lora_wal/wal/wal.rs
1//! `Wal` — the durable log handle.
2//!
3//! Owns a WAL directory of the shape:
4//!
5//! ```text
6//! <dir>/
7//! 0000000001.wal sealed segment
8//! 0000000002.wal sealed segment
9//! 0000000003.wal active segment
10//! ```
11//!
12//! The active segment is identified by the highest numeric file name —
13//! we deliberately do **not** keep a separate `CURRENT` pointer file.
14//! A pointer would be a second source of truth that crashes can
15//! desynchronise from the directory listing without buying anything:
16//! the file names already encode their ordering, and segment headers
17//! are self-describing.
18//!
19//! Lifecycle is `[`Wal::open`] → acquire the directory lock → drain replay
20//! events into the store → resume normal `begin` / `append` / `commit`
21//! traffic. The directory lock is held until the `Wal` drops; a second
22//! live `Wal::open` on the same directory returns [`WalError::AlreadyOpen`].
23//!
24//! All public methods take `&self` and serialise through an internal
25//! [`Mutex`]. The store write lock already serialises query commits in
26//! production, so the inner mutex is uncontested and effectively free.
27
28use std::fs;
29use std::path::Path;
30use std::sync::{Arc, Mutex};
31#[cfg(not(target_arch = "wasm32"))]
32use std::time::Duration;
33
34use lora_store::MutationEvent;
35
36#[cfg(not(target_arch = "wasm32"))]
37use super::group_flusher::{spawn_group_flusher, GroupFlusherHandle};
38use crate::config::SyncMode;
39use crate::dir::{SegmentDir, SegmentId};
40use crate::errors::WalError;
41use crate::lock::DirLock;
42use crate::lsn::Lsn;
43use crate::record::WalRecord;
44use crate::recorder::WroteCommit;
45use crate::replay::{replay_segments, ReplayOutcome};
46use crate::segment::SegmentWriter;
47
48/// State guarded by the inner `Mutex`. Nothing in this struct is
49/// `Send`-unsafe; the lock is purely for `&self`-safe interior
50/// mutation.
51struct WalState {
52 next_lsn: Lsn,
53 durable_lsn: Lsn,
54 active_segment_id: SegmentId,
55 active_writer: SegmentWriter,
56 /// Lowest segment id still on disk. Bumped by `truncate_up_to`.
57 oldest_segment_id: SegmentId,
58}
59
60/// Reserved latch for durability failures that occur outside the immediate
61/// caller path. Wrapped in a `Mutex` instead of an
62/// `AtomicCell<Option<String>>` because failures are rare and we want the
63/// message preserved verbatim for operator-facing reporting
64/// (`/admin/wal/status` `bgFailure`). Once `Some`, every subsequent
65/// commit/flush returns [`WalError::Poisoned`] and the operator is expected to
66/// restart from the last consistent snapshot + WAL.
67type BgFailure = Mutex<Option<String>>;
68
69/// Selects the durability work that [`Wal::flush_inner`] actually does.
70/// Centralising the three modes here means `flush` and `force_fsync`
71/// share one code path and the call sites don't have to remember which
72/// mode advances `durable_lsn` and which does not.
73#[derive(Debug, Clone, Copy)]
74pub(super) enum FlushKind {
75 /// Honour the configured [`SyncMode`]. This is what the recorder's
76 /// `flush()` calls into.
77 PerConfiguredMode,
78 /// Always write the buffer + fsync + advance `durable_lsn`, regardless of
79 /// mode. Used by checkpoints, explicit sync, and clean Group-mode drop.
80 ForceFsync,
81}
82
83/// Live, append-side WAL handle.
84///
85/// Construct via [`Wal::open`]. The returned tuple includes the list of
86/// committed mutation events that need to be re-applied to the
87/// in-memory store before any new traffic is accepted.
88///
89/// `Wal::open` returns `Arc<Self>` because the optional Group-mode
90/// background flusher needs a `Weak<Wal>` to call back into without
91/// taking a strong reference (which would prevent shutdown).
92pub struct Wal {
93 segments: SegmentDir,
94 sync_mode: SyncMode,
95 segment_target_bytes: u64,
96 state: Mutex<WalState>,
97 /// Latched durability failure; surfaced via [`Wal::bg_failure`] and
98 /// propagated to commit/flush/force_fsync as [`WalError::Poisoned`].
