issue from ctrip: windows mesos agent crashed once an application created by marathon

[shiyan2016]

I use the master branch on windows 2016 datacenter desktop system with the commit 499173d395db40e753e86ec5847b2e3944b87c35, and running the mesos agent like this:
.\mesos-agent.exe --master=10.18.11.20:5050 --work_dir=D:\ --isolation=windows/cpu,filesystem/windows --hostname=10.18.62.225 --ip=10.18.62.225 --containerizers="docker,mesos" --log_dir=D:\mesos-log --launcher_dir=D:\mesos-build\mesos-master\build\src --runtime_dir=D:\mesos-runtime

when I create an application with marathon, the windows mesos agent will crashed with the errors:

slave.cpp:4610] Check failed: resource.has_allocation_info()
*** Check failure stack trace: ***

[lilyfang]

I could not repro the agent crash issue. Couple of the questions,

• How do you create an application with marathon?
• What application do you create?

[yongluo2013]

@shiyan2016 can you help post you environment info ?

[shiyan2016]

@lilyfang I try to create the appliation running notepad using marathon with the json:
{
"id": "1",
"cmd": "notepad",
"cpus": 1,
"mem": 128,
"disk": 0,
"instances": 1
}.
And cretae the application running windows container using marathon with the json:
{
"id": "2",
"cmd": "ping -t localhost",
"cpus": 1,
"mem": 2048,
"disk": 20480,
"instances": 1,
"container": {
"docker": {
"image": "hub.cloud.ctripcorp.com/windows_container/windowsservercore",
"network": "HOST"
},
"type": "DOCKER"
}
}. The docker version is: 1.12.
Creating the two applications using marathon will make the agent crashed.

[shiyan2016]

And the mesos master env set up refering this: http://www.cnblogs.com/ee900222/p/docker_2.html.

[lilyfang]

Based on investigation, I can only repro this issue when my setup has a mismatch version between mesos master and agent. Please make sure you install mesos master with mesos-1.2.0(+) if you build your mesos agent with the latest mesos mainstream source code.

[lilyfang]

Here is my steps,

1. Set up the marathon, master and zookeeper(not the agent part) by following the steps at
   https://github.com/Microsoft/mesos-log/blob/master/notes/deployment.md
2. Set up the agent by following the steps at
   https://mesos.apache.org/documentation/latest/windows/
3. Try to create application notepad.exe via marathon,
   {
   "id": "1",
   "cmd": "notepad",
   "cpus": 1,
   "mem": 128,
   "disk": 0,
   "instances": 1
   }.
4. Run the windows agent something as below,
   c:\mesos\build\src>mesos-agent.exe --master=zk://40.118.201.232:2181/mesos --work_dir=c:\mesos\w --runtime_dir=c:\mesos\w --launcher_dir=c:\mesos\build\src --isolation=windows/cpu,filesystem/windows --ip=10.0.0.7

The above combination basically will give you the latest windows agent + master 1.1.0. Then when you try to run windows agent, resource.has_allocation_info() will return NULL, the check which makes sure resource has_allocation_info() will fail.

The issue will go away when master 1.2.0 is used. We need to follow up with Mesosphere to understand how to make sure old master can work with the new agent next.