Multi-region deployment on GKE and GKE on-prem

You are currently viewing version 1.5 of the Apigee hybrid documentation. This version is end of life. You should upgrade to a newer version. For more information, see Supported versions.

Version 1.5 (unsupported)

   Supported versions:

• v1.14 (latest)                        
• v1.13                         
• v1.12                         
• v1.11                         
• List of supported versions

   Unsupported versions:

• v1.10                                      
• v1.9                                         
• v1.8                                         
• v1.7                                         
• v1.6                                         
• v1.5                                         
• v1.4                                         
• v1.3                                         
• v1.2                                         
• v1.1                                         

This topic describes a multi-region deployment for Apigee hybrid on GKE and Anthos GKE deployed on-prem.

Topologies for multi-region deployment include the following:

• Active-Active: When you have applications deployed in multiple geographic locations and you require low latency API response for your deployments. You have the option to deploy hybrid in multiple geographic locations nearest to your clients. For example: US West Coast, US East Coast, Europe, APAC.
• Active-Passive: When you have a primary region and a failover or disaster recovery region.

The regions in a multi-region hybrid deployment communicate via Cassandra, as the following image shows:

Load balancing the MART connection

Each regional cluster must have its own MART IP and hostname; however, you only need to connect the management plane to one of them. Cassandra propagates information to all of the clusters. The best option for high availability for MART is to load balance the individual MART IP addresses and configure your organization to talk to the load balanced MART URL.

Prerequisites

Before configuring hybrid for multiple regions, you must complete the following prerequisites:

• Set up Kubernetes clusters in multiple regions with different CIDR blocks
• Set up cross-region communication
• Cassandra Multi Region requirements:
  • Make sure the pod network namespace has connectivity across the regions, including firewalls, vpn, vpc peering and vNet peering. This is the case for most GKE installations.
  • If the pod network namespace does not have connectivity between pods in different clusters (the clusters are running in "island network mode", for example in GKE on-prem installations), enable the Kubernetes hostNetwork feature by setting cassandra.hostNetwork: true in the overrides file for all of the regions in your Apigee hybrid multi-regions installation.

    For information on the Kubernetes hostNetwork feature, see Host namespaces in the Kubernetes documentation.

  • Enable hostNetwork on existing clusters before expanding your multi-region configuration to new regions.
  • When hostNetwork is enabled, make sure worker nodes can perform reverse DNS lookup. Apigee cassandra uses both forward and reverse DNS lookup to obtain the host IP while starting.
    Tip: If the DNS resolution is not working, try this workaround.

    Add a static host entry in /etc/hosts file of each cassandra worker node similar to the following:

    echo -e "\\n127.0.1.1 ${HOSTNAME}" >> "/etc/hosts"

    or

    "\\nWORKER-NODE-IP ${HOSTNAME}" >> "/etc/hosts"

  • Open Cassandra ports 7000 and 7001 between Kubernetes clusters across all regions to enable worker nodes across regions and datacenters to communicate. See Configure ports.

For detailed information, see Kubernetes documentation.

NOTE: Apigee recommends that you ensure that your servers' times are synchronized.

Several features such as expiration and token revocation rely on accurate system times. If you host the runtime components in different datacenters, then be sure that the system times are synchronized.

You can use a tool such as ntpdate to verify that server times are synchronized.

Configure the multi-region seed host

This section describes how to expand the existing Cassandra cluster to a new region. This setup allows the new region to bootstrap the cluster and join the existing data center. Without this configuration, the multi-region Kubernetes clusters would not know about each other.

1. Run the following kubectl command to identify a seed host address for Cassandra in the current region.

   A seed host address allows a new regional instance to find the original cluster on the very first startup to learn the topology of the cluster. The seed host address is designated as the contact point in the cluster.

   kubectl get pods -o wide -n apigee
   
   NAME                      READY   STATUS      RESTARTS   AGE   IP          NODE                                          NOMINATED NODE
   apigee-cassandra-default-0        1/1     Running     0          5d    10.0.0.11   gke-k8s-dc-2-default-pool-a2206492-p55d
   apigee-cassandra-default-1        1/1     Running     0          5d    10.0.2.4    gke-k8s-dc-2-default-pool-e9daaab3-tjmz
   apigee-cassandra-default-2        1/1     Running     0          5d    10.0.3.5    gke-k8s-dc-2-default-pool-e589awq3-kjch

2. Decide which of the IPs returned from the previous command will be the multi-region seed host.

3. The configuration in this step depends on whether you are on GKE or GKE on-prem:

   GKE Only: In data center 2, configure cassandra.multiRegionSeedHost and cassandra.datacenter in Manage runtime plane components, where multiRegionSeedHost is one of the IPs returned by the previous command:

   cassandra:
     multiRegionSeedHost: seed_host_IP
     datacenter: data_center_name
     rack: rack_name
     hostNetwork: false # Set this to true for Non GKE platforms.

