columns: a corrupt parent state is re-read once per sidecar of a group since #788 met #754

オープン
#839 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 6 日以内に返信

まだ誰も着手していません。

評価

難易度
3/5
見積もり時間
1〜2日
初心者へのやさしさ
78/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
rust

調査の方向性

Start with ChainPrechecks::judge_header in crates/blockchain/state_transition/src/beacon/gossip/column.rs and the StateRebuilds documentation in crates/storage/src/state_writer.rs. Add a test showing that a corrupt parent state is read once for a multi sidecar group, while other read errors still retry. Update the reason="internal" description for lean_data_columns_rejected_total in docs/metrics.md, then run the relevant state transition tests.

索引モデルが issue の本文から書いたものです。

説明

Problem. Merging main into lambdaclass/ethlambda#788 put the PR's "a failed parent-state read drops the sidecar" rule inside main's grouped column prechecks (#754). There, ChainPrechecks::judge_header (crates/blockchain/state_transition/src/beacon/gossip/column.rs) does not remember an Err from get_state. That is deliberate, so that one transient read error does not drop the whole block: the group's next sidecar reads the state again.

For a corrupt parent state, though, every one of those reads fails. A group then re-reads it once per sidecar, one after another on the group's permit, up to the block's 128 columns. Each attempt is a full failed rebuild (about a second of CPU, by the StateRebuilds doc in crates/storage/src/state_writer.rs) plus an error! line. That works against #788's own rule that a corrupt root is read once and shared (Error::CorruptState). Verdicts are unaffected: every sidecar is dropped as Ignore(Internal) either way. Only on-disk corruption triggers it.

A related doc gap: the lean_data_columns_rejected_total section of docs/metrics.md still says reason="internal" counts the sidecars that a panic in the checks cost. Since the merge, it also counts sidecars dropped for a failed parent-state read.

Why it is a follow-up. #788 is approved and in the merge queue. The re-read only costs CPU on a node whose store is already corrupt.

Fix.

  • In judge_header, remember the drop for the block's parent when the error is ethlambda_storage::Error::CorruptState, so the rest of the group is dropped without reading again. Keep reading again on any other Err.
  • Add a test: a corrupt parent state is read once for a multi-sidecar group.
  • Add the failed-read case to the reason="internal" text in docs/metrics.md.
主要言語
Rust
スター
86
フォーク
34
平均マージ
9日 20時間
マージ済み PR(30日)
77

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

lambdaclass/ethlambda のほかの issue

lambdaclass/ethlambda の issue をすべて見る

似ている issue

Rust の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。