Cloudera Docs

Deploying CDP in multiple AWS availability zones

By default, CDP provisions Data Lake, FreeIPA and Data Hubs in a single AWS availability zone (AZ), but you can optionally choose to deploy them across multiple availability zones (multi-AZ). It is possible to enable it either for all or some of these components.

Single-AZ vs multi-AZ deployment

Each AWS region has multiple availability zones, which act as failure domains, preventing small outages from affecting entire regions. By default, a CDP environment (FreeIPA and Data Lake) and Data Hubs running in it are deployed across one subnet, meaning that each of them has VMs spread across a single availability zone (as on AWS every subnet is related to a single availability zone). This is illustrated in the following diagram, where all of the resources run in the eu-central region, and each of them is in a different subnet and availability zone (eu-central-1a for the FreeIPA, eu-central-1b for the Data Lake, and eu-central-1c for the Data Hubs):

If you choose to deploy your CDP environment (FreeIPA and Data Lake) and Data Hubs across multiple subnets and availability zones, each of these components is spread across three or more availability zones, providing high availability and fault tolerance. This is illustrated in the following diagram, where each of the components (FreeIPA, Data Lake, and Data Hubs) is deployed across multiple subnets and multiple availability zones: eu-central-1a, eu-central-1b, and eu-central-1c:

The Data Lake deployed in this setup is the medium duty Data Lake, which along with the enterprise Data Lake has enough nodes to utilize multiple availability zones for providing high availability and fault tolerance.

Multi-AZ can be enabled for Data Lake, FreeIPA, and Data Hubs. It is possible to enable it either for all or some of these components.

Instances within a host group are distributed across availability zones equally, with the least used availability zone(s) preferred. For example, if there is a group of instances across three availability zones with 100 instances in AZ-1, 30 in AZ-2 and 30 in AZ-3, when an upscale of 10 nodes is requested, CDP provisions 5 instances in AZ-2 and 5 in AZ-3.

When a failure happens and a cluster needs to be repaired, the replacement VMs are always provisioned in the same subnet and availability zone as the old ones since the detached disks can only be reattached to a VM in the same availability zone. This means that if there is an availability zone outage, cluster repair is not possible.

Use cases

A multi-AZ Data Lake and FreeIPA constitute a resilient environment that provides a solid basis for multi-AZ Data Hubs and CDP data services. Data Hubs and CDP data services depend on the FreeIPA instance in the Data Lake to provide DNS resolution. Deploying FreeIPA across multiple availability zones ensures that critical DNS resolution is available in the event of an availability zone outage. Furthermore, a medium duty or enterprise Data Lake provides high availability, and additional compute and memory resources for key SDX services and is recommended for production workloads.

Deploying your Data Hubs across multiple availability zones is key if your mission-critical applications depend on HBase and Kafka. Multiple availability zone deployment for operational workloads is considered best practice by the cloud vendors. It ensures that your applications can continue to run in the event of a single availability zone outage.

When an entire availability zone fails, HBase automatically rebalances regions among the remaining instances in the cluster to maintain availability. The write-ahead log (WAL), which is replicated across the three availability zones is automatically replayed by the newly assigned region servers in other availability zones to ensure writes to the database are not lost.

When using the multi availability zone feature, CDP ensures that Kafka replicates partitions across brokers in different availability zones. During an availability zone failure this ensures that no data is lost and applications can continue to access the data they need. Cruise Control, which is deployed alongside every Kafka cluster in CDP Public Cloud, detects that topics need to be rebalanced to the remaining brokers. Once the availability zone is back online, you can repair your Kafka cluster, restoring the initial broker distribution across availability zones. Afterwards Cruise Control kicks in and ensures that all topic partitions are balanced across the cluster.

Limitations

The following limitations apply when deploying a multi-AZ CDP:

  • When an AZ is down, you cannot create a new Data Hub, and create or activate CDP data services within the environment. Existing workloads will continue to work.

  • When an AZ is down, you cannot resize, stop, or restart Data Hubs.

  • When an AZ is down, existing running workloads will continue to run. Some Data Lake services will still be functioning (because they run across more than one AZ) but others may be down if the AZ where they are running is down.

    The following diagram shows how Data Lake services are distributed across AZs (assuming that there are 3 AZs per region). The Data Lake services marked in red run in a single AZ only and will be down if that AZ is down:

    The following table shows which services aren’t affected by an AZ outage (column 1 and 2) and those that may potentially be affected (column 3):

    Services running in all available AZs Services running in two AZs Services running in a single AZ
    Datanode

    HBase Gateway

    HBase Master

    HBase Region Server

    HDFS Gateway

    HDFS Journal Node

    Hive Gateway

    Kafka Broker

    Kafka Gateway

    Solr Server

    		Atlas Server

    HDFS Failover Controller

    Hive Metastore

    IDBroker

    Knox Gateway

    Namenode

    Ranger Admin

    Solr Gateway

    Zookeeper Server

    		 Cloudera Manager

    Ranger UserSync

    Ranger TagSync

    As shown in the diagram and the table, the following three services will be down if the AZ where they are running is down:

    • Cloudera Manager - When Cloudera Manager is down, you cannot perform administrative tasks that you would normally perform via Cloudera Manager.

