#!/bin/sh
# ClinicPro production diagnostics — run ON THE HOST as root (or a user with docker access).
# Collects every piece of evidence needed to pin down why the service goes down
# after hours of uptime (OOM vs full disk vs Redis growth). See docs/ops-restarts.md §7.
#
# Usage: sh collect-diagnostics.sh [container-name-or-id]
set -u
C="${1:-$(docker ps --format '{{.Names}}' | grep -i -m1 clinic || true)}"

echo "=== 1. OOM evidence (kernel, last 48h) ==="
journalctl -k --since "48 hours ago" 2>/dev/null | grep -i -E "oom|killed process" | tail -20 \
  || dmesg 2>/dev/null | grep -i -E "oom|killed process" | tail -20

echo ""
echo "=== 2. Memory / swap now ==="
free -h

echo ""
echo "=== 3. Disk ==="
df -h | head -10
du -sh /var/lib/docker/containers/*/*-json.log 2>/dev/null | sort -h | tail -5

echo ""
echo "=== 4. Container state ==="
docker ps -a --format 'table {{.Names}}\t{{.Status}}'
if [ -n "$C" ]; then
    docker inspect --format \
      'OOMKilled={{.State.OOMKilled}} ExitCode={{.State.ExitCode}} RestartCount={{.RestartCount}} StartedAt={{.State.StartedAt}} FinishedAt={{.State.FinishedAt}}' "$C"
fi

echo ""
echo "=== 5. Live memory per container ==="
docker stats --no-stream

echo ""
echo "=== 6. php-fpm inside app container ==="
if [ -n "$C" ]; then
    docker exec "$C" sh -c 'php -i 2>/dev/null | grep -E "^memory_limit"; ps -o rss,args 2>/dev/null | grep -E "php-fpm|messenger" | grep -v grep'
fi

echo ""
echo "=== 7. Redis memory ==="
R="$(docker ps --format '{{.Names}}' | grep -i -m1 redis || true)"
if [ -n "$R" ]; then
    docker exec "$R" redis-cli info memory 2>/dev/null | grep -E "used_memory_human|maxmemory_human|maxmemory_policy"
    docker exec "$R" redis-cli info server 2>/dev/null | grep -E "uptime_in_seconds|uptime_in_days"
    docker exec "$R" redis-cli xlen messages 2>/dev/null && echo "(messages stream length above)"
    echo "--- last redis logs:"
    docker logs --tail 15 "$R" 2>&1
fi

echo ""
echo "=== 8. Docker log driver config ==="
docker info --format '{{.LoggingDriver}}' 2>/dev/null
cat /etc/docker/daemon.json 2>/dev/null || echo "(no daemon.json — log rotation NOT configured)"

echo ""
echo "=== done — paste this whole output back for diagnosis ==="