SQL Server 2019 Big Data Clusters release notes

Applies to: yesSQL Server 2019 (15.x)

The following release notes apply to SQL Server 2019 Big Data Clusters. This article is broken into sections for each release. Each release has a link to a support article describing the CU changes as well as links to the Linux package downloads. The article also lists known issues for the most recent releases of SQL Server Big Data Clusters (BDC).

Supported platforms

This section explains platforms that are supported with BDC.

Kubernetes platforms

Platform Supported versions
Vanilla (upstream) Kubernetes Deploy BDC on premises using a Kubernetes cluster version minimum 1.13. See Kubernetes version and version skew support policy.
Red Hat OpenShift Deploy BDC on premises using an OpenShift cluster version minimum 4.3. See Red Hat OpenShift Container Platform Life Cycle Policy.

Support introduced in SQL Server 2019 CU5.
Azure Kubernetes Service (AKS) Deploy BDC on AKS cluster version minimum 1.13.
See Supported Kubernetes versions in AKS for version support policy.
Azure Red Hat OpenShift (ARO) Deploy BDC on ARO version minimum 4.3. See Azure Red Hat OpenShift.

Support introduced in SQL Server 2019 CU5.

Host OS for Kubernetes

Platform Host OS Supported versions
Kubernetes Ubuntu 16.04
Kubernetes Red Hat Enterprise Linux 7.3, 7.4, 7.5, 7.6
OpenShift Red Hat Enterprise Linux / CoreOS See OpenShift release notes

SQL Server Editions

Edition Notes
Big Data Cluster edition is determined by the edition of SQL Server master instance. At deployment time Developer edition is deployed by default. You can change the edition after deployment. See Configure SQL Server master instance.


Platform Supported versions
Azure Data CLI (azdata) As a best practice, use the latest version available. Starting with SQL Server 2019 CU5 release, Azure Data CLI (azdata) has an independent semantic version from the server.

Run azdata –-version to validate the version.

See Release history for latest version.
Azure Data Studio Get the latest build of Azure Data Studio.

For a complete list, see Which tools are required?

Release history

The following table lists the release history for SQL Server 2019 Big Data Clusters.

Release 1 BDC Version Azure Data CLI (azdata) version 2 Release date
CU8-GDR 15.0.4083.2 20.2.6 2021-01-12
CU8 15.0.4073.23 20.2.2 2020-10-19
CU6 15.0.4053.23 20.0.1 2020-08-04
CU5 15.0.4043.16 20.0.0 2020-06-22
CU4 15.0.4033.1 15.0.4033 2020-03-31
CU3 15.0.4023.6 15.0.4023 2020-03-12
CU2 15.0.4013.40 15.0.4013 2020-02-13
CU1 15.0.4003.23 15.0.4003 2020-01-07
GDR1 15.0.2070.34 15.0.2070 2019-11-04

1 CU7 is not available for BDC.

2 Azure Data CLI (azdata) version reflects the version of the tool at the time of the CU release. azdata can also release independently of the server release, therefore you might get newer versions when you install the latest packages. Newer versions are compatible with previously released CUs.

How to install updates

To install updates, see How to upgrade SQL Server Big Data Clusters.

CU8-GDR(January 2021)

Cumulative Update 8 GDR (CU8-GDR) release for SQL Server 2019.

Package version Image tag
15.0.4083.2 [2019-CU8-GDR2-ubuntu-16.04]

CU8 (September 2020)

Cumulative Update 8 (CU8) release for SQL Server 2019.

Package version Image tag
15.0.4073.23 [2019-CU8-ubuntu-16.04]

This release includes several fixes and a couple of enhancements.

Added capabilities

CU6 (July 2020)

Cumulative Update 6 (CU6) release for SQL Server 2019.

Package version Image tag
15.0.4053.23 [2019-CU6-ubuntu-16.04]

This release includes minor fixes and enhancements. The following articles include information related to these updates:

CU5 (June 2020)

Cumulative Update 5 (CU5) release for SQL Server 2019.

