Files
deepseek-harness/docs
Yichen Jiang 2425fa4f84 feat(mcp-client): liveness health pings on established connections
Status downgrades were driven by the transport close signal alone. A stdio
child's death closes the transport, but Streamable HTTP holds no connection
between requests, so a dead server kept showing connected until the next
actual call. An established generation now sends a protocol ping every
healthCheckIntervalMs (default 30s, the same value bounding each answer);
a failed or timed-out ping closes the generation into the existing
reconnect path and its diagnostic rides the reconnecting status.
The manager passes healthCheckIntervalMs through per server entry.
2026-08-31 20:03:35 +08:00
..