Option to parse existing Deployment object to generate the one for telepresence about telepresence

[itamarst]

Jean-Paul expressed a preference for doing this on level of config files, rather than talking to Kubernetes cluster, since it gives more visibility over how things are done.

[exarkun]

An idea in the opposite direction might be to have telepresence actually edit the existing Deployment for the user. A command line:

$ telepresence --deployment foo --container bar --new-shell

could patch foo so that its bar container uses the telepresence image and then start a shell in the proxied environment for that container. When the shell exits, telepresence could even roll the deployment back.

That seems kinda snazzy to me though it may be sufficiently automagical that there are some unwanted obscure corner-cases or side-effects.

[itamarst]

Current plan:

$ telepresence --swap-deployment existingservice --run-shell

This swaps out Telepresence proxy for existing pod, and reverts on shutdown.

If you have more than one container:

$ telepresence --swap-deployment existingservice:containername --run-shell

Swap out is done by getting kubectl get deployment --export version of JSON, switching replicas to 1 and image to our image.

My first thought on how revert is to be done was using the apply mechanism: at beginning of process the current revision is recorded, and kubectl rollout undo is used to rollback. Deployments created with both kubectl apply and kubectl create appear to start with a revision, so presumably everything will have a revision. However, this doesn't appear to undo change to number of replicas.

This suggests a different mechanism: the same logic that does the swap out could be used for the swap back, we just need to remember old image name and replicas until shutdown.

[itamarst]

Remaining work for swap-deployment branch:

• Get all tests passing (hopefully just dealing with fiddly bits in Runner.check_call).
• Test against OpenShift, make sure it works.
• Make sure no k8s resources are leaked by telepresence or the test suite.
• Update documentation and asciinema video (that could be done in separate branch, though.)