[BrentOzar]

I haven't yet seen a production problem that Docker containers solve better than IaaS VMs (AWS EC2, Azure VMs, Google Compute Engine.)

I've seen a lot of development problems that containers help solve, like quickly building out Development Edition VMs for continuous integration testing. Since Dev Edition is free, rapidly building/destroying containers makes sense. However...

> dynamically adding Always On Availability Group read-only replicas

Readable AG replicas require SQL Server Enterprise Edition, which is $7k USD per core. That pricing tends to dwarf the hardware it's running on, so it usually makes more sense to run bare metal on these if it's a long-term scale requirement. (I have indeed done temporary scale-out reads on VMs, it's just usually for things like Black Friday workloads for short periods of time.)

(And then at the end, I went to add a disclaimer about what I do and why I'm answering, and then I went, "Oh, wow, it's James Anderson asking this." OK, so we both know each other, but here's my answer anyway for the rest of the HN community.) So source: Microsoft SQL Server consultant.

[james_anderson]

Haha, thanks for your answer Brent. I think you're right about VMs. SQL Server on docker in prod is kind of a thought experiment of mine at the moment.

I'd love to hear from anyone who is even just thinking of doing this.

> dynamically adding Always On Availability Group read-only replicas

Yeah I doubt scaling SQL Server out dynamically will ever make monetary sense, but it might be cool to just get it working.

I'm thinking of creating a Helm chart for K8s that allows you to dynamically scale the SQL Server pod up and down. I think I'll need to use some undocumented DBCC commands (or call the API) to freeze IO, so I can replicate the data files to the new secondary as it spins up.

Horizontally scaling SQL Server would be cool. Maybe if the hosts were licensed so we could run as many containers as we wanted, it could make sense...