test.py

# 运行目录：在项目根目录下运行
import subprocess
import time
import os
import io
import sys
import atexit
import paramiko
import matplotlib.pyplot as plt
import numpy as np
import json
import threading
import logging
import sys
import shutil
import argparse
import csv
import shlex
import re
import signal

# Set LD_LIBRARY_PATH for YashanDB client
lib_path = os.path.expanduser("~/yashandb-client/lib")
if "LD_LIBRARY_PATH" in os.environ:
    os.environ["LD_LIBRARY_PATH"] = lib_path + ":" + os.environ["LD_LIBRARY_PATH"]
else:
    os.environ["LD_LIBRARY_PATH"] = lib_path

logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger("paramiko").setLevel(logging.WARNING)
sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding='utf-8', line_buffering=True)


def kill_server():
    os.makedirs(os.path.dirname(output), exist_ok=True)
    with open(output, "w", encoding="utf-8") as outfile:
        subprocess.run(["pkill", "-9", "-x", "run"], stdout=outfile, stderr=outfile, check=False)
        subprocess.run("rm ./output.txt", stdout=outfile, stderr=outfile, shell=True)
    time.sleep(1)

def build():
    os.makedirs(os.path.dirname(output), exist_ok=True)
    with open(output, "w", encoding="utf-8") as outfile:
        build_type = "Release" if RouterOnly else "Debug (project default)"
        logging.info(f"Rebuilding MP-Router binary with {build_type} configuration.")
        build_dir = os.path.join(workspace, "build")
        shutil.rmtree(build_dir, ignore_errors=True)
        os.makedirs(build_dir, exist_ok=True)
        cmake_cmd = ["cmake"]
        if RouterOnly:
            cmake_cmd.append("-DCMAKE_BUILD_TYPE=Release")
        cmake_cmd.append("..")
        subprocess.run(cmake_cmd, cwd=build_dir, stdout=outfile,
                       stderr=subprocess.STDOUT, check=True)
        subprocess.run(["make", "-j8"], cwd=build_dir, stdout=outfile,
                       stderr=subprocess.STDOUT, check=True)
    time.sleep(1)

original_config_h_text = None
current_built_mlp_mode = None

def set_mlp_prediction(enabled):
    global original_config_h_text
    config_path = os.path.join(workspace, "config.h")
    with open(config_path, "r", encoding="utf-8") as f:
        text = f.read()
    if original_config_h_text is None:
        original_config_h_text = text
    value = "1" if int(enabled) else "0"
    new_text, count = re.subn(
        r"^#define\s+MLP_PREDICTION\s+\d+(\s*//.*)?$",
        lambda m: f"#define MLP_PREDICTION {value}{m.group(1) or ''}",
        text,
        count=1,
        flags=re.MULTILINE,
    )
    if count != 1:
        raise RuntimeError("Unable to find MLP_PREDICTION definition in config.h")
    if new_text != text:
        logging.info(f"Setting MLP_PREDICTION={value} in config.h")
        with open(config_path, "w", encoding="utf-8") as f:
            f.write(new_text)

def restore_config_h():
    if original_config_h_text is None or not RestoreConfigAfterRun:
        return
    config_path = os.path.join(workspace, "config.h")
    with open(config_path, "w", encoding="utf-8") as f:
        f.write(original_config_h_text)
    logging.info("Restored original config.h")

def ensure_build_for_mlp(mlp_enabled):
    global current_built_mlp_mode
    if current_built_mlp_mode == int(mlp_enabled):
        return
    if not RebuildForMLP and current_built_mlp_mode is not None:
        raise RuntimeError("EnableMLP contains multiple values but RebuildForMLP is disabled.")
    kill_server()
    set_mlp_prediction(mlp_enabled)
    build()
    current_built_mlp_mode = int(mlp_enabled)
    logging.info(f"Built MP-Router with MLP_PREDICTION={current_built_mlp_mode}")

def run_cmd(cmd, check=True):
    logging.info(f"Executing: {cmd}")
    result = subprocess.run(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, encoding='utf-8')
    if check and result.returncode != 0:
        logging.error(f"Command failed: {cmd}\nError: {result.stderr}")
        raise Exception(f"Command failed: {cmd}")
    return result

def wait_for_process_stopped(process, timeout):
    logging.info(f"Waiting for process {process.pid} to reach the post-load stop point...")
    deadline = time.time() + timeout
    status_path = f"/proc/{process.pid}/status"
    while time.time() < deadline:
        return_code = process.poll()
        if return_code is not None:
            raise RuntimeError(
                f"Process exited with code {return_code} before reaching the post-load stop point."
            )
        try:
            with open(status_path, "r", encoding="utf-8") as status_file:
                for line in status_file:
                    if line.startswith("State:"):
                        state = line.split()[1]
                        if state in ("T", "t"):
                            logging.info(f"Process {process.pid} is stopped after loading data.")
                            return
                        break
        except FileNotFoundError:
            pass
        time.sleep(0.1)
    raise TimeoutError(
        f"Process {process.pid} did not reach the post-load stop point within {timeout}s."
    )

def run_remote_cmd(cmd, check=True, max_retries=3, allowed_exit_codes=[0], display_cmd=None, host=None):
    remote_host = host or kwr_report_ip
    logging.info(f"Executing Remote on {remote_host}: {display_cmd or cmd}")
    last_exception = None
    
    for attempt in range(1, max_retries + 1):
        ssh = paramiko.SSHClient()
        ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        try:
            # 增加 timeout 防止连接卡死
            ssh.connect(remote_host, username="root", password=kwr_ip_password, timeout=30)
            # 开启 keepalive 防止长时间无数据传输导致连接断开 (特别是在 rsync 过程中)
            ssh.get_transport().set_keepalive(60)
            
            stdin, stdout, stderr = ssh.exec_command(cmd)
            exit_status = stdout.channel.recv_exit_status()
            out = stdout.read().decode('utf-8')
            err = stderr.read().decode('utf-8')
            
            if check and exit_status not in allowed_exit_codes:
                logging.error(f"Remote command failed: {cmd}\nExit Code: {exit_status}\nError: {err}")
                raise Exception(f"Remote command failed: {cmd}")
            
            # 如果是 rsync 返回 24，打印一个警告但视为成功
            if exit_status == 24:
                logging.warning(f"Rsync warning (code 24): Some files vanished during transfer. This is usually safe to ignore.")

            class Result:
                def __init__(self, stdout, stderr, returncode):
                    self.stdout = stdout
                    self.stderr = stderr
                    self.returncode = returncode
            
            return Result(out, err, exit_status)
        except Exception as e:
            logging.warning(f"Remote execution failed (Attempt {attempt}/{max_retries}): {e}")
            last_exception = e
            if attempt < max_retries:
                time.sleep(remote_retry_sleep_seconds)
        finally:
            ssh.close()
            
    logging.error(f"All {max_retries} attempts failed for command: {cmd}")
    raise last_exception

def run_remote_cmd_streaming(cmd, check=True, allowed_exit_codes=(0, 24)):
    logging.info(f"Executing Remote with live progress on {kwr_report_ip}: {cmd}")
    ssh = paramiko.SSHClient()
    ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
    try:
        ssh.connect(kwr_report_ip, username="root", password=kwr_ip_password, timeout=30)
        ssh.get_transport().set_keepalive(60)
        _, stdout, stderr = ssh.exec_command(cmd, get_pty=True)
        while not stdout.channel.exit_status_ready():
            if stdout.channel.recv_ready():
                text = stdout.channel.recv(4096).decode("utf-8", errors="replace")
                print(text, end="", flush=True)
            if stdout.channel.recv_stderr_ready():
                text = stdout.channel.recv_stderr(4096).decode("utf-8", errors="replace")
                print(text, end="", file=sys.stderr, flush=True)
            time.sleep(0.05)
        while stdout.channel.recv_ready():
            print(stdout.channel.recv(4096).decode("utf-8", errors="replace"), end="", flush=True)
        while stdout.channel.recv_stderr_ready():
            print(stdout.channel.recv_stderr(4096).decode("utf-8", errors="replace"), end="", file=sys.stderr, flush=True)
        exit_status = stdout.channel.recv_exit_status()
        if check and exit_status not in allowed_exit_codes:
            raise Exception(f"Remote command failed with exit code {exit_status}: {cmd}")
        return exit_status
    finally:
        ssh.close()

def get_remote_dir_size(path):
    # 获取目录大小 (KB)
    res = run_remote_cmd(f"du -s {shlex.quote(path)} | awk '{{print $1}}'", check=False)
    if res.returncode != 0:
        return -1
    try:
        return int(res.stdout.strip())
    except:
        return -1

def check_remote_exists(path):
    res = run_remote_cmd(f"test -e {shlex.quote(path)}", check=False)
    return res.returncode == 0

def wait_for_remote_path(path, timeout=30, interval=2):
    logging.info(f"Waiting for remote path to be available: {path}")
    start_time = time.time()
    while time.time() - start_time < timeout:
        if check_remote_exists(path):
            return True
        time.sleep(interval)
    return False

def parse_conninfo(conninfo):
    parsed = {}
    for item in shlex.split(conninfo):
        if "=" not in item:
            continue
        key, value = item.split("=", 1)
        parsed[key] = value
    return parsed

def is_postgres_like_db():
    return int(DBType) == 0

def sync_remote_servers_after_case():
    if not is_postgres_like_db():
        return

    hosts = []
    seen = set()
    for conninfo in db_ready_probe_conninfos:
        host = parse_conninfo(conninfo).get("host")
        if host and host not in seen:
            hosts.append(host)
            seen.add(host)

    if not hosts:
        hosts = [kwr_report_ip]

    for host in hosts:
        try:
            result = run_remote_cmd("sync", check=False, max_retries=3, host=host, display_cmd="sync")
        except Exception:
            logging.warning(
                "Post-case sync failed on %s after retries; continuing with saved results.",
                host, exc_info=True,
            )
        else:
            if result.returncode != 0:
                logging.warning(
                    "Post-case sync on %s returned %s; continuing with saved results.",
                    host, result.returncode,
                )

def configured_db_conninfos():
    if int(DBType) == 1:
        return yashan_db_conninfos
    if is_postgres_like_db():
        return db_ready_probe_conninfos
    return []

def build_db_connection_args(node_count=None):
    conninfos = configured_db_conninfos()
    if not conninfos:
        return ""
    if node_count is None:
        node_count = len(conninfos)
    if node_count < 1 or node_count > len(conninfos):
        raise ValueError(
            f"compute node count {node_count} is outside the configured connection range "
            f"[1, {len(conninfos)}] for DB_TYPE={DBType}"
        )
    return "".join(
        f" --db-connection {shlex.quote(conninfo)}"
        for conninfo in conninfos[:node_count]
    )

def drop_public_tables():
    if not shutil.which("psql"):
        raise RuntimeError("psql is required to drop tables after each config group.")

