Skip to main content

velesdb_core/storage/
log_payload.rs

1//! Log-structured payload storage with snapshot support.
2//!
3//! Stores payloads in an append-only log file with an in-memory index.
4//! Supports periodic snapshots for fast cold-start recovery.
5//!
6//! ## WAL Entry Formats
7//!
8//! **CRC32-protected (current, markers 0xC3/0xC4):**
9//! ```text
10//! Store:  [0xC3: 1B] [id: 8B LE] [len: 4B LE] [payload: len B] [crc32: 4B LE]
11//! Delete: [0xC4: 1B] [id: 8B LE] [crc32: 4B LE]
12//! ```
13//!
14//! **Legacy (markers 1/2, read-only for backward compatibility):**
15//! ```text
16//! Store:  [1: 1B] [id: 8B LE] [len: 4B LE] [payload: len B]
17//! Delete: [2: 1B] [id: 8B LE]
18//! ```
19//!
20//! CRC32 covers all bytes preceding the CRC field. On CRC mismatch during
21//! replay, the corrupted entry is skipped and a warning is logged.
22//!
23//! Snapshot format and I/O are handled by the [`super::snapshot`] module.
24
25use super::log_payload_io::{compute_delete_crc, write_store_record, CRC_DELETE_MARKER};
26use super::snapshot;
27use super::traits::PayloadStorage;
28
29// Re-export snapshot items for backward compatibility with existing imports
30#[allow(unused_imports)] // SNAPSHOT_MAGIC/VERSION used only in test modules
31pub(crate) use snapshot::{crc32_hash, SNAPSHOT_MAGIC, SNAPSHOT_VERSION};
32
33use parking_lot::RwLock;
34use rustc_hash::FxHashMap;
35use std::fs::{File, OpenOptions};
36use std::io::{self, BufReader, Read, Seek, SeekFrom, Write};
37use std::path::{Path, PathBuf};
38
39/// Controls how payload WAL writes are synced to disk.
40///
41/// - `Fsync` (default): `flush()` + `sync_all()` — full durability, safe against power loss.
42/// - `FlushOnly`: `flush()` only — data reaches OS kernel but may be lost on power failure.
43/// - `None`: No sync — maximum throughput for bulk imports where data can be re-derived.
44#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
45#[non_exhaustive]
46pub enum DurabilityMode {
47    /// Full durability: flush buffer + fsync to disk.
48    #[default]
49    Fsync,
50    /// Flush buffer to OS only (no fsync). Faster but not power-loss safe.
51    FlushOnly,
52    /// No sync at all. Maximum throughput for bulk imports.
53    None,
54}
55
56/// Log-structured payload storage with snapshot support.
57///
58/// Stores payloads in an append-only log file with an in-memory index.
59/// Supports periodic snapshots for O(1) cold-start recovery instead of O(N) WAL replay.
60#[allow(clippy::module_name_repetitions)]
61pub struct LogPayloadStorage {
62    /// Directory path for storage files
63    path: PathBuf,
64    /// In-memory index: ID -> Offset of length field in WAL
65    index: RwLock<FxHashMap<u64, u64>>,
66    /// Write-Ahead Log writer (append-only)
67    wal: RwLock<io::BufWriter<File>>,
68    /// Independent file handle for reading, protected for seeking
69    reader: RwLock<File>,
70    /// WAL position at last snapshot (0 = no snapshot)
71    last_snapshot_wal_pos: RwLock<u64>,
72    /// Durability mode for WAL writes
73    durability: DurabilityMode,
74    /// Tracked WAL write position (avoids flush+metadata syscall for `DurabilityMode::None`)
75    write_offset: RwLock<u64>,
76}
77
78use super::wal_entry::WalEntry;
79
80impl LogPayloadStorage {
81    /// Creates a new `LogPayloadStorage` with the default durability mode (`Fsync`).
82    ///
83    /// If a snapshot file exists and is valid, loads from snapshot and replays
84    /// only the WAL delta for fast startup. Otherwise, falls back to full WAL replay.
