https://agentoutages.com/ Filter sourced provider incident post-mortems, compound multi-step SLA models, and production failover patterns across model APIs, vector databases, and edge runtimes. en-au Fri, 06 Nov 2026 00:00:00 GMT https://agentoutages.com/incidents/#anthropic-529-vs-429 https://agentoutages.com/incidents/#anthropic-529-vs-429 Tue, 18 Feb 2025 00:00:00 GMT Anthropic API documentation distinguishes regional compute saturation (HTTP 529) from workspace token bucket exhaustion (HTTP 429). Actionable takeaway: Trigger immediate cross-region or cross-model fallback on HTTP 529 rather than burning 60 seconds in local retry sleep. (Source: Anthropic API Error Reference) HTTP-529 Failover SLA-Math https://agentoutages.com/incidents/#openai-k8s-dns-cascade https://agentoutages.com/incidents/#openai-k8s-dns-cascade Thu, 12 Dec 2024 00:00:00 GMT OpenAI documented how cluster-wide control-plane load degraded internal DNS discovery for roughly 4.5 hours, stalling unhedged synchronous agent loops. Actionable takeaway: Enforce strict per-turn connect/read timeouts (under 12s for routing turns) and fail over to a warm secondary provider. (Source: OpenAI Public Status History) Failover Retry-Storms Status-Telemetry https://agentoutages.com/incidents/#github-webhook-replay-idempotency https://agentoutages.com/incidents/#github-webhook-replay-idempotency Thu, 15 Aug 2024 00:00:00 GMT Delayed webhook queues replaying after upstream database recovery caused stateless coding agents without idempotency ledgers to execute duplicate write actions. Actionable takeaway: Hash (event_id + tool_name + canonical_args) in durable storage before executing any external mutation tool. (Source: GitHub Status Incident Archive) Idempotency Webhook-Replay