    drop_tables_sql = """
DO $$
DECLARE
    r record;
BEGIN
    FOR r IN
        SELECT schemaname, tablename
        FROM pg_tables
        WHERE schemaname = 'public'
    LOOP
        EXECUTE format('DROP TABLE IF EXISTS %I.%I CASCADE', r.schemaname, r.tablename);
    END LOOP;
END $$;
"""

    for conninfo in db_ready_probe_conninfos:
        parsed = parse_conninfo(conninfo)
        env = os.environ.copy()
        if parsed.get("password"):
            env["PGPASSWORD"] = parsed["password"]

        logging.info(f"Dropping public tables on {mask_conninfo_password(conninfo)}")
        res = subprocess.run(
            build_local_sql_exec_command(conninfo, drop_tables_sql),
            stdout=subprocess.PIPE,
            stderr=subprocess.PIPE,
            text=True,
            env=env,
            timeout=db_drop_tables_timeout_seconds,
        )
        if res.returncode != 0:
            logging.error(
                f"Failed to drop public tables on {mask_conninfo_password(conninfo)}: "
                f"stdout={res.stdout.strip()}, stderr={res.stderr.strip()}"
            )
            raise RuntimeError("Failed to drop public tables after config group.")

def truncate_kwr_tables():
    if not shutil.which("psql"):
        raise RuntimeError("psql is required to truncate KWR tables.")

    truncate_kwr_sql = """
TRUNCATE TABLE perf.kwr_last_sql_stmt_all;
TRUNCATE TABLE perf.kwr_stmt_list;
"""

    for conninfo in db_ready_probe_conninfos:
        parsed = parse_conninfo(conninfo)
        env = os.environ.copy()
        if parsed.get("password"):
            env["PGPASSWORD"] = parsed["password"]

        logging.info(f"Truncating KWR tables on {mask_conninfo_password(conninfo)}")
        res = subprocess.run(
            build_local_sql_exec_command(conninfo, truncate_kwr_sql),
            stdout=subprocess.PIPE,
            stderr=subprocess.PIPE,
            text=True,
            env=env,
            timeout=db_drop_tables_timeout_seconds,
        )
        if res.returncode != 0:
            logging.error(
                f"Failed to truncate KWR tables on {mask_conninfo_password(conninfo)}: "
                f"stdout={res.stdout.strip()}, stderr={res.stderr.strip()}"
            )
            raise RuntimeError("Failed to truncate KWR tables.")

def build_db_probe_command(conninfo):
    parsed = parse_conninfo(conninfo)
    host = parsed.get("host", "127.0.0.1")
    port = parsed.get("port", "5432")
    user = parsed.get("user", "system")
    password = parsed.get("password", "")
    dbname = parsed.get("dbname", parsed.get("database", workload))
    return (
        "if command -v ksql >/dev/null 2>&1; then client=ksql; "
        "elif command -v psql >/dev/null 2>&1; then client=psql; "
        "else exit 127; fi; "
        f"PGPASSWORD={shlex.quote(password)} timeout 5 \"$client\" "
        f"-h {shlex.quote(host)} -p {shlex.quote(str(port))} "
        f"-U {shlex.quote(user)} -d {shlex.quote(dbname)} "
        "-Atqc 'select 1' >/dev/null 2>&1"
    )

def build_local_pg_isready_command(conninfo):
    parsed = parse_conninfo(conninfo)
    host = parsed.get("host", "127.0.0.1")
    port = parsed.get("port", "5432")
    user = parsed.get("user", "system")
    dbname = parsed.get("dbname", parsed.get("database", workload))
    return [
        "pg_isready",
        "-h", host,
        "-p", str(port),
        "-U", user,
        "-d", dbname,
        "-t", str(db_pg_isready_timeout_seconds),
    ]

def build_local_sql_probe_command(conninfo):
    parsed = parse_conninfo(conninfo)
    host = parsed.get("host", "127.0.0.1")
    port = parsed.get("port", "5432")
    user = parsed.get("user", "system")
    dbname = parsed.get("dbname", parsed.get("database", workload))
    return [
        "psql",
        "-h", host,
        "-p", str(port),
        "-U", user,
        "-d", dbname,
        "-Atqc", "select 1",
    ]

def build_local_sql_exec_command(conninfo, sql):
    parsed = parse_conninfo(conninfo)
    host = parsed.get("host", "127.0.0.1")
    port = parsed.get("port", "5432")
    user = parsed.get("user", "system")
    dbname = parsed.get("dbname", parsed.get("database", workload))
    return [
        "psql",
        "-v", "ON_ERROR_STOP=1",
        "-h", host,
        "-p", str(port),
        "-U", user,
        "-d", dbname,
        "-Atqc", sql,
    ]

def build_tcp_probe_command(conninfo):
    parsed = parse_conninfo(conninfo)
    host = parsed.get("host", "127.0.0.1")
    port = parsed.get("port", "5432")
    return (
        f"timeout {db_tcp_probe_timeout_seconds} "
        f"bash -lc '</dev/tcp/{shlex.quote(host)}/{shlex.quote(str(port))}'"
    )

def mask_conninfo_password(conninfo):
    return re.sub(r"password=[^ ]+", "password=***", conninfo)

def local_database_probe_ready(conninfo, verbose=False):
    parsed = parse_conninfo(conninfo)
    password = parsed.get("password", "")
    env = os.environ.copy()
    if password:
        env["PGPASSWORD"] = password

    if shutil.which("pg_isready"):
        cmd = build_local_pg_isready_command(conninfo)
        if verbose:
            logging.info(f"Local pg_isready probe: {mask_conninfo_password(conninfo)}")
        res = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, env=env)
        if res.returncode != 0:
            logging.info(f"pg_isready not ready yet: rc={res.returncode}, stdout={res.stdout.strip()}, stderr={res.stderr.strip()}")
            return False, True

    if shutil.which("psql"):
        cmd = build_local_sql_probe_command(conninfo)
        if verbose:
            logging.info(f"Local SQL readiness probe: {mask_conninfo_password(conninfo)}")
        try:
            res = subprocess.run(
                cmd,
                stdout=subprocess.PIPE,
                stderr=subprocess.PIPE,
                text=True,
                env=env,
                timeout=db_sql_probe_timeout_seconds,
            )
        except subprocess.TimeoutExpired:
            logging.info(
                f"SQL probe timed out after {db_sql_probe_timeout_seconds}s: "
                f"{mask_conninfo_password(conninfo)}"
            )
            return False, True
        if res.returncode != 0:
            logging.info(f"SQL probe not ready yet: rc={res.returncode}, stdout={res.stdout.strip()}, stderr={res.stderr.strip()}")
            return False, True
        return True, True

    return False, False

def probe_database_ready(verbose=False):
    if not db_ready_probe_conninfos:
        logging.warning("No db_ready_probe_conninfos configured; using CRM status only.")
        return True, True

    if use_local_db_readiness_probe:
        for conninfo in db_ready_probe_conninfos:
            ready, probe_available = local_database_probe_ready(conninfo, verbose=verbose)
            if not probe_available:
                logging.warning("Neither pg_isready nor psql was found locally; falling back to remote TCP readiness probe.")
                break
            if not ready:
                return False, True
        else:
            return True, True

    for conninfo in db_ready_probe_conninfos:
        cmd = build_tcp_probe_command(conninfo)
        res = run_remote_cmd(
            cmd,
            check=False,
            max_retries=1,
            display_cmd=f"db tcp probe: {mask_conninfo_password(conninfo)}",
        )
        if res.returncode != 0:
            return False, True

    if not enable_sql_readiness_probe:
        return True, True

    probe_available = True
    for conninfo in db_ready_probe_conninfos:
        cmd = build_db_probe_command(conninfo)
        res = run_remote_cmd(
            cmd,
            check=False,
            max_retries=1,
            display_cmd=f"db readiness probe: {mask_conninfo_password(conninfo)}",
        )
        if res.returncode == 127:
            probe_available = False
            break
        if res.returncode != 0:
            return False, True
    return True, probe_available

def wait_for_db_start(timeout=3000):
    if not is_postgres_like_db():
        logging.info("Skipping PostgreSQL/KES database readiness probe for non-PostgreSQL DB_TYPE.")
        return

    logging.info("Waiting for database to start...")
    start_time = time.time()
    probe_attempt = 0
    while time.time() - start_time < timeout:
        # 检查数据库资源状态
        res = run_remote_cmd(f"crm resource status {db_resource_name}", check=False)
        # 根据实际 crm 输出调整，通常 running 表示已启动
        status_text = res.stdout + res.stderr
        if res.returncode == 0 and "is NOT running" not in status_text and "not found" not in status_text.lower():
            probe_attempt += 1
            ready, probe_available = probe_database_ready(verbose=False)
            if ready:
                logging.info(f"Database connection probe succeeded ({len(db_ready_probe_conninfos)} endpoints).")
                return
            if not probe_available:
                logging.warning("Neither ksql nor psql was found on remote host; falling back to a short grace wait after CRM start.")
                time.sleep(db_start_probe_fallback_sleep_seconds)
                return
            logging.info("Database resource is running, but SQL connection is not ready yet.")
        time.sleep(db_status_poll_seconds)
    raise Exception("Database failed to start within timeout")

def wait_for_db_stop(timeout=3000):
    if not is_postgres_like_db():
        return

    logging.info("Waiting for database to stop...")
    start_time = time.time()
    while time.time() - start_time < timeout:
        # 检查数据库资源状态
        res = run_remote_cmd(f"crm resource status {db_resource_name}", check=False)
        # 如果不包含 "is running on"，则认为已停止
        status_text = res.stdout + res.stderr
        if "is running on" not in status_text and "Started:" not in status_text:
            logging.info("Database resource appears to be stopped.")
            return
        time.sleep(db_status_poll_seconds)
    raise Exception("Database failed to stop within timeout")

def restart_database_resource(reason="between config groups"):
    if not is_postgres_like_db():
        logging.info(f"Skipping PostgreSQL/KES resource restart for DB_TYPE={DBType}.")
        return

    logging.info(f"Restarting database resource {reason}...")
    run_remote_cmd(f"crm resource stop {db_resource_name}")
    wait_for_db_stop()
    sync_remote_servers_after_case()
    run_remote_cmd(f"crm resource start {db_resource_name}")
    wait_for_db_start()
    logging.info("Database resource restart completed.")

def is_tpcc_workload(workload_name):
    return workload_name in ("tpcc", "tpcc-standard")

def restart_database_before_tpcc_mode(case):
    if UseDataCache or not RestartDBBeforeEachTPCCMode or not is_tpcc_workload(case["workload"]):
        return
    logging.info(
        f"Restarting database before TPC-C mode {case['run_mode']} "
        f"(case {case['case_id']})."
    )
    restart_database_resource("before TPC-C mode")

def reset_db_data(backup_path):
    logging.info(f">>> Resetting Database Data from {backup_path} <<<")
    if backup_path.rstrip("/") == database_data_path.rstrip("/"):
        raise Exception("Data source directory must differ from the active database data directory.")
    
    # Check backup size first
    backup_size = get_remote_dir_size(backup_path)
    logging.info(f"Backup size: {backup_size} KB")
    
    # 修复：空目录大小通常为4KB，原先 <= 0 的判断会导致空目录被当作有效备份
    # 从而导致 rsync --delete 删空数据库。这里设置一个最小阈值，例如 100MB (102400 KB)
    min_backup_size = 100 * 1024 
    if backup_size < min_backup_size:
        error_msg = f"Critical Error: Backup size at {backup_path} is too small ({backup_size} KB). Refusing to restore to avoid data loss."
        logging.error(error_msg)
        raise Exception(error_msg)

    # 1. Stop Database
    run_remote_cmd(f"crm resource stop {db_resource_name}")
    wait_for_db_stop()
    
    # 2. Restore Data (使用 rsync 保证权限和完整性)
    # --delete 确保目标目录中多余的文件被删除，保持与源目录完全一致
    # -a 归档模式，保留权限、所有者等
    logging.info("Restoring data from backup...")
    
    # Check disk space before restore (optional debug)
    run_remote_cmd(f"df -h {shlex.quote(database_data_path)}", check=False)

    source = shlex.quote(backup_path.rstrip("/") + "/")
    destination = shlex.quote(database_data_path)
    run_remote_cmd_streaming(
        f"rsync -a --delete --info=progress2 --human-readable --timeout=3000 {source} {destination}"
    )
    
    # Verify restore
    restored_size = get_remote_dir_size(database_data_path)
    logging.info(f"Restored size: {restored_size} KB")
    
    if abs(restored_size - backup_size) > backup_size * 0.1:
        logging.error(f"CRITICAL: Restore size mismatch! Backup: {backup_size}, Restored: {restored_size}")
        # raise Exception("Restore failed: Size mismatch")

    # Fix permissions just in case (assuming kingbase user)
    run_remote_cmd(f"chown -R kingbase:kingbase {database_data_path}", check=False)
    run_remote_cmd("sync", check=False)

    # 3. Start Database
    run_remote_cmd(f"crm resource start {db_resource_name}")
    
    # 4. Wait for startup
    wait_for_db_start()
    logging.info(">>> Database Reset Complete <<<")

def workload_scale_value(workload_name, account_count=None, warehouse_count=None):
    if workload_name in ("tpcc", "tpcc-standard"):
        return warehouse_count
    return account_count

def workload_scale_arg(workload_name, account_count=None, warehouse_count=None):
    if workload_name in ("tpcc", "tpcc-standard"):
        return f"--warehouse-count {warehouse_count}"
    return f"--account-count {account_count}"

def workload_scale_label(workload_name, account_count=None, warehouse_count=None):
    if workload_name in ("tpcc", "tpcc-standard"):
        return f"wh{warehouse_count}"
    return f"acc{account_count}"

def data_cache_path(workload_name, account_count=None, warehouse_count=None):
    scale_value = workload_scale_value(workload_name, account_count, warehouse_count)
    if workload_name == "smallbank":
        return f"/sharedata/kingbase/{workload_name}_{scale_value}"
    return f"/sharedata/kingbase/{workload_name}_{workload_scale_label(workload_name, account_count, warehouse_count)}"

def data_cache_is_valid(backup_path):
    if not check_remote_exists(backup_path):
        return False
    backup_size = get_remote_dir_size(backup_path)
    min_backup_size = 100 * 1024
    if backup_size < min_backup_size:
        logging.warning(
            f"Ignoring invalid data cache {backup_path}: size is only {backup_size} KB"
        )
        return False
    return True

def ensure_data_cache(case, force_reload=False):
    backup_path = case.get("data_cache_path") or data_cache_path(
        case["workload"], case.get("account_count"), case.get("warehouse_count")
    )
    if force_reload or not data_cache_is_valid(backup_path):
        reason = "--force-reload" if force_reload else "no valid cache was detected"
        logging.info(f"Preparing data cache {backup_path}: {reason}")
        prepare_backup_data(case, backup_path)
    else:
        logging.info(f"Automatically using existing data cache: {backup_path}")
    return backup_path

def list_for_workload(mapping, workload_name, default_values):
    if isinstance(mapping, dict):
        return mapping.get(workload_name, default_values)
    return default_values

def access_patterns_for_workload(workload_name):
    return list_for_workload(WorkloadAccessPatterns, workload_name, AccessPattern)

def default_access_pattern_for_workload(workload_name):
    if isinstance(DefaultAccessPattern, dict):
        return DefaultAccessPattern.get(workload_name, access_patterns_for_workload(workload_name)[0])
    return DefaultAccessPattern

def default_scale_for_workload(workload_name):
    if workload_name in ("tpcc", "tpcc-standard"):
        return None, list_for_workload(WarehouseCount, workload_name, WarehouseCount)[0]
    return list_for_workload(AccountCount, workload_name, AccountCount)[0], None

def scale_values_for_workload(workload_name):
    if workload_name in ("tpcc", "tpcc-standard"):
        for warehouse_count in list_for_workload(WarehouseCount, workload_name, WarehouseCount):
            yield None, warehouse_count
    else:
        account_counts = list_for_workload(AccountCount, workload_name, AccountCount)
        if SweepMode == "axis":
            account_counts = account_counts[:1]
        for account_count in account_counts:
            yield account_count, None

def default_access_config(workload_name):
    access_pattern = default_access_pattern_for_workload(workload_name)
    if access_pattern == 1:
        return {
            "access_pattern": access_pattern,
            "zipfian_theta": DefaultZipfianTheta,
            "zipfian_generator": ZipfianGenerator,
            "hotspot_fraction": None,
            "hotspot_prob": None,
        }
    if access_pattern == 2:
        return {
            "access_pattern": access_pattern,
            "zipfian_theta": None,
            "zipfian_generator": None,
            "hotspot_fraction": DefaultHotspotFraction,
            "hotspot_prob": DefaultHotspotProb,
        }
    return {
        "access_pattern": access_pattern,
        "zipfian_theta": None,
        "zipfian_generator": None,
        "hotspot_fraction": None,
        "hotspot_prob": None,
    }

def access_configs_for_workload(workload_name):
    configs = []
    for access_pattern in access_patterns_for_workload(workload_name):
        if access_pattern == 1:
            for theta in ZipfianTheta:
                configs.append({
                    "access_pattern": access_pattern,
                    "zipfian_theta": theta,
                    "zipfian_generator": ZipfianGenerator,
                    "hotspot_fraction": None,
                    "hotspot_prob": None,
                })
        elif access_pattern == 2:
            for hotspot_fraction in HotspotFraction:
                for hotspot_prob in HotspotProb:
                    configs.append({
                        "access_pattern": access_pattern,
                        "zipfian_theta": None,
                        "zipfian_generator": None,
                        "hotspot_fraction": hotspot_fraction,
                        "hotspot_prob": hotspot_prob,
                    })
        else:
            configs.append({
                "access_pattern": access_pattern,
                "zipfian_theta": None,
                "zipfian_generator": None,
                "hotspot_fraction": None,
                "hotspot_prob": None,
            })
    return configs

def base_case_config(workload_name, account_count=None, warehouse_count=None):
    case = {
        "workload": workload_name,
        "account_count": account_count,
        "warehouse_count": warehouse_count,
        "worker_threads": DefaultWorkerThreads,
        "affinity_txn_ratio": DefaultAffinityTxnRatio,
        "batch_size": DefaultBatchSize,
        "num_bucket": DefaultNumBucket,
        "compute_node_count": DefaultComputeNodeCount,
        "tpcc_partition_warehouses": DefaultTPCCPartitionWarehouses if workload_name in ("tpcc", "tpcc-standard") else 0,
        "long_txn_length": DefaultLongTxnLength if EnableLongTxn else None,
        "long_txn_write_pct": LongTxnWritePct if EnableLongTxn else None,
        "key_page_ratio": DefaultKeyPageMapCapacity,
        "mlp_enabled": DefaultEnableMLP,
        "use_data_cache": UseDataCache,
        "scan_axis": "base",
    }
    case.update(default_access_config(workload_name))
    return case

def dedupe_case_configs(configs):
    deduped = []
    seen = set()
    for case in configs:
        key = (
            case["workload"],
            case.get("account_count"),
            case.get("warehouse_count"),
            case["access_pattern"],
            case.get("zipfian_theta"),
            case.get("zipfian_generator"),
            case.get("hotspot_fraction"),
            case.get("hotspot_prob"),
            case["worker_threads"],
            case["affinity_txn_ratio"],
            case["batch_size"],
            case["num_bucket"],
            case["compute_node_count"],
            case.get("tpcc_partition_warehouses"),
            case.get("long_txn_length"),
            case.get("long_txn_write_pct"),
            case.get("key_page_ratio"),
            case.get("mlp_enabled"),
        )
        if key not in seen:
            seen.add(key)
            deduped.append(case)
    return deduped

def values_except_default(values, default_value):
    return [value for value in values if value != default_value]

def contains_int_value(values, target_value):
    return any(int(value) == int(target_value) for value in values)

def access_config_key(config):
    return (
        config["access_pattern"],
        config.get("zipfian_theta"),
        config.get("zipfian_generator"),
        config.get("hotspot_fraction"),
        config.get("hotspot_prob"),
    )

def build_axis_case_configs(workload_name, account_count=None, warehouse_count=None):
    base = base_case_config(workload_name, account_count, warehouse_count)
    include_baseline_cases = contains_int_value(EnableMLP, DefaultEnableMLP)
    configs = [base] if include_baseline_cases else []

    if include_baseline_cases:
        default_access_key = access_config_key(default_access_config(workload_name))
        for access_config in access_configs_for_workload(workload_name):
            if access_config_key(access_config) == default_access_key:
                continue
            case = dict(base)
            case.update(access_config)
            case["scan_axis"] = "access"
            configs.append(case)