   For example:

   cassandra:
     multiRegionSeedHost: 10.0.0.11
     datacenter: "dc-2"
     rack: "ra-1"
     hostNetwork: false

   GKE on-prem Only: In data center 2, configure cassandra.multiRegionSeedHost in your overrides file, where multiRegionSeedHost is one of the IPs returned by the previous command:

   cassandra:
     hostNetwork: true
     multiRegionSeedHost: seed_host_IP
     datacenter: data_center_name
   

   For example:

   cassandra:
     hostNetwork: true
     multiRegionSeedHost: 10.0.0.11
     datacenter: "dc-2"
   

4. In the new data center/region, before you install hybrid, set the same TLS certificates and credentials in overrides.yaml as you set in the first region.
   NOTE: Be sure to use the same Cassandra TLS certificates and credentials in the second data center as you provided in the original data center. The credentials you set in overrides.yaml in the first data center must match the ones you specify in overrides.yaml in the second data center. For details see Configuring TLS for Cassandra.

Set up the new region

After you configure the seed host, you can set up the new region.

To set up the new region:

1. Copy your certificate from the existing cluster to the new cluster. The new CA root is used by Cassandra and other hybrid components for mTLS. Therefore, it is essential to have consistent certificates across the cluster.
   1. Set the context to the original namespace:

      kubectl config use-context original-cluster-name

   2. Export the current namespace configuration to a file:

      kubectl get namespace namespace -o yaml > apigee-namespace.yaml

   3. Export the apigee-ca secret to a file:

      kubectl -n cert-manager get secret apigee-ca -o yaml > apigee-ca.yaml

   4. Set the context to the new region's cluster name:

      kubectl config use-context new-cluster-name

   5. Import the namespace configuration to the new cluster. Be sure to update the "namespace" in the file if you're using a different namespace in the new region:

      kubectl apply -f apigee-namespace.yaml

   6. Import the secret to the new cluster:

      kubectl -n cert-manager apply -f apigee-ca.yaml

2. Install hybrid in the new region. Be sure that the overrides-DC_name.yaml file includes the same TLS certificates that are configured in the first region, as explained in the previous section.

   Execute the following two commands to install hybrid in the new region:

   apigeectl init -f overrides/overrides-DC_name.yaml

   apigeectl apply -f overrides/overrides-DC_name.yaml

3. Verify the hybrid installation is successful by running the following command:

   apigeectl check-ready -f overrides_your_cluster_name.yaml

4. Verify the Cassandra cluster setup by running the following command. The output should show both the existing and new data centers.
   Warning: If the output of this command does not show both the existing and new data centers, the setup failed. In this case, do not proceed further.

   To fix the problem, the hybrid installation on the new region must be deleted and recreated. See Uninstall hybrid runtime. If the setup fails, do not attempt to proceed further or attempt to join the cluster.

   kubectl exec apigee-cassandra-default-0 -n apigee  \
     -- nodetool -u JMX_user -pw JMX_password status

   Example showing a successful setup:

   Datacenter: dc-1
   ====================
   Status=Up/Down
   |/ State=Normal/Leaving/Joining/Moving
   --  Address        Load       Tokens  Owns  Host ID                               Rack
   UN  10.132.87.93   68.07 GiB  256     ?     fb51465c-167a-42f7-98c9-b6eba1de34de  c
   UN  10.132.84.94   69.9 GiB   256     ?     f621a5ac-e7ee-48a9-9a14-73d69477c642  b
   UN  10.132.84.105  76.95 GiB  256     ?     0561086f-e95b-4232-ba6c-ad519ff30336  d
   
   Datacenter: dc-2
   ====================
   Status=Up/Down
   |/ State=Normal/Leaving/Joining/Moving
   --  Address        Load       Tokens  Owns  Host ID                               Rack
   UN  10.132.0.8     71.61 GiB  256     ?     8894a98b-8406-45de-99e2-f404ab10b5d6  c
   UN  10.132.9.204   75.1 GiB   256     ?     afa0ffa3-630b-4f1e-b46f-fc3df988092e  a
   UN  10.132.3.133   68.08 GiB  256     ?     25ae39ab-b39e-4d4f-9cb7-de095ab873db  b

5. Set up Cassandra on all the pods in the new data centers.
   1. Get apigeeorg from the cluster with the following command:

      kubectl get apigeeorg -n apigee -o json | jq .items[].metadata.name
      

      For example:

      Ex: kubectl get apigeeorg -n apigee -o json | jq .items[].metadata.name
      "rg-hybrid-b7d3b9c"
      

   2. Create a cassandra data replication custom resource (YAML) file. The file can have any name. In the following examples the file will have the name datareplication.yaml.

      The file must contain the following:

      apiVersion: apigee.cloud.google.com/v1alpha1
      kind: CassandraDataReplication
      metadata:
        name: REGION_EXPANSION
        namespace: NAMESPACE
      spec:
        organizationRef: APIGEEORG_VALUE
        force: false
        source:
          region: SOURCE_REGION

      Where:

      • REGION_EXPANSION is the name you are giving this metadata. You can use any name.
      • NAMESPACE is the same namespace that is provided in overrides.yaml. This is usually "apigee".
      • APIGEEORG_VALUE is the value output from the kubectl get apigeeorg -n apigee -o json | jq .items[].metadata.name command in the previous step. For example, rg-hybrid-b7d3b9c
      • SOURCE_REGION is the datacenter name in the source region. This is the value set for cassandra:datacenter: in your overrides.yaml.

      For example:

      apiVersion: apigee.cloud.google.com/v1alpha1
      kind: CassandraDataReplication
      metadata:
        name: region-expansion
        namespace: apigee
      spec:
        organizationRef: rg-hybrid-b7d3b9c
        force: false
        source:
          region: "dc-1"

   3. Apply the CassandraDataReplication with the following command:

      kubectl apply -f datareplication.yaml

   4. Verify the rebuild status using the following command.
      Note: The rebuild operation can take up to several hours depending on the data size.

      kubectl -n apigee get apigeeds -o json | jq .items[].status.cassandraDataReplication

      The results should look something like:

      {
        "rebuildDetails": {
          "apigee-cassandra-default-0": {
            "state": "complete",
            "updated": 1623105760
          },
          "apigee-cassandra-default-1": {
            "state": "complete",
            "updated": 1623105765
          },
          "apigee-cassandra-default-2": {
            "state": "complete",
            "updated": 1623105770
          }
        },
        "state": "complete",
        "updated": 1623105770
      }

6. Verify the rebuild processes from the logs. Also, verify the data size using the nodetool status command:

   kubectl logs apigee-cassandra-default-0 -f -n apigee

   kubectl exec apigee-cassandra-default-0 -n apigee  -- nodetool -u JMX_user -pw JMX_password status

   The following example shows example log entries:

   INFO  01:42:24 rebuild from dc: dc-1, (All keyspaces), (All tokens)
   INFO  01:42:24 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Executing streaming plan for Rebuild
   INFO  01:42:24 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Starting streaming to /10.12.1.45
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889, ID#0] Beginning stream session with /10.12.1.45
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Starting streaming to /10.12.4.36
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889 ID#0] Prepare completed. Receiving 1 files(0.432KiB), sending 0 files(0.000KiB)
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Session with /10.12.1.45 is complete
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889, ID#0] Beginning stream session with /10.12.4.36
   INFO  01:42:25 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Starting streaming to /10.12.5.22
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889 ID#0] Prepare completed. Receiving 1 files(0.693KiB), sending 0 files(0.000KiB)
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Session with /10.12.4.36 is complete
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889, ID#0] Beginning stream session with /10.12.5.22
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889 ID#0] Prepare completed. Receiving 3 files(0.720KiB), sending 0 files(0.000KiB)
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] Session with /10.12.5.22 is complete
   INFO  01:42:26 [Stream #3a04e810-580d-11e9-a5aa-67071bf82889] All sessions completed

7. Update the seed hosts. Remove multiRegionSeedHost: 10.0.0.11 from overrides-DC_name.yaml and reapply.
   Seed hosts are local cluster members. To boot up a new region an external seed host is required. Once a region boots up you need to change the seed hosts back to their local clusters in overrides.yaml and then reapply the configuration.

   apigeectl apply -f overrides/overrides-DC_name.yaml

Check the Cassandra cluster status

The following command is useful to see if the cluster setup is successful in two data centers. The command checks the nodetool status for the two regions.

kubectl exec apigee-cassandra-default-0 -n apigee  -- nodetool -u JMX_user -pw JMX_password status


Datacenter: dc-1
================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
--  Address     Load       Tokens       Owns (effective)  Host ID                               Rack
UN  10.12.1.45  112.09 KiB  256          100.0%            3c98c816-3f4d-48f0-9717-03d0c998637f  ra-1
UN  10.12.4.36  95.27 KiB  256          100.0%            0a36383d-1d9e-41e2-924c-7b62be12d6cc  ra-1
UN  10.12.5.22  88.7 KiB   256          100.0%            3561f4fa-af3d-4ea4-93b2-79ac7e938201  ra-1
Datacenter: dc-2
================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
--  Address     Load       Tokens       Owns (effective)  Host ID                               Rack
UN  10.0.4.33   78.69 KiB  256          0.0%              a200217d-260b-45cd-b83c-182b27ff4c99  ra-1
UN  10.0.0.21   78.68 KiB  256          0.0%              9f3364b9-a7a1-409c-9356-b7d1d312e52b  ra-1
UN  10.0.1.26   15.46 KiB  256          0.0%              1666df0f-702e-4c5b-8b6e-086d0f2e47fa  ra-1

Troubleshooting

See Cassandra data replication failure.