Using AKS and SOCKS to connect to a Private Azure DB
I ran into a roadblock recently where I wanted to be able to conveniently connect to a managed postgres database within Azure that was not running on public subnets. And by conveniently, I mean that I’d rather not have to spin up an ephemeral virtual machine running in the same network and proxy the connection, and I’d like to use a local client (preferably with a GUI). After several web searches, it became evident that Azure does not readily provide much tooling to support this.
Should the database be migrated to public subnets? Ideally not, since it is good practice to host internal infrastructure in restricted subnets.
How do others handle this?#
With GCP, connecting to a private db instance from any machine can be achieved with cloud-sql-proxy. This works by proxying requests from your machine to the SQL database instance in the cloud, while the authentication is handled by GCP’s IAM.
So what about Azure? Is there any solution that is as elegant as cloud-sql-proxy?
Similar to what AWS has recommended, perhaps a bastion is the way forward?
Azure has a fully-managed service called Azure Bastion that provides secure access to virtual machines that do not have public IPs. This looks interesting, but unfortunately it costs money and requires an additional virtual machine.
Because this adds cost (and complexity), it does not seem like a desirable option in its current state. If it provided a more seamless connection to the database, it would be more appealing.
SOCKS is a protocol that enables a way to proxy connections by exchanging network packets between the client and the server. There are many implementations and many readily available container images that can run a SOCKS server.
It’s possible to use this sort of proxy to connect to a private DB, but is it any simpler than using a virtual machine as a jumphost? It wasn’t until I stumbled upon kubectl-plugin-socks5-proxy that I was convinced that using SOCKS could be made simple.
So how does it work? By installing the kubectl plugin and then running
kubectl socks5-proxy, a SOCKS proxy server is spun up in a pod and then opens up port-forwarding session using kubectl.
As you can see below, this k8s plugin is wrapped up nicely:
$ kubectl socks5-proxy using: namespace=default using: port=1080 using: name=davegallant-proxy using: image=serjs/go-socks5-proxy Creating SOCKS5 Proxy (Pod)... pod/davegallant-proxy created
With the above proxy connection open, it is possible to access both the DNS and private IPs accessible within the k8s cluster. In this case, I am able to access the private database, since there is network connectivity between the k8s cluster and the database.
Caveats and Conclusion#
The above outlined solution makes some assumptions:
- there is a k8s cluster
- the k8s cluster has network connectivity to the desired private database
If these stars align, than this solution might work as a stopgap for accessing a private Azure DB (and I’m assuming this could work similarly on AWS).
It would be nice if Azure provided tooling similar to cloud-sql-proxy, so that using private databases would be more of a convenient experience.
One other thing to note is that some clients (such as dbeaver) do not provide DNS resolution over SOCKS. So in this case, you won’t be able to use DNS as if you were inside the cluster, but instead have to rely on knowing private ip addresses.