Idle app is significantly more epensive than mentioned in cost calculator

[syonip]

I've deployed the demo and used it once or twice with a few queries, but the costs seem to be quite steep, even with zero usage.

Expected monthly cost is ~180$ with few queries, while the pricing calculator mentions ~100$ for 30K semantic queries.

Am I doing something wrong? Why is this costing so much with no use?
Thanks.

[zedhaque]

Here you go. Item 4 from the URL below describes the issue:

https://github.com/Azure-Samples/azure-search-openai-demo/blob/main/docs/deploy_lowcost.md

You will need to delete your existing AI Search and recreate it under the Free tier, but know there are some limitations.

Cheers

[syonip]

Thanks @zedhaque.
I am more interested in getting a realistic estimation of the costs of a non-free solution in order to reflect it to the client.
What could explain the difference between the calculator's estimation of "Azure Container Apps" (fore example) at 0$ Monthly, while my deployment is ~60$ monthly?
One of these must be wrong.

[zedhaque]

Maybe you have a higher number of minimum replicas running for your container apps.

The default in this repo is "1" for container apps, so it's not $0 per month.

https://github.com/Azure-Samples/azure-search-openai-demo/blob/main/infra/core/host/container-app.bicep

@description('The minimum number of replicas to run. Must be at least 1.')
param containerMinReplicas int = 1

[syonip]

I don't know enough about the costs of replicas, I'm just trying to understand the difference between what the price calculator shows. Am I reading this wrong?

This is from the link in docs: https://azure.com/e/e3490de2372a4f9b909b0d032560e41b

[pamelafox]

I am seeing similar, I'm investigating. Thanks for filing!

[pamelafox]

See my PR:
#2440

I discovered the extra cost was due to two reasons: higher than default CPU and memory (which we can't turn down, as our app needs the higher memory), and a scaling rule of 1 replica minimum.
I've adjusted the scaling rule to 0 replicas minimum, which should reduce the costs to near 0, since it will only need to scale when you're actually using it.

[syonip]

Thanks @pamelafox this is good news, how do I update my deployment to reflect this change?

[pamelafox]

Add this line to main.bicep:
containerMinReplicas: 0

Right under the containerMemory line