feat(profiling/tool): add command hijack launcher

installpydep.sh

@@ -0,0 +1,3 @@
+#! /usr/bin/env bash
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+pip install --no-index --find-links=${SCRIPT_DIR}/offline_packages -r requirements.txt

module/profiling/tool/cmd_hijack_launcher/cmd_hijack_launcher.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+import os
+import sys
+import shutil
+import subprocess
+import datetime
+
+def main():
+    script_path = os.path.realpath(__file__)
+    script_dir = os.path.dirname(script_path)
+    config_path = os.path.join(script_dir, "config.yaml")
+
+    cmd = os.path.basename(sys.argv[0])
+    cmd_args = sys.argv[1:]
+
+    # 清除自身环境变量
+    env_path = os.environ.get("PATH", "")
+    paths = []
+    for path in env_path.split(os.path.pathsep):
+        if os.path.realpath(path) != script_dir:
+            paths.append(path)
+    env_path_clean = os.path.pathsep.join(paths)
+
+    # 查找被劫持的原始命令
+    hijacked_cmd = shutil.which(cmd, path=env_path_clean)
+    if not hijacked_cmd:
+        print(f"Error: can't find hijacked command \"{hijacked_cmd}\" in system PATH", file=sys.stderr)
+        sys.exit(127)
+
+    # 打印基础信息
+    pid = os.getpid()
+    print(f"Pid: {pid}")
+    print(f"Hijacked cmd: {cmd}")
+    print(f"Current path: {os.getcwd()}")
+
+    # 解析配置文件
+    launched_cmds = []
+    if os.path.isfile(config_path):
+        try:
+            import yaml
+            with open(config_path, "r", encoding="utf-8") as f:
+                config = yaml.load(f, Loader=yaml.SafeLoader)
+                if config and isinstance(config, dict):
+                    launched_cmds = config.get("launch", [])
+        except Exception as e:
+            print(f"Error: failed to load config.yaml. (code: {e.errno}. {e.lower()})", file=sys.stderr)
+            exit()
+
+    # 构建命令格式化map
+    now = datetime.datetime.now()
+    timestamp = str(int(now.timestamp())).zfill(10)
+    datetime_fmt_chars = "YymdHMS"
+    fmt = {"pid": str(pid), "cmd": cmd, "timestamp": timestamp}
+    for ch in datetime_fmt_chars:
+        fmt[ch] = now.strftime(f"%{ch}")
+
+    # 发射命令
+    for i, lcmd in enumerate(launched_cmds):
+        if not isinstance(lcmd, str) or not lcmd.strip():
+            continue
+
+        try:
+            lcmd = lcmd.format_map(fmt)
+            lcmd_name = os.path.splitext(os.path.basename(lcmd.split()[0]))[0]
+            log_file = f"launched-{now.strftime(f"%y%m%d%H%M%S")}-{pid}-{i}-{lcmd_name}-screen.log"
+            with open(log_file, "w", encoding="utf-8") as f:
+                # TODO(resserops): 是否支持shell解释，通过yaml配置决策
+                subprocess.Popen(lcmd, stdout=f, stderr=subprocess.STDOUT, shell=True, close_fds=True)
+                print(f"Cmd '{lcmd_name}' successfully launched")
+        except Exception as e:
+            print(f"Error: failed to launch command '{lcmd}' (code: {e.errno}. {e.lower()})", file=sys.stderr)
+
+    # 执行被劫持的原始命令
+    print()
+    os.execv(hijacked_cmd, [hijacked_cmd] + cmd_args)
+
+if __name__ == "__main__":
+    main()

module/profiling/tool/cmd_hijack_launcher/config.yaml

@@ -0,0 +1,10 @@
+# launch (list(str)): 发射命令配置，每个字符串必须是可执行的shell命令，支持{}格式化
+#   {pid}:  命令pid
+#   {cmd}:  命令名称
+#   {timestamp}: 10位时间戳
+#   {Y}, {y}, {m}, {d}, {H}, {M}, {S}: 调起命令的日期时间
+launch:
+  - "while kill -0 {pid} 2>/dev/null; do ps -p {pid} -o pid,%cpu,%mem,stat,command; sleep 1; done"
+
+# screen_log_path (str): 发射命令屏幕输出日志保存路径，支持{}格式化
+screen_log_path: "."

