Allow the namespace path to be container ID#224
Allow the namespace path to be container ID#224laijs wants to merge 1 commit intoopencontainers:masterfrom
Conversation
In numerous cases, we need to share the container's namespace with an existing container. In these cases, we have to resolve the namespace file step by step. For example, the docker resolves it from the information of the target container in several steps and saves it to the config (now it is in-memory config, it will be runtime.json in future). We can move the resolving operation to the oci implementations and allow the namespace path to be container ID in the spec. Reasons: 0) it really repsents "sharing the namespace with existing container". (it is the most important reason.) 1) the administrator writes the runtime.json much conveniently and less error. And it is better readability and the administrator is easier to check the runtime.json which is written by himself or by other human or by other tools. 2) we may have more orchestra tools than oci implementations, the resolving operation is better in implementations than in the tools. Signed-off-by: Lai Jiangshan <jiangshanlai@gmail.com>
|
I really don't want to have magic resolution in the runtime.json like this. If we consider this feature we should have the runtime resolve the naming. |
|
This sounds useful, but I'd turn it around. Instead of making it a #!/bin/sh Then you could use: "path": "/run/opencontainer/containers/{container-id}/ns/{namespace}" in another container's runtime.json to join any of the initial |
|
Ok, lets close this one seeing that higher level tooling can generate these paths based on container IDs |
|
On Thu, Oct 22, 2015 at 04:16:29PM -0700, Michael Crosby wrote:
We should probably start collecting a list of these punted features, |
In numerous cases, we need to share the container's namespace with
an existing container. In these cases, we have to resolve the namespace
file step by step. For example, the docker resolves it from the
information of the target container in several steps and saves it to the
config (now it is in-memory config, it will be runtime.json in future).
We can move the resolving operation to the oci implementations
and allow the namespace path to be container ID in the spec.
Reasons:
(it is the most important reason.)
less error. And it is better readability and the administrator
is easier to check the runtime.json which is written by himself
or by other human or by other tools.
the resolving operation is better in implementations than
in the tools.
Signed-off-by: Lai Jiangshan jiangshanlai@gmail.com