99 bg_failure: Arc<BgFailure>,
100 /// Background flusher for `SyncMode::Group`. `Drop` joins the
101 /// thread, so a `Wal` going out of scope is a clean shutdown
102 /// signal. Absent on `wasm32`, where Group mode falls back to the
103 /// drop-time flush.
104 #[cfg(not(target_arch = "wasm32"))]
105 flusher: Mutex<Option<GroupFlusherHandle>>,
106 /// Held for the lifetime of the WAL so a second handle cannot append
107 /// to the same active segment concurrently.
108 _dir_lock: DirLock,
109}
110
111impl Wal {
112 /// Open or create the WAL directory at `dir`.
113 ///
114 /// `checkpoint_lsn` is the LSN stamped into the most recent
115 /// snapshot the caller is restoring from (or [`Lsn::ZERO`] if
116 /// there is no snapshot). Replay skips records at or below this
117 /// fence — they are already represented in the loaded state.
118 ///
119 /// Returns `(wal, committed_events)`. The caller is expected to
120 /// apply every event in `committed_events` to its in-memory store
121 /// in order before issuing any new `begin` / `append` calls.
122 pub fn open(
123 dir: impl Into<std::path::PathBuf>,
124 sync_mode: SyncMode,
125 segment_target_bytes: u64,
126 checkpoint_lsn: Lsn,
127 ) -> Result<(Arc<Self>, Vec<MutationEvent>), WalError> {
128 let segments = SegmentDir::new(dir);
129 fs::create_dir_all(segments.root())?;
130 let dir_lock = DirLock::acquire(segments.root())?;
131
132 let entries = segments.list()?;
133 let (active_id, active_writer, replay) = if entries.is_empty() {
134 Self::open_fresh(&segments)?
135 } else {
136 Self::open_existing(&segments, &entries, checkpoint_lsn)?
137 };
138
139 let next_lsn = if replay.max_lsn.is_zero() {
140 Lsn::new(1)
141 } else {
142 replay.max_lsn.next()
143 };
144 // Treat everything readable at open time as the recovered
145 // durability fence. This does not prove the bytes were
146 // fsync-confirmed before the previous process died; it means
147 // they survived to this open and future appends must start
148 // after them.
149 let durable_lsn = replay.max_lsn;
150
151 let oldest_segment_id = entries.first().map(|e| e.id).unwrap_or(active_id);
152
153 let state = WalState {
154 next_lsn,
155 durable_lsn,
156 active_segment_id: active_id,
157 active_writer,
158 oldest_segment_id,
159 };
160
161 let wal = Arc::new(Self {
162 segments,
163 sync_mode,
164 segment_target_bytes,
165 state: Mutex::new(state),
166 bg_failure: Arc::new(Mutex::new(None)),
167 #[cfg(not(target_arch = "wasm32"))]
168 flusher: Mutex::new(None),
169 _dir_lock: dir_lock,
170 });
171
172 // Spawn the Group flusher *after* the Arc exists so it can hold a
173 // `Weak<Wal>` that drops when the last strong ref does. The flusher's
174 // own Drop joins the thread, so removing the field on `Wal::drop` is
175 // a clean shutdown signal. Wasm has no real fsync boundary and no
176 // thread support, so Group there relies on the drop-time flush.
177 #[cfg(not(target_arch = "wasm32"))]
178 if let SyncMode::Group { interval_ms } = sync_mode {
179 let interval = Duration::from_millis(u64::from(interval_ms.max(1)));
180 let handle = spawn_group_flusher(Arc::downgrade(&wal), interval);
181 *wal.flusher.lock().unwrap() = Some(handle);
182 }
183
184 Ok((wal, replay.committed_events))
185 }
186
187 /// Brand-new WAL directory. Create segment 1 with `base_lsn = 1`
188 /// so LSN 0 stays reserved for "empty / never written".
189 fn open_fresh(
190 segments: &SegmentDir,
191 ) -> Result<(SegmentId, SegmentWriter, ReplayOutcome), WalError> {
192 let id = SegmentId::FIRST;
193 let writer = SegmentWriter::create(segments.path_for(id), Lsn::new(1))?;
194 segments.sync_dir()?;
195 let replay = ReplayOutcome {
196 committed_events: Vec::new(),
197 max_lsn: Lsn::ZERO,
198 torn_tail: None,
199 checkpoint_lsn_observed: None,
200 };
201 Ok((id, writer, replay))
202 }
203
204 /// Existing directory. Replay every segment to surface committed
205 /// events + detect a torn tail; reopen the highest-id segment
206 /// for append; truncate it if the torn tail is in *that* segment.
