feat(taiko-client-rs): migrate prover from taiko-client#21788
Draft
davidtaikocha wants to merge 31 commits into
Draft
feat(taiko-client-rs): migrate prover from taiko-client#21788davidtaikocha wants to merge 31 commits into
davidtaikocha wants to merge 31 commits into
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…ntions Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…encoding Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…genesis anchor read Address final review findings: - request_proof retries on any transient raiko/RPC/enrichment error instead of dropping the proposal, matching the Go prover's inner constant backoff. - read genesis anchor state at block number 0 (not block hash 0x0) and propagate errors; short-circuit the batch-0 engine lookup like Go ProposalLastBlockID. - commit the dedup cursor only after the fallible Proposed-handler reads succeed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Port #21782: skip the ZK proof path and fall back to the base proof when a proposal is more than maxZKProofProposalDistance ahead of the last finalized proposal (default 30), so a slow ZK proof does not block catch-up. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Drop the two dead --backoff.* flags (never read; request retries use --prover.proofPollingInterval) and --prove.gasLimit (prove gas is always estimated by the tx-manager). No behavior change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Behavior-preserving cleanups (no public interface or behavior changes): - state: fetch_update replaces the hand-rolled compare_exchange loop - cache: extract shared `cloned` helper for the duplicated map fetch - producer/compose: extract `join_base_and_geth` for the base/sgxgeth join - submitter: add ProofRequestMeta::from_request + unexpected_proof_type helpers and a shared test-harness builder - buffer: simplify the overflow guard - prover, tx_manager_adapter, commands/prover, raiko/prove_e2e tests: drop redundant bindings, reuse helpers, inline a trivial wrapper fmt + clippy (incl. missing_docs) clean. `just test` passes except the pre-existing prover::prove_e2e::single_proposal_is_proven timeout, which fails identically on the unmodified baseline. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The prover e2e tests proposed from a different account than the prover with an empty localProposerAddresses. With the deployed 4-hour proving window, the prover routed each proposal to WaitForExpiry(window + 72s) and never proved within the 90s timeout, so single_proposal_is_proven timed out. Register the proposer as a local proposer so the prover treats its proposals as assigned and proves immediately, mirroring the Go prover test's LocalProposerAddresses setup. Also restore the per-branch routing INFO logs the Rust port had dropped (received event / submit-now / wait-for-expiry) to match the Go prover's observability. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…oofs aggregate_proofs_by_type cleared the buffer and returned on any non-Pending error, permanently dropping already-generated proofs whose Proposed events were already marked handled (no scanner replay). Go retries all aggregation errors with constant backoff and only clears on shutdown. Mirror that: retry non-terminal errors, keeping the buffer. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
davidtaikocha
force-pushed
the
feat/taiko-client-rs-prover
branch
from
229f05f to
90fa970
Compare
base-tx-manager returns Ok(receipt) even when a prove transaction reached confirmation depth but reverted on-chain, so batch_submit_proofs counted a revert as a successful submission. Check receipt.status(): on a revert, record taiko_prover_submission_reverted, log, and return ProverError::SubmissionReverted so the batch consumer resends the proof requests (matching Go prover.go:302-314) instead of treating them as sent. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Mirror Go's withRetry submit contract: bounded in-place retry of the validate/wait/build read phase with proofs buffered, falling back to a re-request only on revert, unretryable send, or exhaustion. Treat a not-found L1 block during validation as transient (retry) instead of a permanent skip, bound wait_transition_verified to 60s, and cache a contiguous proof on buffer overflow instead of dropping it. Also: - Share the getCoreState read via rpc::Client::core_state and hoist a rpc::base_tx_manager_config builder used by prover and proposer. - Pin fee_limit_multiplier=10 to match Go's --tx.feeLimitMultiplier (the base tx-manager default was 5). - Report latest_verified_id as the finalized L2 block number (Go parity) and log the Proved checkpoint. - Unify the Proposed/Proved scanner reconnect loops; drop dead code and unused raiko payload fields. Verified: prover/proposer/rpc lib tests pass; clippy and fmt clean. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Record Go's updateProvingMetrics per-proof-type latency/count series in the Rust prover: thread a request_at start instant through the ProofProducer trait and request_validated, and emit generation_time / generation_time_sum / generated for sgx_geth, sgx, r0, and sp1 in both single and aggregation modes. Add a float_counter helper for cumulative fractional-second sums. Also clarify in docs/comments that the ZK proposal-distance gate is a Rust-only catch-up optimization with no Go equivalent, and document the absence of a periodic proposal re-scan. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Ports the prover — the last Go-only component — from
packages/taiko-clientto a newcrates/proverinpackages/taiko-client-rs, plus ataiko-client proversubcommand. This unblocks the eventual sunset of the Go client (the driver and proposer are already at parity; the driver retirement plan covers the rest).This PR is implementation + tests + docs (phases 1–4). The gated operational rollout (devnet → testnet → mainnet shadow → cutover → Go-client teardown) is tracked separately and not part of this diff.
Approach
Rust-idiomatic rebuild with the same external behavior as the Go prover (raiko API,
Inbox.provecalldata, scheduling/aggregation semantics), reusing the workspace's existing primitives:Proposed/Proved) viaevent-scanner(HTTP/WS, reorg-aware) — same pattern as the driver.base-tx-manager— same pattern as the proposer.crates/bindings(prove,encodeProveInput,getCoreState,getProposalHash).The genuinely new code is the raiko HTTP client, the per-type proof buffer/cache, the compose producers (sgxgeth + sgx/zk), and the request → buffer → aggregate → submit pipeline. The RPC-free routing core (
submitter::Pipeline) is split out so it's unit-testable without a node.Parity highlights
cast abi-encodegolden fixture.file:linereferences in comments.--l1.privateprivate-mempool submission is deferred post-cutover.--prover.shadowMode(full pipeline, no L1 submit) for the mainnet shadow gate.Scope decisions (intentional)
--l1.privateprivate-mempool tx manager yet (deferred).op/nativeproof types,anchor_tx_validator, and ontake/pacaya bindings not ported (dead/unused in the Go prover).Tests
crates/prover(raiko status mapping, buffer/cache semantics, producers via stub raiko, calldata golden fixture, submitter pipeline routing/aggregation, scheduling matrix, startup-cursor clamp) + 13 client-bin tests, all green;clippy -D warnings -D missing_docsclean.crates/prover/tests/prove_e2e.rs): driver + proposer + prover in dummy mode (single-proposal proof, two-proposal aggregation). Requires the Docker harness — runs in CI viataiko-client-rs--test.yml; not exercised locally.A final adversarial review (Go-vs-Rust, consensus-adjacent paths) found and fixed 3 correctness issues before this PR: transient proof-request failures now retry instead of dropping the proposal; genesis anchor state reads at block number 0 (not block hash
0x0); the dedup cursor commits only after fallible reads succeed.See
docs/prover-migration-runbook.mdfor the flag mapping and rollout/rollback procedure.🤖 Generated with Claude Code