Re-schedule problem

[atibon]

I've 10 PC, and I submit a topology with 20 container, 176 component, 3 cpu per container, and topology.component.rammap is A:1073741824,B:1073741824,C:2147483648,D:2147483648,E:1073741824. If need any other config, pls tell me.

A fewer minutes later, I found that:

upload picture failed...

Resources cost is not the same as I think. I check Mesos-UI, found that, some PC has been assigned 5 container, and some PC has been assigned 1 container.

And Total CPU used in a container is bigger than 3. Maybe something is wrong.

[kramasamy]

@maosongfu - can you take a look at this?

[maosongfu]

@atibon
First question, are u using aurora scheduler or mesos scheduler?

" some PC has been assigned 5 container, and some PC has been assigned 1 container."
We rely on aurora/mesos to do the actual scheduling. So how to assign containers for PC is determined by aurora/mesos.

"And Total CPU used in a container is bigger than 3. Maybe something is wrong."
For using aurora/mesos scheduler, we pack topology request into actual aurora/mesos request to ask them to do the actual scheduling. Could u share the resource requirement on aurora ui or mesos ui? Then we can check whether it is an issue related to resource request transformation, or it is something related to aurora/mesos.

Also, I don't understand "upload picture failed..."