[YARN-9518] can't use CGroups with YARN in centos7

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.7.7
• Labels: cgroup patch
• Flags: Patch, Important
• Assignee: Unassigned
• Reporter: Shurong Mai
• Created: 29/Apr/19 08:43
• Updated: 14/May/19 18:59
• Resolved: 14/May/19 18:36

Description

The os version is centos7. 

cat /etc/redhat-release
CentOS Linux release 7.3.1611 (Core)

When I had set configuration variables  for cgroup with yarn, nodemanager could be start without any matter. But when I ran a job, the job failed with these exceptional nodemanager logs in the end.

In these logs, the important logs is " Can't open file /sys/fs/cgroup/cpu as node manager - Is a directory "

After I analysed, I found the reason. In centos6, the cgroup "cpu" and "cpuacct" subsystem are as follows: 

/sys/fs/cgroup/cpu
/sys/fs/cgroup/cpuacct

But in centos7, as follows:

/sys/fs/cgroup/cpu -> cpu,cpuacct
/sys/fs/cgroup/cpuacct -> cpu,cpuacct
/sys/fs/cgroup/cpu,cpuacct

"cpu" and "cpuacct" have merge as "cpu,cpuacct".  "cpu"  and  "cpuacct"  are symbol links. 

As I look at source code, nodemamager get the cgroup subsystem info by reading /proc/mounts. So It get the cpu and cpuacct subsystem path are also "/sys/fs/cgroup/cpu,cpuacct". 

The resource description arguments of container-executor is such as follows: 

cgroups=/sys/fs/cgroup/cpu,cpuacct/hadoop-yarn/container_1554210318404_0057_02_000001/tasks

There is a comma in the cgroup path, but the comma is separator of multi resource. Therefore, the cgroup path is truncated by container-executor as "/sys/fs/cgroup/cpu" rather than correct cgroup path " /sys/fs/cgroup/cpu,cpuacct/hadoop-yarn/container_1554210318404_0057_02_000001/tasks " and report the error in the log  " Can't open file /sys/fs/cgroup/cpu as node manager - Is a directory "

Hence I modify the source code and submit a patch. The idea of patch is that nodemanager get the cgroup cpu path as "/sys/fs/cgroup/cpu" rather than "/sys/fs/cgroup/cpu,cpuacct". As a result, the  resource description arguments of container-executor is such as follows: 

cgroups=/sys/fs/cgroup/cpu/hadoop-yarn/container_1554210318404_0057_02_000001/tasks

Note that there is no comma in the path, and is a valid path because "/sys/fs/cgroup/cpu" is symbol link to "/sys/fs/cgroup/cpu,cpuacct". 

After applied the patch, the problem is resolved and the job can run successfully.

The patch is compatible with  cgroup path of history os version such as centos6, centos7 , and universally applicable to cgroup subsystem paths such as cgroup network subsystem as follows:  

/sys/fs/cgroup/net_cls -> net_cls,net_prio
/sys/fs/cgroup/net_prio -> net_cls,net_prio
/sys/fs/cgroup/net_cls,net_prio

 

 

##################################################################################################################################

exceptional nodemanager logs:

2019-04-19 20:17:20,095 INFO org.apache.hadoop.yarn.server.nodemanager.containermanager.container.ContainerImpl: Container container_1554210318404_0042_01_000001 transitioned from LOCALIZED to RUNNING
2019-04-19 20:17:20,101 WARN org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor: Exit code from container container_1554210318404_0042_01_000001 is : 27
2019-04-19 20:17:20,103 WARN org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor: Exception from container-launch with container ID: container_155421031840
4_0042_01_000001 and exit code: 27
ExitCodeException exitCode=27:
        at org.apache.hadoop.util.Shell.runCommand(Shell.java:585)
        at org.apache.hadoop.util.Shell.run(Shell.java:482)
        at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:776)
        at org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor.launchContainer(LinuxContainerExecutor.java:299)
        at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:302)
        at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:82)
        at java.util.concurrent.FutureTask.run(FutureTask.java:266)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
        at java.lang.Thread.run(Thread.java:745)
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Exception from container-launch.
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Container id: container_1554210318404_0042_01_000001
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Exit code: 27
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Stack trace: ExitCodeException exitCode=27:
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.util.Shell.runCommand(Shell.java:585)
2019-04-19 20:17:20,108 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.util.Shell.run(Shell.java:482)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.util.Shell$ShellCommandExecutor.execute(Shell.java:776)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.yarn.server.nodemanager.LinuxContainerExecutor.launchContainer(LinuxContainerExecutor.java:299)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:302)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch.call(ContainerLaunch.java:82)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at java.util.concurrent.FutureTask.run(FutureTask.java:266)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:       at java.lang.Thread.run(Thread.java:745)
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor:
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Shell output: main : command provided 1
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: main : user is test_hadoop
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: main : requested yarn user is datadev
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Writing to cgroup task files...
2019-04-19 20:17:20,109 INFO org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor: Can't open file /sys/fs/cgroup/cpu as node manager - Is a directory
2019-04-19 20:17:20,131 WARN org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch: Container exited with a non-zero exit code 27
2019-04-19 20:17:20,133 INFO org.apache.hadoop.yarn.server.nodemanager.containermanager.container.ContainerImpl: Container container_1554210318404_0042_01_000001 transitioned from RUNNING to EXITED_WITH_FAILURE
2019-04-19 20:17:20,133 INFO org.apache.hadoop.yarn.server.nodemanager.containermanager.launcher.ContainerLaunch: Cleaning up container container_1554210318404_0042_01_000001
 

Issue Links

• duplicates: YARN-2194 Cgroups cease to work in RHEL7