Package version Image tag
15.0.4043.16 [2019-CU5-ubuntu-16.04]

Added capabilities

  • Support for Big Data Clusters deployment on Red Hat OpenShift. Support includes OpenShift container platform deployed on premises version 4.3 and up and Azure Red Hat OpenShift. See Deploy SQL Server Big Data Clusters on OpenShift
  • Updated the BDC deployment security model so privileged containers deployed as part of BDC are no longer required. In addition to non-privileged, containers are running as non-root user by default for all new deployments using SQL Server 2019 CU5.
  • Added support for deploying multiple big data clusters against an Active Directory domain.
  • Azure Data CLI (azdata) has its own semantic version, independent from the server. Any dependency between the client and the server version of azdata is removed. We recommend using the latest version for both client and server to ensure you are benefiting from latest enhancements and fixes.
  • Introduced two new stored procedures, sp_data_source_objects and sp_data_source_table_columns, to support introspection of certain External Data Sources. They can be used by customers directly via T-SQL for schema discovery and to see what tables are available to be virtualized. We leverage these changes in the External Table Wizard of the Data Virtualization Extension for Azure Data Studio, which allows you to create external tables from SQL Server, Oracle, MongoDB, and Teradata.
  • Added support to persist customizations performed in Grafana. Before CU5 customers would notice that any edits in Grafana configurations would be lost upon metricsui pod (that hosts Grafana dashboard) restart. This issue is fixed and all configurations are now persisted.
  • Fixed security issue related to the API used to collect pod and node metrics using Telegraf (hosted in the metricsdc pods). As a result of this change, Telegraf now requires a service account, cluster role and cluster bindings to have the necessary permissions to collect the pod and node metrics. See Custer role required for pods and nodes metrics collection for more details.
  • Added two feature switches to control the collection of pod and node metrics. In case you are using different solutions for monitoring your Kubernetes infrastructure, you can turn off the built-in metrics collection for pods and host nodes by setting allowNodeMetricsCollection and allowPodMetricsCollection to false in control.json deployment configuration file. For OpenShift environments, these settings are set to false by default in the built-in deployment profiles, since collecting pod and node metrics required privileged capabilities.

CU4 (April 2020)

Cumulative Update 4 (CU4) release for SQL Server 2019. The SQL Server Database Engine version for this release is 15.0.4033.1.

Package version Image tag
15.0.4033.1 [2019-CU4-ubuntu-16.04]

CU3 (March 2020)

Cumulative Update 3 (CU3) release for SQL Server 2019. The SQL Server Database Engine version for this release is 15.0.4023.6.

Package version Image tag
15.0.4023.6 [2019-CU3-ubuntu-16.04]

Resolved issues

SQL Server 2019 CU3 resolves the following issues from previous releases.

CU2 (February 2020)

Cumulative Update 2 (CU2) release for SQL Server 2019. The SQL Server Database Engine version for this release is 15.0.4013.40.

Package version Image tag
15.0.4013.40 [2019-CU2-ubuntu-16.04]

CU1 (January 2020)

Cumulative Update 1 (CU1) release for SQL Server 2019. The SQL Server Database Engine version for this release is 15.0.4003.23.

Package version Image tag
15.0.4003.23 [2019-CU1-ubuntu-16.04]

GDR1 (November 2019)

SQL Server 2019 General Distribution Release 1 (GDR1) - introduces general availability for Big Data Clusters. The SQL Server Database Engine version for this release is 15.0.2070.34.

Package version Image tag
15.0.2070.34 [2019-GDR1-ubuntu-16.04]

SQL Server 2019 servicing updates

For current information about SQL Server servicing updates, see https://support.microsoft.com/help/4518398.

Known issues

MSDTC capabilities can not be enabled for SQL Server master instance running within BDC

  • Affected releases: All big data cluster deployment configurations, irrespective of the release.

  • Issue and customer impact: With SQL Server deployed within BDC as SQL Server master instance, the MSDTC feature can not be enabled. There is no workaround to this issue.

HA SQL Server Database Encryption key encryptor rotation

  • Affected releases: All big data cluster HA deployments irrespective of the release.

  • Issue and customer impact: With SQL Server deployed with HA, the certificate rotation for the encrypted database fails. When the following command is executed on the master pool, an error message will appear:

    CERTIFICATE <NewCertificateName>

    There is no impact, the command fails and the target database encryption is preserved using the previous certificate.

Empty Livy jobs before you apply cumulative updates

  • Affected releases: All version up to CU6. Resolved for CU8.

  • Issue and customer impact: During an upgrade, sparkhead returns 404 error.

  • Workaround: Before upgrading BDC, ensure that there are no active Livy sessions or batch jobs. Follow the instructions under Upgrade from supported release to avoid this.

    If Livy returns a 404 error during the upgrade process, restart the Livy server on both sparkhead nodes. For example:

    kubectl -n <clustername> exec -it sparkhead-0/sparkhead-1 -c hadoop-livy-sparkhistory -- exec supervisorctl restart livy

Big data cluster generated service accounts passwords expiration

  • Affected releases: All big data cluster deployments with Active Directory integration, irrespective of the release

  • Issue and customer impact: During big data cluster deployment, the workflow generates a set of service accounts.Depending on the password expiration policy set in the Domain Controller, passwords for these accounts can expire (default is 42 days). At this time, there is no mechanism to rotate credentials for all accounts in BDC, so the cluster will become inoperable once the expiration period is met.