        for compute_node_count in values_except_default(ComputeNodeCounts, DefaultComputeNodeCount):
            case = dict(base)
            case["compute_node_count"] = compute_node_count
            case["scan_axis"] = "compute_node_count"
            configs.append(case)

    if workload_name != "smallbank":
        if include_baseline_cases and workload_name in ("tpcc", "tpcc-standard"):
            for partition_warehouses in values_except_default(TPCCPartitionWarehouse, DefaultTPCCPartitionWarehouses):
                case = dict(base)
                case["tpcc_partition_warehouses"] = partition_warehouses
                case["scan_axis"] = "tpcc_partition_warehouses"
                configs.append(case)
        return dedupe_case_configs(configs)

    if include_baseline_cases:
        for account_count_value in values_except_default(
            list_for_workload(AccountCount, workload_name, AccountCount),
            account_count,
        ):
            case = dict(base)
            case["account_count"] = account_count_value
            case["scan_axis"] = "account_count"
            configs.append(case)

        for worker_threads in values_except_default(WorkerThreadCount, DefaultWorkerThreads):
            case = dict(base)
            case["worker_threads"] = worker_threads
            case["scan_axis"] = "worker_threads"
            configs.append(case)

        for affinity_ratio in values_except_default(AffinityTxnRatio, DefaultAffinityTxnRatio):
            case = dict(base)
            case["affinity_txn_ratio"] = affinity_ratio
            case["scan_axis"] = "affinity_txn_ratio"
            configs.append(case)

        for batch_size in values_except_default(BatchSize, DefaultBatchSize):
            case = dict(base)
            case["batch_size"] = batch_size
            case["scan_axis"] = "batch_size"
            configs.append(case)

        for num_bucket in values_except_default(NumBucket, DefaultNumBucket):
            case = dict(base)
            case["num_bucket"] = num_bucket
            case["scan_axis"] = "num_bucket"
            configs.append(case)

        if EnableLongTxn:
            for long_txn_length in values_except_default(LongTxnSize, DefaultLongTxnLength):
                case = dict(base)
                case["long_txn_length"] = long_txn_length
                case["scan_axis"] = "long_txn_length"
                configs.append(case)

        for key_page_ratio in values_except_default(KeyPageMapCapacity, DefaultKeyPageMapCapacity):
            case = dict(base)
            case["key_page_ratio"] = key_page_ratio
            case["scan_axis"] = "key_page_capacity"
            configs.append(case)

    for mlp_enabled in values_except_default(EnableMLP, DefaultEnableMLP):
        for theta in ZipfianTheta:
            case = dict(base)
            case.update({
                "access_pattern": 1,
                "zipfian_theta": theta,
                "zipfian_generator": ZipfianGenerator,
                "hotspot_fraction": None,
                "hotspot_prob": None,
            })
            case["mlp_enabled"] = mlp_enabled
            case["scan_axis"] = "mlp_zipfian"
            configs.append(case)

    return dedupe_case_configs(configs)

def build_full_case_configs(workload_name, account_count=None, warehouse_count=None):
    configs = []
    long_txn_sizes = LongTxnSize if EnableLongTxn else [None]
    tpcc_partition_values = TPCCPartitionWarehouse if workload_name in ("tpcc", "tpcc-standard") else [0]
    for access_config in access_configs_for_workload(workload_name):
        for worker_threads in WorkerThreadCount:
            for batch_size in BatchSize:
                for key_page_ratio in KeyPageMapCapacity:
                    for mlp_enabled in EnableMLP:
                        for affinity_ratio in AffinityTxnRatio:
                            for num_bucket in NumBucket:
                                for compute_node_count in ComputeNodeCounts:
                                    for partition_warehouses in tpcc_partition_values:
                                        for long_txn_length in long_txn_sizes:
                                            case = {
                                                "workload": workload_name,
                                                "account_count": account_count,
                                                "warehouse_count": warehouse_count,
                                                "worker_threads": worker_threads,
                                                "affinity_txn_ratio": affinity_ratio,
                                                "batch_size": batch_size,
                                                "num_bucket": num_bucket,
                                                "compute_node_count": compute_node_count,
                                                "tpcc_partition_warehouses": partition_warehouses,
                                                "long_txn_length": long_txn_length,
                                                "long_txn_write_pct": LongTxnWritePct if EnableLongTxn else None,
                                                "key_page_ratio": key_page_ratio,
                                                "mlp_enabled": mlp_enabled,
                                                "use_data_cache": UseDataCache,
                                                "scan_axis": "full",
                                            }
                                            case.update(access_config)
                                            configs.append(case)
    return dedupe_case_configs(configs)

def build_case_configs_for_workload(workload_name, account_count=None, warehouse_count=None):
    if SweepMode == "axis":
        return build_axis_case_configs(workload_name, account_count, warehouse_count)
    if SweepMode == "full":
        return build_full_case_configs(workload_name, account_count, warehouse_count)
    raise ValueError(f"Unknown SweepMode: {SweepMode}")

def build_case_plan():
    main_case_pairs = []
    mlp_case_pairs = []
    seen = set()
    baseline_mlp = int(DefaultEnableMLP)
    mlp_run_modes = MLPRunModeType if MLPRunModeType else RunModeType
    key_page_capacity_run_modes = KeyPageCapacityRunModeType if KeyPageCapacityRunModeType else RunModeType
    batch_size_run_modes = BatchSizeRunModeType if BatchSizeRunModeType else RunModeType

    def case_key(case_config, run_mode):
        return (
            case_config["workload"],
            case_config.get("account_count"),
            case_config.get("warehouse_count"),
            case_config["access_pattern"],
            case_config.get("zipfian_theta"),
            case_config.get("zipfian_generator"),
            case_config.get("hotspot_fraction"),
            case_config.get("hotspot_prob"),
            case_config["worker_threads"],
            case_config["affinity_txn_ratio"],
            case_config["batch_size"],
            case_config["num_bucket"],
            case_config["compute_node_count"],
            case_config.get("tpcc_partition_warehouses"),
            case_config.get("long_txn_length"),
            case_config.get("long_txn_write_pct"),
            case_config.get("key_page_ratio"),
            case_config.get("mlp_enabled"),
            case_config.get("smallbank_key_permutation_seed"),
            case_config.get("tpcc_warehouse_permutation_seed"),
            case_config.get("tpcc_warehouse_load_seed"),
            run_mode,
        )

    def case_group_key(case_config):
        return case_key(case_config, None)[:-1]

    case_id = 0
    case_group_id = 0
    case_group_ids = {}
    for workload_name in Workloads:
        for account_count, warehouse_count in scale_values_for_workload(workload_name):
            for case_config in build_case_configs_for_workload(workload_name, account_count, warehouse_count):
                case_config["smallbank_key_permutation_seed"] = (
                    SmallBankKeyPermutationSeed if workload_name == "smallbank" else None
                )
                case_config["tpcc_warehouse_permutation_seed"] = (
                    TPCCWarehousePermutationSeed if workload_name in ("tpcc", "tpcc-standard") else None
                )
                case_config["tpcc_warehouse_load_seed"] = (
                    TPCCShuffleWarehouseLoadSeed
                    if workload_name in ("tpcc", "tpcc-standard") else None
                )
                load_seed = case_config["tpcc_warehouse_load_seed"]
                if load_seed is not None and (not isinstance(load_seed, int) or not 0 <= load_seed <= 2**64 - 1):
                    raise ValueError("TPCCShuffleWarehouseLoadSeed must be None or a uint64")
                group_key = case_group_key(case_config)
                if group_key not in case_group_ids:
                    case_group_id += 1
                    case_group_ids[group_key] = case_group_id
                is_mlp_delta = (
                    case_config.get("scan_axis") in ("mlp", "mlp_zipfian")
                    and int(case_config.get("mlp_enabled", baseline_mlp)) != baseline_mlp
                )
                if case_config.get("scan_axis") == "batch_size":
                    run_modes = batch_size_run_modes
                elif case_config.get("scan_axis") == "key_page_capacity":
                    run_modes = key_page_capacity_run_modes
                elif is_mlp_delta:
                    run_modes = mlp_run_modes
                else:
                    run_modes = RunModeType
                if workload_name in ("tpcc", "tpcc-standard"):
                    run_modes = list(run_modes) + [
                        mode for mode in TPCCRuleRunModeType if mode not in run_modes
                    ]
                for run_mode in run_modes:
                    key = case_key(case_config, run_mode)
                    if key in seen:
                        continue
                    seen.add(key)
                    case = dict(case_config)
                    case["run_mode"] = run_mode
                    case["seron_conflict_detection"] = None
                    case["seron_rule_high_conflict_threshold"] = None
                    if run_mode == 33:
                        if SeronConflictDetection not in ("page", "rule"):
                            raise ValueError("SeronConflictDetection must be page or rule")
                        if not 0 <= SeronRuleHighConflictThreshold <= 1:
                            raise ValueError("SeronRuleHighConflictThreshold must be in [0,1]")
                        case["seron_conflict_detection"] = SeronConflictDetection
                        case["seron_rule_high_conflict_threshold"] = SeronRuleHighConflictThreshold
                    case["hot_page_weighted_affinity"] = bool(HotPageWeightedAffinity and run_mode == 11)
                    case["case_group_id"] = case_group_ids[group_key]
                    case["data_cache_path"] = data_cache_path(
                        workload_name,
                        case_config.get("account_count"),
                        case_config.get("warehouse_count"),
                    )
                    if case.get("tpcc_partition_warehouses"):
                        case["data_cache_path"] += "_whpart"
                    if case.get("tpcc_warehouse_load_seed") is not None:
                        case["data_cache_path"] += f"_whload{case['tpcc_warehouse_load_seed']}"
                    if is_mlp_delta:
                        mlp_case_pairs.append(case)
                    else:
                        main_case_pairs.append(case)

    cases = main_case_pairs + mlp_case_pairs
    for case in cases:
        case_id += 1
        case["case_id"] = case_id
    return cases

def write_case_plan(cases, plan_path):
    columns = list(cases[0].keys()) if cases else []
    with open(plan_path, "w", newline="", encoding="utf-8") as f:
        writer = csv.DictWriter(f, fieldnames=columns, delimiter="\t")
        writer.writeheader()
        writer.writerows(cases)
    markdown_path = os.path.splitext(plan_path)[0] + ".md"
    with open(markdown_path, "w", encoding="utf-8") as f:
        f.write(f"# MP-Router Case Plan\n\nTotal cases: {len(cases)}\n\n")
        if columns:
            f.write("| " + " | ".join(columns) + " |\n")
            f.write("| " + " | ".join(["---"] * len(columns)) + " |\n")
            for case in cases:
                f.write("| " + " | ".join(str(case.get(column, "")) for column in columns) + " |\n")
    logging.info(f"Case plan written: {plan_path} ({len(cases)} cases)")

