feat(server): improve doc gc (#15363)

#### PR Dependency Tree


* **PR #15363** 👈

This tree was auto-generated by
[Charcoal](https://github.com/danerwilliams/charcoal)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Bug Fixes**
* Improved validation of workspace roots and document projections, with
clearer failures for malformed or incomplete data.
  * Improved document reference rebuilding and cleanup reliability.
* Updated document update merging to better handle invalid binary data.

* **Performance**
* Avoided unnecessary document reconstruction when no updates are
pending.

* **Tests**
* Updated coverage for malformed workspace roots and document snapshot
parsing.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
This commit is contained in:
DarkSky
2026-07-28 11:20:12 +08:00
committed by GitHub
parent b05f165820
commit cfc7bbb90f
6 changed files with 80 additions and 72 deletions
@@ -1,4 +1,5 @@
use affine_doc_loader as doc_loader;
use chrono::{DateTime, Utc};
use sqlx::PgPool;
use super::{
@@ -8,11 +9,7 @@ use super::{
const PARSER_VERSION: i32 = 1;
struct ExtractedRef {
blob_key: String,
block_id: String,
flavour: String,
}
type ExtractedRef = doc_loader::BlobRef;
#[derive(Default)]
struct ProjectionState {
@@ -153,23 +150,9 @@ async fn purge_removed_doc_refs(pool: &PgPool, workspace_id: &str, current_doc_i
Ok(result.rows_affected() as i64)
}
fn extract_refs(snapshot: &CurrentDoc) -> RuntimeResult<Vec<ExtractedRef>> {
let parsed = doc_loader::parse_doc_from_binary(snapshot.blob.clone(), snapshot.doc_id.clone())
.map_err(|err| RuntimeError::invalid_state(format!("Doc blob refs parse failed: {err}")))?;
let mut refs = Vec::new();
for block in parsed.blocks {
let Some(blob_keys) = block.blob else {
continue;
};
for blob_key in blob_keys {
refs.push(ExtractedRef {
blob_key,
block_id: block.block_id.clone(),
flavour: block.flavour.clone(),
});
}
}
Ok(refs)
fn extract_refs(blob: Vec<u8>) -> RuntimeResult<Vec<ExtractedRef>> {
doc_loader::get_blob_refs_from_binary(blob)
.map_err(|err| RuntimeError::invalid_state(format!("Doc blob refs parse failed: {err}")))
}
#[cfg(test)]
@@ -191,7 +174,7 @@ mod tests {
updated_at: Utc::now(),
};
let refs = extract_refs(&snapshot).expect("refs should parse");
let refs = extract_refs(snapshot.blob).expect("refs should parse");
assert!(
refs
@@ -233,19 +216,25 @@ mod tests {
updated_at: Utc::now(),
};
assert!(extract_refs(&snapshot).is_err());
assert!(extract_refs(snapshot.blob).is_err());
}
}
async fn replace_doc_refs(pool: &PgPool, snapshot: &CurrentDoc, refs: Vec<ExtractedRef>) -> RuntimeResult<(i64, i64)> {
async fn replace_doc_refs(
pool: &PgPool,
workspace_id: &str,
doc_id: &str,
updated_at: DateTime<Utc>,
refs: Vec<ExtractedRef>,
) -> RuntimeResult<(i64, i64)> {
let mut tx = pool
.begin()
.await
.map_err(|err| RuntimeError::database("Doc blob refs transaction failed", err))?;
let deleted = sqlx::query("DELETE FROM doc_blob_refs WHERE workspace_id = $1 AND doc_id = $2")
.bind(&snapshot.workspace_id)
.bind(&snapshot.doc_id)
.bind(workspace_id)
.bind(doc_id)
.execute(&mut *tx)
.await
.map_err(|err| RuntimeError::database("Doc blob refs delete failed", err))?
@@ -267,12 +256,12 @@ async fn replace_doc_refs(pool: &PgPool, snapshot: &CurrentDoc, refs: Vec<Extrac
error = NULL
"#,
)
.bind(&snapshot.workspace_id)
.bind(&snapshot.doc_id)
.bind(workspace_id)
.bind(doc_id)
.bind(reference.blob_key)
.bind(reference.block_id)
.bind(reference.flavour)
.bind(snapshot.updated_at)
.bind(updated_at)
.bind(PARSER_VERSION)
.execute(&mut *tx)
.await
@@ -330,9 +319,15 @@ async fn rebuild_doc_blob_refs_inner(
return Ok(result);
};
match extract_refs(&snapshot) {
let CurrentDoc {
workspace_id,
doc_id,
blob,
updated_at,
} = snapshot;
match extract_refs(blob) {
Ok(refs) => {
let (written, deleted) = replace_doc_refs(&pool, &snapshot, refs).await?;