207 fn open_existing(
208 segments: &SegmentDir,
209 entries: &[crate::dir::SegmentEntry],
210 checkpoint_lsn: Lsn,
211 ) -> Result<(SegmentId, SegmentWriter, ReplayOutcome), WalError> {
212 let paths: Vec<_> = entries.iter().map(|e| e.path.clone()).collect();
213 let replay = replay_segments(&paths, checkpoint_lsn)?;
214
215 // The active segment is whichever file has the highest
216 // numeric id — segment file names are self-describing, so
217 // there is no separate CURRENT pointer.
218 let active = entries.last().expect("entries non-empty in open_existing");
219 let (mut writer, _torn_from_writer) =
220 SegmentWriter::open_for_append(segments.path_for(active.id))?;
221
222 // A torn tail in a *sealed* segment is impossible (sealed
223 // segments are never appended to), so we only need to handle
224 // the active one.
225 if let Some(t) = &replay.torn_tail {
226 if t.segment_path == active.path {
227 writer.truncate_to(t.last_good_offset)?;
228 } else {
229 return Err(WalError::Malformed(format!(
230 "torn tail found in sealed segment {}",
231 t.segment_path.display()
232 )));
233 }
234 }
235
236 Ok((active.id, writer, replay))
237 }
238
239 pub fn dir(&self) -> &Path {
240 self.segments.root()
241 }
242
243 pub fn sync_mode(&self) -> SyncMode {
244 self.sync_mode
245 }
246
247 pub fn durable_lsn(&self) -> Lsn {
248 self.state.lock().unwrap().durable_lsn
249 }
250
251 /// Latched durability failure, if any. `None` means the WAL is healthy.
252 /// Once set, every commit / flush / force_fsync starts returning
253 /// [`WalError::Poisoned`] and the WAL stops accepting new
254 /// transactions until the operator restarts from the last
255 /// consistent snapshot + WAL.
256 pub fn bg_failure(&self) -> Option<String> {
257 self.bg_failure.lock().unwrap().clone()
258 }
259
260 /// Direct handle to the latched-failure mutex. Used by the bg
261 /// flusher to record an fsync failure exactly once. Hidden from
262 /// outside the module so the latch stays single-writer.
263 #[cfg(not(target_arch = "wasm32"))]
264 pub(super) fn bg_failure_slot(&self) -> &BgFailure {
265 &self.bg_failure
266 }
267
268 fn check_healthy(&self) -> Result<(), WalError> {
269 if self.bg_failure.lock().unwrap().is_some() {
270 return Err(WalError::Poisoned);
271 }
272 Ok(())
273 }
274
275 /// LSN that the *next* `begin` / `append` call will allocate.
276 /// Exposed for tests and for sanity checks at boot; not part of
277 /// any durability contract.
278 pub fn next_lsn(&self) -> Lsn {
279 self.state.lock().unwrap().next_lsn
280 }
281
282 pub fn oldest_segment_id(&self) -> u64 {
283 self.state.lock().unwrap().oldest_segment_id.raw()
284 }
285
286 pub fn active_segment_id(&self) -> u64 {
287 self.state.lock().unwrap().active_segment_id.raw()
288 }
289
290 // -------------------------------------------------------------
291 // Low-level record primitives.
292 //
293 // Production code does **not** use these directly — every commit
294 // goes through [`Self::commit_tx`], which writes the begin/batch/
295 // commit triple atomically and routes durability through the
296 // configured single-thread flush policy. The methods below remain
297 // `pub` for the crate's own integration tests and for the rare
298 // admin path (`checkpoint_marker`) that needs to insert a single record.
299 // Mixing them with `commit_tx` against the same WAL is supported
300 // but unnecessary; if you find yourself calling `begin` /
301 // `append` / `commit` from a new caller, prefer `commit_tx`
302 // unless you specifically need the partial-write shape.
303 // -------------------------------------------------------------
304
305 /// Allocate a `TxBegin` record and return its LSN. *Test/admin
306 /// primitive.* Production commits use [`Self::commit_tx`].
