`eth_getWitness` only errors

[2xic]

System information

Erigon version: erigon version 3.3.0-dev-f3a2b0d3 (I run into this issue on other 3.1.0+,3.2.0+ builds also)

OS & Version: Ubuntu 24.04.3 LTS

Commit hash: f3a2b0d

Erigon Command (with flags/config):

Consensus Layer: Caplin

Consensus Layer Command (with flags/config): default

Chain/Network: Mainnet

Expected behaviour

eth_getWitness gives me a response.

Actual behaviour

eth_getWitness gives me a error.

Steps to reproduce the behaviour

Start rpcdeamon and try to curl a recent block

./build/bin/rpcdaemon --http.api=eth,debug,trace --datadir /home/erigon-datadir

The request

curl -X POST http://localhost:8545 -H "Content-Type: application/json" --data '{"jsonrpc":"2.0","method":"eth_getWitness","params":["0x16a48a4"],"id":1}' 

Response

{"jsonrpc":"2.0","id":1,"error":{"code":-32000,"message":"method handler crashed"}}

Backtrace

EROR[11-09|13:57:44.216] RPC method eth_getWitness crashed: interface conversion: *temporal.Tx is not kv.TemporalRwTx: missing method Append
[service.go:222 panic.go:787 iface.go:102 iface.go:474 memory_mutation.go:810 stage_execute.go:304 stage_execute.go:204 stage_execute.go:420 stage_execute.go:377 stage_witness.go:117 eth_call.go:667 eth_call.go:553 value.go:581 value.go:365 service.go:227 handler.go:529 handler.go:479 handler.go:420 handler.go:240 handler.go:333 asm_amd64.s:1700]

[2xic]

I already tried to debug this myself as I was hoping it would be a simple fix, but there are some layer of indirection that I don't understand. For instance the RPC will start with as a RO tx

erigon/rpc/jsonrpc/eth_call.go

Lines 655 to 661 in f3a2b0d

	roTx2, err := db.BeginRo(ctx)
	if err != nil {
	return nil, err
	}
	defer roTx2.Rollback()
	txBatch2 := membatchwithdb.NewMemoryBatch(roTx2, "", logger)
	defer txBatch2.Rollback()

Which then calls into

erigon/execution/stagedsync/stage_witness.go

Line 94 in f3a2b0d

func RewindStagesForWitness(batch *membatchwithdb.MemoryMutation, blockNr, latestBlockNr uint64, cfg *WitnessCfg, regenerateHash bool, ctx context.Context, logger log.Logger) error {

Then

erigon/execution/stagedsync/stage_execute.go

Line 348 in f3a2b0d

func UnwindExecutionStage(u *UnwindState, s *StageState, doms *execctx.SharedDomains, rwTx kv.TemporalRwTx, ctx context.Context, cfg ExecuteBlockCfg, logger log.Logger) (err error) {

So now we are marked as rwTx kv.TemporalRwTx, alright, I guess that is fine as long as the operations are meant to be handled in memory only and not actually commit anything.

After a few steps we end up at

erigon/execution/stagedsync/stage_execute.go

Lines 304 to 306 in f3a2b0d

	if err := tx.Unwind(ctx, txUnwindTo, changeset); err != nil {
	return err
	}

Which then actually calls into

erigon/db/kv/membatchwithdb/memory_mutation.go

Lines 809 to 811 in f3a2b0d

	func (m *MemoryMutation) Unwind(ctx context.Context, txNumUnwindTo uint64, changeset *[kv.DomainLen][]kv.DomainEntryDiff) error {
	return m.db.(kv.TemporalRwTx).Unwind(ctx, txNumUnwindTo, changeset)
	}

Now we are trying to cast a RO into RW which fails ... I thought something as simple as not doing the rewind could work (just do return nil instead return m.db.(kv.TemporalRwTx).Unwind(ctx, txNumUnwindTo, changeset)), but then I get

{"jsonrpc":"2.0","id":1,"error":{"code":-32000,"message":"mismatch in the short node keys node1.Key(000f0a0e080a05060e0e04000d060b0708050c040a0302070b00090a0b0f05040b050d0c0707030405030a0d050b080a0004060e04060209080a0d00000e0c08050f0e0e020e010e0405020a0e060f020808020c0607070f090a090f09060a020b0d0107030c0a0903090000040e0b0905090300040c0608)!=node2.Key(000f0a0e080a05060e0e04000d060b0708050c040a0302070b00090a0b0f05040b050d0c0707030405030a0d050b080a0004060e0406020910090300040c0608100f0e0e020e010e0405020a0e060f020808020c0607070f090a090f09060a020b0d0107030c0a0903090000040e0b0905090300040c060810)"}}

So I'm assuming that there is some operation that needs to be done. Although because of some of the abstractions here I'm not able to fully understand the entire flow here.

Happy to try to open a PR with a fix if sufficient pointers are given!

[yperbasis]

TODO: check if geth/nethermind support eth_getWitness. If not, we should remove it.

[VBulikov]

yperbasis added Imp2 as Importance

[yperbasis]

Also is it required for zkEVM?

[yperbasis]

@2xic how do you use eth_getWitness? We've recently patched eth_getProof. Can you switch to it?

[2xic]

@2xic how do you use eth_getWitness? We've recently patched eth_getProof. Can you switch to it?

I was mainly using it as a proxy for implementing debug_executionWitness in Erigon. This is what I and others commonly use for ZkVMs. The codepath would be similiar, but the output format is for instance different from the existing endpoint.

Last time I checked eth_getProof had a different code path, but will take another look later today.

TODO: check if geth/nethermind support eth_getWitness. If not, we should remove it.

They don't as far as I know, they all implement a variant of it which is the debug_executionWitness. Geth + Reth + EthRex has is on default branch. Nethermind not yet, but I have seen it implemented on some of the stateless branches they have.

[awskii]

fixed with #22077 #22099