def parse_optional_value(value, parser):
    if value is None or value == "" or value == "None":
        return None
    return parser(value)

def parse_bool_value(value):
    if isinstance(value, bool):
        return value
    return str(value).strip().lower() in ("1", "true", "yes", "y")

def normalize_case_from_plan(row):
    case = dict(row)
    int_fields = [
        "case_id", "case_group_id", "run_mode", "access_pattern",
        "worker_threads", "batch_size", "num_bucket", "compute_node_count", "mlp_enabled",
        "account_count", "warehouse_count", "tpcc_partition_warehouses", "long_txn_length",
        "long_txn_write_pct", "smallbank_key_permutation_seed",
        "tpcc_warehouse_permutation_seed",
        "tpcc_warehouse_load_seed",
    ]
    float_fields = [
        "zipfian_theta", "hotspot_fraction", "hotspot_prob",
        "affinity_txn_ratio", "key_page_ratio",
        "seron_rule_high_conflict_threshold",
    ]
    none_string_fields = ["zipfian_generator", "seron_conflict_detection"]

    for field in int_fields:
        if field in case:
            case[field] = parse_optional_value(case[field], int)
    for field in float_fields:
        if field in case:
            case[field] = parse_optional_value(case[field], float)
    for field in none_string_fields:
        if field in case and (case[field] == "" or case[field] == "None"):
            case[field] = None
    if "use_data_cache" in case:
        case["use_data_cache"] = parse_bool_value(case["use_data_cache"])
    case["hot_page_weighted_affinity"] = parse_bool_value(
        case.get("hot_page_weighted_affinity", False)
    )
    case.setdefault("compute_node_count", DefaultComputeNodeCount)
    return case

def read_case_plan(plan_path):
    with open(plan_path, newline="", encoding="utf-8") as f:
        return [normalize_case_from_plan(row) for row in csv.DictReader(f, delimiter="\t")]

def write_case_metadata(case, dest_dir, kwr_report_name, command):
    metadata = dict(case)
    if RouterOnly:
        metadata["configured_worker_threads"] = case["worker_threads"]
        metadata["worker_threads"] = runtime_worker_threads(case)
        metadata["router_threads"] = runtime_router_threads(case)
        metadata["generator_threads"] = runtime_generator_threads(case)
    metadata["kwr_report_name"] = kwr_report_name
    metadata["kwr_scope"] = KWRScope
    metadata["router_only"] = RouterOnly
    metadata["tpcc_routing_diagnostics"] = TPCCRoutingDiagnostics
    metadata["virtual_arrival_rate"] = VirtualArrivalRate
    metadata["load_restart_run"] = LoadRestartRun
    metadata["build_type"] = "Release" if RouterOnly else "Debug"
    metadata["preprocess_batch_concurrency"] = runtime_preprocess_batch_concurrency()
    metadata["command"] = command
    with open(os.path.join(dest_dir, "metadata.json"), "w", encoding="utf-8") as f:
        json.dump(metadata, f, ensure_ascii=False, indent=2)

def download_kwr_reports(case, dest_dir, kwr_report_name):
    conninfos = configured_db_conninfos()[:case["compute_node_count"]]
    targets = list(enumerate(conninfos)) if KWRScope == "all" else list(enumerate(conninfos[:1]))
    for node_id, conninfo in targets:
        host = parse_conninfo(conninfo).get("host", kwr_report_ip)
        report_file = (
            f"{kwr_report_name}_node{node_id}_end.html"
            if KWRScope == "all"
            else f"{kwr_report_name}_end.html"
        )
        remote_run_report = os.path.join(kwr_report_path, report_file)
        local_run_report = os.path.join(dest_dir, report_file)
        local_temp_report = local_run_report + ".part"

        ssh = paramiko.SSHClient()
        ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
        try:
            ssh.connect(host, username="root", password=kwr_ip_password, timeout=30)
            sftp = ssh.open_sftp()
            try:
                if os.path.exists(local_temp_report):
                    os.remove(local_temp_report)
                sftp.get(remote_run_report, local_temp_report)
                os.replace(local_temp_report, local_run_report)
                logging.info(f"Collected node {node_id} KWR report from {host}: {report_file}")
            finally:
                sftp.close()
        except Exception as e:
            if os.path.exists(local_temp_report):
                os.remove(local_temp_report)
            logging.warning(
                f"Node {node_id} KWR report is unavailable on {host}: {e}"
            )
        finally:
            ssh.close()

def value_for_path(value):
    return str(value).replace("/", "_")

def runtime_worker_threads(case):
    return RouterOnlyWorkerThreads if RouterOnly else case["worker_threads"]

def runtime_router_threads(case):
    return RouterOnlyRouterThreads if RouterOnly else case["worker_threads"]

def runtime_preprocess_batch_concurrency():
    return RouterOnlyPreprocessBatchConcurrency if RouterOnly else PreprocessBatchConcurrency

def runtime_generator_threads(case):
    return RouterOnlyGeneratorThreads if RouterOnly else case["worker_threads"] * 2

def case_group_dir_name(case):
    parts = [
        case["workload"],
        f"p{case['access_pattern']}",
    ]
    if case.get("zipfian_theta") is not None:
        parts.append(f"ZipfTheta{value_for_path(case['zipfian_theta'])}")
    if case.get("zipfian_generator") is not None:
        parts.append(f"ZipfGen{value_for_path(case['zipfian_generator'])}")
    if case.get("hot_page_weighted_affinity"):
        parts.append("HotPageAffinity")
    if case.get("seron_conflict_detection") == "rule":
        parts.append(f"RuleDelay{value_for_path(case['seron_rule_high_conflict_threshold'])}")
    if case.get("hotspot_fraction") is not None:
        parts.append(f"HsFrac{value_for_path(case['hotspot_fraction'])}")
    if case.get("hotspot_prob") is not None:
        parts.append(f"HsProb{value_for_path(case['hotspot_prob'])}")
    parts.extend([
        workload_scale_label(case["workload"], case.get("account_count"), case.get("warehouse_count")),
        f"t{runtime_worker_threads(case)}",
        f"r{value_for_path(case['affinity_txn_ratio'])}",
        f"b{case['batch_size']}",
        f"n{case.get('compute_node_count', DefaultComputeNodeCount)}",
        f"nb{case['num_bucket']}",
        f"whpart{case.get('tpcc_partition_warehouses', 0)}",
        f"kp{value_for_path(case['key_page_ratio'])}",
        f"mlp{case['mlp_enabled']}",
    ])
    if RouterOnly:
        parts.append(f"rt{runtime_router_threads(case)}")
        parts.append(f"gt{runtime_generator_threads(case)}")
    if case.get("long_txn_length") is not None:
        parts.append(f"lt{case['long_txn_length']}")
    if case.get("long_txn_write_pct") is not None:
        parts.append(f"wr{case['long_txn_write_pct']}")
    return "_".join(parts)

def case_progress_label(case, case_index, total_cases, attempt):
    scale_label = workload_scale_label(case["workload"], case.get("account_count"), case.get("warehouse_count"))
    extras = []
    if case["access_pattern"] == 1:
        extras.append(f"zipf={case['zipfian_theta']}/{case['zipfian_generator']}")
    elif case["access_pattern"] == 2:
        extras.append(f"hotspot={case['hotspot_fraction']}/{case['hotspot_prob']}")
    extras.extend([
        f"threads={runtime_worker_threads(case)}",
        f"nodes={case.get('compute_node_count', DefaultComputeNodeCount)}",
        f"batch={case['batch_size']}",
        f"affinity={case['affinity_txn_ratio']}",
        f"kp={case['key_page_ratio']}",
    ])
    if RouterOnly:
        extras.append(f"router_threads={runtime_router_threads(case)}")
    if case.get("tpcc_partition_warehouses"):
        extras.append("tpcc_wh_part=1")
    if case.get("long_txn_length") is not None:
        extras.append(f"long={case['long_txn_length']}")
    if case.get("long_txn_write_pct") is not None:
        extras.append(f"write={case['long_txn_write_pct']}%")
    return (
        f"Case {case_index + 1}/{total_cases} "
        f"(id={case['case_id']}, attempt={attempt}/{max_try}) "
        f"{case['workload']} {scale_label} mode={case['run_mode']} access={case['access_pattern']} "
        f"{', '.join(extras)}"
    )

def summarize_result_dir(result_dir):
    summary_script = os.path.join(workspace, "draw", "revison_draw", "summarize_mp_router_results.py")
    subprocess.run([sys.executable, summary_script, result_dir], check=False)

def run_git_capture(args, output_path):
    with open(output_path, "w", encoding="utf-8") as f:
        result = subprocess.run(
            ["git"] + args,
            cwd=workspace,
            stdout=f,
            stderr=subprocess.STDOUT,
            encoding="utf-8",
            check=False,
        )
    return result.returncode

def backup_remote_kingbase_conf(result_dir):
    remote_conf = os.path.join(database_data_path.rstrip("/"), "kingbase.conf")
    local_conf = os.path.join(result_dir, "kingbase.conf")
    ssh = paramiko.SSHClient()
    ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
    try:
        ssh.connect(kwr_report_ip, username="root", password=kwr_ip_password, timeout=30)
        sftp = ssh.open_sftp()
        try:
            sftp.get(remote_conf, local_conf)
            logging.info(f"Backed up remote kingbase.conf to {local_conf}")
        finally:
            sftp.close()
    except Exception as e:
        logging.warning(f"Failed to back up remote kingbase.conf from {remote_conf}: {e}")
    finally:
        ssh.close()

def capture_run_reproducibility_metadata(result_dir):
    os.makedirs(result_dir, exist_ok=True)
    logging.info(f"Capturing reproducibility metadata under {result_dir}")
    git_dir = os.path.join(result_dir, "git")
    os.makedirs(git_dir, exist_ok=True)

    captures = [
        (["rev-parse", "HEAD"], "head.txt"),
        (["branch", "--show-current"], "branch.txt"),
        (["status", "--short"], "status_short.txt"),
        (["diff", "--stat"], "diff_stat.txt"),
        (["diff"], "diff.patch"),
        (["diff", "--cached"], "diff_cached.patch"),
        (["log", "-1", "--decorate=full", "--stat"], "last_commit.txt"),
    ]
    for args, filename in captures:
        rc = run_git_capture(args, os.path.join(git_dir, filename))
        if rc != 0:
            logging.warning(f"git {' '.join(args)} exited with code {rc}")

    backup_remote_kingbase_conf(result_dir)

