#azure-openai 6 articles

• "Your data stays in the region you chose" — for Azure OpenAI the deployment TYPE decides where prompts are processed, not the regiongotchas / azure-functionsA GlobalStandard deployment in an EU region may process prompts in any Azure region; only Data Zone (EU Data Boundary) or regional Standard keep processing in the zone or geography — write data-location claims from the…
• A "public mode" flag that disables the token check opens every endpoint in the Function App, not just the chatgotchas / azure-functionsOne Function App serves a signed-in SPFx app and an anonymous website chatbot; the website instance switches off the token check, and every other HTTP function in the same app (usage stats, storage read/write, file…
• Azure OpenAI — a pinned model+version in your deploy script is a time bombgotchas / azure-functions"Deprecating" blocks NEW deployments well before retirement; resolve the newest GA version at deploy time, and match the deployment name to the model family
• Azure OpenAI 429 on a single request is a TPM ceiling, not a busy service — measure size, not frequencygotchas / azure-functionsOne request whose prompt exceeds the deployment's TPM allowance can never fit the window, so retrying is futile — a sub-second rejection means "over the ceiling", not "busy"; separate the two with a small/large request…
• RAG finds nothing that SharePoint Search finds — the reasoning model spent the whole token cap thinkinggotchas / searchGPT-5/o-series count hidden reasoning in max_completion_tokens; a small cap on the query-generator call returns empty text (HTTP 200), and a raw-question fallback ANDs every word in KQL
• Retry only fast failures — a gateway error after 230 seconds is a timeout, not a hiccupgotchas / azure-functionsThe 502/503/504 a client rightly retries for transient failures also comes back when a request ran out of time (Azure Load Balancer returns 502 after 230 s), and retrying that multiplies the wait (3 × ~230 s) and the…