Api-Key

2.2.12.1

Proxy:

• Add status checker
• Add ability to validate requests via api-key
• Add ability to call backend via api-key
• Allow bakend api keys to be configured outside Hostn so Hostn_api_key can be backed via keyvault.

RequestAPI:

• Add status checker

APIM Policy:

• Version changed to v2.1.0
• Backend field names changed:
  • priority -> priorityGroup
  • ModelType -> label
  • api-key -> auth
  • LimitConcurrency, BufferResponse, and Timeout are now read as limitConcurrency, bufferResponse, and timeout
• Authentication is now explicit:
  • In v2.0.1, an empty api-key meant "use Managed Identity".
  • In v2.1.0, auth: "MI" means Managed Identity, auth: "<key>" means send api-key: <key>, and auth: "" means send no auth header.
• Backend URLs are now composed from url plus optional path:
  • In v2.0.1, the policy appended /openai when building backendUrl.
  • In v2.1.0, the policy combines url and path during normalization and uses the result as-is.
  • If you relied on the automatic /openai append, add "path": "/openai" or include /openai directly in url.
• Missing backend settings now get defaults:
  • If limitConcurrency is omitted, the policy sets it to off.
  • If bufferResponse is omitted, the policy sets it to true.
  • If timeout is omitted, the policy sets it to 10 seconds.
• Retry budget handling bug fix:
  • v2.0.1 allowed the request path to keep going while RetryCount >= 0.
  • v2.1.0 only retries while RetryCount > 0.
  • If you previously used retryCount: 1 the policy retried twice. For the same behaviour increase it to 2.
• PTU skip-on-context-window now keys off label:
  • In v2.0.1, the context-window-exceeded path skipped PTU backends when ModelType == "PTU".
  • In v2.1.0, it skips them when label == "PTU".