Use azdata distcp to perform data movement between SQL Server big data clusters

Applies to: yesSQL Server 2019 (15.x)

This article explains how to use azdata to perform high performant distributed data copies between SQL Server Big Data Clusters.

Prerequisites

Introduction to distributed data copies on SQL Server Big Data Clusters

Hadoop HDFS DistCP is a command-line tool used to perform distributed parallel copies of files and folders from one HDFS cluster to another. Distributed parallel copying enables fast transfer of Data Lake scale files and folders between two different clusters, enabling migrations, the creation of segmented environments, high-availability, and disaster recovery scenarios.

Hadoop HDFS DistCP uses an internal MapReduce job to expand a list of files and directories into input to multiple map tasks, each of which will copy a partition of the files specified in the source list to the destination. This allows multiple data nodes in the source cluster to send data directly to multiple data nodes in the destination clusters, creating a truly distributed parallel copy scenario.

On SQL Server Big Data Clusters CU13 and above, the distributed copy functionality is integrated into the product and is exposed through the azdata bdc hdfs distcp command. The command is an asynchronous operation, it returns immediately while the copy job executes in the background. Monitor the copy job using either azdata or the provided monitoring user interfaces.

Only SQL Server Big Data Clusters sources and destinations are supported.

Clusters may be deployed in both Active Directory enabled mode or basic security modes. Copies may be performed between any combination of security modes. For Active Directory enabled clusters, it is required that they are in the same domain.

In this guide we will cover the following data copy scenarios:

  • Active Directory enabled cluster to Active Directory enabled cluster
  • Basic security cluster to Active Directory enabled cluster
  • Basic security cluster to basic security cluster

Required steps to all scenarios

Certificates are required to create a trusted relationship between source and destination clusters. These steps are required only once per source/destination cluster combination.

Important

If a SQL Server big data cluster with basic authentication (non-AD) is upgraded to CU13, the distcp functionality won't work.

To enable the distcp functionality in this scenario execute the following additional steps once:

kubectl -n $CLUSTER_NAME exec -it nmnode-0-0 -- bash
export HADOOP_USER_NAME=hdfs
hadoop fs -mkdir -p /tmp/hadoop-yarn/staging/history
hadoop fs -chown yarn /tmp/hadoop-yarn
hadoop fs -chown yarn /tmp/hadoop-yarn/staging
hadoop fs -chown yarn /tmp/hadoop-yarn/staging/history
hadoop fs -chmod 733 /tmp/hadoop-yarn
hadoop fs -chmod 733 /tmp/hadoop-yarn/staging
hadoop fs -chmod 733 /tmp/hadoop-yarn/staging/history

The required notebooks in the next steps are part of the Operational notebooks for SQL Server Big Data Clusters. For more information how to install and use the notebooks, see Operational notebooks

Step 1 - Certificate creation and installation

Connect to your source cluster using Azure Data Studio. This is where the files will be copied from. Perform the following steps:

  1. Create new cluster root certificate in source cluster using notebook cer001-create-root-ca.ipynb

  2. Download the new cluster root certificate to your machine using cer002-download-existing-root-ca.ipynb

  3. Install the new root certificate to the source cluster using cer005-install-existing-root-ca.ipynb. This step will take 10-15 minutes.

  4. Generate new knox certificate on the source cluster cer021-create-knox-cert.ipynb

  5. Sign with the new knox certificate with the new root certificate using cer031-sign-knox-generated-cert.ipynb

  6. Install the new knox certificate on the source cluster using cer041-install-knox-cert.ipynb

    Important

    The certificate generation commands will create the root CA file (cluster-ca-certificate.crt) and Knox certificate file (cacert.pem) in "C:\Users\{Username}\AppData\Local\Temp\1\mssql-cluster-root-ca\" on Windows and "/tmp/mssql-cluster-root-ca/ on Unix.

Step 2 - Certificate installation on destination

Connect to your destination cluster using Azure Data Studio. This is where the files will be copied to. Perform the following steps:

Warning

If you're using Azure Data Studio on different machines to perform this tasks, copy the cluster-ca-certificate.crt and cacert.pem files generated on Step 1 to the right locations on the other machine before running the notebook in the next step.

  1. Install the new root certificate from the source cluster to the destination cluster using cer005-install-existing-root-ca.ipynb. This step will take 10-15 minutes.

Step 3 - Keytab file creation

You need to create a keytab file if any of the clusters is Active Directory enabled. The file will perform authentication to enable the copy to take place.

Create the keytab file using ktutil. An example follows. Make sure to define or replace the environment variables DOMAIN_SERVICE_ACCOUNT_USERNAME, DOMAIN_SERVICE_ACCOUNT_PASSWORD, and REALM with the appropriate values.

ktutil
> add_entry -password -p $DOMAIN_SERVICE_ACCOUNT_USERNAME@$REALM -k 1 -e arcfour-hmac-md5
$DOMAIN_SERVICE_ACCOUNT_PASSWORD
> add_entry -password -p $DOMAIN_SERVICE_ACCOUNT_USERNAME@$REALM -k 1 -e aes256-cts
$DOMAIN_SERVICE_ACCOUNT_PASSWORD
> add_entry -password -p $DOMAIN_SERVICE_ACCOUNT_USERNAME@$REALM -k 1 -e aes256-cts-hmac-sha1-96
$DOMAIN_SERVICE_ACCOUNT_PASSWORD
> wkt /tmp/$DOMAIN_SERVICE_ACCOUNT_USERNAME.keytab
> exit

Step 4 - Copy the keytab to the HDFS location

This step is only required if any of the clusters is Active Directory enabled.

