[tmpnet] Enable single node networks #3003
Conversation
| nodes, err := tmpnet.NewNodes(tmpnet.DefaultNodeCount) | ||
| require.NoError(ginkgo.GinkgoT(), err) |
There was a problem hiding this comment.
q: why panicking instead of using the require infra? No action required
There was a problem hiding this comment.
If this were only for e2e it makes sense, but I was looking to support antithesis testing as well as e2e and we're not using assertions because failing fast isn't the pattern. The process running the tests needs to run continuously for an arbitrary length of time and error logging is the signal that something is wrong.
There was a problem hiding this comment.
Hmm, I think I'm confusing things with that mention of 'failing fast'. That's the reason to not use require, but it doesn't explain why I'm using panic. I wanted to enable a simpler declarative form for network definition e.g.
network := tmpnet.Network{
...
Nodes: tmpnet.NewNodesOrPanic(tmpnet.DefaultNodeCount),
} The counterpoint is that any network with subnets requires associating nodes to subnets so this simple declarative form isn't sufficient.
So maybe this approach is best described as an experiment to see if simpler panic-based test setup helpers are a good idea or not. My thinking is that this approach is compatible so long as we're either going to use these functions at the start of an antithesis or load test, where failure to perform test setup should exit the process, or with a test runner like ginkgo where a panic at any point during test execution will fail the test but not halt execution of subsequent tests.
Why this should be merged
Single-node networks can be useful for development.
How this works
Allow the number of nodes to be configured with
--node-countto the e2e suite and tmpnetctl and ensure--sybil-protection-enabled=falseis configured when a network is started with a single node.How this was tested
CI