Mesh status
All regions operational
Checked continuously from outside the mesh, from three vantage points per region. Last update —.
Incident history
Published within 72 hours, whatever the cause.
2026-06-14
Resolved · 47 minutes
eu-central
Replay over-delivery after cursor buffer rotation
A change to how retention buffers rotate caused resumed subscribers in eu-central to receive up to 40 duplicated messages immediately after a replay frame. Ordering held throughout — the duplicates were contiguous and carried repeated cursors — so SDK clients acknowledging by cursor were unaffected, and clients acknowledging by application-level ID were not. The change was reverted 47 minutes after the first report.
This one was ours, and it is precisely the failure mode our delivery contract exists to prevent. The full postmortem is in the namespace log for every affected team, with the exact cursor ranges involved.
2026-05-02
Resolved · 1 hour 12 minutes
ap-southeast
Elevated publish latency, ap-southeast
p99 publish acknowledgement in ap-southeast rose from 12 ms to 340 ms after an upstream transit provider began dropping packets between two availability zones. Fan-out latency to already-attached subscribers was unaffected, and no publish was rejected — they were slow, not lost. We shifted the region's publish plane to a second transit path and have since made that path primary.
2026-03-19
Resolved · 22 minutes
us-east
Subscriber token minting returned 503 in us-east
The token minting endpoint in us-east returned 503 for approximately 22 minutes following a certificate rotation that raced its own config reload. Existing subscriber connections were unaffected — only new attaches failed, and clients holding a valid token rode through it. Rotation and reload are now serialized behind a single lease.
Incidents are published within 72 hours of resolution regardless of cause or severity, including the ones we caused. Subscribe to updates by emailing [email protected].