#!/bin/bash
# Monitor I2P router health endpoint during live test.
# Polls every 30 seconds, writes timestamped JSON to a log file.
# Usage: bash tools/monitor_router.sh [duration_minutes] [interval_seconds]

DURATION_MIN=${1:-60}
INTERVAL_SEC=${2:-30}
LOG_FILE="$(dirname "$0")/router_monitor.log"
HEALTH_URL="http://localhost:9701"

TOTAL_POLLS=$(( DURATION_MIN * 60 / INTERVAL_SEC ))

echo "=== I2P Router Monitor ===" | tee "$LOG_FILE"
echo "Started: $(date -Iseconds)" | tee -a "$LOG_FILE"
echo "Duration: ${DURATION_MIN}min, Interval: ${INTERVAL_SEC}s, Polls: ${TOTAL_POLLS}" | tee -a "$LOG_FILE"
echo "---" | tee -a "$LOG_FILE"

for i in $(seq 1 $TOTAL_POLLS); do
    TS=$(date -Iseconds)
    RESPONSE=$(curl -s --max-time 5 "$HEALTH_URL" 2>&1)
    EXIT_CODE=$?

    if [ $EXIT_CODE -ne 0 ]; then
        echo "[$TS] poll=$i CURL_FAILED exit=$EXIT_CODE" | tee -a "$LOG_FILE"
    else
        # Extract key fields for the summary line
        STATE=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('state','?'))" 2>/dev/null)
        UPTIME=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('uptime_seconds',0))" 2>/dev/null)
        PEERS=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('peer_count',0))" 2>/dev/null)
        ATTEMPTED=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('peers_attempted',0))" 2>/dev/null)
        SUCCESSFUL=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('peers_successful',0))" 2>/dev/null)
        FAILED=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('peers_failed',0))" 2>/dev/null)
        NETDB=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('netdb_routerinfos',0))" 2>/dev/null)
        NETDB_DISK=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('netdb_disk',0))" 2>/dev/null)
        RSS=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('resources',{}).get('rss_mb',0))" 2>/dev/null)
        CPU=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('resources',{}).get('cpu_percent','?'))" 2>/dev/null)
        I2NP_RX=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('i2np_messages_received',0))" 2>/dev/null)
        I2NP_TX=$(echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('i2np_messages_sent',0))" 2>/dev/null)

        echo "[$TS] poll=$i state=$STATE uptime=${UPTIME}s peers=$PEERS attempted=$ATTEMPTED ok=$SUCCESSFUL fail=$FAILED netdb=$NETDB disk=$NETDB_DISK rss=${RSS}MB cpu=${CPU}% i2np_rx=$I2NP_RX i2np_tx=$I2NP_TX" | tee -a "$LOG_FILE"

        # Write full JSON every 5th poll for detailed analysis
        if [ $(( i % 5 )) -eq 0 ]; then
            echo "  FULL_JSON: $RESPONSE" >> "$LOG_FILE"
        fi
    fi

    # Check if container is still running
    if ! podman ps --filter name=i2p-router-test --format "{{.Status}}" | grep -q "Up"; then
        echo "[$TS] CONTAINER STOPPED — aborting monitor" | tee -a "$LOG_FILE"
        echo "Last container logs:" | tee -a "$LOG_FILE"
        podman logs --tail 20 i2p-router-test 2>&1 | tee -a "$LOG_FILE"
        break
    fi

    sleep $INTERVAL_SEC
done

echo "---" | tee -a "$LOG_FILE"
echo "Finished: $(date -Iseconds)" | tee -a "$LOG_FILE"
echo "Final health:" | tee -a "$LOG_FILE"
curl -s --max-time 5 "$HEALTH_URL" 2>&1 | python3 -m json.tool 2>&1 | tee -a "$LOG_FILE"