigMigration CR to migrate the selected Kubernetes resources. Verify that the correct MigPlan is referenced in migPlanRef:

    1. apiVersion: migration.openshift.io/v1alpha1
    2. kind: MigMigration
    3. metadata:
    4.   generateName: <migplan>
    5.   namespace: openshift-migration
    6. spec:
    7.   migPlanRef:
    8.     name: <migplan>
    9.     namespace: openshift-migration
    10.   stage: false

Migration controller options

You can edit migration plan limits, enable persistent volume resizing, or enable cached Kubernetes clients in the MigrationController custom resource (CR) for large migrations and improved performance.

Increasing limits for large migrations

You can increase the limits on migration objects and container resources for large migrations with the Migration Toolkit for Containers (MTC).

You must test these changes before you perform a migration in a production environment.

Procedure

  1. Edit the MigrationController custom resource (CR) manifest:

    1. $ oc edit migrationcontroller -n openshift-migration

  2. Update the following parameters:

    1. ...
    2. mig_controller_limits_cpu: "1" (1)
    3. mig_controller_limits_memory: "10Gi" (2)
    4. ...
    5. mig_controller_requests_cpu: "100m" (3)
    6. mig_controller_requests_memory: "350Mi" (4)
    7. ...
    8. mig_pv_limit: 100 (5)
    9. mig_pod_limit: 100 (6)
    10. mig_namespace_limit: 10 (7)
    11. ...
    1Specifies the number of CPUs available to the MigrationController CR.
    2Specifies the amount of memory available to the MigrationController CR.
    3Specifies the number of CPU units available for MigrationController CR requests. 100m represents 0.1 CPU units (100 * 1e-3).
    4Specifies the amount of memory available for MigrationController CR requests.
    5Specifies the number of persistent volumes that can be migrated.
    6Specifies the number of pods that can be migrated.
    7Specifies the number of namespaces that can be migrated.

  3. Create a migration plan that uses the updated parameters to verify the changes.

    If your migration plan exceeds the MigrationController CR limits, the MTC console displays a warning message when you save the migration plan.

Enabling persistent volume resizing for direct volume migration

You can enable persistent volume (PV) resizing for direct volume migration to avoid running out of disk space on the destination cluster.

When the disk usage of a PV reaches a configured level, the MigrationController custom resource (CR) compares the requested storage capacity of a persistent volume claim (PVC) to its actual provisioned capacity. Then, it calculates the space required on the destination cluster.

A pv_resizing_threshold parameter determines when PV resizing is used. The default threshold is 3%. This means that PV resizing occurs when the disk usage of a PV is more than 97%. You can increase this threshold so that PV resizing occurs at a lower disk usage level.

PVC capacity is calculated according to the following criteria:

  • If the requested storage capacity (spec.resources.requests.storage) of the PVC is not equal to its actual provisioned capacity (status.capacity.storage), the greater value is used.

  • If a PV is provisioned through a PVC and then subsequently changed so that its PV and PVC capacities no longer match, the greater value is used.

Prerequisites

  • The PVCs must be attached to one or more running pods so that the MigrationController CR can execute commands.

Procedure

  1. Log in to the host cluster.

  2. Enable PV resizing by patching the MigrationController CR:

    1. $ oc patch migrationcontroller migration-controller -p '{"spec":{"enable_dvm_pv_resizing":true}}' \ (1)
    2.   --type='merge' -n openshift-migration
    1Set the value to false to disable PV resizing.

  3. Optional: Update the pv_resizing_threshold parameter to increase the threshold:

    1. $ oc patch migrationcontroller migration-controller -p '{"spec":{"pv_resizing_threshold":41}}' \ (1)
    2.   --type='merge' -n openshift-migration
    1The default value is 3.

    When the threshold is exceeded, the following status message is displayed in the MigPlan CR status:

    1. status:
    2.   conditions:
    3. ...
    4.   - category: Warn
    5.     durable: true
    6.     lastTransitionTime: "2021-06-17T08:57:01Z"
    7.     message: 'Capacity of the following volumes will be automatically adjusted to avoid disk capacity issues in the target cluster:  [pvc-b800eb7b-cf3b-11eb-a3f7-0eae3e0555f3]'
    8.     reason: Done
    9.     status: "False"
    10.     type: PvCapacityAdjustmentRequired

    For AWS gp2 storage, this message does not appear unless the pv_resizing_threshold is 42% or greater because of the way gp2 calculates volume usage and size. (BZ#1973148)

Enabling cached Kubernetes clients

You can enable cached Kubernetes clients in the MigrationController custom resource (CR) for improved performance during migration. The greatest performance benefit is displayed when migrating between clusters in different regions or with significant network latency.

Delegated tasks, for example, Rsync backup for direct volume migration or Velero backup and restore, however, do not show improved performance with cached clients.

Cached clients require extra memory because the MigrationController CR caches all API resources that are required for interacting with MigCluster CRs. Requests that are normally sent to the API server are directed to the cache instead. The cache watches the API server for updates.

You can increase the memory limits and requests of the MigrationController CR if OOMKilled errors occur after you enable cached clients.

Procedure

  1. Enable cached clients by running the following command:

    1. $ oc -n openshift-migration patch migrationcontroller migration-controller --type=json --patch \
    2.   '[{ "op": "replace", "path": "/spec/mig_controller_enable_cache", "value": true}]'

  2. Optional: Increase the MigrationController CR memory limits by running the following command:

    1. $ oc -n openshift-migration patch migrationcontroller migration-controller --type=json --patch \
    2.   '[{ "op": "replace", "path": "/spec/mig_controller_limits_memory", "value": <10Gi>}]'

  3. Optional: Increase the MigrationController CR memory requests by running the following command:

    1. $ oc -n openshift-migration patch migrationcontroller migration-controller --type=json --patch \
    2.   '[{ "op": "replace", "path": "/spec/mig_controller_requests_memory", "value": <350Mi>}]'