Skip to content

op-deployer: run live bootstrap and apply through netchef's proxyd before the v0.9 RC #23971

Closed

Description

@nonsense

#23664 moves live apply and bootstrap onto forge: a keyless run on a fork of the L1 head, then a replay of the recorded transactions through txmgr. Three of its behaviours only show up behind a load-balanced RPC such as proxyd, and CI covers none of them, because CI sends only to anvil:

  1. proxyd answers 403 to some methods that forge calls.
  2. A read can reach a backend that lags the one that took the send. The next script forks at or after the block of the send.
  3. A backend can report a stale pending nonce. op-deployer waits up to 2 min for the deployer's pending transactions.

Also watch one more spot. txmgr reads the nonce once per script with NonceAt(latest), for the script's first transaction. That read comes after op-deployer's own nonce check. A backend that is behind the last send makes the signer refuse (deployment nonce changed) and stops apply. That fails safe, but it fails the run.

Run

op-deployer from develop or the RC, against Sepolia through the proxyd L1 endpoint that netchef uses, with a throwaway funded key:

  • bootstrap superchain
  • bootstrap implementations
  • init and apply --deployment-target live for one chain with additional dispute games
  • apply again, which should send nothing

Pass

  • Every step completes, with no deployment nonce changed or nonce mismatch.
  • Fork reruns (forge run on the L1 fork failed; running it again) and waits (waiting for the L1 head, waiting for the deployer's pending transactions) are rare and short.
  • No 403 or 429 on a method that op-deployer needs.

This blocks the v0.9 op-deployer release candidate, before netchef picks it up.

Context: #23664 (comment)

Activity

  1. nonsense commented on Nov 12, 2026

    @nonsense
    ContributorAuthor

    Ran it on 2026-11-12: op-deployer built from develop at 72fcf94, against Sepolia through the proxyd L1 endpoint that netchef uses, with a throwaway key. A local proxy recorded every JSON-RPC call (1,853 in total) with its HTTP status and error.

    Step Result Time Sends
    bootstrap superchain ok 81 s 4
    bootstrap implementations ok 257 s 11 (the other 21 implementations were already at their CREATE2 addresses on Sepolia)
    init + apply --deployment-target live, with CANNON (254) and ALPHABET (255, made respected) as additional games ok 250 s 6; both games are registered in the DisputeGameFactory
    apply again ok, sent nothing 3 s 0 (deployer nonce 21 before and after)

    Against the pass criteria:

    • Every step completes: yes. No deployment nonce changed. nonce mismatch appears once, only as the reason for one pending-transaction wait (latest 17, pending 18), which cleared on the next poll.
    • Fork reruns and waits are rare and short: no fork reruns and no re-simulations. All waits were in apply: three waiting for the L1 head (a receipt reported a block one ahead of the head the RPC returned) and one waiting for the deployer's pending transactions. Each cleared on the next 2 s poll.
    • No 403 or 429 on a method op-deployer needs: no 429s. The only 403s were forge's probes anvil_nodeInfo (138) and eth_getAccountInfo (9), which the endpoint doesn't allow. forge fell back to standard methods without failing. Every other method returned 200.

    Other RPC errors, all absorbed:

    • 8 already known and 1 nonce too low on txmgr rebroadcasts of transactions that had already been mined.
    • 1 block not found on the deployer nonce read at the head block, which the wait loop retried.

    Cost: 385M gas at 1 gwei, or 0.385 Sepolia ETH.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions