Reference RCA-663F3DD00948AFE9
OpenAI API endpoint
Measurement point: Stealth NYC, 32 Avenue of the Americas.
Published investigation. Updates run automatically.
Recorded endpoint latency history is not available for this event. Missing data remains unknown.
Choose a measurement type to see the closest of the available sampled observations to the selected chart point. Each probe is separate and may follow a different path.
The endpoint measurements and collection limitations remain below. Missing data cannot establish a path or a fault.
Compare one selected trace from before the event with one during it. Values are individual replies, not a statistical comparison.
No same-address, same-method pair is available among the selected replay samples. This does not mean no other matching traces were recorded.
RTTs include the return path. Different responders, missing replies and independent probe flows limit what a change can localize. MTR snapshots remain separate in the replay.
Latency anomaly. Times shown are UTC.
Recorded recovery describes this event, not current reachability.
Endpoint round-trip time was elevated during part of the event. The available path observations do not identify where the increase occurred.
Recording counts describe collected measurements, not application availability or packet loss. Exact packet paths and the return path remain unverified.
RTT is round trip time; p95 describes the upper range of successful recordings. Endpoint measurements differ from individual trace replies and are not an uptime percentage.
Event and source timestamps are UTC. The latency chart uses its selected timezone. Page refresh times are separate from measurement timestamps.
Comparisons use matching destinations before and during the event. ICMP traces use echo probes. TCP/443 traces send SYN probes to port 443 only; they do not complete a TCP handshake or perform a TLS or HTTP exchange.
The Stealth gNMI interface verifies the active route-table choice and maps its primary next hop to the Stealth logical interface, configured attachment and, where applicable, the known LAG member set at each snapshot. Stable Stealth replies, together with mapped interface evidence showing no corresponding errors or sustained capacity pressure, reduce the likelihood of a Stealth forwarding issue. An 80% load warning warrants investigation; it does not prove congestion. Lower interval averages can conceal bursts or queue pressure. Optical readings assess the physical link, not congestion.
Probe reply handling can differ from packet forwarding. A slow intermediate reply that disappears at later hops does not locate a forwarding problem. A destination that does not respond to a trace is not thereby shown to be unavailable.
Stealth does not use ECMP in this monitored scope. The route snapshot does not identify which physical LAG member carried a particular independent probe, attest line-card forwarding at that packet instant, prove the exact application packet flow or observe the return path. RTT includes the return path, which may differ. No remote reverse traceroute is available. Tunnels and silent routers may hide parts of the path. Address registration alone does not establish the forwarding operator or responsibility.
References: ICMP reply handling / RFC 1812 · Round trip measurements / RFC 2681 · Full methodology
