Advanced migration options | Migrating from version 3 to 4

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
Term	Definition
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 an application from on-premises to a cloud-based cluster

You can migrate from a source cluster that is behind a firewall to a cloud-based destination cluster by establishing a network tunnel between the two clusters. The crane tunnel-api command establishes such a tunnel by creating a VPN tunnel on the source cluster and then connecting to a VPN server running on the destination cluster. The VPN server is exposed to the client using a load balancer address on the destination cluster.

A service created on the destination cluster exposes the source cluster’s API to MTC, which is running on the destination cluster.

Prerequisites

The system that creates the VPN tunnel must have access and be logged in to both clusters.
It must be possible to create a load balancer on the destination cluster. Refer to your cloud provider to ensure this is possible.
Have names prepared to assign to namespaces, on both the source cluster and the destination cluster, in which to run the VPN tunnel. These namespaces should not be created in advance. For information about namespace rules, see https://kubernetes.io/docs/concepts/overview/working-with-objects/names/#dns-subdomain-names.
When connecting multiple firewall-protected source clusters to the cloud cluster, each source cluster requires its own namespace.
OpenVPN server is installed on the destination cluster.
OpenVPN client is installed on the source cluster.
When configuring the source cluster in MTC, the API URL takes the form of https://proxied-cluster.<namespace>.svc.cluster.local:8443.
- If you use the API, see Create a MigCluster CR manifest for each remote cluster.
- If you use the MTC web console, see Migrating your applications using the MTC web console.
The MTC web console and Migration Controller must be installed on the target cluster.

Procedure

Install the crane utility:

$ podman cp $(podman create registry.redhat.io/rhmtc/openshift-migration-controller-rhel8:v1.8):/crane ./

Log in remotely to a node on the source cluster and a node on the destination cluster.
Obtain the cluster context for both clusters after logging in:
```
$ oc config view
```

Establish a tunnel by entering the following command on the command system:

$ crane tunnel-api [--namespace <namespace>] \
      --destination-context <destination-cluster> \
      --source-context <source-cluster>

If you don’t specify a namespace, the command uses the default value openvpn.

For example:

$ crane tunnel-api --namespace my_tunnel \
      --destination-context openshift-migration/c131-e-us-east-containers-cloud-ibm-com/admin \
      --source-context default/192-168-122-171-nip-io:8443/admin

See all available parameters for the crane tunnel-api command by entering crane tunnel-api --help.

The command generates TSL/SSL Certificates. This process might take several minutes. A message appears when the process completes.

The OpenVPN server starts on the destination cluster and the OpenVPN client starts on the source cluster.

After a few minutes, the load balancer resolves on the source node.

You can view the log for the OpenVPN pods to check the status of this process by entering the following commands with root privileges:

# oc get po -n <namespace>

Example output

NAME            READY     STATUS      RESTARTS    AGE
<pod_name>    2/2       Running     0           44s

# oc logs -f -n <namespace> <pod_name> -c openvpn

When the address of the load balancer is resolved, the message Initialization Sequence Completed appears at the end of the log.

On the OpenVPN server, which is on a destination control node, verify that the openvpn service and the proxied-cluster service are running:
```
$ oc get service -n <namespace>
```
On the source node, get the service account (SA) token for the migration controller:
```
# oc sa get-token -n openshift-migration migration-controller
```
Open the MTC web console and add the source cluster, using the following values:
- Cluster name: The source cluster name.
- URL: proxied-cluster.<namespace>.svc.cluster.local:8443. If you did not define a value for <namespace>, use openvpn.
- Service account token: The token of the migration controller service account.
- Exposed route host to image registry: proxied-cluster.<namespace>.svc.cluster.local:5000. If you did not define a value for <namespace>, use openvpn.

After MTC has successfully validated the connection, you can proceed to create and run a migration plan. The namespace for the source cluster should appear in the list of namespaces.

Additional resources

For information about creating a MigCluster CR manifest for each remote cluster, see Migrating an application by using the MTC API.
For information about adding a cluster using the web console, see Migrating your applications by using 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.

