#!/usr/bin/env bash
# Sprint 27: snapshot стейджа каждые 5 минут на протяжении soak'a.
#
# Запуск:
#   tests/load/monitor-soak.sh /tmp/soak-metrics.csv > /dev/null 2>&1 &
#   (затем k6 run soak-4h.js — оба идут параллельно)
#
# Записывает CSV с колонками:
#   ts,api_mem_mb,api_cpu_pct,pg_connections,disk_free_gb,p95_db_ms
#
# Источники:
#   - docker stats food-market-stage-api-1 (на 192.168.1.190)
#   - psql pg_stat_activity (через docker exec postgres)
#   - df -h (на 192.168.1.190)
#   - /metrics → histogram_quantile для p95 DB

set -uo pipefail

OUT="${1:-/tmp/soak-metrics.csv}"
INTERVAL="${INTERVAL:-300}"   # 5 минут default
DURATION="${DURATION:-14400}" # 4 часа default
STAGE_HOST="${STAGE_HOST:-192.168.1.190}"

# Header
if [ ! -f "$OUT" ]; then
  echo "ts,api_mem_mb,api_cpu_pct,pg_connections,disk_free_gb,me_p95_ms,products_p95_ms" > "$OUT"
fi

end=$(($(date +%s) + DURATION))
while [ "$(date +%s)" -lt $end ]; do
  TS=$(date -Iseconds)

  # API container stats (mem MB, CPU %)
  STATS=$(ssh -o ConnectTimeout=5 nns@$STAGE_HOST \
    "docker stats --no-stream --format '{{.MemUsage}} {{.CPUPerc}}' food-market-stage-api-1" 2>/dev/null || echo "0MiB / 0MiB 0%")
  MEM=$(echo "$STATS" | awk '{print $1}' | sed 's/MiB//;s/GiB//')
  # Если в GiB — конвертим в MiB (×1024)
  if echo "$STATS" | awk '{print $1}' | grep -q GiB; then
    MEM=$(python3 -c "print(int(float('$MEM')*1024))")
  fi
  CPU=$(echo "$STATS" | awk '{print $4}' | sed 's/%//')

  # PG connections
  PG_CONN=$(ssh -o ConnectTimeout=5 nns@$STAGE_HOST \
    "docker exec food-market-stage-postgres-1 psql -U food_market -d food_market -tA -c 'SELECT count(*) FROM pg_stat_activity'" 2>/dev/null || echo "0")

  # Disk
  DISK=$(ssh -o ConnectTimeout=5 nns@$STAGE_HOST "df -BG --output=avail / | tail -1 | tr -d 'G '" 2>/dev/null || echo "0")

  # P95 latency (rough — из /metrics histogram)
  METRICS=$(curl -fsS --max-time 5 https://test.admin.food-market.kz/metrics 2>/dev/null || echo "")
  # Парсим через python (histogram_quantile сложно в shell)
  P95_ME=$(echo "$METRICS" | python3 -c "
import sys
buckets = []
total = 0
for line in sys.stdin:
    if 'http_request_duration_seconds_bucket' in line and 'action=\"GetMe\"' in line:
        try:
            le = float(line.split('le=\"')[1].split('\"')[0])
            val = float(line.rsplit(' ', 1)[1])
            buckets.append((le, val))
        except: pass
buckets.sort()
if buckets:
    total = buckets[-1][1]
    p95_target = 0.95 * total
    for le, v in buckets:
        if v >= p95_target:
            print(int(le * 1000)); break
    else: print(0)
else: print(0)
" 2>/dev/null || echo 0)

  P95_PRODUCTS=$(echo "$METRICS" | python3 -c "
import sys
buckets = []
for line in sys.stdin:
    if 'http_request_duration_seconds_bucket' in line and 'action=\"List\"' in line and 'controller=\"Products\"' in line:
        try:
            le = float(line.split('le=\"')[1].split('\"')[0])
            val = float(line.rsplit(' ', 1)[1])
            buckets.append((le, val))
        except: pass
buckets.sort()
if buckets:
    total = buckets[-1][1]
    p95_target = 0.95 * total
    for le, v in buckets:
        if v >= p95_target:
            print(int(le * 1000)); break
    else: print(0)
else: print(0)
" 2>/dev/null || echo 0)

  echo "$TS,$MEM,$CPU,$PG_CONN,$DISK,$P95_ME,$P95_PRODUCTS" >> "$OUT"
  echo "[$TS] mem=${MEM}MiB cpu=${CPU}% pg_conn=$PG_CONN disk=${DISK}G p95_me=${P95_ME}ms p95_prod=${P95_PRODUCTS}ms"

  sleep "$INTERVAL"
done