  • Workaround: Update the expiration policy for the BDC service accounts to “Password never expires” in the Domain Controller. For a complete list of these accounts see Auto generated Active Directory objects. This action can be done before or after the expiration time. In the latter case, Active Directory will reactivate the expired passwords.

Credentials for accessing services through gateway endpoint

  • Affected releases: New clusters deployed starting with CU5.

  • Issue and customer impact: For new big data clusters deployed using SQL Server 2019 CU5, gateway username is not root. If the application used to connect to gateway endpoint is using the wrong credentials, you will see an authentication error. This change is a result of running applications within the big data cluster as non-root user (a new default behavior starting with SQL Server 2019 CU5 release, when you deploy a new big data cluster using CU5, the username for the gateway endpoint is based on the value passed through AZDATA_USERNAME environment variable. It is the same username used for the controller and SQL Server endpoints. This is only impacting new deployments, existing big data clusters deployed with any of the previous releases will continue to use root. There is no impact to credentials when the cluster is deployed to use Active Directory authentication.

  • Workaround: Azure Data Studio will handle the credentials change transparently for the connection made to gateway to enable HDFS browsing experience in the ObjectExplorer. You must install latest Azure Data Studio release that includes the necessary changes that address this use case. For other scenarios where you must provide credentials for accessing service through the gateway (e.g. logging in with Azure Data CLI (azdata), accessing web dashboards for Spark), you must ensure the correct credentials are used. If you are targeting an existing cluster deployed before CU5 you will continue using root username to connect to gateway, even after upgrading the cluster to CU5. If you deploy a new cluster using CU5 build, log in by providing the username corresponding to AZDATA_USERNAME environment variable.

Pods and nodes metrics not being collected

  • Affected releases: New and existing clusters that are using CU5 images

  • Issue and customer impact: As a result of a security fix related to the API that telegraf was using to collect metrics pod and host node metrics, customers may noticed that the metrics are not being collected. This is possible in both new and existing deployments of BDC (after upgrade to CU5). As a result of the fix, Telegraf now requires a service account with cluster wide role permissions. The deployment attempts to create the necessary service account and cluster role, but if the user deploying the cluster or performing the upgrade does not have sufficient permissions, deployment/upgrade proceeds with a warning and succeeds, but the pod & node metrics will not be collected.

  • Workaround: You can ask an administrator to create the role and service account (either before or after the deployment/upgrade), and BDC will use them. This article describes how to create the required artifacts.

azdata bdc copy-logs command failure

  • Affected releases: Azure Data CLI (azdata) version 20.0.0

  • Issue and customer impact: Implementation of copy-logs command is assuming kubectl client tool is installed on the client machine from which the command is issued. If you are issuing the command against a BDC cluster installed on OpenShift, from a client where only oc tool is installed, you will get an error: An error occurred while collecting the logs: [WinError 2] The system cannot find the file specified.

  • Workaround: Install kubectl tool on the same client machine and re-issue the azdata bdc copy-logs command. See instructions here how to install kubectl.

Deployment with private repository

  • Affected releases: GDR1, CU1, CU2. Resolved for CU3.

  • Issue and customer impact: Upgrade from private repository has specific requirements

  • Workaround: If you use a private repository to pre-pull the images for deploying or upgrading BDC, ensure that the current build images as well as the target build images are in the private repository. This enables successful rollback, if necessary. Also, if you changed the credentials of the private repository since the original deployment, update the corresponding secret in Kubernetes before you upgrade. Azure Data CLI (azdata) does not support updating the credentials through AZDATA_PASSWORD and AZDATA_USERNAME environment variables. Update the secret using kubectl edit secrets.

Upgrading using different repositories for current and target builds is not supported.

Upgrade may fail due to timeout

  • Affected releases: GDR1, CU1, CU2. Resolved for CU 3.

  • Issue and customer impact: An upgrade may fail due to timeout.

    The following code shows what the failure might look like:

    >azdata.EXE bdc upgrade --name <mssql-cluster>
    Upgrading cluster to version 15.0.4003
    NOTE: Cluster upgrade can take a significant amount of time depending on
    configuration, network speed, and the number of nodes in the cluster.
    Upgrading Control Plane.
    Control plane upgrade failed. Failed to upgrade controller.

