@nonsense here's a suggested approach for removing Redis and replacing it by a custom implementation. I left 2 comments on the places that I'm most unsure about. I didn't test it with testground yet, but as tests are passing here it should work. We also need to check the performance.

Do you think you could leave a review up until Wednesday?

Looking at the PR on the main repository, it seems that all tests work out well with this version. We still need to do the stress tests to make sure there's no performance regressions.

@hacdias this LGTM. I added a few comments. I also think we should test with Testground as well, and not rely solely on the tests here. This is a new service, and not yet battle-tested, so that we rely so much on the unit tests.

Could you test with:

• local:docker
• local:exec
• cluster:k8s (lotus-soup and storm and network/ping-pong testplans should be enough - I can help with that)

I will test the service with the different runners next week. I may need some help with k8s. For the record, I tested with local:docker before opening this PR and it worked. But I will make sure to test again with both locals.

In addition to testing, I want to mention that testground/testground#1187 is using this PR and testing those runners. However, I am now going to test the local runners.

Tested with:

• local:docker
  • network ping-pong 2 instances
  • splitbrain accept 20 / 50 instances
• local:exec
  • example sync 20 / 50 / 100 instances

(can be updated)

@nonsense executed network/ping-pong and example/sync (100 instances) on k8s without issues. Lotus _compositions/baseline-k8s-1-1.toml also run well.