The overall goal of this PR is to simplify the networking architecture without loss of any security measures.
This PR incorporates several changes that were unfortunately difficult to separate into different PRs. They include:
Support for both Developer and StandardV2 APIM tiers. The Developer tier is cheaper but requires more time to deploy. The StandardV2 tier is more expensive but reduces the time to deploy. Pricing for each tier can be found here. Development teams will find the StandardV2 tier easier to work with due to the faster deployment time. Default behavior has been (and will remain to be) to deploy the Developer tier unless a user overrides this decision (as documented in the deployment guide).
Simplifies the overall deployment of multiple vnets (requiring network peering) down to a single vnet
Simplifies the deployment of graphrag in AKS be reducing the number of pods down to 1 (previous the accelerator would deploy graphrag as two separate AKS pods). Autoscaling via AKS is enabled.
Pushes more Azure deployment logic out of the main deploy.sh script and into a bicep implementation. This will allow users to leverage Azure in deploying resources in parallel and improve the investigation of any deployment errors.
The overall goal of this PR is to simplify the networking architecture without loss of any security measures.
This PR incorporates several changes that were unfortunately difficult to separate into different PRs. They include:
Developer
andStandardV2
APIM tiers. TheDeveloper
tier is cheaper but requires more time to deploy. TheStandardV2
tier is more expensive but reduces the time to deploy. Pricing for each tier can be found here. Development teams will find theStandardV2
tier easier to work with due to the faster deployment time. Default behavior has been (and will remain to be) to deploy theDeveloper
tier unless a user overrides this decision (as documented in the deployment guide).deploy.sh
script and into a bicep implementation. This will allow users to leverage Azure in deploying resources in parallel and improve the investigation of any deployment errors.