    This error is more likely to occur when you upgrade BDC in Azure Kubernetes Service (AKS).

  • Workaround: Increase the timeout for the upgrade.

    To increase the timeouts for an upgrade, edit the upgrade config map. To edit the upgrade config map:

    1. Run the following command:

      kubectl edit configmap controller-upgrade-configmap
    2. Edit the following fields:

      controllerUpgradeTimeoutInMinutes Designates the number of minutes to wait for the controller or controller db to finish upgrading. Default is 5. Update to at least 20.

      totalUpgradeTimeoutInMinutes: Designates the combines amount of time for both the controller and controller db to finish upgrading (controller + controllerdb upgrade). Default is 10. Update to at least 40.

      componentUpgradeTimeoutInMinutes: Designates the amount of time that each subsequent phase of the upgrade has to complete. Default is 30. Update to 45.

    3. Save and exit

    The python script below is another way to set the timeout:

    from kubernetes import client, config
    import json
    def set_upgrade_timeouts(namespace, controller_timeout=20, controller_total_timeout=40, component_timeout=45):
         """ Set the timeouts for upgrades
         The timeout settings are as follows
         controllerUpgradeTimeoutInMinutes: sets the max amount of time for the controller
             or controllerdb to finish upgrading
         totalUpgradeTimeoutInMinutes: sets the max amount of time to wait for both the
             controller and controllerdb to complete their upgrade
         componentUpgradeTimeoutInMinutes: sets the max amount of time allowed for
             subsequent phases of the upgrade to complete
         upgrade_config_map = client.CoreV1Api().read_namespaced_config_map("controller-upgrade-configmap", namespace)
         upgrade_config = json.loads(upgrade_config_map.data["controller-upgrade"])
         upgrade_config["controllerUpgradeTimeoutInMinutes"] = controller_timeout
         upgrade_config["totalUpgradeTimeoutInMinutes"] = controller_total_timeout
         upgrade_config["componentUpgradeTimeoutInMinutes"] = component_timeout
         upgrade_config_map.data["controller-upgrade"] = json.dumps(upgrade_config)
         client.CoreV1Api().patch_namespaced_config_map("controller-upgrade-configmap", namespace, upgrade_config_map)

Livy job submission from Azure Data Studio (ADS) or curl fail with 500 error

  • Issue and customer impact: In an HA configuration, Spark shared resources sparkhead are configured with multiple replicas. In this case, you might experience failures with Livy job submission from Azure Data Studio (ADS) or curl. To verify, curl to any sparkhead pod results in refused connection. For example, curl https://sparkhead-0:8998/ or curl https://sparkhead-1:8998 returns 500 error.

    This happens in the following scenarios:

    • Zookeeper pods, or processes for each zookeeper instance, are restarted a few times.
    • When networking connectivity is unreliable between sparkhead pod and Zookeeper pods.
  • Workaround: Restarting both Livy servers.

    kubectl -n <clustername> exec sparkhead-0 -c hadoop-livy-sparkhistory supervisorctl restart livy
    kubectl -n <clustername> exec sparkhead-1 -c hadoop-livy-sparkhistory supervisorctl restart livy

Create memory optimized table when master instance in an availability group

  • Issue and customer impact: You cannot use the primary endpoint exposed for connecting to availability group databases (listener) to create memory optimized tables.

  • Workaround: To create memory optimized tables when SQL Server master instance is an availability group configuration, connect to the SQL Server instance, expose an endpoint, connect to the SQL Server database, and create the memory optimized tables in the session created with the new connection.

Insert to external tables Active Directory authentication mode

  • Issue and customer impact: When SQL Server master instance is in Active Directory authentication mode, a query that selects only from external tables, where at least one of the external tables is in a storage pool, and inserts into another external table, the query returns:

    Msg 7320, Level 16, State 102, Line 1
    Cannot execute the query "Remote Query" against OLE DB provider "SQLNCLI11" for linked server "SQLNCLI11". Only domain logins can be used to query Kerberized storage pool.
  • Workaround: Modify the query in one of the following ways. Either join the storage pool table to a local table, or insert into the local table first, then read from the local table to insert into the data pool.

Transparent Data Encryption capabilities can not be used with databases that are part of the availability group in the SQL Server master instance

  • Issue and customer impact: In an HA configuration, databases that have encryption enabled can't be used after a failover since the master key used for encryption is different on each replica.

  • Workaround: There is no workaround for this issue. We recommend to not enable encryption in this configuration until a fix is in place.

Next steps

For more information about SQL Server Big Data Clusters, see What are SQL Server 2019 Big Data Clusters?