Currently, testground has 3 runners:

• local:exec which "just" starts a binary on the machine,
• local:docker which relies on docker containers running on the machine,
• cluster:k8s which relies on Kubernetes.

One of Testground promises (to be clarified) is to let a user run a test instance on a "mock network". You might use these feature to simulate custom topologies, network latency, etc (see https://docs.testground.ai/concepts-and-architecture/sidecar).

But:

• it might be too difficult to hold that promise in 100% of cases (bugs such as limit packet size to typical MTUs used on the public internet #1385),
• we might also want to run tests going through the "real" internet (benchmarks, ProbeLab, etc),
• and we might want to compare testground results on a mocked network with testground results on real networks ("testing the testground").

One solution might be to add remote runners to testground which would let you (the user) run a test by ssh'ing on different machines you control and use the internet and real networks as the test.

Steps

• Clarify what we need from remote:exec, remote:docker,
• Define an architecture (are we using simple ssh, docker swarm, etc),
• Plan and implement.