def access_extra_args(case):
    permutation = ""
    if case.get("workload") == "smallbank" and case.get("smallbank_key_permutation_seed") is not None:
        permutation = f" --smallbank-key-permutation-seed {case['smallbank_key_permutation_seed']}"
    if case.get("workload") in ("tpcc", "tpcc-standard") and case.get("tpcc_warehouse_permutation_seed") is not None:
        permutation = f" --tpcc-warehouse-permutation-seed {case['tpcc_warehouse_permutation_seed']}"
    if case.get("workload") in ("tpcc", "tpcc-standard") and case.get("tpcc_warehouse_load_seed") is not None:
        permutation += f" --tpcc-warehouse-load-seed {case['tpcc_warehouse_load_seed']}"
    if case.get("run_mode") == 33:
        # Old saved plans retain page detection rather than inheriting new defaults.
        permutation += f" --seron-conflict-detection {case.get('seron_conflict_detection', 'page')}"
        permutation += f" --seron-rule-high-conflict-threshold {case.get('seron_rule_high_conflict_threshold', 0.10)}"
    if case["access_pattern"] == 1:
        return permutation + f" --zipfian-theta {case['zipfian_theta']} --zipfian-generator {case['zipfian_generator']}"
    if case["access_pattern"] == 2:
        return permutation + f" --hotspot-fraction {case['hotspot_fraction']} --hotspot-prob {case['hotspot_prob']}"
    return permutation

def extend_sizes_for_case(case):
    if case["workload"] not in ("tpcc", "tpcc-standard") and case.get("account_count") is not None:
        account_count = case["account_count"]
        if account_count <= 2000000:
            return 100000, 10000
        if account_count <= 5000000:
            return 200000, 30000
        return 800000, 80000
    return sys_extend_size, sys_index_extend_size

def build_load_only_command(case):
    sys_extend, sys_index_extend = extend_sizes_for_case(case)
    cmd = (
        f"./run --workload {case['workload']} --load-data-only --system-mode {case['run_mode']} "
        f"--db-type {DBType} --use-sp {UseStoredProcedures} "
        f"--access-pattern {case['access_pattern']}{access_extra_args(case)} "
        f"{workload_scale_arg(case['workload'], case.get('account_count'), case.get('warehouse_count'))} "
        f"{build_db_connection_args(case['compute_node_count'])} "
        f"--worker-threads {case['worker_threads']} "
        f"--sys_extend_size {sys_extend} --sys_index_extend_size {sys_index_extend}"
    )
    if case.get("tpcc_partition_warehouses"):
        cmd += " --tpcc-partition-warehouses"
    if Unlog:
        cmd += " --unlog"
    return cmd

def prepare_backup_data(case, backup_path):
    logging.info(f">>> Preparing Backup Data (Load & Backup) to {backup_path} <<<")

    # 1. 确保数据库是启动状态
    run_remote_cmd(f"crm resource start {db_resource_name}", check=False)
    wait_for_db_start()

    # 2. 运行导入数据 (使用 --load-data-only)
    # 使用 build 目录下的 run
    cmd = build_load_only_command(case)
    logging.info(f"Loading data with command: {cmd}")
    # 切换到运行目录执行
    cwd_backup = os.getcwd()
    os.chdir(Run_Path)
    run_cmd(cmd)
    os.chdir(cwd_backup)

    # 3. Stop Database
    logging.info("Stopping database for offline backup...")
    run_remote_cmd(f"crm resource stop {db_resource_name}")
    wait_for_db_stop()  # 确保数据库完全停止

    # 检查数据目录是否存在，如果不存在说明 stop 把磁盘也卸载了
    if not wait_for_remote_path(database_data_path, timeout=data_path_wait_timeout_seconds):
        logging.error(f"CRITICAL: Data path '{database_data_path}' disappeared after stopping database!")
        logging.error("It seems 'crm resource stop' unmounted the storage.")
        logging.info("Attempting to restart database to recover mount point...")
        run_remote_cmd(f"crm resource start {db_resource_name}")
        wait_for_db_start()
        raise Exception(f"Cannot perform offline backup: Storage unmounted when DB stops.")

    # 4. Backup Data
    logging.info(f"Backing up data to {backup_path} ...")
    run_remote_cmd(f"mkdir -p {backup_path}")
    # 再次检查源目录防止 rsync 报错
    if check_remote_exists(database_data_path):
        source = shlex.quote(database_data_path.rstrip("/") + "/")
        destination = shlex.quote(backup_path.rstrip("/") + "/")
        run_remote_cmd_streaming(
            f"rsync -a --delete --info=progress2 --human-readable --timeout=3000 {source} {destination}"
        )
    else:
        raise Exception(f"Source path {database_data_path} not found")
    
    logging.info(">>> Backup Complete <<<")

# run settings
# const
max_try = 5
workspace = os.getcwd()
output = workspace + "/build/output.txt"
result = workspace + "/build/serve/test/result.txt"
log = workspace + "/build/serve/test/partitioning_log.log"
Run_Path = workspace + "/build/serve/test/"
kwr_report_ip = "172.16.0.108"
kwr_ip_password = "Hcyhcy123."
kwr_report_path = "/home/kingbase/MP-Router/kwr/"
KWRScope = "all" # single: node0 only; all: one snapshot/report per active database node
database_data_path = "/sharedata/kingbase/data-hot/"
db_resource_name = "clone-DB"
remote_retry_sleep_seconds = 3
db_status_poll_seconds = 5
db_start_probe_fallback_sleep_seconds = 5
db_tcp_probe_timeout_seconds = 2
db_pg_isready_timeout_seconds = 2
db_sql_probe_timeout_seconds = 5
db_drop_tables_timeout_seconds = 300
use_local_db_readiness_probe = True
enable_sql_readiness_probe = True
data_path_wait_timeout_seconds = 30
load_pause_timeout_seconds = 12 * 60 * 60
test_interval_seconds = 5
config_group_interval_seconds = 30
db_ready_probe_conninfos = [
    "host=172.16.0.108 port=44321 user=system password=123456 dbname=smallbank",
    "host=172.16.0.110 port=44321 user=system password=123456 dbname=smallbank",
    "host=172.16.0.109 port=44321 user=system password=123456 dbname=smallbank",
    "host=172.16.0.111 port=44321 user=system password=123456 dbname=smallbank",
    # "host=172.16.0.30 port=44321 user=system password=123456 dbname=smallbank",
    # "host=172.16.0.29 port=44321 user=system password=123456 dbname=smallbank",
    # "host=172.16.0.31 port=44321 user=system password=123456 dbname=smallbank",
    # "host=172.16.0.32 port=44321 user=system password=123456 dbname=smallbank",
]

yashan_db_conninfos = [
    "ip_port=172.16.0.24:1688 user=sys password=Wljwlj123.",
    "ip_port=172.16.0.26:1688 user=sys password=Wljwlj123.",
    "ip_port=172.16.0.27:1688 user=sys password=Wljwlj123.",
    "ip_port=172.16.0.25:1688 user=sys password=Wljwlj123.",
]
DBType = 0 # 0: PostgreSQL/KES, 1: YashanDB, 2: MySQL
UseStoredProcedures = 1
RouterOnly = False
LoadRestartRun = False


# !      !       !            注意：根据实际环境修改以上路径和参数                  !        !        !
# -------------------------------------------- # test parameters -------------------------------------------- #
# dynamic
# RunModeType = [0, 25, 2, 23, 28, 33, 11]
# RunModeType = [33]
# RunModeType = [33]
# RunModeType = [23]
RunModeType = [11]
# RunModeType = [28]
TPCCRuleRunModeType = [31]
# RunModeType = [11, 13, 2]
# RunModeType = [28]
# ! system: 0 random, 2 page hash, 11 MP-Router, 13 without scheduling, 23 metis,
# ! 25 load balance, 28 chimera-like
# ! 32 PAR + page-conflict delay; 33 independent column ranges + same page-conflict delay
# ! To compare only these variants, explicitly select RunModeType = [23, 32, 33].
# ! TPCCRuleRunModeType: 31 TPC-C warehouse rule router, auto-added for TPC-C workloads
# RunModeType = [13]
# RunModeType = [1]
Workloads = ["smallbank"] # one script run can cover multiple workloads
WorkloadAccessPatterns = {
    "smallbank": [1],
    "tpcc": [2,0],
    "tpcc-standard": [2,0],
}
SweepMode = "axis" # axis: vary one dimension from defaults; full: Cartesian product
AccessPattern = [1, 2, 0] # 0 uniform, 1 zipfian, 2 hotspot
# AccessPattern = [1]
# ZipfianTheta = [0.4]
ZipfianTheta = [1.1, 0.9, 0.8, 0.6, 0.1]
# ZipfianTheta = [0.9, 0.8, 0.6]
# ZipfianTheta = [0.8]
ZipfianGenerator = "finite" # options: finite, legacy
SmallBankKeyPermutationSeed = None # None: unchanged workload; integer: enable seeded key permutation
TPCCWarehousePermutationSeed = None # None: unchanged workload; integer: enable seeded warehouse permutation
TPCCShuffleWarehouseLoadSeed = 42 # None: original load order; integer: shuffle whole-warehouse load tasks
SeronConflictDetection = "page" # mode 33 only: "page" preserves the existing detector
SeronRuleHighConflictThreshold = 0.10 # rule risk > threshold triggers intra-rule detection
HotPageWeightedAffinity = False # mode11 only: weight ownership affinity by window-local RW page heat
TPCCRoutingDiagnostics = False # diagnostic output only; no routing policy changes
HotspotFraction = [0.25]
# HotspotFraction = [0.001, 0.01, 0.1, 1]
# HotspotProb = [0.8]
HotspotProb = [0.5, 0.8, 0.9]
# account = 100W, 单个表大概14W个页面, 每个页面8KB, 大小约1.1GB
AccountCount = [10000000]
WarehouseCount = [200]
WorkerThreadCount = [16]
# WorkerThreadCount = [16, 2, 4, 8, 32, 64]
try_count = 35000
TimeRun = 1 # 0:disable, 1:enable
EnableDynamicWorkload = 0 # 0:disable, 1:enable dynamic workload phases in ./run
# WarmupSeconds = 120
# RunSeconds = 150
WarmupSeconds = 15
RunSeconds = 30
VirtualArrivalRate = 150000 # TPS for virtual-arrival latency only; 0 disables it, no pacing
FillPipelineBubble = 0
PreprocessBatchConcurrency = 1
RouterOnlyWorkerThreads = 4
RouterOnlyRouterThreads = 16
RouterOnlyPreprocessBatchConcurrency = 2
RouterOnlyGeneratorThreads = 8
Unlog = 1
UseDataCache = False # True: restore workload data cache before each case; False: do not restart DB, load data in each run
RestartDBBeforeEachTPCCMode = True # True: for TPC-C, restart DB before each mode when UseDataCache=False
workload = Workloads[0]
sys_extend_size = 300000
sys_index_extend_size = 30000
AffinityTxnRatio = [0.8]
# AffinityTxnRatio = [0.8, 1, 0.6, 0.4, 0.2, 0]
# AffinityTxnRatio = [1, 0.8, 0.6, 0.4, 0.2, 0]
# BatchSize = [10000, 5000, 1000, 500, 100, 50000, 100000, 10, 50] # default 10000
BatchSize = [10000]
NumBucket = [1]
ComputeNodeCounts = [4] # first value is the default; remaining values form the node-count axis
# ComputeNodeCounts = [8, 6, 4, 2, 1, 7, 5, 3] # first value is the default; remaining values form the node-count axis
TPCCPartitionWarehouse = [0] # 0:disable, 1:partition warehouse by w_id range
EnableLongTxn = 0 # 0:disable, 1:enable
# LongTxnSize = [8]
LongTxnSize = [2, 4, 8, 12, 16, 20] # only valid when EnableLongTxn=1
LongTxnWritePct = 50 # only valid when EnableLongTxn=1; per-operation write probability
KeyPageMapCapacity = [1.1]
# KeyPageMapCapacity = [1.1, 1.0, 0.8, 0.6, 0.4, 0.2] # passed to --key-page-ratio
EnableMLP = [0]
# EnableMLP = [0, 1] # 0:disable, 1:enable; changing this requires rebuilding with MLP_PREDICTION
MLPRunModeType = [11] # MLP-delta cases run only these modes; baseline MLP=0 reuses normal sweep results
BatchSizeRunModeType = [11] # BatchSize axis runs only these modes
KeyPageCapacityRunModeType = [2, 23, 11, 28] # KeyPageMapCapacity axis runs only these modes
RebuildForMLP = True
RestoreConfigAfterRun = True