Make sure to define or replace the environment variables DOMAIN_SERVICE_ACCOUNT_USERNAME with the appropriate value.

Upload the keytab to destination (secure cluster):

azdata bdc hdfs mkdir -p /user/$DOMAIN_SERVICE_ACCOUNT_USERNAME
azdata bdc hdfs cp -f /tmp/$DOMAIN_SERVICE_ACCOUNT_USERNAME.keytab -t /user/$DOMAIN_SERVICE_ACCOUNT_USERNAME/$DOMAIN_SERVICE_ACCOUNT_USERNAME.keytab

Data copy scenarios

Scenario 1 - Active Directory enabled cluster to Active Directory enabled cluster

  1. Export the required environment variable DISTCP_KRB5KEYTAB:

    export DISTCP_KRB5KEYTAB=/user/$DOMAIN_SERVICE_ACCOUNT_USERNAME/$DOMAIN_SERVICE_ACCOUNT_USERNAME.keytab
    
  2. Either define or replace CLUSTER_CONTROLLER, DESTINATION_CLUSTER_NAMESPACE, and PRINCIPAL variables with appropriate vales. Then use azdata to authenticate to destination cluster:

    azdata login --auth ad --endpoint $CLUSTER_CONTROLLER --namespace $DESTINATION_CLUSTER_NAMESPACE --principal $PRINCIPAL
    
  3. Run the distributed data copy:

    azdata bdc hdfs distcp submit -f https://knox.$CLUSTER_NAME.$DOMAIN:30443/file.txt -t  /file.txt
    

Scenario 2 - Basic security cluster to Active Directory enabled cluster

  1. Export the required environment variable DISTCP_KRB5KEYTAB, DISTCP_AZDATA_USERNAME, and DISTCP_AZDATA_PASSWORD:

    export DISTCP_KRB5KEYTAB=/user/$DOMAIN_SERVICE_ACCOUNT_USERNAME/$DOMAIN_SERVICE_ACCOUNT_USERNAME.keytab
    export DISTCP_AZDATA_USERNAME=<your basic security bdc username> 
    export DISTCP_AZDATA_PASSWORD=<your basic security bdc password>
    
  2. Either define or replace CLUSTER_CONTROLLER, DESTINATION_CLUSTER_NAMESPACE, and PRINCIPAL variables with appropriate vales. Then use azdata to authenticate to destination cluster:

    azdata login --auth ad --endpoint $CLUSTER_CONTROLLER --namespace $DESTINATION_CLUSTER_NAMESPACE --principal $PRINCIPAL
    
  3. Run the distributed data copy

    azdata bdc hdfs distcp submit --from-path https://$SOURCE_CLUSTER_IP:30443/file.txt --to-path /file.txt
    

Scenario 3 - Basic security cluster to basic security cluster

  1. Export the required environment variable DISTCP_AZDATA_USERNAME and DISTCP_AZDATA_PASSWORD:

    export DISTCP_AZDATA_USERNAME=<your basic security bdc username> 
    export DISTCP_AZDATA_PASSWORD=<your basic security bdc password>
    
  2. Use azdata to authenticate to destination cluster

  3. Run the distributed data copy

    azdata bdc hdfs distcp submit --from-path https://$SOURCE_CLUSTER_IP:30443/file.txt --to-path /file.txt
    

Monitoring the distributed copy job

The azdata bdc hdfs distcp submit command is an asynchronous operation, it returns immediately while the copy job is running on the background. Monitor the copy job using either azdata or the provided monitoring user interfaces.

List all distributed copy jobs

azdata bdc hdfs distcp list

Take note of the job-id of job you and to track. All other commands depend on it.

Get the state of a distributed copy job

azdata bdc hdfs distcp state [--job-id | -i] <JOB_ID>

Get the complete status of a distributed copy job

azdata bdc hdfs distcp status [--job-id | -i] <JOB_ID>

Get the logs of a distributed copy job

azdata bdc hdfs distcp log [--job-id | -i] <JOB_ID>

Distributed copy hints

  1. In order to copy entire directories and their contents, end the path argument with a directory, without the '/'. The following example copies the entire sourceDirectories directory to the root HDFS destination:

    azdata bdc hdfs distcp submit --from-path https://$SOURCE_CLUSTER_IP:30443/sourceDirectories --to-path /
    
  2. The command may contain options, supporting the --overwrite, --preserve, --update, and --delete modifiers. The modifier should be placed just after the submit keyword, like below:

    azdata bdc hdfs distcp submit {options} --from-path https://$SOURCE_CLUSTER_IP:30443/sourceDirectories --to-path /
    

Next steps

For more information, see Introducing SQL Server Big Data Clusters.

For complete reference of the new command, see azdata bdc hdfs distcp.