Internal images

If your application uses internal images from the openshift namespace, you must ensure that the required versions of the images are present on the target cluster.

You can manually update an image stream tag in order to use a deprecated OKD 3 image on an OKD 4.13 cluster.

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 3 cluster:
- 8443 (API server)
- 443 (routes)
- 53 (DNS)
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.

The OKD 3 registry must be exposed manually.

Procedure

To create a route to an OKD 3 registry, run the following command:

$ oc create route passthrough --service=docker-registry -n default

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

$ 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 4.13, the 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:

apiVersion: migration.openshift.io/v1alpha1
kind: MigrationController
metadata:
  name: migration-controller
  namespace: openshift-migration
spec:
  [...]
  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:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-egress-from-rsync-pods
spec:
  podSelector:
    matchLabels:
      owner: directvolumemigration
      app: directvolumemigration-rsync-transfer
  egress:
  - {}
  policyTypes:
  - Egress

Ingress traffic to Rsync pods

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-all-egress-from-rsync-pods
spec:
  podSelector:
    matchLabels:
      owner: directvolumemigration
      app: directvolumemigration-rsync-transfer
  ingress:
  - {}
  policyTypes:
  - 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:

apiVersion: network.openshift.io/v1
kind: EgressNetworkPolicy
metadata:
  name: test-egress-policy
  namespace: <namespace>
spec:
  egress:
  - to:
      cidrSelector: <cidr_of_source_or_target_cluster>
    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:

apiVersion: migration.openshift.io/v1alpha1
kind: MigrationController
metadata:
  name: migration-controller
  namespace: openshift-migration
spec:
  [...]
  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
Variable	Type	Default	Description
`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:

spec:
  src_supplemental_groups: "1000,2000"
  target_supplemental_groups: "2000,3000"

Configuring proxies

Prerequisites

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

Procedure

Get the MigrationController CR manifest:

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

Update the proxy parameters:
```
apiVersion: migration.openshift.io/v1alpha1
kind: MigrationController
metadata:
  name: <migration_controller>
  namespace: openshift-migration
...
spec:
  stunnel_tcp_proxy: http://<username>:<password>@<ip>:<port> (1)
  noProxy: example.com (2)
```
1 Stunnel proxy URL for direct volume migration.

2 Comma-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.
Save the manifest as migration-controller.yaml.

Apply the updated manifest:

$ 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

Create a MigCluster CR manifest for the host cluster:

$ cat << EOF | oc apply -f -
apiVersion: migration.openshift.io/v1alpha1
kind: MigCluster
metadata:
  name: <host_cluster>
  namespace: openshift-migration
spec:
  isHostCluster: true
EOF

Create a Secret object manifest for each remote cluster:

$ cat << EOF | oc apply -f -
apiVersion: v1
kind: Secret
metadata:
  name: <cluster_secret>
  namespace: openshift-config
type: Opaque
data:
  saToken: <sa_token> (1)
EOF

1	Specify the base64-encoded `migration-controller` service account (SA) token of the remote cluster. You can obtain the token by running the following command:

$ oc sa get-token migration-controller -n openshift-migration | base64 -w 0

Create a MigCluster CR manifest for each remote cluster:

$ cat << EOF | oc apply -f -
apiVersion: migration.openshift.io/v1alpha1
kind: MigCluster
metadata:
  name: <remote_cluster> (1)
  namespace: openshift-migration
spec:
  exposedRegistryPath: <exposed_registry_route> (2)
  insecure: false (3)
  isHostCluster: false
  serviceAccountSecretRef:
    name: <remote_cluster_secret> (4)
    namespace: openshift-config
  url: <remote_cluster_url> (5)
EOF

1	Specify the `Cluster` CR of the remote cluster.
2	Optional: For direct image migration, specify the exposed registry route.
3	SSL verification is enabled if `false`. CA certificates are not required or checked if `true`.
4	Specify the `Secret` object of the remote cluster.
5	Specify the URL of the remote cluster.

