Productionize blockchain reactor v2 [Tracking Issue]

[tessr]

The blockchain reactor v2 has been around for almost a year, but it hasn't been fully tested, and there are a still a number of outstanding follow-ups around it:

• Queue configuration [blockchain] Routine queue configuration in v2 #4206
• Event documentation [blockchain] Document events in blockchain v2 #4201
• Possible invariant violation blockchain/v2: (*scheduler).maxHeight() internals disagree with its API do/invariant #5844
• Lock misuse blockchain/v2: (*BlockchainReactor).{Add, Remove}Peer misuse lock by acquiring a read lock then writes to the critical section acquired channel #5842
• Lock misuse 2 blockchain/v2: (*BlockchainReactor).Receive can leave code locked before exiting, and also misuses Read locks #5843
• Issues unearthed in formal verification blockchain/v2: issues with the current protocol #5496

Perhaps:

• Consider light client verification for fast sync Light client verification for fast sync #4457

(I think there are more; will add to this tracking issue as I find them.)

We'd like to make v2 the default blockchain reactor in Tendermint 1.0, but we need to handle these outstanding issues first.

[alexanderbez]

I would be happy to tackle these :)

[tessr]

OK, nice! I will put your face on it, speculatively

[cmwaters]

I think we should also consider light client verification for fast sync: #4457

[melekes]

#5496 is probably the most important issue

[tessr]

Thanks, will add these both ^

[tac0turtle]

I'll be the one to say it.

Let's remove v2. We can focus on v0 and look into a simplification of v0 if need be.

[alexanderbez]

I support that. Unless someone really wants to take ownership and gain domain expertise over v2, I say we should be OK with removing it.

[tac0turtle]

Seems like there is general agreement for removing the protocol. I would propose we close all issues, mark v2 as deprecated in 0.35 and remove it in 0.36?

@ebuchman, as your team spent many months on this work, I would love to have your support in this work.

[ebuchman]

What is the motivating reason for abandoning v2 for v0? It seems there is some missing context between the comment about #5496 being the biggest issue (a protocol issue which i believe also exists in v0 and would be easier to solve in v2), and the comment about removing it and "simplifying" v0. Is there an issue somewhere that can fill in this context? Maybe something about problems/reliability issues using it to sync on a real network?

[tac0turtle]

there have been a few discussions within the tendermint team throughout the past year. Back when anton and erik were still working on the project, they agreed v2 is unnecessarily complex, does not add performance or understanding to the protocol. V0 is simple and has fewer issues than v2 in the time it has been in the repo. We have had many messages from users that they are running into issues with v2, while with v0 we still get messages, but way less frequent.

This led us to removing users the ability to use v2 (merged in 0.35), shortly after that a few team members started thinking what if we removed v2 and there seem to be agreement around it.

[tac0turtle]

V2 has been removed