let (written, deleted) = replace_doc_refs(&pool, &workspace_id, &doc_id, updated_at, refs).await?;
result.parsed_docs = 1;
result.refs_written = written;
result.refs_deleted = deleted;
@@ -311,9 +311,12 @@ async fn current_activity(
}
fn root_contains(root: CurrentDoc, doc_id: &str) -> RuntimeResult<bool> {
let ids = affine_doc_loader::get_doc_ids_from_binary(root.blob, true)
let projection = affine_doc_loader::project_workspace_root(root.blob, true)
.map_err(|err| RuntimeError::invalid_state(format!("Document cleanup root parse failed: {err}")))?;
Ok(ids.iter().any(|id| id == doc_id))
if !projection.complete {
return Err(RuntimeError::invalid_state("Document cleanup root doc is incomplete"));
}
Ok(projection.doc_ids.iter().any(|id| id == doc_id))
}
async fn delete_doc_rows(tx: &mut Transaction<'_, Postgres>, candidate: &Candidate) -> RuntimeResult<i64> {
@@ -101,6 +101,9 @@ fn merge_current_doc(
if snapshot.is_none() && updates.is_empty() {
return Ok(None);
}
if updates.is_empty() {
return Ok(snapshot);
}
let mut doc = Doc::default();
let mut updated_at = snapshot
.as_ref()
@@ -136,8 +139,12 @@ async fn load_workspace_live_doc_ids(pool: &PgPool, workspace_id: &str) -> Runti
fn workspace_live_doc_ids(root: Option<CurrentDoc>) -> RuntimeResult<Vec<String>> {
let root = root.ok_or_else(|| RuntimeError::invalid_state("Workspace root doc is missing"))?;
let mut ids = affine_doc_loader::get_doc_ids_from_binary(root.blob, true)
let projection = affine_doc_loader::project_workspace_root(root.blob, true)
.map_err(|err| RuntimeError::invalid_state(format!("Workspace root doc parse failed: {err}")))?;
if !projection.complete {
return Err(RuntimeError::invalid_state("Workspace root doc is incomplete"));
}
let mut ids = projection.doc_ids;
ids.sort();
ids.dedup();
Ok(ids)
@@ -1556,6 +1563,18 @@ mod tests {
}))
.is_err()
);
assert!(
workspace_live_doc_ids(Some(CurrentDoc {
workspace_id: "workspace".to_string(),
doc_id: "workspace".to_string(),
blob: vec![
1, 1, 1, 1, 40, 0, 1, 0, 11, 115, 117, 98, 95, 109, 97, 112, 95, 107, 101, 121, 1, 119, 13, 115, 117, 98, 95,
109, 97, 112, 95, 118, 97, 108, 117, 101, 0,
],
updated_at: Utc::now(),
}))
.is_err()
);
}
#[test]
+13 -12
View File
@@ -3,7 +3,7 @@ use memory_indexer::{SearchHit, SnapshotData};
use napi_derive::napi;
use serde::Serialize;
use sqlx::Row;
use y_octo::DocOptions;
use y_octo::merge_updates_v1;
// Increment this whenever there is a breaking change in the index format or how
// updates are applied
@@ -120,7 +120,7 @@ impl SqliteDocStorage {
}
segments.extend(updates.into_iter().map(|update| update.bin.to_vec()));
merge_updates(segments, doc_id).map(Some)
merge_updates(segments).map(Some)
}
pub async fn init_index(&self) -> Result<()> {
@@ -249,7 +249,7 @@ impl SqliteDocStorage {
}
}
fn merge_updates(mut segments: Vec<Vec<u8>>, guid: &str) -> Result<Vec<u8>> {
fn merge_updates(mut segments: Vec<Vec<u8>>) -> Result<Vec<u8>> {
if segments.is_empty() {
return Err(ParseError::DocNotFound.into());
}
@@ -258,15 +258,9 @@ fn merge_updates(mut segments: Vec<Vec<u8>>, guid: &str) -> Result<Vec<u8>> {
return segments.pop().ok_or(ParseError::DocNotFound.into());
}
let mut doc = DocOptions::new().with_guid(guid.to_string()).build();
for update in segments.iter() {
doc
.apply_update_from_binary_v1(update)
.map_err(|_| ParseError::InvalidBinary)?;
}
let buffer = doc
.encode_update_v1()
let update = merge_updates_v1(segments).map_err(|_| ParseError::InvalidBinary)?;
let buffer = update
.encode_v1()
.map_err(|err| ParseError::ParserError(err.to_string()))?;
Ok(buffer)
@@ -317,6 +311,13 @@ mod tests {
.execute(&storage.pool)
.await
.unwrap();
sqlx::query(r#"INSERT INTO updates (doc_id, data, created_at) VALUES (?, ?, ?)"#)
.bind("demo-doc")
.bind(&[0, 0][..])
.bind(Utc::now().naive_utc())
.execute(&storage.pool)
.await
.unwrap();
let result = storage.crawl_doc_data("demo-doc").await.unwrap();