    • Ranger UserSync - When Ranger UserSync is down, existing users are not impacted, but synchronizing new users to Ranger is disrupted. For example, if you add a new user and sync it to FreeIPA while Ranger UserSync is down, the newly added user will not be synced to Ranger and as a result the user will not be able to access workload data.

    • Ranger TagSync - When Ranger TagSync is down, existing tags and policies are not impacted, but synchronizing new, updated, or removed tags with an external metadata service such as Apache Atlas and creating Ranger policies with these tags is disrupted.

  • Single AZ environments or clusters cannot be converted to multi-AZ.
  • While CDP CLI allows you to specify multiple subnets per Data Hub’s host group, due to load balancer requirements only one subnet per AZ can be configured for the Cloudera Manager node group of the Data Hub cluster.
  • While by default CDP deploys Data Lake, FreeIPA, and Data Hubs using CloudFormation, in case of the multi-AZ setup no CloudFormation is used. This affects vertical scaling: After vertical scaling (for example during cluster repair or upgrade) CDP always provisions the original instance types. For example, if you are using vertical scaling runbooks, repair and upgrade operations will use the previous instance type and root volume settings.

Enabling multi-AZ

Multi-AZ can be enabled for Data Lake, FreeIPA, and Data Hubs running on AWS. It is possible to enable it either for all or some of these components.

To enable multi-AZ, you should register a multi-AZ CDP environment with a Medium Duty Data Lake. You can also create multi-AZ Data Hubs within any existing environment. Detailed steps are provided below.

Register a multi-AZ environment

You can register a multi-AZ AWS environment via CDP UI or CDP CLI. You may choose to enable multi-AZ for Data Lake only or for FreeIPA only. There is no requirement to enable both.

Steps

Register your environment as usual, just make sure to do the following:

  1. On the Data Access and Data Lake Scaling page, select the Medium Duty Data Lake.

  2. On the same page, scroll down and in the bottom of the page enable the Advanced Options.

  3. In the Network and Availability section enable the Enable Multiple Availability Zones for Data Lake toggle button in order to enable multi-AZ for Data Lake. The option is disabled by default. The option only appears when a Medium Duty Data Lake is selected.

  4. On the Region, Networking, and Security page, scroll down and in the bottom of the page enable the Advanced Options.

  5. In the Network and Availability section enable the Enable Multiple Availability Zones for Data Lake toggle button in order to enable multi-AZ for FreeIPA. The option is disabled by default.

Register your AWS environment via CDP CLI as usual, but when creating a template for the environment registration commands, add the following to enable multi-AZ:

  • In the environment creation command, add the multiAz=true to the --free-ipa option.

  • In the Data Lake creation command, add the --multi-az option.

  • In the Data Lake creation command, make sure to specify the Medium Duty Data Lake.

The following example illustrates these necessary additions, which are emphasized using bold typeface: 

cdp environments create-aws-environment \
  --environment-name tb-multiaz-env \
  --credential-name tb-cldr-acc \
  --region "eu-central-1" \
  --security-access cidr=0.0.0.0/0 \
  --no-enable-tunnel \
  --authentication publicKeyId="test" \
  --log-storage storageLocationBase=s3a://cb-group/test,instanceProfile=arn:aws:iam::152413716728:instance-profile/mock-idbroker-admin-role \
  --vpc-id vpc-03e505817e3619238 \
  --subnet-ids subnet-013855b2gc32c2cd8 subnet-02b9054ef829374fe subnet-085c9ff36b38c0b35 \
  --free-ipa instanceCountByGroup=3,multiAz=true
cdp environments set-id-broker-mappings \
  --environment-name tb-multiaz-env \
  --data-access-role arn:aws:iam::152413716728:role/mock-idbroker-admin-role \
  --ranger-audit-role arn:aws:iam::152413716728:role/mock-idbroker-admin-role \
  --set-empty-mappings
cdp datalake create-aws-datalake \
  --datalake-name tb-multiaz-dl \
  --environment-name tb-multiaz-env \
  --cloud-provider-configuration instanceProfile=arn:aws:iam::152413716728:instance-profile/mock-idbroker-assumer,storageBucketLocation=s3a://cb-group/test \
  --scale MEDIUM_DUTY_HA \
  --runtime 7.2.12 \
  --no-enable-ranger-raz \
  --profile mowdev \
  --multi-az

Create multi-AZ Data Hubs

If you would like to create multi-AZ Data Hubs, see Creating a multi-AZ Data Hub on AWS.