[Epic]: server/v2 production readiness via simapp/v2 and testing

[kocubinski]

Recent PRs (including #20412) have introduced server/v2 infrastructure and an example app, simapp/v2, using this code. In short server/v2 is a decomposition of baseapp and friends into modular, composable, parts. What is merged into main at the time of writing this issue should be considered a functional PoC. Completion of the issues enumerated here will harden the code quality of server/v2 and elevate it from PoC to production ready.

Testing

For comprehensive testing, we should adapt all integration, e2e, and sims tests currently covering runtime/v1 for runtime/v2. Perhaps the clever use of a build flag and interfaces in the right places could make this easier.

• convert simapp-v2-init to a tests/systemtests #20471
• server/v2/cometbft needs unit tests with mocks #20484
• server/v2: integration tests  #20799
• server/v2: e2e/system tests #20800
• tests/sims support for runtime/v2 #20469
  • simulation.SimulateFromSeed expects a BaseApp. Is it possible for this function to accept an abstraction which both BaseApp and something from runtime/v2 satisfy?
• v2/cometbft unit tests don't compile

  cosmos-sdk/server/v2/cometbft/abci_test.go

  Line 3 in 5bf88be

  /* TODO this has been failing for months, it needs to be re-written

• server/v2: wire grpc and gateway #20798

Breaking

• simapp/v2 start crashes with panic: proto:not found when using go.mod replaces instead of go.work #20492
• server/v2 handles genesis by processing gentxs as a genesis block, but store/v2 and iavl do not support the creation of genesis block at height 0.
  • store/v2 and iavl v1 must support the creation of state at version 0 to support server/v2 style genesis #20443
  • this hack in the init script is required to overcome it (initial height = 2)

  • cosmos-sdk/scripts/simapp-v2-init.sh

    Line 25 in 5bf88be

    jq '.initial_height = 2' genesis.json > temp.json && mv temp.json genesis.json

• refactor(x/**): rewrite ante handlers as tx validators #19949 (merged into server_modular) and possibly follow up PRs around TxValidators must be rebased onto main
• sdk.Context is not provided by runtime/v2, but required for certain tx handlers using gogo interface types Environment in context #19640

Feature Parity

• simapp/v2 is started using an ad-hoc start command handler, neither an api nor grpc server is started. this should be refactored to use ServerModule, i.e. comet node as a module, api server as a module, etc.

  cosmos-sdk/server/v2/server.go

  Line 16 in 5bf88be

  type ServerModule interface {

• server/v2/comebft queries for the paths "app", "store", "p2p", "custom" are not yet implemented #20494
• server/v2/cometbft commands should support both json and yaml output #20495
• simapp/v2 needs an implementation of CLI pruning commands #20511
• simapp/v2 needs an implementation of CLI snapshot commands #20512
• server/v1 may produce a hash of empty string instead of nil, should server/v2? #20513
• simapp/v2 needs an export implementation #20514

Code quality

• deferred initialization of routers is not thread-safe but required to resolve an initialization circular dependency.

  • cosmos-sdk/server/v2/stf/core_router_service.go

    Lines 98 to 105 in 5bf88be

    	// TODO threadsafe?
    	if m.handler == nil {
    	var err error
    	m.handler, err = m.builder.Build()
    	if err != nil {
    	return fmt.Errorf("cannot create queryRouter: %w", err)
    	}
    	}

• simapp/v2 still uses server/v1 CLI helpers for commands and config generation/parsing. These should be migrated to server/v2. server/v2 start commands and helpers #19720 has more detail.
• server/v2/stf should not manage the storage and retrieval of header info #20510

STF Todos

Beta Give feedback

1. feat(stf): investigate if recurseChanges is needed
2. feat(stf): MsgBuilder should not be lazy building handlers

[julienrbrt]

Closing this epic as https://github.com/orgs/cosmos/projects/26/views/38 is tracking what's left.