Teams error 504 Gateway Timeout: 504 - gateway timeout

A service in the delivery chain waited on another and gave up. The Teams status table lists 504 as "Gateway Timeout," retry Yes with exponential backoff, and the rate-limit guide includes 504 among the codes that "must also be retried." For your outbound sends this is a transient platform-side timeout: the request may or may not have been applied, so idempotency matters - a retried send after an ambiguous 504 is how duplicate messages happen.

Keep the mirror image in mind too: when your endpoint is the slow party on inbound delivery, the channel-side symptom is the 502 bot-returned-an-error surface and, for invoke-style interactions, the "Unable to reach the app" client message. A 504 in your own reverse-proxy logs in front of the bot means your app server, not Microsoft, is timing out.

Ambiguity is the defining property here: unlike a clean 429 or 503, a 504 means the operation's outcome is unknown. Design for at-least-once delivery with deduplication rather than exactly-once assumptions, and log both the attempt and the retry decision so duplicates can be traced when they do slip through.

HTTP/1.1 504 Gateway Timeout

{
  "error": {
    "code": "GatewayTimeout",
    "message": "Gateway Timeout."
  }
}

Same machinery as 503: durable queue, centralized retry with backoff and jitter, alerting on rates. Add duplicate suppression for user-visible messages so an ambiguous timeout never posts twice. Keep handler latency low enough that the inbound mirror case cannot occur - acknowledge first, work later.