DefaultAccessPattern = {
    "smallbank": 1,
    "tpcc": 2,
    "tpcc-standard": 2,
}
DefaultZipfianTheta = 0.8
DefaultHotspotFraction = 0.25
DefaultHotspotProb = 0.8
DefaultWorkerThreads = WorkerThreadCount[0]
DefaultBatchSize = BatchSize[0]
DefaultKeyPageMapCapacity = KeyPageMapCapacity[0]
DefaultEnableMLP = 0
DefaultAffinityTxnRatio = AffinityTxnRatio[0]
DefaultNumBucket = NumBucket[0]
DefaultComputeNodeCount = ComputeNodeCounts[0]
DefaultTPCCPartitionWarehouses = TPCCPartitionWarehouse[0]
DefaultLongTxnLength = LongTxnSize[0]

# -------------------------------------------- # main test logic -------------------------------------------- #

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--force-reload", action="store_true", help="Force reload data even if backup exists")
    parser.add_argument(
        "--no-data-cache",
        action="store_true",
        help="Override UseDataCache and disable automatic workload data-cache detection, creation, and restore.",
    )
    parser.add_argument(
        "--plan-only",
        action="store_true",
        help="Generate case_plan.tsv and case_plan.md without building or running experiments.",
    )
    parser.add_argument("--enable-long-txn", action="store_true", help="Enable long transactions")
    parser.add_argument("--long-txn-length", type=int, default=None, help="Length of long transactions")
    parser.add_argument("--zipfian-generator", choices=("finite", "legacy"), default=None, help="Zipfian generator type")
    parser.add_argument("--smallbank-key-permutation-seed", type=int, default=None,
                        help="Enable seeded SmallBank account-ID permutation for every selected mode")
    parser.add_argument("--tpcc-warehouse-permutation-seed", type=int, default=None,
                        help="Enable seeded TPC-C warehouse-ID permutation for every selected mode")
    parser.add_argument("--tpcc-routing-diagnostics", action="store_true", default=TPCCRoutingDiagnostics,
                        help="Record mode11 TPC-C assignment/selection diagnostics")
    parser.add_argument("--db-type", type=int, choices=(0, 1, 2), default=None, help="0: PostgreSQL/KES, 1: YashanDB, 2: MySQL")
    parser.add_argument("--warmup-seconds", type=int, default=None, help="Override WarmupSeconds for --time-run")
    parser.add_argument("--run-seconds", type=int, default=None, help="Override RunSeconds for --time-run")
    parser.add_argument("--num-bucket", type=int, default=None, help="Override NumBucket for this script run")
    parser.add_argument("--node-count", type=int, default=None, help="Run with only the first N configured database nodes")
    parser.add_argument("--kwr-scope", choices=("single", "all"), default=None, help="Collect KWR from node0 only or every active DB node")
    parser.add_argument(
        "--preprocess-batch-concurrency",
        type=int,
        default=None,
        help="Number of mode11 batches preprocessed concurrently [default: 2].",
    )
    parser.add_argument(
        "--router-only",
        action="store_true",
        help="Measure SmallBank router capacity with empty consumers; automatically builds Release/O3.",
    )
    parser.add_argument(
        "--load-restart-run",
        action="store_true",
        help="For every case, pause the same process after loading, restart PostgreSQL/KES, then resume the benchmark.",
    )
    parser.add_argument("--disable-dynamic", action="store_true", help="Disable --enable-dynamic for this run")
    parser.add_argument(
        "--resume-result-dir",
        default=None,
        help="Resume into an existing result directory and reuse its case_plan.tsv.",
    )
    parser.add_argument(
        "--resume-from-case",
        type=int,
        default=None,
        help="Resume from this case_id in the selected case plan.",
    )
    args = parser.parse_args()
    TPCCRoutingDiagnostics = args.tpcc_routing_diagnostics

    # Handle LongTxnSize override
    if args.long_txn_length is not None:
        LongTxnSize = [args.long_txn_length]
    if args.zipfian_generator is not None:
        ZipfianGenerator = args.zipfian_generator
    if args.smallbank_key_permutation_seed is not None:
        if not 0 <= args.smallbank_key_permutation_seed <= 2**64 - 1:
            parser.error("--smallbank-key-permutation-seed must be a uint64")
        SmallBankKeyPermutationSeed = args.smallbank_key_permutation_seed
    if args.tpcc_warehouse_permutation_seed is not None:
        if not 0 <= args.tpcc_warehouse_permutation_seed <= 2**64 - 1:
            parser.error("--tpcc-warehouse-permutation-seed must be a uint64")
        TPCCWarehousePermutationSeed = args.tpcc_warehouse_permutation_seed
    if args.db_type is not None:
        DBType = args.db_type
    if args.warmup_seconds is not None:
        WarmupSeconds = args.warmup_seconds
    if args.run_seconds is not None:
        RunSeconds = args.run_seconds
    if args.num_bucket is not None:
        NumBucket = [args.num_bucket]
        DefaultNumBucket = args.num_bucket
    if args.node_count is not None:
        ComputeNodeCounts = [args.node_count]
        DefaultComputeNodeCount = args.node_count
    if args.kwr_scope is not None:
        KWRScope = args.kwr_scope
    if args.preprocess_batch_concurrency is not None:
        if args.preprocess_batch_concurrency <= 0:
            raise ValueError("--preprocess-batch-concurrency must be greater than 0")
        PreprocessBatchConcurrency = args.preprocess_batch_concurrency
    RouterOnly = args.router_only
    LoadRestartRun = args.load_restart_run
    if args.disable_dynamic:
        EnableDynamicWorkload = 0
    if RouterOnly and any(workload_name != "smallbank" for workload_name in Workloads):
        raise ValueError("--router-only currently supports SmallBank workloads only")
    configured_node_count = len(configured_db_conninfos())
    if configured_node_count:
        invalid_node_counts = [
            node_count for node_count in ComputeNodeCounts
            if node_count < 1 or node_count > configured_node_count
        ]
        if invalid_node_counts:
            raise ValueError(
                f"Invalid ComputeNodeCounts {invalid_node_counts}; DB_TYPE={DBType} has "
                f"{configured_node_count} configured database connections"
            )
    UseDataCache = UseDataCache and not args.no_data_cache
    if UseDataCache and not is_postgres_like_db():
        raise RuntimeError("UseDataCache is only wired for the PostgreSQL/KES data directory workflow.")
    if LoadRestartRun and UseDataCache:
        raise ValueError("--load-restart-run cannot be combined with UseDataCache; use --no-data-cache.")
    if LoadRestartRun and not is_postgres_like_db():
        raise ValueError("--load-restart-run currently supports PostgreSQL/KES only.")
    resume_result_dir = os.path.abspath(args.resume_result_dir) if args.resume_result_dir else None
    logging.info(f"UseDataCache = {UseDataCache}")
    logging.info(f"DBType = {DBType}")
    logging.info(f"EnableDynamicWorkload = {EnableDynamicWorkload}")
    logging.info(f"ZipfianGenerator = {ZipfianGenerator}")
    logging.info(f"ComputeNodeCounts = {ComputeNodeCounts}")
    logging.info(f"KWRScope = {KWRScope}")
    logging.info(f"RouterOnly = {RouterOnly}")
    logging.info(f"LoadRestartRun = {LoadRestartRun}")
    logging.info(f"BuildType = {'Release' if RouterOnly else 'Debug'}")
    if RouterOnly:
        logging.info(
            "RouteOnlyThreads = "
            f"workers:{RouterOnlyWorkerThreads}, router:{RouterOnlyRouterThreads}, "
            f"preprocess_batches:{RouterOnlyPreprocessBatchConcurrency}, "
            f"generators:{RouterOnlyGeneratorThreads}"
        )
    else:
        logging.info(f"PreprocessBatchConcurrency = {PreprocessBatchConcurrency}")