module/profiling/tool/cmd_hijack_launcher/hijack.py

@@ -0,0 +1,174 @@
+#!/usr/bin/env python3
+import argparse
+import os
+import sys
+import shutil
+import filecmp
+import subprocess
+
+def remove_entry(entry):
+    if os.path.isdir(entry) and not os.path.islink(entry):
+        shutil.rmtree(entry)
+    else:
+        os.remove(entry)
+
+def main():
+    wrapper_name = "cmd_hijack_launcher.py"
+    setup_script_name = "setup.sh"
+    critical_files = (wrapper_name, setup_script_name, "config.yaml")
+
+    # 参数解析
+    program = argparse.ArgumentParser(description="deploy symlinks for command hijacking to the specified prefix directory")
+    program.add_argument("-p", "--prefix", default=os.path.expanduser(os.path.join("~", ".local", "bin", os.path.splitext(wrapper_name)[0])), help=f"installation prefix directory")
+    program.add_argument("-c", "--clean", action="store_true", help=f"clean up obsolete hijack symlinks in the prefix directory that are not specified in the current command list")
+    program.add_argument("-f", "--force", action="store_true", help=f"force deployment by overwriting existing files, symlinks, or directories if conflicts occur")
+    program.add_argument("commands", nargs="*", help="commands to hijack. must be executable in the system PATH")
+    args = program.parse_args()
+
+    # 打印基础信息
+    script_dir = os.path.dirname(os.path.realpath(__file__))
+    prefix_dir = os.path.abspath(args.prefix)
+
+    print(f"Script dir: {script_dir}")
+    print(f"Prefix dir: {prefix_dir}\n")
+
+    # 创建prefix（安装路径）
+    if os.path.exists(prefix_dir) or os.path.islink(prefix_dir):
+        if not os.path.isdir(prefix_dir):
+            if args.force:
+                try:
+                    os.remove(prefix_dir)
+                    os.makedirs(prefix_dir, exist_ok=True)
+                except OSError as e:
+                    print(f"Error: failed to force replace prefix directory '{prefix_dir}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+                    sys.exit(1)
+            else:
+                print(f"Error: prefix path exists but is not a directory '{prefix_dir}'. use -f to overwrite", file=sys.stderr)
+                sys.exit(1)
+    else:
+        try:
+            os.makedirs(prefix_dir, exist_ok=True)
+        except OSError as e:
+            print(f"Error: cannot create directory '{prefix_dir}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+            sys.exit(1)
+
+    # 复制关键脚本到prefix
+    def copy_to_prefix(f):
+        src = os.path.join(script_dir, f)
+        # 检查src存在
+        if not os.path.exists(src):
+            print(f"Error: critical file '{f}' not found, hijack failed", file=sys.stderr)
+            sys.exit(1)
+
+        dst = os.path.join(prefix_dir, f)
+        if os.path.exists(dst) or os.path.islink(dst):
+            if os.path.isfile(dst) and filecmp.cmp(src, dst):
+                # 如果已存在文件相同，直接跳过
+                return
+            if args.force:
+                try:
+                    remove_entry(dst)
+                except OSError as e:
+                    print(f"Error: failed to force replace critical file '{f}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+                    return
+            else:
+                print(f"Warning: critical file '{f}' already exists. use -f to overwrite", file=sys.stderr)
+                return
+
+        try:
+            shutil.copy2(src, dst) # 使用copy2保留原始文件权限
+        except OSError as e:
+            print(f"Error: failed to copy critical file. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+            return
+
+    for f in critical_files:
+        copy_to_prefix(f)
+
+    # 清理未在commands中的软链接
+    exists_cmds = []
+    link_target = os.path.join(".", wrapper_name)
+
+    try:
+        for entry in os.scandir(prefix_dir):
+            if entry.is_symlink() and entry.name not in args.commands and os.readlink(entry.path) == link_target:
+                if args.clean:
+                    try:
+                        os.remove(entry.path)
+                        print(f"Command '{entry.name}' unhijacked")
+                    except OSError:
+                        print(f"Error: failed to remove symlink for '{entry.name}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+                        continue
+                else:
+                    exists_cmds.append(entry.name)
+    except OSError as e:
+        print(f"Error: failed to scan prefix directory for cleanup. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+
+    # 设置软链接
+    succ_cmds = []
+    for cmd in args.commands:
+        if cmd == wrapper_name:
+            print(f"Warning: skipping self-hijack of '{cmd}'", file=sys.stderr)
+            continue
+
+        if cmd == "python" or cmd == "python3":
+            print(f"Warning: skipping hijack command '{cmd}' to prevent infinite recursion loop", file=sys.stderr)
+            continue
+
+        if not shutil.which(cmd):
+            print(f"Warning: command '{cmd}' not found in system PATH. skipping hijack", file=sys.stderr)
+            continue
+
+        try:
+            # 不允许劫持shell内建命令和关键字
+            res = subprocess.run(["bash", "-c", f"type -t {cmd}"], stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True)
+            cmd_type = res.stdout.strip()
+
+            if cmd_type == "keyword":
+                print(f"Warning: skipping hijack command '{cmd}' because it is a shell keyword", file=sys.stderr)
+                continue
+            if cmd_type == "builtin":
+                print(f"Warning: skipping hijack command '{cmd}' because it is a shell builtin command", file=sys.stderr)
+                continue
+        except OSError as e:
+            print(f"Error: failed to check command type for '{cmd}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+
+        link_name = os.path.join(prefix_dir, cmd)
+        if os.path.islink(link_name) and os.readlink(link_name) == link_target:
+            # 已经存在且正确的链接可以忽略
+            exists_cmds.append(cmd)
+            continue
+
+        if os.path.exists(link_name) or os.path.islink(link_name):
+            # 不是链接，或者是链接但并未链接到目标
+            if args.force:
+                try:
+                    remove_entry(link_name)
+                except OSError as e:
+                    print(f"Error: failed to force replace hijack link '{cmd}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+                    continue
+            elif os.path.islink(link_name):
+                print(f"Error: hijack link '{f}' already exists. use -f to overwrite '{cmd}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+                continue
+
+        # 建立软链接
+        try:
+            os.symlink(link_target, link_name)
+            succ_cmds.append(cmd)
+        except OSError as e:
+            print(f"Error: failed to create symlink for '{cmd}'. (code: {e.errno}, {e.lower()})", file=sys.stderr)
+
+    exists_cmds.sort()
+    for cmd in exists_cmds:
+        print(f"Command '{cmd}' already hijacked")
+
+    succ_cmds.sort()
+    for cmd in succ_cmds:
+        print(f"Command '{cmd}' successfully hijacked")
+
+    if len(succ_cmds) == 0 and len(exists_cmds) == 0:
+        print("No commands hijacked")
+    else:
+        print(f"To enable and activate the hijacked commands, please run: 'source {os.path.join(prefix_dir, setup_script_name)}'")
+
+if __name__ == "__main__":
+    main()

