rdp-proxy/scripts/ops/test-docker-disk-guard.sh

#!/usr/bin/env sh
set -eu

WARN_PERCENT="${WARN_PERCENT:-85}"
CLEANUP_PERCENT="${CLEANUP_PERCENT:-85}"
CRITICAL_PERCENT="${CRITICAL_PERCENT:-95}"
MIN_TMP_AGE_MINUTES="${MIN_TMP_AGE_MINUTES:-360}"
MOUNT_PATH="${MOUNT_PATH:-/}"
STATUS_DIR="${STATUS_DIR:-/tmp/rap-web-admin/html/downloads/ops}"
LOG_DIR="${LOG_DIR:-/tmp/rap-ops}"
WEBHOOK_URL="${WEBHOOK_URL:-}"

mkdir -p "$STATUS_DIR" "$LOG_DIR"

started_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
log_file="$LOG_DIR/test-docker-disk-guard.log"
status_file="$STATUS_DIR/test-docker-disk-guard-status.json"

log() {
  printf '%s %s\n' "$(date -u +%Y-%m-%dT%H:%M:%SZ)" "$*" >>"$log_file"
}

json_escape() {
  printf '%s' "$1" | sed 's/\\/\\\\/g; s/"/\\"/g'
}

disk_used_percent() {
  df -P "$MOUNT_PATH" | awk 'NR==2 { gsub("%", "", $5); print $5 }'
}

disk_avail_human() {
  df -hP "$MOUNT_PATH" | awk 'NR==2 { print $4 }'
}

disk_used_human() {
  df -hP "$MOUNT_PATH" | awk 'NR==2 { print $3 }'
}

disk_size_human() {
  df -hP "$MOUNT_PATH" | awk 'NR==2 { print $2 }'
}

docker_df_summary() {
  if command -v docker >/dev/null 2>&1; then
    docker system df 2>&1 | tr '\n' '; ' | sed 's/"/\\"/g'
  else
    printf 'docker cli not found'
  fi
}

cleanup_safe() {
  log "cleanup started: docker build cache and old RAP tmp artifacts"
  if command -v docker >/dev/null 2>&1; then
    docker builder prune -af >>"$log_file" 2>&1 || true
    docker image prune -f >>"$log_file" 2>&1 || true
  fi
  find /tmp -maxdepth 1 \( \
    -name 'rap-android-build-*' -o \
    -name 'rap-agent-build*' -o \
    -name 'rap-backend-build*' -o \
    -name 'rap-c*-build*' -o \
    -name 'rap-node-agent-*' -o \
    -name 'rap-*vpnfarm*' -o \
    -name 'rap-build-*' \
  \) -mmin +"$MIN_TMP_AGE_MINUTES" -exec rm -rf {} + >>"$log_file" 2>&1 || true
  sync || true
  log "cleanup finished"
}

expansion_hint() {
  root_source="$(df -P "$MOUNT_PATH" | awk 'NR==2 { print $1 }')"
  vg_free=""
  root_lv_bytes=""
  parent_part_bytes=""
  if command -v vgs >/dev/null 2>&1; then
    vg_free="$(vgs --noheadings --units g -o vg_free 2>/dev/null | awk '{print $1}' | tr '\n' ' ' | sed 's/^ *//;s/ *$//')"
  fi
  if command -v lsblk >/dev/null 2>&1; then
    root_lv_bytes="$(lsblk -b -n -o TYPE,SIZE,MOUNTPOINTS 2>/dev/null | awk '$1 == "lvm" && $3 == "/" { print $2; exit }')"
    parent_part_bytes="$(lsblk -b -n -o TYPE,SIZE 2>/dev/null | awk '$1 == "part" { last=$2 } $1 == "lvm" { print last; exit }')"
  fi
  if printf '%s' "$root_source" | grep -q '^/dev/mapper/'; then
    if [ -n "$root_lv_bytes" ] && [ -n "$parent_part_bytes" ] && [ "$parent_part_bytes" -gt "$((root_lv_bytes + 1073741824))" ]; then
      printf 'LVM root detected on %s. Backing partition is larger than root LV, so expansion is likely available. Run with sudo: sudo lvextend -r -l +100%%FREE %s' "$root_source" "$root_source"
      return
    fi
    if [ -n "$vg_free" ] && ! printf '%s' "$vg_free" | grep -Eq '(^| )0(\.00)?g( |$)'; then
      printf 'LVM root detected on %s. If approved, extend inside existing VG: sudo lvextend -r -l +100%%FREE %s' "$root_source" "$root_source"
    else
      printf 'LVM root detected on %s. No obvious free VG space. Expand VM disk, then run pvresize on the PV and lvextend -r for root LV.' "$root_source"
    fi
  else
    printf 'Root filesystem is %s. Expand underlying disk/volume, then grow filesystem according to host partition layout.' "$root_source"
  fi
}

notify() {
  level="$1"
  message="$2"
  if [ -n "$WEBHOOK_URL" ] && command -v curl >/dev/null 2>&1; then
    payload="$(printf '{"level":"%s","message":"%s","host":"%s","observed_at":"%s"}' \
      "$(json_escape "$level")" \
      "$(json_escape "$message")" \
      "$(json_escape "$(hostname)")" \
      "$(date -u +%Y-%m-%dT%H:%M:%SZ)")"
    curl -fsS -m 5 -H 'Content-Type: application/json' -d "$payload" "$WEBHOOK_URL" >>"$log_file" 2>&1 || true
  fi
}

before_percent="$(disk_used_percent)"
action="none"
level="ok"

if [ "$before_percent" -ge "$CLEANUP_PERCENT" ]; then
  action="cleanup_safe"
  cleanup_safe
fi

after_percent="$(disk_used_percent)"
if [ "$after_percent" -ge "$CRITICAL_PERCENT" ]; then
  level="critical"
elif [ "$after_percent" -ge "$WARN_PERCENT" ]; then
  level="warning"
fi

hint="$(expansion_hint)"
summary="$(docker_df_summary)"
finished_at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"

cat >"$status_file.tmp" <<EOF_STATUS
{
  "schema_version": "rap.ops.test_docker_disk_guard.v1",
  "host": "$(json_escape "$(hostname)")",
  "mount_path": "$(json_escape "$MOUNT_PATH")",
  "started_at": "$(json_escape "$started_at")",
  "finished_at": "$(json_escape "$finished_at")",
  "level": "$(json_escape "$level")",
  "action": "$(json_escape "$action")",
  "thresholds": {
    "warn_percent": $WARN_PERCENT,
    "cleanup_percent": $CLEANUP_PERCENT,
    "critical_percent": $CRITICAL_PERCENT
  },
  "disk": {
    "before_used_percent": $before_percent,
    "after_used_percent": $after_percent,
    "size": "$(json_escape "$(disk_size_human)")",
    "used": "$(json_escape "$(disk_used_human)")",
    "available": "$(json_escape "$(disk_avail_human)")"
  },
  "docker_system_df": "$(json_escape "$summary")",
  "expansion_hint": "$(json_escape "$hint")",
  "log_file": "$(json_escape "$log_file")"
}
EOF_STATUS
mv "$status_file.tmp" "$status_file"

message="test-docker disk ${level}: ${after_percent}% used after action=${action}. ${hint}"
log "$message"
if [ "$level" != "ok" ]; then
  notify "$level" "$message"
fi

if [ "$level" = "critical" ]; then
  exit 2
fi
exit 0