使用 Mesos 运行 dockered Spark 会破坏 Mesos答案

【问题标题】：Running dockered Spark with Mesos breaks Mesos使用 Mesos 运行 dockered Spark 会破坏 Mesos
【发布时间】：2019-03-12 08:08:51
【问题描述】：

我试图在 Mesos 上使用 Spark 在 DC/OS 上运行 Jupyter Notebook docker（Ubuntu 16.04 版本）。 Python 输出了大量无用的错误消息，但在连接到容器并尝试从容器运行 spark-submit 作业后，我收到了很多关于连接问题的错误。

Spark 驱动程序无法正确连接到 Mesos，似乎在大多数情况下设置 LIBPROCESS_IP 就足够了。但是，在我的情况下，使用它会完全挂起 Mesos。

这是我在 docker 容器中运行的：

export LIBPROCESS_ADVERTISE_IP=172.16.6.105; export SPARK_HOME=spark-2.3.2-bin-hadoop2.6; export JAVA_HOME=/usr/lib/jvm/java-8-openjdk-amd64; export MESOS_NATIVE_JAVA_LIBRARY=/usr/local/lib/libmesos.so; export LIBPROCESS_IP=172.19.0.4; ./spark-2.3.2-bin-hadoop2.6/bin/spark-submit --master mesos://leader.mesos:5050 --class org.apache.spark.examples.SparkPi https://downloads.mesosphere.com/spark/assets/spark-examples_2.11-2.0.1.jar 30

Spark 驱动程序挂在这部分：

I0312 07:18:13.722151  3764 sched.cpp:232] Version: 1.2.3
I0312 07:18:13.732707  3758 sched.cpp:336] New master detected at master@172.16.6.103:5050
I0312 07:18:13.733749  3758 sched.cpp:352] No credentials provided. Attempting to register without authentication

在这一步中，Mesos 挂起。根本无法访问 UI，DCOS 启动后检查显示错误。

我检查了 Mesos 日志，这是我看到的：

mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.911664 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.911737 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.911801 32335 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.911841 32335 master.cpp:3048] Framework f1731b0f-a064-434f-8f15-2225a57ce2de-0014 (Spark Pi) at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534 already subscribed, resending acknowledgement
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912062 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912149 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912243 32335 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912281 32335 master.cpp:3048] Framework f1731b0f-a064-434f-8f15-2225a57ce2de-0014 (Spark Pi) at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534 already subscribed, resending acknowledgement
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912369 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912441 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912499 32335 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912534 32335 master.cpp:3048] Framework f1731b0f-a064-434f-8f15-2225a57ce2de-0014 (Spark Pi) at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534 already subscribed, resending acknowledgement
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912771 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912860 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912921 32335 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:19:36 centos-master-01 mesos-master[32324]: I0312 08:19:31.912957 32335 master.cpp:3048] Framework f1731b0f-a064-434f-8f15-2225a57ce2de-0014 (Spark Pi) at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534 already subscribed, resending acknowledgement

我有时也会看到：

mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638309   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638342   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.638381   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638442   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638475   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.638514   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638572   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638605   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.638644   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638715   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638751   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.638790   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638847   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638881   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.638921   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.638978   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.639011   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement
mar 12 08:35:13 centos-master-01 mesos-master[837]: W0312 08:35:11.639060   855 master.hpp:2322] Master attempted to send message to disconnected framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.639118   855 master.cpp:3038] Subscribing framework Spark Pi with checkpointing disabled and capabilities [  ]
mar 12 08:35:13 centos-master-01 mesos-master[837]: I0312 08:35:11.639153   855 master.cpp:3048] Framework e40238eb-4b82-4883-be2c-54103b84dea0-0009 (Spark Pi) at scheduler-0ae862ca-bf59-4f80-8d95-9d244c796547@172.16.6.105:35139 already subscribed, resending acknowledgement

这一直在重复。当我停止驱动程序时，Mesos 仍然损坏并且一直在输出这些消息：

mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871507 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871595 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871671 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871744 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871811 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871911 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.871979 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.872048 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534
mar 12 08:19:49 centos-master-01 mesos-master[32324]: I0312 08:19:41.872140 32335 master.cpp:2958] Received SUBSCRIBE call for framework 'Spark Pi' at scheduler-e6d4dc88-8470-4519-967d-c86c2fee1c39@172.16.6.105:38534

看来，Spark Driver 向 Mesos 发送订阅调用的速度非常快，以至于 Mesos 无法跟上处理它们的速度。试过 Spark 2.3.2 和 2.4.0，结果一样。

我尝试将 Spark 连接到 Spark Mesos Dispatcher，但是即使设置了这些 LIBPROCESS 变量，我也会遇到常见的连接错误：

E0312 08:01:55.658208  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.658838  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.659353  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.660073  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.660650  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.661358  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.662775  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.663313  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.663964  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected
E0312 08:01:55.664711  4874 process.cpp:2431] Failed to shutdown socket with fd 279: Transport endpoint is not connected

有人遇到过这样的问题吗？我该如何解决？

【问题讨论】：

标签： docker apache-spark jupyter-notebook mesos

【解决方案1】：

我使用 docker compose 在 mesos 上运行 spark。我已经有一个安装了 Mesos 的 docker 镜像并配置了 mesos 集群，例如，我确定了 Master 和 worker。然后，我为 master 和 slave 编写了这些 docker compose。他们没有错误地工作。

作曲大师：

version: '3.7'
services:
  master:
   image: ubuntu_mesos_spark
   command: bash -c "sleep 40; /home/mesos-1.7.0/build/bin/mesos-master.sh --ip=150.20.11.136 --work_dir=/var/run/mesos --hostname=x.x.x.x"  ##hostname : IP of the master node
   restart: always
   network_mode: host
   environment:
    - MESOS_HOSTNAME="150.20.11.136"
    - MESOS_QUORUM=1
    - MESOS_LOG_DIR=/var/log/mesos
   expose:
    - 5050
    - 4040
    - 7077
    - 8080
   ports:
    - 5050:5050
    - 4040:4040
    - 7077:7077
    - 8080:8080

组成奴隶：

  version: '3.7'
  services: 
    slave:
      image: ubuntu_mesos_spark
      command: bash -c "sleep 40; /home/mesos-1.7.0/build/bin/mesos-slave.sh 
      --master=150.20.11.136:5050 --work_dir=/var/run/mesos 
      --systemd_enable_support=false"
      restart: always
      privileged: true
      network_mode: host
      environment:
      - MESOS_HOSTNAME="150.20.11.157"
      - MESOS_EXECUTOR_REGISTRATION_TIMEOUT=5mins #also in Dockerfile
      - MESOS_LOG_DIR=/var/log/mesos
      - MESOS_LOGGING_LEVEL=INFO
      expose:
      - 5051
      ports:
      - 5051:5051

我希望这很有用。

【讨论】：

@m-gh 谢谢！ network_mode: host 就是答案。尝试在桥接模式下运行 spark 并且忘记了我可以使用主机模式解决所有问题（我很少使用它），我感到很愚蠢。答案就是这么简单。
@Tomasz，很高兴它有帮助。