Verify that all clusters are in a Ready state:
```
$ oc describe MigCluster <cluster>
```
Create a Secret object manifest for the replication repository:
```
$ cat << EOF | oc apply -f -
apiVersion: v1
kind: Secret
metadata:
  namespace: openshift-config
  name: <migstorage_creds>
type: Opaque
data:
  aws-access-key-id: <key_id_base64> (1)
  aws-secret-access-key: <secret_key_base64> (2)
EOF
```
1 Specify the key ID in base64 format.

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

Create a MigStorage CR manifest for the replication repository:

$ cat << EOF | oc apply -f -
apiVersion: migration.openshift.io/v1alpha1
kind: MigStorage
metadata:
  name: <migstorage>
  namespace: openshift-migration
spec:
  backupStorageConfig:
    awsBucketName: <bucket> (1)
    credsSecretRef:
      name: <storage_secret> (2)
      namespace: openshift-config
  backupStorageProvider: <storage_provider> (3)
  volumeSnapshotConfig:
    credsSecretRef:
      name: <storage_secret> (4)
      namespace: openshift-config
  volumeSnapshotProvider: <storage_provider> (5)
EOF

1	Specify the bucket name.
2	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.
3	Specify the storage provider.
4	Optional: 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.
5	Optional: If you are copying data by using snapshots, specify the storage provider.

Verify that the MigStorage CR is in a Ready state:
```
$ oc describe migstorage <migstorage>
```

Create a MigPlan CR manifest:

$ cat << EOF | oc apply -f -
apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
  destMigClusterRef:
    name: <host_cluster>
    namespace: openshift-migration
  indirectImageMigration: true (1)
  indirectVolumeMigration: true (2)
  migStorageRef:
    name: <migstorage> (3)
    namespace: openshift-migration
  namespaces:
    - <source_namespace_1> (4)
    - <source_namespace_2>
    - <source_namespace_3>:<destination_namespace> (5)
  srcMigClusterRef:
    name: <remote_cluster> (6)
    namespace: openshift-migration
EOF

1	Direct image migration is enabled if `false`.
2	Direct volume migration is enabled if `false`.
3	Specify the name of the `MigStorage` CR instance.
4	Specify one or more source namespaces. By default, the destination namespace has the same name.
5	Specify a destination namespace if it is different from the source namespace.
6	Specify the name of the source cluster `MigCluster` instance.

Verify that the MigPlan instance is in a Ready state:

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

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

$ cat << EOF | oc apply -f -
apiVersion: migration.openshift.io/v1alpha1
kind: MigMigration
metadata:
  name: <migmigration>
  namespace: openshift-migration
spec:
  migPlanRef:
    name: <migplan> (1)
    namespace: openshift-migration
  quiescePods: true (2)
  stage: false (3)
  rollback: false (4)
EOF

1	Specify the `MigPlan` CR name.
2	The pods on the source cluster are stopped before migration if `true`.
3	A stage migration, which copies most of the data without stopping the application, is performed if `true`.
4	A completed migration is rolled back if `true`.

Verify the migration by watching the MigMigration CR progress:

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

The output resembles the following:

