Replace local testnet script with Kurtosis

[jimmygchen]

Issue Addressed

The local testnet script has become quite difficult to maintain over time and using Kurtosis for local testing is now the prefer approach.

This PR replaces the scripts with a kurtosis config file and runs the same test on CI.

TODO

• Migrate doppelganger tests to kurtosis
• Remove lcli commands that are no longer needed (e.g. new-testnet, insecure-validators etc)
• Update Lighthouse book and README

Future considerations

• Add assertoor tests and utilise assertoor github action for CI.

[realbigsean]

legend

[jimmygchen]

I didn't include DEBUG_LEVEL and VC_COUNT, as they are already in network_parameters.yaml and is easy to modify if needed rather than adding a bunch of flags.

Some reasoning for the default values:

• BUILD_IMAGE=true: when testing locally, we usually want to test the code in working directory, so this is on by default. If anyone wants to use latest image, can also set this to false and update the image in network_parameters.yaml file
• CI=false: when testing locally, a block explorer and metrics explorer is usually handy but not useful on CI.

[pawanjay176]

Wondering if its easier from the end user perspective if BUILD_IMAGE is the actual docker image that the user wants to use, and if nothing is specified, then we build the local image and use it?

e.g.

# Uses the remote docker image
./start_local_testnet -b ethpandaops/lighthouse:unstable
# Builds local image and uses that
./start_local_testnet

[jimmygchen]

I think the boolean flag may still be useful in the case of local testing, because if a user already have a local image and want to re-use it, then they could run it without rebuilding, and sometimes they would want to rebuild.

My thoughts around designing the params is to prefer the use the network_parameters.yaml file where possible - this essentially replaces the vars.env - and avoid having config in two places for simplicity and not having to worry about priority.

Same reason why I removed DEBUG_LEVEL and VC_COUNT from the scripts. I'm not a big fan of maintaining large shell scripts and would love to keep functionalities as small as possible.

[pawanjay176]

The issue with the doppelganger tests was that the doppelganger VC was starting before epoch 1 and we currently don't support that prior to epoch 1 . See

lighthouse/validator_client/src/doppelganger_service.rs

Lines 407 to 416 in 3058b96

	let remaining_epochs = if current_epoch <= genesis_epoch {
	// Disable doppelganger protection when the validator was initialized before genesis.
	//
	// Without this, all validators would simply miss the first
	// `DEFAULT_REMAINING_DETECTION_EPOCHS` epochs and then all start at the same time. This
	// would be pointless.
	//
	// The downside of this is that no validators have doppelganger protection at genesis.
	// It's an unfortunate trade-off.
	0

Fixed that to start the VC exactly on epoch 1 and its working fine. Although we still need to capture the exit code correctly i think.

[pawanjay176]

Looks good to me overall. Just a few minor nits.
I'm happy with merging this as is and addressing them in the subsequent docs PR.

[pawanjay176]

Wondering if its easier from the end user perspective if BUILD_IMAGE is the actual docker image that the user wants to use, and if nothing is specified, then we build the local image and use it?

e.g.

# Uses the remote docker image
./start_local_testnet -b ethpandaops/lighthouse:unstable
# Builds local image and uses that
./start_local_testnet

[jimmygchen]

@mergify queue

[mergify]

queue

✅ The pull request has been merged automatically

The pull request has been merged automatically at 5fc0145

[jimmygchen]

Thanks for the review @pawanjay176 , happy to discuss / address the comments above further.