Monitor agent health via periodic heartbeats and detect failures using timeout-based detection.

Commands

Send heartbeat to peers

pilotctl --json publish "registry-hostname" "heartbeat:$SWARM_NAME" \
  --data "{\"agent\":\"$AGENT_ID\",\"timestamp\":\"$(date -u +%s)\",\"status\":\"alive\"}"

Detect failed agents

TIMEOUT=30
CURRENT_TIME=$(date -u +%s)
FAILED_AGENTS=$(pilotctl --json inbox \
  | jq --arg now "$CURRENT_TIME" --arg timeout "$TIMEOUT" \
    '[.messages[] | select(.topic == "heartbeat:'$SWARM_NAME'") | {agent: .payload.agent, last_seen: .payload.timestamp}]
    | group_by(.agent)
    | map(select(($now | tonumber) - (map(.last_seen) | max) > ($timeout | tonumber)))
    | .[].agent')

Verify failure with direct ping

for agent in $FAILED_AGENTS; do
  AGENT_ADDR=$(pilotctl --json peers | jq -r '.[] | select(.node_id == "'$agent'") | .address')
  PING_RESULT=$(pilotctl --json ping "$AGENT_ADDR" --count 3 --timeout 2s)
  LOSS=$(echo "$PING_RESULT" | jq -r '.packet_loss_pct')
  if [ "$LOSS" = "100" ]; then
    echo "Agent $agent CONFIRMED DOWN"
  fi
done

Workflow Example

Health monitor for worker pool with automatic task redistribution:

#!/bin/bash
SWARM_NAME="worker-pool"
HEARTBEAT_INTERVAL=5
FAILURE_TIMEOUT=15
REGISTRY_HOST="registry.example.com"
# Background: Send own heartbeats
(
  while true; do
    pilotctl --json publish "$REGISTRY_HOST" "heartbeat:$SWARM_NAME" \
      --data "{\"agent\":\"$AGENT_ID\",\"timestamp\":\"$(date -u +%s)\"}"
    sleep $HEARTBEAT_INTERVAL
  done
) &
# Monitor peer heartbeats
while true; do
  CURRENT_TIME=$(date -u +%s)
  # Detect timeouts and trigger recovery
  # ...
  sleep $HEARTBEAT_INTERVAL
done

Dependencies

Requires pilot-protocol skill, jq, and bc.