Katyusha's blog

1
#!/usr/bin/env bash
2
# 性能开销基准测试：对每个场景，分别在「未加载工具(基线)」与「加载工具」两态下
3
# 跑等量固定负载，测量：
4
#   - 工具自身 CPU 占用%        -> 对应评分"CPU 开销"
5
#   - 工具峰值常驻内存 RSS       -> 对应评分"内存开销"
6
#   - 负载完成耗时增幅(slowdown%) -> 对应评分"时延/吞吐影响"
7
#
8
# 用法： bash scripts/benchmark.sh [scenario|all] [输出文件]
9
#   scenario ∈ cpu|io|mem|lock|syscall|all（默认 all）
10
set -uo pipefail
11

12
SC="${1:-all}"
13
OUT="${2:-}"
14
BIN="$(cd "$(dirname "$0")/.." && pwd)/bin/ebpf-rca"
15
CLK="$(getconf CLK_TCK 2>/dev/null || echo 100)"
16

17
[ -x "$BIN" ] || { echo "未找到二进制 $BIN，请先 make build" >&2; exit 1; }
18

19
# 各场景的固定工作量负载（固定 ops/数据量，保证两态可比）
20
workload_cmd() {
21
  case "$1" in
22
  cpu)     echo "stress-ng --cpu $(nproc) --cpu-method matrixprod --cpu-ops 8000" ;;
23
  lock)    echo "stress-ng --mutex 8 --mutex-ops 300000" ;;
24
  mem)     echo "stress-ng --vm 4 --vm-bytes 75% --vm-ops 3000" ;;
25
  io)      echo "fio --name=bench --filename=./fio-bench.img --size=1G --rw=randrw --rwmixread=70 --bs=4k --iodepth=64 --numjobs=4 --group_reporting" ;;
26
  syscall) echo "dd if=/dev/zero of=/dev/null bs=1 count=30000000" ;;
27
  esac
28
}
29

30
# 计时执行一条命令，打印墙钟秒数
31
time_cmd() {
32
  local s e
33
  s=$(date +%s.%N)
34
  eval "$1" >/dev/null 2>&1 || true
35
  e=$(date +%s.%N)
36
  awk "BEGIN{print $e-$s}"
37
}
38

39
bench_one() {
40
  local sc="$1" cmd base withv tpid tstart tend st rest u s hwm cpu slow rss
41
  cmd="$(workload_cmd "$sc")"
42
  [ -n "$cmd" ] || { echo "跳过未知场景 $sc" >&2; return; }
43
  echo "[bench] $sc：测基线..." >&2
44
  base=$(time_cmd "$cmd")
45

46
  echo "[bench] $sc：加载工具后再测..." >&2
47
  sudo "$BIN" --scenario "$sc" --duration 3600s >/dev/null 2>&1 &
48
  sleep 1.5
49
  tpid=$(pgrep -n -x ebpf-rca || true)
50
  tstart=$(date +%s.%N)
51
  withv=$(time_cmd "$cmd")
52
  tend=$(date +%s.%N)
53

54
  cpu=0; rss=0
55
  if [ -n "$tpid" ] && [ -r "/proc/$tpid/stat" ]; then
56
    st=$(cat "/proc/$tpid/stat")
57
    rest="${st##*) }"          # 跳过 comm，剩余首字段为 state(field3)
58
    u=$(echo "$rest" | awk '{print $12}')  # utime = field14
59
    s=$(echo "$rest" | awk '{print $13}')  # stime = field15
60
    hwm=$(awk '/VmHWM/{print $2}' "/proc/$tpid/status" 2>/dev/null || echo 0)
61
    cpu=$(awk "BEGIN{el=$tend-$tstart; if(el<=0)el=1; print (($u+$s)/$CLK)/el*100}")
62
    rss=$(awk "BEGIN{print ${hwm:-0}/1024}")
63
  fi
64
  [ -n "$tpid" ] && sudo kill "$tpid" 2>/dev/null || true
65
  [ "$sc" = "io" ] && rm -f ./fio-bench.img
66

67
  slow=$(awk "BEGIN{if($base<=0)print 0; else print ($withv-$base)/$base*100}")
68
  printf "| %-7s | %8.2f | %8.2f | %10.2f | %7.2f | %9.1f |\n" \
69
    "$sc" "$base" "$withv" "$slow" "$cpu" "$rss"
70
}
71

72
scenarios=()
73
if [ "$SC" = "all" ]; then scenarios=(cpu io mem lock syscall); else scenarios=("$SC"); fi
74

75
{
76
  echo "# ebpf-rca 性能开销基准 ($(date '+%Y-%m-%d %H:%M'))"
77
  echo
78
  echo "| 场景 | 基线(s) | 加载后(s) | 负载变慢% | 工具CPU% | 工具RSS(MB) |"
79
  echo "|------|---------|-----------|-----------|----------|-------------|"
80
  for sc in "${scenarios[@]}"; do bench_one "$sc"; done
81
  echo
82
  echo "> 负载变慢% = (加载后耗时-基线耗时)/基线 ×100，越小越好；工具CPU%/RSS 为 ebpf-rca 自身开销。"
83
} | { [ -n "$OUT" ] && tee "$OUT" || cat; }