With the imminent return of containers in ROS 7.4, I'm interest to revisit a plan I had last year.
I'm tired of supporting customer internet services and the assortment of routers customers bring to the party.
What I'm looking towards is to host a large central router and deliver the customers LAN via VLAN to their home, no router required at customer premises.
The central router should be partitioned into separate router instances. I like ROS, and would prefer to use it for customer router instances.
This sounds like MetaRouter from times long gone... will we be able to achieve that again with the release of containers? Can I run many ROS instance in containers in a single large router?
Will this work well? Perhaps most importantly, why *shiouldn't* I do this?

7.4 has landed. Does anybody know if it's possible to launch an ROS instance in a container?

7.4 has landed. Does anybody know if it's possible to launch an ROS instance in a container?

I doubt a docker image exists for ROS or will ever exist.
Too much dependencies with kernel.

Isn't that the whole point of a container compared to a VM; it literally shares the running kernel instance with host OS... dependency on ROS kernel should be satisfied by definition, and that's the whole point using ROS as container host for ROS containers.

7.4 has landed.

…with container support removed, undermining the premise behind this thread.

However, why insist on using RouterOS as the host? The biggest badass of them all is a pipsqueak compared to a serious x86 server. 16 cores at 2 GHz? Puh-lease. Dell can outdo that, and for probably less money, even after you add all the NIC cards to get 16 interfaces.

They can't outdo it for the same wattage, but TANSTAAFL.

Point is, you can design a host to run a hypervisor, which in turn lets you deploy a pile of CHRs, which works today.

Isn't that the whole point of a container compared to a VM; it literally shares the running kernel instance with host OS... dependency on ROS kernel should be satisfied by definition, and that's the whole point using ROS as container host for ROS containers.

Not if certain functionality is too dependent from version to version.
As ROS7 is now, it would be a management nightmare to keep everything in sync with the right versions without breaking anything above or below.
Not to mention the combined resources required ...
I am still convinced there will never be a docker image of ROS.
Besides, who would make/maintain it ? The only ones able to do that, got more important things to take care of (put it differently: they should be occupied with other stuff, I certainly hope they will not waste time on this).

I fully agree with tangent's suggestion.
Clean, can be done already for a long time and a lot less restrictions.

Well MetaRouter used to be a thing. ROS as container is just that. The use cases are as compelling or even greater than before.
Hypervisor with CHR VM's feels particularly shit. Running a couple hundred VM instances compared to router containers is a HUGE distinction in resources. Also, CHR licensing is way too expensive!
The pricing for CHR is just wrong... you get an ROS license for free with cheap 'tik devices that are similarly priced to a software license alone.