Tag Archives: availability groups

Always On Availability Groups using containers

Posted by on December 11, 2020 No comments | This post has 2,829 views.

The complete code can be found on my GitHub account

As a SQL Server person, I usually need to work with full blown Availability Groups for my various test scenarios.
I need to have a reliable and consistent way to rebuild the whole setup, multiple times a day.

For this purpose, docker containers are a perfect fit.
This approach will serve multiple scenarios (tsql development, performance tuning, infrastructure changes, etc.)

Target

Using the process I’ll explain below, I will deploy:

  • 3 nodes running SQL Server 2019 Dev on top of Ubuntu 18.04
  • 1 Clusterless Availability Group (also known as Read-Scale Availability Group)

Note that our Clusterless AG is not a high availability or disaster recovery solution.
It only provides a mechanism to synchronize databases across multiple servers (containers).
Only manual failover without data loss and forced failover with data loss is possible when using Read-Scale availability groups.

For production ready and true HA and DR one should look into traditional availability groups running on top of Windows Failover Cluster.
Another viable solution is to run SQL Server instance on Kubernetes in Azure Kubernetes Service (AKS), with persistent storage for high availability.

How

The workflow consist of the following steps:

  • prepare a custom docker image running Ubuntu 18.04 and SQL Server 2019
  • create a configuration file that will be used by docker-compose to spin up the 3 nodes

 

 

 

 

 

The actual build of the Availability Group will be performed by the entrypoint.sh script that will run on all the containers based on the image we just created.

The entrypoint.sh file is used to configure the container.
We just need to add a few .sql scripts that will get executed using sqlcmd utility.
In this case is the ag.sql file that contains the commands to create logins, certificates, endpoints and finally the Availability Group.

Remember, we’re using a Clusterless Availability Group, so the SQL Server service on Linux uses certificates to authenticate communication between the mirroring endpoints.

In a matter of minutes I have a fully working AG.

Credentials

During the build of the docker image and to create the AG I will need to specify various variables and credentials.

For production environments the recommended approach to manage secrets is to use a vault.

For my case I’m storing various variables and credentials in plain text files in the env folder.
Docker will parse those files and they will be available as environment variables.

  • sapassword.env – this contains the SA password and it’s needed when the custom image is built.

  • sqlserver.env – various variables are set here and are needed when the custom image is built.

  • miscpassword.env – will be needed to create the login and certificate needed by the Availability Group. This file is actually added to the container and it will be deleted after the Availability Group is created.

The advantage of this approach is that I have only one place where I store all these variables and credentials, but as I mentioned earlier, it’s not a proper solution from a security standpoint.

A few alternative approaches would be:
– use a tool to manage secrets, like Vault
– multi-stage builds
– use BuildKit

Conclusion

From a testing and development point of view, this solution works very well for me as I can rebuild the environment in a fast and consistent way.

It’s not by any means the best option out there, but it’s really simple to use and reproduce.

See it in action

Click on the image for the full gif

Wrong network location profile causes issues with Windows Failover Cluster

Posted by on July 25, 2020 2 comments | This post has 39,088 views.

Hi folks,

The other day I was pulling hair from my head trying to configure a Windows Failover Cluster intended for an SQL Server Availability Group setup.

During the cluster validation stage I always got this message:

The Windows Firewall on node node01.domain.local is not properly configured for failover clustering.
In particular, the ‘Public’ firewall profile is enabled on adapter ‘node01.domain.local – SLOT 1 PORT 2’.
The ‘Failover Clusters’ rule group is not enabled in firewall profile ‘Public’.
This may prevent some network communication between cluster nodes.

The OS install and networking part was already configured by a someone else and it was a pretty straightforward installation.

The issue turned out to be caused by the 2 NICs we have for iSCSI traffic which did not have a gateway configured.

Windows uses gateways to identify networks. If it doesn’t have a gateway configured, or if it can’t successfully ping it, it will not be able to identify the network it’s connected to and will assume it’s a public one.

Network cards in Windows can be connected to one of these type of networks:
– Public
– Private
– DomainAuthenticated

By default, the public network location type is assigned to any new networks when they are first connected.

A public network is considered to be shared with the world, with no protection between the local computer and any other computer. Therefore, the Windows Firewall rules associated with the public profile are the most restrictive.

As part of the Windows Failover Cluster validation/creation there are checks to verify connectivity (between cluster nodes, active directory, etc.).

These were the settings I had:
Before

All I needed to do was to move all non-domain network interfaces into the private profile:

After

After the change the cluster creation went without issue.

This small detail be easily missed and can cause a lot of headaches and lost time investigating failover clusters.

Cheers!