SSSP framework not seen as one cluster on Mesos

[MiLk]

Hi,

I'm trying to run SSSP on Mesos.

I have downloaded the source code from the git repo and run play universal:package-zip-tarball to build a tarball from the source.

The tarball has been extracted on each node of my Mesos cluster.

I run SSSP with the following command:

LIBPROCESS_IP=172.16.0.180 \
MESOS_NATIVE_LIBRARY=/usr/local/lib/libmesos.so \
/opt/sssp/bin/sssp \
-Dhttp.port=9000 \
-Dconfig.file=/etc/sssp.conf

And /etc/sssp.conf contains:

include "application.conf"

mesos.master="zk://172.16.0.180:2181,172.16.0.181:2181,172.16.0.182:2181/mesos"
mesos.workers=1
mesos.mem=256
mesos.cpus=0.2

But in Mesos, I see 3 frameworks instead of one, like it is with Marathon. What is wrong?

[solidsnack]

Did you run the command more than once? You shouldn't need to unpack the tarball in more than one place...

[MiLk]

I run the command on each node, because I supposed it works like Marathon.

After more tests, I was able to run only one Marathon by using the -Dhttp.address option. Before I installed SSSP on each, because the task try to download SSSP from localhost. By setting this option, it works.

But SSSP use more memory and cpus than what is configured in the configuration file. It uses all the resources of the Mesos cluster and I'm not able to run anything with Marathon.

[solidsnack]

When you say it uses all the resources on the cluster...how much does it use? What does the Mesos console look like?

[MiLk]

I put the logs created after starting the framework here: https://gist.github.com/MiLk/10949600

I attached two screens of the Mesos panel. As you can see, it uses more than what is configured in /etc/mesos.conf.

After addind a new slave: