Problem
When a TCP connect event hasn't been processed yet, L7 events queue in pendingL7Events (capped at 500 in registry.go). If the TCP event never arrives (lost from perf buffer), those L7 events are silently dropped after retry.
No metric tracks:
- How many events are pending
- How many events were dropped due to missing TCP connection
- How many events were dropped due to the 500 cap
Fix
- Add a
node_agent_pending_l7_events gauge
- Add a
node_agent_l7_events_dropped_total counter with reason label (no_connection, cap_exceeded, retry_exhausted)
Impact
Medium — silent data loss with no operator visibility.
Problem
When a TCP connect event hasn't been processed yet, L7 events queue in
pendingL7Events(capped at 500 in registry.go). If the TCP event never arrives (lost from perf buffer), those L7 events are silently dropped after retry.No metric tracks:
Fix
node_agent_pending_l7_eventsgaugenode_agent_l7_events_dropped_totalcounter with reason label (no_connection,cap_exceeded,retry_exhausted)Impact
Medium — silent data loss with no operator visibility.