    # 删除之前的结果
    if os.path.exists(result):
        os.remove(result)
    # 创建图像文件夹
    if not os.path.exists("./result"):
        os.mkdir("./result")
    os.chdir("./result")
    if resume_result_dir:
        figure_path = resume_result_dir
        time_str = os.path.basename(os.path.normpath(figure_path))
        plan_path = os.path.join(figure_path, "case_plan.tsv")
        if not os.path.exists(plan_path):
            raise FileNotFoundError(f"Cannot resume: case_plan.tsv not found in {figure_path}")
        planned_cases = read_case_plan(plan_path)
        logging.info(f"Resuming from existing result directory: {figure_path}")
    else:
        # 创建此次测试的结果文件夹，以时间命名
        base_time_str = time.strftime("%Y%m%d%H%M%S", time.localtime())
        for suffix in range(1000):
            time_str = base_time_str if suffix == 0 else f"{base_time_str}_{suffix}"
            try:
                os.mkdir(time_str)
                break
            except FileExistsError:
                continue
        else:
            raise RuntimeError(f"Unable to create unique result directory for {base_time_str}")
        figure_path = os.path.join(os.getcwd(), time_str)
        planned_cases = build_case_plan()
        write_case_plan(planned_cases, os.path.join(figure_path, "case_plan.tsv"))

    if args.resume_from_case is not None:
        planned_cases = [case for case in planned_cases if int(case["case_id"]) >= args.resume_from_case]
        if not planned_cases:
            raise ValueError(f"No cases found with case_id >= {args.resume_from_case}")
        logging.info(
            f"Resume filter applied: starting from case {planned_cases[0]['case_id']} "
            f"({len(planned_cases)} cases remaining)."
        )
    if args.plan_only:
        logging.info(f"Plan-only mode completed: {figure_path}")
        raise SystemExit(0)

    capture_run_reproducibility_metadata(figure_path)
    if TPCCRoutingDiagnostics and any(c["run_mode"] != 11 or c["workload"] not in ("tpcc", "tpcc-standard") for c in planned_cases):
        raise ValueError("TPC-C routing diagnostics require mode11 TPC-C cases only")

    # !开始本次的测试
    os.chdir(workspace)
    atexit.register(restore_config_h)
    if not resume_result_dir and is_postgres_like_db():
        logging.info("Dropping public tables once before starting this script run.")
        wait_for_db_start()
        drop_public_tables()
        if not RouterOnly:
            truncate_kwr_tables()
            sync_remote_servers_after_case()
    elif not resume_result_dir:
        logging.info(f"Skipping PostgreSQL/KES table cleanup for DB_TYPE={DBType}.")
    
    # 标记是否已经准备好备份数据
    current_backup_key = None
    router_only_loaded_data_key = None

    for case_index, case in enumerate(planned_cases):
        ensure_build_for_mlp(case["mlp_enabled"])
        backup_path = case["data_cache_path"]
        if UseDataCache and backup_path != current_backup_key:
            backup_path = ensure_data_cache(case, force_reload=args.force_reload)
            current_backup_key = backup_path

        attempt = 0
        success = False

        router_only_data_key = (case["workload"], case.get("account_count"), case.get("warehouse_count"))
        reuse_router_only_data = (
            RouterOnly and not LoadRestartRun and
            router_only_loaded_data_key == router_only_data_key
        )

        if UseDataCache and not reuse_router_only_data:
            reset_db_data(backup_path)
        # 确保 server 进程被清理
        kill_server()
        if not RouterOnly and not LoadRestartRun:
            restart_database_before_tpcc_mode(case)

        # 删除之前的结果文件，防止误判
        if os.path.exists(result):
            os.remove(result)

        os.chdir(Run_Path)
        while attempt < max_try and not success:
            attempt += 1
            if os.path.exists(result):
                os.remove(result)
            extra_part_log = ""
            if case["access_pattern"] == 1:
                extra_part_log = f", ZipfianTheta={case['zipfian_theta']}, ZipfianGenerator={case['zipfian_generator']}"
            elif case["access_pattern"] == 2:
                extra_part_log = f", HotspotFraction={case['hotspot_fraction']}, HotspotProb={case['hotspot_prob']}"

            scale_label = workload_scale_label(case["workload"], case.get("account_count"), case.get("warehouse_count"))
            logging.info(case_progress_label(case, case_index, len(planned_cases), attempt))

            kwr_report_name = (
                f"kwr_{time_str}_case{case['case_id']:03d}"
                f"_mode{case['run_mode']}_access{case['access_pattern']}"
                f"_{case['workload']}_{scale_label}_n{case['compute_node_count']}"
                f"_thd{runtime_worker_threads(case)}"
            )

            sys_extend, sys_index_extend = extend_sizes_for_case(case)
            cmd = (
                f"./run --workload {case['workload']} --system-mode {case['run_mode']} "
                f"--db-type {DBType} --use-sp {UseStoredProcedures} "
                f"--access-pattern {case['access_pattern']}{access_extra_args(case)} "
                f"{workload_scale_arg(case['workload'], case.get('account_count'), case.get('warehouse_count'))} "
                f"{build_db_connection_args(case['compute_node_count'])} "
                f"--worker-threads {runtime_worker_threads(case)} --kwr-name {kwr_report_name}"
                f" --kwr-scope {KWRScope}"
                f" --sys_extend_size {sys_extend} --sys_index_extend_size {sys_index_extend} "
                f"--affinity-txn-ratio {case['affinity_txn_ratio']} "
                f" --batch-size {case['batch_size']} --num-bucket {case['num_bucket']}"
                f" --key-page-ratio {case['key_page_ratio']}"
                f" --preprocess-batch-concurrency {runtime_preprocess_batch_concurrency()}"
            )
            cmd += f" --virtual-arrival-rate {VirtualArrivalRate}"
            if case.get("tpcc_partition_warehouses"):
                cmd += " --tpcc-partition-warehouses"
            if case.get("hot_page_weighted_affinity"):
                cmd += " --hot-page-weighted-affinity"
            if TPCCRoutingDiagnostics:
                cmd += " --tpcc-routing-diagnostics"
                for name in ("tpcc_window_assignments.csv", "tpcc_plan_decisions.csv"):
                    path = os.path.join(Run_Path, name)
                    if os.path.exists(path):
                        os.remove(path)
            if UseDataCache or reuse_router_only_data:
                cmd += " --skip-load-data"
            if LoadRestartRun:
                cmd += " --pause-after-load"

            if TimeRun:
                cmd += f" --time-run --warmup-seconds {WarmupSeconds} --run-seconds {RunSeconds} --fill-pipeline-bubble {FillPipelineBubble}"
                if Unlog:
                    cmd += " --unlog"
            else:
                current_try_count = try_count
                if EnableLongTxn and case.get("long_txn_length"):
                    current_try_count = 35000 // case["long_txn_length"]
                cmd += f" --try-count {current_try_count}"

            if EnableLongTxn and case.get("long_txn_length"):
                long_txn_write_pct = case.get("long_txn_write_pct", LongTxnWritePct)
                cmd += (
                    f" --enable-long-txn --long-txn-length {case['long_txn_length']}"
                    f" --long-txn-write-pct {long_txn_write_pct}"
                )
            if EnableDynamicWorkload:
                cmd += " --enable-dynamic"
            if RouterOnly:
                cmd += (
                    f" --router-threads {runtime_router_threads(case)}"
                    f" --generator-threads {runtime_generator_threads(case)}"
                    " --router-only --important-router-batch-log 0"
                )

            if LoadRestartRun:
                run_remote_cmd(f"crm resource start {db_resource_name}", check=False)
            wait_for_db_start()

            with open(output, "w", encoding="utf-8") as outfile:
                process = subprocess.Popen(
                    shlex.split(cmd) if LoadRestartRun else cmd,
                    shell=not LoadRestartRun,
                    stdout=outfile,
                    stderr=subprocess.STDOUT,
                )
                if LoadRestartRun:
                    try:
                        wait_for_process_stopped(process, load_pause_timeout_seconds)
                        restart_database_resource(
                            f"after in-process load for case {case['case_id']} attempt {attempt}"
                        )
                        logging.info(f"Resuming process {process.pid} after database restart.")
                        process.send_signal(signal.SIGCONT)
                        return_code = process.wait()
                    except Exception as exc:
                        logging.warning(
                            f"Case {case['case_id']} in-process load/restart handshake failed: {exc}"
                        )
                        if process.poll() is None:
                            process.kill()
                            process.wait()
                        return_code = -1
                else:
                    return_code = process.wait()
            if return_code == 0 and os.path.exists(result):
                success = True
                if RouterOnly:
                    router_only_loaded_data_key = router_only_data_key
                logging.info(f"Case {case['case_id']} completed successfully.")

                # 同一组负载参数放在一个目录下，mode 作为下一层目录，避免顶层结果过散。
                dest_dir = os.path.join(
                    figure_path,
                    case_group_dir_name(case),
                    f"m{case['run_mode']}",
                )
                os.makedirs(dest_dir, exist_ok=True)
                write_case_metadata(case, dest_dir, kwr_report_name, cmd)

                # 使用shutil复制文件，保持内容一致
                dest_file = os.path.join(dest_dir, "result.txt")
                shutil.copy2(result, dest_file)
                log_file = os.path.join(dest_dir, "partitioning_log.log")
                shutil.copy2(log, log_file)
                if TPCCRoutingDiagnostics:
                    for name in ("tpcc_window_assignments.csv", "tpcc_plan_decisions.csv"):
                        shutil.copy2(os.path.join(Run_Path, name), os.path.join(dest_dir, name))

                if is_postgres_like_db() and not RouterOnly:
                    download_kwr_reports(case, dest_dir, kwr_report_name)
                summarize_result_dir(figure_path)
            else:
                if return_code != 0:
                    logging.warning(f"Run exited with code {return_code}, retrying...")
                elif not os.path.exists(result):
                    logging.warning("Result file not found, retrying...")
        if not success:
            logging.error(
                f"Test failed after {max_try} attempts for case {case['case_id']}: "
                f"Workload={case['workload']}, RunMode={case['run_mode']}, AccessPattern={case['access_pattern']}, "
                f"Scale={workload_scale_label(case['workload'], case.get('account_count'), case.get('warehouse_count'))}"
            )

        if not RouterOnly:
            sync_remote_servers_after_case()
            time.sleep(test_interval_seconds)
        next_case = planned_cases[case_index + 1] if case_index + 1 < len(planned_cases) else None
        if (not RouterOnly and next_case is not None and
                next_case.get("case_group_id") != case.get("case_group_id")):
            logging.info(
                f"Completed config group {case.get('case_group_id')}; "
                f"sleeping {config_group_interval_seconds}s before the next config group."
            )
            if not UseDataCache:
                if is_postgres_like_db():
                    wait_for_db_start()
                    drop_public_tables()
                    truncate_kwr_tables()
                    sync_remote_servers_after_case()
                    if LoadRestartRun:
                        logging.info("Next case will restart the database after its in-process load phase.")
                    elif RestartDBBeforeEachTPCCMode and is_tpcc_workload(next_case["workload"]):
                        logging.info("Next config group is TPC-C; database restart will run before its mode.")
                    else:
                        restart_database_resource()
                else:
                    logging.info(f"Skipping PostgreSQL/KES cleanup between config groups for DB_TYPE={DBType}.")
            time.sleep(config_group_interval_seconds)
    kill_server()
    summarize_result_dir(figure_path)
    logging.info("All tests completed.")