307 ///
308 /// Rotation happens here so a transaction is always wholly within
309 /// one segment.
310 pub fn begin(&self) -> Result<Lsn, WalError> {
311 self.check_healthy()?;
312 let mut state = self.state.lock().unwrap();
313 self.maybe_rotate(&mut state)?;
314 Self::alloc_and_append(&mut state, |lsn| WalRecord::TxBegin { lsn })
315 }
316
317 /// Append a single mutation to the active segment's pending
318 /// buffer. *Test/admin primitive.* Not durable until `flush()`
319 /// runs; production commits use [`Self::commit_tx`].
320 pub fn append(&self, tx_begin_lsn: Lsn, event: &MutationEvent) -> Result<Lsn, WalError> {
321 self.check_healthy()?;
322 let mut state = self.state.lock().unwrap();
323 Self::alloc_and_append(&mut state, |lsn| WalRecord::Mutation {
324 lsn,
325 tx_begin_lsn,
326 event: event.clone(),
327 })
328 }
329
330 /// Append many mutations as one framed record. *Test/admin
331 /// primitive.* Production commits use [`Self::commit_tx`], which
332 /// writes the begin/batch/commit triple in a single critical
333 /// section.
334 pub fn append_batch(
335 &self,
336 tx_begin_lsn: Lsn,
337 events: Vec<MutationEvent>,
338 ) -> Result<Lsn, WalError> {
339 self.check_healthy()?;
340 if events.is_empty() {
341 return Err(WalError::Encode(
342 "mutation batch must contain at least one event".into(),
343 ));
344 }
345 let mut state = self.state.lock().unwrap();
346 Self::alloc_and_append(&mut state, |lsn| WalRecord::MutationBatch {
347 lsn,
348 tx_begin_lsn,
349 events,
350 })
351 }
352
353 /// Append a standalone `TxCommit` marker. *Test/admin primitive.*
354 /// Production commits use [`Self::commit_tx`].
355 pub fn commit(&self, tx_begin_lsn: Lsn) -> Result<Lsn, WalError> {
356 self.check_healthy()?;
357 let mut state = self.state.lock().unwrap();
358 Self::alloc_and_append(&mut state, |lsn| WalRecord::TxCommit { lsn, tx_begin_lsn })
359 }
360
361 /// Append a `TxAbort` marker. *Test/admin primitive.* Production
362 /// code never writes `TxAbort`: [`Self::commit_tx`] writes the
363 /// begin/batch/commit triple atomically, so an aborted query has
364 /// nothing on disk to mark as aborted.
365 pub fn abort(&self, tx_begin_lsn: Lsn) -> Result<Lsn, WalError> {
366 self.check_healthy()?;
367 let mut state = self.state.lock().unwrap();
368 Self::alloc_and_append(&mut state, |lsn| WalRecord::TxAbort { lsn, tx_begin_lsn })
369 }
370
371 /// One-shot transaction commit.
372 ///
373 /// Encodes `TxBegin` + `MutationBatch` + `TxCommit` as a single
374 /// contiguous run inside one short critical section, then applies the
375 /// configured flush policy. Compared to the legacy
376 /// `begin → append_batch → commit → flush` sequence this collapses
377 /// four separate state-lock acquisitions into one while preserving the
378 /// release's single-writer execution model. Future concurrent commit
379 /// plumbing can build around this one-shot boundary without changing the
380 /// recorder contract.
381 ///
382 /// Returns [`WroteCommit::No`] for an empty event list (no records
383 /// are written, no fsync is issued).
384 pub fn commit_tx(&self, events: Vec<MutationEvent>) -> Result<WroteCommit, WalError> {
385 self.check_healthy()?;
386 if events.is_empty() {
387 return Ok(WroteCommit::No);
388 }
389
390 // Phase 1: allocate the LSN window and encode all three
391 // records into the active segment's pending buffer in one
392 // critical section. Collapsing what was four separate state
393 // lock acquisitions (begin / append_batch / commit / flush)
394 // into one is the lock-side win that pairs with the
395 // lock-free emit short-circuit on the recorder side.
