You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
SSE now survives non-200 responses (qfg-41nh.9). The SDK owns SSE reconnection instead of
delegating it to the eventsource package, which treats any non-200 response as terminal — a
single 502 from the edge during a deploy used to kill real-time updates for the process lifetime
(silently degrading to the 60s fallback poll). Every SSE failure now tears down and recreates the
connection with jittered exponential backoff (500ms doubling to a 30s cap, retrying forever),
matching sdk-go. This also replaces the previous constant ~3s reconnect delay on network failures.
The stream stays pinned to the primary stream URL; there is no rotation.
Immediate fetch when the fallback poller engages after SSE loss (qfg-41nh.9). When the Layer 2
HTTP poller engages after the 120s disconnect grace window, it now fetches immediately instead of
waiting a further full poll interval — first data after SSE loss arrives at ~120s (matching
sdk-go/java/net) instead of ~180s.