90-day availability per service.
Routing-pool health and median latency per region.
Streaming time-to-first-token rose to ~340 ms for approximately 22 minutes on a single upstream pool. Auto-failover redirected affected traffic to APAC · Tokyo within 90 seconds; remaining latency was upstream-side. No request failures recorded.
14:18 – 14:40 UTC · 22 min · 0 failures
A bad config push to one US-West replica raised the 5xx rate to 0.8% for 4 minutes
before auto-rollback triggered. Affected requests returned idc-failover: true
and were retried successfully against the US East pool.
09:12 – 09:16 UTC · 4 min · 318 retries
Capacity upgrade in EU · Frankfurt. Traffic was drained to neighboring regions for
roughly 11 minutes. No user-visible failures; some EU customers saw a temporary
idc-region header value of eu-ams.
03:00 – 03:11 UTC · 11 min · 0 failures