396 {
397 let mut state = self.state.lock().unwrap();
398 self.maybe_rotate(&mut state)?;
399 let begin_lsn = state.next_lsn;
400 let batch_lsn = begin_lsn.next();
401 let commit_lsn = batch_lsn.next();
402 state.next_lsn = commit_lsn.next();
403 state
404 .active_writer
405 .append(&WalRecord::TxBegin { lsn: begin_lsn })?;
406 state.active_writer.append(&WalRecord::MutationBatch {
407 lsn: batch_lsn,
408 tx_begin_lsn: begin_lsn,
409 events,
410 })?;
411 state.active_writer.append(&WalRecord::TxCommit {
412 lsn: commit_lsn,
413 tx_begin_lsn: begin_lsn,
414 })?;
415 }
416
417 // Phase 2: durability per sync mode. PerCommit fsyncs inline;
418 // Group is cooperative in this single-threaded release (no bg
419 // flusher thread); None just pushes bytes to the page cache.
420 match self.sync_mode {
421 SyncMode::PerCommit => self.flush_inner(FlushKind::ForceFsync)?,
422 SyncMode::Group { .. } | SyncMode::None => {
423 self.flush_inner(FlushKind::PerConfiguredMode)?;
424 }
425 }
426
427 Ok(WroteCommit::Yes)
428 }
429
430 /// Append a `Checkpoint` marker. `snapshot_lsn` should equal the
431 /// LSN written into the snapshot file's header — replay uses
432 /// it to defend against the snapshot-rename-but-no-marker race.
433 pub fn checkpoint_marker(&self, snapshot_lsn: Lsn) -> Result<Lsn, WalError> {
434 self.check_healthy()?;
435 let mut state = self.state.lock().unwrap();
436 Self::alloc_and_append(&mut state, |lsn| WalRecord::Checkpoint {
437 lsn,
438 snapshot_lsn,
439 })
440 }
441
442 /// Single-source-of-truth for "allocate the next LSN, build the
443 /// record, push it onto the active segment's pending buffer".
444 /// The five public append paths (`begin / append / commit / abort
445 /// / checkpoint_marker`) all funnel through here so the LSN
446 /// allocation never gets out of sync with the encoded record.
447 #[inline]
448 fn alloc_and_append(
449 state: &mut WalState,
450 build: impl FnOnce(Lsn) -> WalRecord,
451 ) -> Result<Lsn, WalError> {
452 let lsn = state.next_lsn;
453 state.next_lsn = lsn.next();
454 state.active_writer.append(&build(lsn))?;
455 Ok(lsn)
456 }
457
458 /// Flush the active segment's pending buffer.
459 ///
460 /// What "flush" means depends on [`SyncMode`]:
461 ///
462 /// - `PerCommit` — write the buffer to the OS, `fsync`, and
463 /// advance `durable_lsn`. The strongest contract: every
464 /// record up to `next_lsn - 1` is on disk.
465 /// - `Group` — write the buffer to the OS, but leave
466 /// `durable_lsn` unchanged until an explicit `force_fsync`,
467 /// checkpoint, sync, or clean drop.
468 /// - `None` — write the buffer to the OS only, but advance
469 /// `durable_lsn` anyway. The mode opts out of crash
470 /// durability, so the checkpoint fence reports
471 /// "what's been written" instead of "what's actually safe".
472 pub fn flush(&self) -> Result<(), WalError> {
473 self.check_healthy()?;
474 self.flush_inner(FlushKind::PerConfiguredMode)
475 }
476
477 /// Unconditionally write the buffer to the OS, `fsync`, and
478 /// advance `durable_lsn`. Used by callers that need a durability
479 /// point right now regardless of the configured cadence (e.g.
480 /// checkpoint). Returns [`WalError::Poisoned`] if the WAL has already
481 /// latched a durability failure.
482 pub fn force_fsync(&self) -> Result<(), WalError> {
483 self.check_healthy()?;
484 self.flush_inner(FlushKind::ForceFsync)
485 }
486
487 /// Single source of truth for the flush state machine. Skips the
488 /// `check_healthy` gate so clean shutdown can force a final Group-mode
489 /// sync even if callers are otherwise done with the handle.
490 pub(super) fn flush_inner(&self, kind: FlushKind) -> Result<(), WalError> {
491 let mut state = self.state.lock().unwrap();
492 let written_lsn = Lsn::new(state.next_lsn.raw().saturating_sub(1));
493
494 // Decide whether this call is allowed to advance `durable_lsn`.
