Support CRUSH locations about ceph-container HOT 4 CLOSED

@Ulexus yes it's a restricted set but using several variables for each type might be really lengthy and will provide the exact same thing. Type are not going to change anytime soon, I don't think new type will come up either BUT having a single variable will be more flexible.
Thus we give the freedom to the users to build their own location.

If we go with CRUSH_ROOT, CRUSH_REGION, CRUSH_DATACENTER etc etc we will have an entrypoint like this:

root=$CRUSH_ROOT region=$CRUSH_REGION datacenter=$CRUSH_DATACENTER host=$(hostname).

In the situation where region is not (simply not required by the operator) set we have to come up with a default value to not this will also add unnecessary complexity to the crush map.
We could also have a set of if statements to check if the variable is empty or not and if so remove it from the command line. But we agree that's a terrible solution.

That's why I prefer to use:

CRUSH_LOCATION='root=default datacenter=dc1 room=1 rack=10 host=$(hostname)'

This will let the user decide and will remove complex solutions in the entrypoint.
It can even be as simple as that:

CRUSH_LOCATION='root=default host=$(hostname)', and this will likely be the default for the CRUSH_LOCATION variable.

from ceph-container.

@leseb I don't know what I was thinking. If the user can read enough to get the environment variable names, they can read enough to get the list of options. Yes, the free-form CRUSH_LOCATION makes much more sense.

from ceph-container.

hi @leseb,
I think put all bucket type and value in a CRUSH_LOCATION is good.
And here I have a question on the use case of ceph/daemon:
what's the goal of ceph/daemon? A unified docker images for all daemons?

from ceph-container.

@tangzhankun the ceph/daemon contains all the Ceph daemons, it just allows you to enable the one you want. You can only run one daemon per container. If you want a container with all the daemons running you should look at ceph/demo.

from ceph-container.