perf(workspace): 流式处理重命名与删除恢复快照

This commit is contained in:
2026-09-09 06:10:49 +08:00
parent b15e7a10a2
commit bd6ccc4237
3 changed files with 165 additions and 10 deletions
@@ -605,3 +605,13 @@ Core 的独立数据目录目前不等于已授权 Vault。Python 旧笔记写
- 新增实际文件在 open_verified 后被同长度修改的复制测试,要求 copy_verified 返回 SYNC_SPOOL_CORRUPT,调用者仅在复制校验成功时发布临时文件。该证据针对复制内容校验,不代表所有重解析点/目录替换攻击均已覆盖。 - 新增实际文件在 open_verified 后被同长度修改的复制测试,要求 copy_verified 返回 SYNC_SPOOL_CORRUPT,调用者仅在复制校验成功时发布临时文件。该证据针对复制内容校验,不代表所有重解析点/目录替换攻击均已覆盖。
- rename/delete 的 file_ops payload 恢复、冲突保留/解决、外部变更扫描及部分 API 仍有整块读取;因此完整大附件低内存与全崩溃边界验收继续未完成。 - rename/delete 的 file_ops payload 恢复、冲突保留/解决、外部变更扫描及部分 API 仍有整块读取;因此完整大附件低内存与全崩溃边界验收继续未完成。
- 最终 desktop 全目标 133 通过、12 ignored(库 118、Host 8、其余集成 7),包括真实 Sync 服务有序推送/重复提交及故障恢复集成;日志 `.build/sync-stream-journal-full.log`。desktop 全目标 Clippy -D warnings 通过,日志 `.build/sync-stream-journal-clippy.log`。ignored 长时沙箱验收本轮未重跑,完整生产化目标保持未完成。 - 最终 desktop 全目标 133 通过、12 ignored(库 118、Host 8、其余集成 7),包括真实 Sync 服务有序推送/重复提交及故障恢复集成;日志 `.build/sync-stream-journal-full.log`。desktop 全目标 Clippy -D warnings 通过,日志 `.build/sync-stream-journal-clippy.log`。ignored 长时沙箱验收本轮未重跑,完整生产化目标保持未完成。
## 增量:重命名与删除的流式快照和恢复
- 新增 store_payload_file:以有界摘要校验源文件,通过同一已校验句柄将快照流式写入 spool 临时文件,复制时复核摘要与长度,fsync/无覆盖发布后才登记 payload 引用。已有 spool 仍做有界校验;源文件超出 100 MiB 或内容不符合预期时不接受该快照。
- prepare_file_op 的源摘要和 payload 保存不再使用整块 Vec。apply_file_op 保持旧 inline 日志迁移兼容,随后只使用摘要/长度引用与已校验文件句柄;源/目标 CAS 检查改为有界 hash_file,目标不存在时按 64 KiB 缓冲复制并复核内容,再按原顺序发布目标、删除源文件、提交身份/观察索引/outbox/回执事务。持久化引用摘要必须与 file_ops 的预期摘要一致。
- 原有 Workspace 测试 10 项通过,日志 `.build/file-ops-stream-tests.log`。新增实际 100 MiB 文件测试,以分块写入建立源文件,依次模拟重命名前目标不存在、目标已复制且源仍在、目标已复制且源已删除;每次重开后核验摘要、原 file_id、已提交回执及仅一条 outbox,再次重开 revision 不递增。删除场景在源已删除后从 payload 恢复 trash,最后验证准备删除后发生的外部修改保持原样并进入 conflict,不产生额外 outbox。
- 大文件测试通过,耗时 4.36 秒,日志 `.build/file-ops-stream-large.log`。该测试覆盖人工构造的持久化恢复阶段,不等于完整强杀/断电矩阵,也不是 RSS 测量;文件系统路径替换与全部持久化保证仍按原验收要求继续验证。
- 冲突保留/解决、外部变更发现和部分记录/API 仍有整块内容读取;大附件端到端内存验收和整个生产化目标仍未完成。
- desktop 全目标累计 134 通过、12 ignored(库 119、Host 8、其余集成 7),包括真实 Sync 服务及故障恢复集成,日志 `.build/file-ops-stream-full.log`。Clippy 首次指出流式改造后 workspace.rs 的 Write 导入仅供测试使用,已收为 cfg(test)desktop 全目标 Clippy -D warnings 随后通过,日志 `.build/file-ops-stream-clippy.log`。未把这组结果当作全部大附件 RSS、断电或完整生产化验收通过。
+34
View File
@@ -24,6 +24,40 @@ impl Workspace {
} }
self.register_payload(operation, &digest, content.len() as u64) self.register_payload(operation, &digest, content.len() as u64)
} }
pub(crate) fn store_payload_file(
&self,
operation: &str,
source: &Path,
expected: &str,
) -> Result<()> {
let size = fs::metadata(source)?.len();
if size > 100 * 1024 * 1024 {
return Err(HostError::new("FILE_TOO_LARGE"));
}
let mut input = open_verified(source, expected, size).map_err(|error| {
if error.code == "SYNC_SPOOL_CORRUPT" {
HostError::new("REVISION_CONFLICT")
} else {
error
}
})?;
let target = self.sync_spool(expected)?;
if target.exists() {
verify(&target, expected, size)?;
} else {
let parent = target
.parent()
.ok_or_else(|| HostError::new("UNSAFE_PATH"))?;
let mut temp = tempfile::NamedTempFile::new_in(parent)?;
copy_verified(&mut input, &mut temp, expected, size)?;
temp.as_file().sync_all()?;
temp.persist_noclobber(&target)
.map_err(|_| HostError::new("SYNC_SPOOL_FAILED"))?;
#[cfg(unix)]
fs::File::open(parent)?.sync_all()?;
}
self.register_payload(operation, expected, size)
}
fn register_payload(&self, operation: &str, digest: &str, size: u64) -> Result<()> { fn register_payload(&self, operation: &str, digest: &str, size: u64) -> Result<()> {
let old: Option<(String, i64)> = self let old: Option<(String, i64)> = self
.db .db
+120 -9
View File
@@ -5,6 +5,7 @@ use rusqlite::{params, Connection, OptionalExtension};
use serde::{Deserialize, Serialize}; use serde::{Deserialize, Serialize};
use sha2::{Digest, Sha256}; use sha2::{Digest, Sha256};
use std::fs::{self, File, OpenOptions}; use std::fs::{self, File, OpenOptions};
#[cfg(test)]
use std::io::Write; use std::io::Write;
use std::path::{Component, Path, PathBuf}; use std::path::{Component, Path, PathBuf};
use uuid::Uuid; use uuid::Uuid;
@@ -872,13 +873,12 @@ impl Workspace {
if !source.is_file() { if !source.is_file() {
return Err(HostError::new("FILE_NOT_FOUND")); return Err(HostError::new("FILE_NOT_FOUND"));
} }
let content = fs::read(source)?; if crate::payloads::hash_file(&source)? != expected {
if hash(&content) != expected {
return Err(HostError::new("REVISION_CONFLICT")); return Err(HostError::new("REVISION_CONFLICT"));
} }
self.entry(path)? self.entry(path)?
.ok_or_else(|| HostError::new("FILE_NOT_FOUND"))?; .ok_or_else(|| HostError::new("FILE_NOT_FOUND"))?;
self.store_payload(id, &content)?; self.store_payload_file(id, &source, expected)?;
let tx = self.db.transaction()?; let tx = self.db.transaction()?;
tx.execute( tx.execute(
"INSERT INTO operations VALUES (?1,?2,'pending',NULL)", "INSERT INTO operations VALUES (?1,?2,'pending',NULL)",
@@ -922,13 +922,22 @@ impl Workspace {
)) ))
}, },
)?; )?;
let content = self.payload(id, &content)?; if self.payload_ref(id)?.is_none() {
self.store_payload(id, &content)?; self.store_payload(id, &content)?;
}
let (digest, size) = self
.payload_ref(id)?
.ok_or_else(|| HostError::new("SYNC_SPOOL_CORRUPT"))?;
if digest != expected || !(0..=100 * 1024 * 1024).contains(&size) {
return Err(HostError::new("SYNC_SPOOL_CORRUPT"));
}
let mut payload =
crate::payloads::open_verified(&self.sync_spool(&digest)?, &digest, size as u64)?;
let source = self.resolve(&path)?; let source = self.resolve(&path)?;
let previous = self let previous = self
.entry(&path)? .entry(&path)?
.ok_or_else(|| HostError::new("FILE_NOT_FOUND"))?; .ok_or_else(|| HostError::new("FILE_NOT_FOUND"))?;
let source_conflict = source.exists() && hash(&fs::read(&source)?) != expected; let source_conflict = source.exists() && crate::payloads::hash_file(&source)? != expected;
let target = if kind == "rename" { let target = if kind == "rename" {
self.resolve(&destination)? self.resolve(&destination)?
} else { } else {
@@ -939,8 +948,8 @@ impl Workspace {
fs::create_dir_all(&trash)?; fs::create_dir_all(&trash)?;
trash.join(id) trash.join(id)
}; };
let target_conflict = let target_conflict = target.exists()
target.exists() && (linked(&target)? || hash(&fs::read(&target)?) != expected); && (linked(&target)? || crate::payloads::hash_file(&target)? != expected);
if source_conflict || target_conflict { if source_conflict || target_conflict {
let tx = self.db.transaction()?; let tx = self.db.transaction()?;
tx.execute("UPDATE file_ops SET state='conflict' WHERE id=?1", [id])?; tx.execute("UPDATE file_ops SET state='conflict' WHERE id=?1", [id])?;
@@ -951,14 +960,14 @@ impl Workspace {
tx.commit()?; tx.commit()?;
return Err(HostError::new("RECOVERY_CONFLICT")); return Err(HostError::new("RECOVERY_CONFLICT"));
} }
// journal 保留完整内容,目标刷盘后才删除来源;两处崩溃均可幂等重放。 // The durable payload remains available after removing the source.
if !target.exists() { if !target.exists() {
let parent = target let parent = target
.parent() .parent()
.ok_or_else(|| HostError::new("UNSAFE_PATH"))?; .ok_or_else(|| HostError::new("UNSAFE_PATH"))?;
fs::create_dir_all(parent)?; fs::create_dir_all(parent)?;
let mut temp = tempfile::NamedTempFile::new_in(parent)?; let mut temp = tempfile::NamedTempFile::new_in(parent)?;
temp.write_all(&content)?; crate::payloads::copy_verified(&mut payload, &mut temp, &digest, size as u64)?;
temp.as_file().sync_all()?; temp.as_file().sync_all()?;
temp.persist_noclobber(&target) temp.persist_noclobber(&target)
.map_err(|_| HostError::new("PATH_CONFLICT"))?; .map_err(|_| HostError::new("PATH_CONFLICT"))?;
@@ -1043,6 +1052,108 @@ mod tests {
assert_eq!(ws.pending_count().unwrap(), 3); assert_eq!(ws.pending_count().unwrap(), 3);
} }
#[test]
fn hundred_mib_file_operations_recover_each_copy_stage_without_duplicate_events() {
let dir = tempfile::tempdir().unwrap();
let mut ws = Workspace::open(dir.path()).unwrap();
fs::create_dir(dir.path().join("attachments")).unwrap();
let mut path = "attachments/source.bin".to_owned();
let mut source = File::create(dir.path().join(&path)).unwrap();
let block = vec![23u8; 64 * 1024];
for _ in 0..1600 {
source.write_all(&block).unwrap();
}
source.sync_all().unwrap();
drop(source);
let digest = crate::payloads::hash_file(&dir.path().join(&path)).unwrap();
let seed = Uuid::new_v4().to_string();
ws.store_payload_file(&seed, &dir.path().join(&path), &digest)
.unwrap();
let identity = Uuid::new_v4().to_string();
ws.write_spooled_with_identity(
&path,
&digest,
(&digest, 100 * 1024 * 1024),
"remote",
&Uuid::new_v4().to_string(),
Some(&identity),
)
.unwrap();
for stage in 0..3 {
let target = format!("attachments/stage-{stage}.bin");
let operation = ws
.prepare_file_op("rename", &path, &target, &digest)
.unwrap();
if stage >= 1 {
fs::copy(dir.path().join(&path), dir.path().join(&target)).unwrap();
}
if stage == 2 {
fs::remove_file(dir.path().join(&path)).unwrap();
}
drop(ws);
ws = Workspace::open(dir.path()).unwrap();
assert!(!dir.path().join(&path).exists());
assert_eq!(
crate::payloads::hash_file(&dir.path().join(&target)).unwrap(),
digest
);
assert_eq!(ws.entry(&target).unwrap().unwrap().file_id, identity);
assert_eq!(
ws.operation(&operation).unwrap().unwrap()["state"],
"committed"
);
let revision = ws.entry(&target).unwrap().unwrap().revision;
drop(ws);
ws = Workspace::open(dir.path()).unwrap();
assert_eq!(ws.entry(&target).unwrap().unwrap().revision, revision);
assert_eq!(
ws.db
.query_row(
"SELECT COUNT(*) FROM outbox WHERE operation_id=?1",
[&operation],
|r| r.get::<_, i64>(0)
)
.unwrap(),
1
);
path = target;
}
let deletion = ws.prepare_file_op("delete", &path, "", &digest).unwrap();
fs::remove_file(dir.path().join(&path)).unwrap();
drop(ws);
ws = Workspace::open(dir.path()).unwrap();
assert_eq!(
crate::payloads::hash_file(&dir.path().join(".ainote/trash").join(&deletion)).unwrap(),
digest
);
assert!(ws.entry(&path).unwrap().unwrap().deleted);
ws.recover().unwrap();
assert_eq!(ws.pending_count().unwrap(), 4);
let restored = "attachments/conflict.bin";
ws.write_spooled_with_identity(
restored,
"",
(&digest, 100 * 1024 * 1024),
"remote",
&Uuid::new_v4().to_string(),
None,
)
.unwrap();
let conflict = ws.prepare_file_op("delete", restored, "", &digest).unwrap();
fs::write(dir.path().join(restored), b"external edit").unwrap();
drop(ws);
ws = Workspace::open(dir.path()).unwrap();
assert_eq!(
fs::read(dir.path().join(restored)).unwrap(),
b"external edit"
);
assert_eq!(
ws.operation(&conflict).unwrap().unwrap()["state"],
"conflict"
);
assert_eq!(ws.pending_count().unwrap(), 4);
}
#[test] #[test]
fn saves_conflict_and_reopen() { fn saves_conflict_and_reopen() {
let dir = tempfile::tempdir().unwrap(); let dir = tempfile::tempdir().unwrap();