Example output

Name:         c8b034c0-6567-11eb-9a4f-0bc004db0fbc
Namespace:    openshift-migration
Labels:       migration.openshift.io/migplan-name=django
Annotations:  openshift.io/touch: e99f9083-6567-11eb-8420-0a580a81020c
API Version:  migration.openshift.io/v1alpha1
Kind:         MigMigration
...
Spec:
  Mig Plan Ref:
    Name:       migplan
    Namespace:  openshift-migration
  Stage:        false
Status:
  Conditions:
    Category:              Advisory
    Last Transition Time:  2021-02-02T15:04:09Z
    Message:               Step: 19/47
    Reason:                InitialBackupCreated
    Status:                True
    Type:                  Running
    Category:              Required
    Last Transition Time:  2021-02-02T15:03:19Z
    Message:               The migration is ready.
    Status:                True
    Type:                  Ready
    Category:              Required
    Durable:               true
    Last Transition Time:  2021-02-02T15:04:05Z
    Message:               The migration registries are healthy.
    Status:                True
    Type:                  RegistriesHealthy
  Itinerary:               Final
  Observed Digest:         7fae9d21f15979c71ddc7dd075cb97061895caac5b936d92fae967019ab616d5
  Phase:                   InitialBackupCreated
  Pipeline:
    Completed:  2021-02-02T15:04:07Z
    Message:    Completed
    Name:       Prepare
    Started:    2021-02-02T15:03:18Z
    Message:    Waiting for initial Velero backup to complete.
    Name:       Backup
    Phase:      InitialBackupCreated
    Progress:
      Backup openshift-migration/c8b034c0-6567-11eb-9a4f-0bc004db0fbc-wpc44: 0 out of estimated total of 0 objects backed up (5s)
    Started:        2021-02-02T15:04:07Z
    Message:        Not started
    Name:           StageBackup
    Message:        Not started
    Name:           StageRestore
    Message:        Not started
    Name:           DirectImage
    Message:        Not started
    Name:           DirectVolume
    Message:        Not started
    Name:           Restore
    Message:        Not started
    Name:           Cleanup
  Start Timestamp:  2021-02-02T15:03:18Z
Events:
  Type    Reason   Age                 From                     Message
  ----    ------   ----                ----                     -------
  Normal  Running  57s                 migmigration_controller  Step: 2/47
  Normal  Running  57s                 migmigration_controller  Step: 3/47
  Normal  Running  57s (x3 over 57s)   migmigration_controller  Step: 4/47
  Normal  Running  54s                 migmigration_controller  Step: 5/47
  Normal  Running  54s                 migmigration_controller  Step: 6/47
  Normal  Running  52s (x2 over 53s)   migmigration_controller  Step: 7/47
  Normal  Running  51s (x2 over 51s)   migmigration_controller  Step: 8/47
  Normal  Ready    50s (x12 over 57s)  migmigration_controller  The migration is ready.
  Normal  Running  50s                 migmigration_controller  Step: 9/47
  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

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.
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.
Clone application manifests to the target cluster.

You can use Argo CD to clone the application manifests to the target cluster.
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.
If the cloned application is in a quiesced state, unquiesce it.
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

See Excluding PVCs from migration to select PVCs for state migration.
See Mapping PVCs to migrate source PV data to provisioned PVCs on the destination cluster.
See Migrating Kubernetes objects to migrate the Kubernetes objects that constitute an application’s state.

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

- hosts: localhost
  gather_facts: false
  tasks:
  - name: get pod name
    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

- hosts: localhost
  gather_facts: false
  tasks:
  - name: Get pod
    k8s_info:
      kind: pods
      api: v1
      namespace: openshift-migration
      name: "{{ lookup( 'env', 'HOSTNAME') }}"
    register: pods

  - name: Print pod name
    debug:
      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

- hosts: localhost
  gather_facts: false
  tasks:
  - name: Set a boolean
    set_fact:
      do_fail: true

  - name: "fail"
    fail:
      msg: "Cause a failure"
    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

- hosts: localhost
  gather_facts: false
  tasks:
  - set_fact:
      namespaces: "{{ (lookup( 'env', 'MIGRATION_NAMESPACES')).split(',') }}"

  - debug:
      msg: "{{ item }}"
    with_items: "{{ namespaces }}"

  - debug:
      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

Edit the MigrationController custom resource manifest:

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

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:

apiVersion: migration.openshift.io/v1alpha1
kind: MigrationController
metadata:
  name: migration-controller
  namespace: openshift-migration
spec:
  disable_image_migration: true (1)
  disable_pv_migration: true (2)
  additional_excluded_resources: (3)
  - resource1
  - resource2
  ...

1	Add `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.
2	Add `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.
3	You can add OKD resources that you want to exclude to the `additional_excluded_resources` list.

