Gather and/or duplicate metrics container

[marminthibaut]

We have to choose between:

• gather all statistics (cpu, blkio, container, etc.) into a unique and big event
• remove the container event, and duplicate container information to all other metrics

We should compare to topbeat and packetbeat to see their logics.

[erwanncloarec]

Well in both Topbeat & Packetbeat, the events are separated. In Topbeat, the events are even seperated per process! I don't see why we should not do the same thing.
It is a benefit because it is clearer in ES when it is seperated, and more easy to query.
We can also have a benefit if there is a metric that no longer exist, we can still have the others.

[marminthibaut]

I think we could adopt the second choice: remove the container event, and duplicate container information to all other metrics.

[marminthibaut]

Nothing to do:

we keep the five types:

• container
• memory
• cpu
• net
• blkio