495 // PerCommit and forced syncs advance after the fsync boundary. None
496 // advances after write because the mode opts out of crash durability.
497 // Group is cooperative for this release: normal `flush()` writes bytes
498 // but leaves the durable fence alone until `force_fsync`, checkpoint,
499 // sync(), or drop.
500 let do_fsync = matches!(
501 (kind, self.sync_mode),
502 (FlushKind::ForceFsync, _) | (_, SyncMode::PerCommit)
503 );
504 let advance_durable = matches!(
505 (kind, self.sync_mode),
506 (FlushKind::ForceFsync, _) | (_, SyncMode::PerCommit) | (_, SyncMode::None)
507 );
508
509 if do_fsync {
510 state.active_writer.flush_and_sync()?;
511 } else {
512 state.active_writer.flush_buffer()?;
513 }
514 if advance_durable {
515 state.durable_lsn = written_lsn;
516 }
517 Ok(())
518 }
519
520 /// Drop sealed segments whose entire LSN range is at or below
521 /// `fence_lsn`. Idempotent and safe to call repeatedly.
522 ///
523 /// The active segment is never deleted — even if every record in
524 /// it predates the fence, it is still the rotation target for
525 /// new appends. The segment immediately before the active one
526 /// is also kept as a tombstone so a subsequent crash before the
527 /// next checkpoint still finds a self-describing log start.
528 pub fn truncate_up_to(&self, fence_lsn: Lsn) -> Result<(), WalError> {
529 let mut state = self.state.lock().unwrap();
530 let active_id = state.active_segment_id;
531 let entries = self.segments.list()?;
532
533 let mut to_drop: Vec<crate::dir::SegmentEntry> = Vec::new();
534 for (i, entry) in entries.iter().enumerate() {
535 // Active segment and the one immediately preceding it
536 // are kept by policy.
537 if entry.id >= active_id.saturating_prev() {
538 break;
539 }
540 // Segment `i` covers `[base_i, base_{i+1} - 1]`. We are
541 // safe to drop only when `base_{i+1} - 1 <= fence_lsn`.
542 let next = match entries.get(i + 1) {
543 Some(n) => n,
544 None => break,
545 };
546 let next_base = SegmentDir::base_lsn(&next.path)?;
547 if next_base.raw().saturating_sub(1) <= fence_lsn.raw() {
548 to_drop.push(entry.clone());
549 }
550 }
551
552 for entry in to_drop {
553 fs::remove_file(&entry.path)?;
554 if entry.id >= state.oldest_segment_id {
555 state.oldest_segment_id = entry.id.next();
556 }
557 }
558 if state.oldest_segment_id != entries.first().map(|e| e.id).unwrap_or(active_id) {
559 self.segments.sync_dir()?;
560 }
561 Ok(())
562 }
563
564 /// Rotate the active segment when it has grown past
565 /// `segment_target_bytes`. Called from `begin()` so rotation only
566 /// ever lands at a transaction boundary.
567 fn maybe_rotate(&self, state: &mut WalState) -> Result<(), WalError> {
568 if state.active_writer.bytes_written() < self.segment_target_bytes {
569 return Ok(());
570 }
571 // Seal the current segment (forces a flush + fsync) and open
572 // a fresh one with `base_lsn = next_lsn` so the segment file
573 // names line up with the record LSNs they contain.
574 state.active_writer.flush_and_sync()?;
575 state.active_writer.seal()?;
576
577 let next_id = state.active_segment_id.next();
578 let writer = SegmentWriter::create(self.segments.path_for(next_id), state.next_lsn)?;
579 self.segments.sync_dir()?;
580 state.active_writer = writer;
581 state.active_segment_id = next_id;
582 Ok(())
583 }
584}
585
586impl Drop for Wal {
587 fn drop(&mut self) {
588 if matches!(self.sync_mode, SyncMode::Group { .. }) {
589 let _ = self.flush_inner(FlushKind::ForceFsync);
590 }
591 // Join the group flusher, if any, before the directory lock is
592 // released. That keeps the "one live append owner" boundary intact
593 // through shutdown.
594 #[cfg(not(target_arch = "wasm32"))]
595 if let Ok(slot) = self.flusher.get_mut() {
596 let _ = slot.take();
597 }
598 }
599}