hoodi gas mismatch

[sudeepdino008]

May 04 08:50:14 snapshotter-bm-v34-hoodi-n1 erigon[351005]: [WARN] [05-04|08:50:14.642] [4/6 Execution] Execution failed         block=2740282 txNum=99022106 header-hash=0xd402cceeb035d561d71ec726bd2f743bbbc5bc39d669fcf7f1268e7a597297f3 err="invalid block, txnIdx=33, gas used by execution: 2973809, in header: 2901285, headerNum=2740282, d402cceeb035d561d71ec726bd2f743bbbc5bc39d669fcf7f1268e7a597297f3" isForkValidation=false isBlockProduction=false
May 04 08:50:14 snapshotter-bm-v34-hoodi-n1 erigon[351005]: [INFO] [05-04|08:50:14.643] [4/6 Execution] serial done              in=15.889734ms buf=383.9KB/1.0GB blk=2740281 blks=1 blk/s=62 txs=0 tx/s=0 gas/s=281.01M stepsInDB=0.50 step=253.5 alloc=10.6GB sys=16.4GB isForkValidation=false isBlockProduction=false isApplyingBlocks=true
May 04 08:50:14 snapshotter-bm-v34-hoodi-n1 erigon[351005]: [WARN] [05-04|08:50:14.643] Cannot update chain head                 hash=0x5c2cf5d601bb13f43e3eecf7b63ad248d85655e013adfcfc4d17f4cca82768ad err="updateForkChoice: [4/6 Execution] invalid block, txnIdx=33, gas used by execution: 2973809, in header: 2901285, headerNum=2740282, d402cceeb035d561d71ec726bd2f743bbbc5bc39d669fcf7f1268e7a597297f3"

hoodi v34 snapshotter...

[VBulikov]

yperbasis added Imp1 as Importance

[VBulikov]

yperbasis changed Importance from Imp1 to Imp0

[awskii]

confirmed on another node with v3.4

  [WARN] [4/6 Execution] Execution failed
         block=2789993
         txNum=100285188
         header-hash=0xc4c754565911f13b96a841782095a21a32806f927307ffa71695ecd99801c236
         err="invalid block, txnIdx=34,
              gas used by execution: 4352410,
              in header:             4279886,
              headerNum=2789993,
              c4c754565911f13b96a841782095a21a32806f927307ffa71695ecd99801c236"
         isForkValidation=false
         isBlockProduction=false

  [WARN] Cannot update chain head
         hash=<varies — caplin keeps offering new head hashes>
         err="updateForkChoice: [4/6 Execution] invalid block, txnIdx=34,
              gas used by execution: 4352410, in header: 4279886,
              headerNum=2789993, c4c754565911f13b96a841782095a21a32806f927307ffa71695ecd99801c236"

  [WARN] error executing clstage
         app=caplin
         stage=ForkChoice
         err="failed to compute and notify services of new fork choice:
              failed to run forkchoice: bad block as forkchoice"

Key facts to reproduce:

• chain: hoodi
• block: 2,789,993
• header hash: 0xc4c754565911f13b96a841782095a21a32806f927307ffa71695ecd99801c236
• txnIdx: 34
• txNum (cumulative): 100,285,188
• gas Erigon computes: 4,352,410
• gas in header: 4,279,886
• Δ: 72,524 gas (Erigon overcounts)
• erigon version: 7acc94e (v3.4.1-5-g7acc94e, release/3.4)

[JkLondon]

Root cause (hoodi snapshotter, release/3.4, block 2 789 993 invalid)

release/3.4 is missing the unwind-correctness fixes that landed in main (notably bfa03df parallel-commitment correctness for reorg/unwind + SD recreate; e0851b7 stale slot resurrection fix was applied on r3.4 and then reverted twice). As a result, when a sidechain block at the tip is unwound, contracts created in it via CREATE/CREATE2 are not removed from the latest-state domains (AccountsDomain/CodeDomain) in MDBX — code+nonce remain.

What happened concretely on hoodi:

At height 2 789 992 the node briefly applied a sidechain block 0x706b073b… containing a CREATE2 deploy of an EIP-1167 minimal proxy at the counterfactual address 0xd3b61948….
Canonical 0x72c42d17… then took over; unwind ran but left code (45 bytes, the EIP-1167 stub pointing to impl 0x6bb86b3b…) and nonce=1 behind on 0xd3b61948….
Canonical block 2 789 993 has a tx (0x73652205…, idx 7) that performs the same CREATE2 to the same address. With the phantom state present, EIP-684/EIP-1014 collision triggers consume-all-gas: status=0, gasUsed=243 528 (== gas limit) instead of canonical 171 004 → block gas overshoot by exactly +72 524 → gas used mismatch → forkchoice rejected.
Proofs:

eth_getCode(0xd3b61948…, 2789992) = 45-byte EIP-1167 stub locally; 0x on every canonical block up to and including 2 789 992.
Local hash at 2 789 990 / 2 789 991 matches canon; 2 789 992 mismatches (local sidechain 0x706b073b…).
The phantom appeared between blocks 2 789 991 and 2 789 992 — recent enough to still be in MDBX, not yet merged into any sealed .kv snapshot.
Triggers: any reorg at the tip touching a CREATE2 deploy to a counterfactual address (safe wallets, clone factories / EIP-1167, ERC-4337 accounts, deterministic deployers). Will recur on r3.4 on every such reorg.

Recovery without backport: integration reset_state (clears MDBX domain tables only, leaves .kv snapshots intact, re-executes from snapshot tip along canonical chain — bypasses the broken unwind path entirely).

Proper fix: backport bfa03df (adapted, since r3.4 lacks the parallel-commitment infra) + investigate why #20494 was reverted on r3.4 before re-applying its equivalent.

[JkLondon]

root cause fix merged, snapshotter is fine. Feel free to reopen