module/profiling/tool/cmd_hijack_launcher/setup.sh

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+if [ "${BASH_SOURCE[0]}" = "$0" ]; then
+    echo "Error: The script must be sourced. please run: 'source $0'"
+    exit 1
+fi
+
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+export PATH="${SCRIPT_DIR}:${PATH}"

module/profiling/tool/monitor/monitor.cpp

@@ -392,8 +392,8 @@ void PlotCpu() {
         }
     }
 
+    py::scoped_interpreter guard{}; // 必须放在try-catch块外侧，否则无法捕获py::error_already_set异常
     try {
-        py::scoped_interpreter guard{}; // 绘图函数仅执行单次
         auto plt{py::module_::import("matplotlib.pyplot")};
         auto subplots{plt.attr("subplots")(
             valid_idxs.size(), 1, py::arg("sharex") = true,
@@ -440,10 +440,7 @@ void PlotCpu() {
         plt.attr("xlabel")("Time (s)");
         plt.attr("tight_layout")();
         plt.attr("show")();
-    } catch (const py::error_already_set &e) {
-        std::cerr << "Error: " << e.what() << std::endl;
-        exit(1);
-    }
+    } catch (const py::error_already_set &e) { std::cerr << "Error: " << e.what() << std::endl; }
 }
 
 void Report() {

packpydep.sh

@@ -0,0 +1,4 @@
+#! /usr/bin/env bash
+SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+mkdir -p ${SCRIPT_DIR}/pydep
+pip download -d ${SCRIPT_DIR}/pydep -r ${SCRIPT_DIR}/requirements.txt