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

Verify that the resource is excluded:

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

The output contains the excluded resources:

Example output

name: EXCLUDED_RESOURCES
value:
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

spec:
  namespaces:
    - namespace_2
    - 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

spec:
  namespaces:
    - namespace_1:namespace_1

Correct namespace reference

spec:
  namespaces:
    - 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:

apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
...
  persistentVolumes:
  - capacity: 10Gi
    name: <pv_name>
    pvc:
...
    selection:
      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:

apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
...
  persistentVolumes:
  - capacity: 10Gi
    name: <pv_name>
    pvc:
      name: <source_pvc>:<destination_pvc> (1)

1	Specify 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:

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

apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
  persistentVolumes:
  - capacity: 10Gi
    name: pvc-095a6559-b27f-11eb-b27f-021bddcaf6e4
    proposedCapacity: 10Gi
    pvc:
      accessModes:
      - ReadWriteMany
      hasReference: true
      name: mysql
      namespace: mysql-persistent
    selection:
      action: <copy> (1)
      copyMethod: <filesystem> (2)
      verify: true (3)
      storageClass: <gp2> (4)
      accessMode: <ReadWriteMany> (5)
    storageClass: cephfs

1	Allowed 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`.
2	Allowed values are `snapshot` and `filesystem`. Default value is `filesystem`.
3	The `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`.
4	You 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.
5	Allowed 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.

Additional resources

For details about the move and copy actions, see MTC workflow.
For details about the skip action, see Excluding PVCs from migration.
For details about the file system and snapshot copy methods, see About data copy methods.

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

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

To update the MigPlan CR to include Kubernetes resources:

apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
  includedResources:
  - kind: <kind> (1)
    group: ""
  - kind: <kind>
    group: ""

1	Specify the Kubernetes object, for example, `Secret` or `ConfigMap`.

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

apiVersion: migration.openshift.io/v1alpha1
kind: MigPlan
metadata:
  name: <migplan>
  namespace: openshift-migration
spec:
  includedResources:
  - kind: <kind> (1)
    group: ""
  - kind: <kind>
    group: ""
...
  labelSelector:
    matchLabels:
      <label> (2)

1	Specify the Kubernetes object, for example, `Secret` or `ConfigMap`.
2	Specify the label of the resources to migrate, for example, `app: frontend`.

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

apiVersion: migration.openshift.io/v1alpha1
kind: MigMigration
metadata:
  generateName: <migplan>
  namespace: openshift-migration
spec:
  migPlanRef:
    name: <migplan>
    namespace: openshift-migration
  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

Edit the MigrationController custom resource (CR) manifest:
```
$ oc edit migrationcontroller -n openshift-migration
```

Update the following parameters:

...
mig_controller_limits_cpu: "1" (1)
mig_controller_limits_memory: "10Gi" (2)
...
mig_controller_requests_cpu: "100m" (3)
mig_controller_requests_memory: "350Mi" (4)
...
mig_pv_limit: 100 (5)
mig_pod_limit: 100 (6)
mig_namespace_limit: 10 (7)
...

1	Specifies the number of CPUs available to the `MigrationController` CR.
2	Specifies the amount of memory available to the `MigrationController` CR.
3	Specifies the number of CPU units available for `MigrationController` CR requests. `100m` represents 0.1 CPU units (100 * 1e-3).
4	Specifies the amount of memory available for `MigrationController` CR requests.
5	Specifies the number of persistent volumes that can be migrated.
6	Specifies the number of pods that can be migrated.
7	Specifies the number of namespaces that can be migrated.

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

Enable PV resizing by patching the MigrationController CR:

$ oc patch migrationcontroller migration-controller -p '{"spec":{"enable_dvm_pv_resizing":true}}' \ (1)
  --type='merge' -n openshift-migration

1	Set the value to `false` to disable PV resizing.

Optional: Update the pv_resizing_threshold parameter to increase the threshold:

$ oc patch migrationcontroller migration-controller -p '{"spec":{"pv_resizing_threshold":41}}' \ (1)
  --type='merge' -n openshift-migration

1	The default value is `3`.

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

status:
  conditions:
...
  - category: Warn
    durable: true
    lastTransitionTime: "2021-06-17T08:57:01Z"
    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]'
    reason: Done
    status: "False"
    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

Enable cached clients by running the following command:

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

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

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

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

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

1	Stunnel proxy URL for direct volume migration.
2	Comma-separated list of destination domain names, domains, IP addresses, or other network CIDRs to exclude proxying.

1	Specify the key ID in base64 format.
2	Specify the secret key in base64 format.