From d40b2afae07e6f44ee26eabb2278f59e3ecc19e5 Mon Sep 17 00:00:00 2001 From: Bogdan Szabo Date: Sun, 1 Nov 2026 09:09:06 +0100 Subject: [PATCH] A pushed document's signatures are found through the evidence's own document column `signature_keys_of_document` reached the document through `chunks`, and `fact_evidence.chunk_id` has no index, so every extracted pushed document scanned all of `fact_evidence`: 12 s a document on 8.3 M rows, 28 to 30 copies at once, the database at its CPU limit for the length of a backfill. `fact_evidence.document_id` is indexed and equals the chunk's document, so reading it gives the same rows in 0.16 s. Refs deeplethe/utopia#1471. Co-Authored-By: Claude Fable 5.1 --- crates/utopia-store/src/phrase_bindings.rs | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/crates/utopia-store/src/phrase_bindings.rs b/crates/utopia-store/src/phrase_bindings.rs index 502a739ca..928692d83 100644 --- a/crates/utopia-store/src/phrase_bindings.rs +++ b/crates/utopia-store/src/phrase_bindings.rs @@ -115,7 +115,12 @@ pub async fn signatures<'e>( } /// 一篇文档里活着的开放陈述落在哪些签名上（键同 [`PhraseSignature::key`]）。推送来的 -/// 文档抽完，本体只判这些签名（#1242）；别的签名还是对齐器的 +/// 文档抽完，本体只判这些签名（#1242）；别的签名还是对齐器的。 +/// +/// 证据按 `fact_evidence.document_id` 找（`fact_evidence_document_idx`），不经 `chunks`： +/// `fact_evidence.chunk_id` 没有索引，经 chunk 找要全表扫一遍证据，830 万行的库上每篇 +/// 文档 12 秒，推送一阵时几十份同时跑，把数据库顶在 CPU 上限（#1471）；按文档列走索引 +/// 是 0.16 秒，行一样——证据的 document_id 写入时就等于它 chunk 的文档 pub async fn signature_keys_of_document( pool: &PgPool, kb_id: Uuid, @@ -128,8 +133,8 @@ pub async fn signature_keys_of_document( LEFT JOIN entities o ON o.id = f.object_id WHERE f.kb_id = $1 AND f.layer = 'open' AND f.invalidated_at IS NULL AND f.phrase IS NOT NULL AND btrim(f.phrase) <> '' - AND EXISTS (SELECT 1 FROM fact_evidence fe JOIN chunks c ON c.id = fe.chunk_id - WHERE fe.fact_id = f.id AND c.document_id = $2)", + AND EXISTS (SELECT 1 FROM fact_evidence fe + WHERE fe.fact_id = f.id AND fe.document_id = $2)", phrase = phrase_sql("f.phrase") ); Ok(sqlx::query_as(&sql)