85    ///
86    /// # Errors
87    ///
88    /// Returns an error if file operations fail.
89    pub fn new<P: AsRef<Path>>(path: P) -> io::Result<Self> {
90        Self::new_with_durability(path, DurabilityMode::default())
91    }
92
93    /// Creates a new `LogPayloadStorage` with the specified durability mode.
94    ///
95    /// See [`DurabilityMode`] for available modes and their trade-offs.
96    ///
97    /// # Errors
98    ///
99    /// Returns an error if file operations fail.
100    pub fn new_with_durability<P: AsRef<Path>>(
101        path: P,
102        durability: DurabilityMode,
103    ) -> io::Result<Self> {
104        let path = path.as_ref().to_path_buf();
105        std::fs::create_dir_all(&path)?;
106        let log_path = path.join("payloads.log");
107
108        let wal = Self::open_wal_writer(&log_path)?;
109        let (reader, wal_len) = Self::open_wal_reader(&log_path)?;
110        let (index, last_snapshot_wal_pos) = Self::load_or_replay_index(&path, &log_path, wal_len)?;
111
112        Ok(Self {
113            path,
114            index: RwLock::new(index),
115            wal: RwLock::new(wal),
116            reader: RwLock::new(reader),
117            last_snapshot_wal_pos: RwLock::new(last_snapshot_wal_pos),
118            durability,
119            write_offset: RwLock::new(wal_len),
120        })
121    }
122
123    /// Opens the WAL file for append-mode writing.
124    fn open_wal_writer(log_path: &Path) -> io::Result<io::BufWriter<File>> {
125        let writer_file = OpenOptions::new()
126            .create(true)
127            .append(true)
128            .open(log_path)?;
129        Ok(io::BufWriter::new(writer_file))
130    }
131
132    /// Opens the WAL file for random-access reading, creating it if absent.
133    ///
134    /// Returns the reader handle and the current WAL length in bytes.
135    fn open_wal_reader(log_path: &Path) -> io::Result<(File, u64)> {
136        if !log_path.exists() {
137            File::create(log_path)?;
138        }
139        let reader = File::open(log_path)?;
140        let wal_len = reader.metadata()?.len();
141        Ok((reader, wal_len))
142    }
143
144    /// Loads the payload index, trying a snapshot first, falling back to full WAL replay.
145    ///
146    /// Returns `(index, last_snapshot_wal_position)`.
147    fn load_or_replay_index(
148        dir: &Path,
149        log_path: &Path,
150        wal_len: u64,
151    ) -> io::Result<(FxHashMap<u64, u64>, u64)> {
152        let snapshot_path = dir.join("payloads.snapshot");
153        if let Ok((snapshot_index, snapshot_wal_pos)) = snapshot::load_snapshot(&snapshot_path) {
154            let index = Self::replay_wal_from(log_path, snapshot_index, snapshot_wal_pos, wal_len)?;
155            Ok((index, snapshot_wal_pos))
156        } else {
157            let index = Self::replay_wal_from(log_path, FxHashMap::default(), 0, wal_len)?;
158            Ok((index, 0))
159        }
160    }
161
162    /// Applies the configured durability mode to a WAL writer.
163    fn sync_wal(wal: &mut io::BufWriter<File>, mode: DurabilityMode) -> io::Result<()> {
164        match mode {
165            DurabilityMode::Fsync => {
166                wal.flush()?;
167                wal.get_ref().sync_all()?;
168            }
169            DurabilityMode::FlushOnly => {
170                wal.flush()?;
171            }
172            DurabilityMode::None => {}
173        }
174        Ok(())
175    }
176
177    /// Syncs the WAL according to durability mode, resyncing `write_offset`
178    /// with the actual file length on failure to prevent desync on subsequent
179    /// writes.
180    ///
181    /// RF-2: Shared by `store` and `delete` to eliminate duplicated
182    /// sync-and-resync-offset error handling.
183    fn sync_wal_or_resync(
184        wal: &mut io::BufWriter<File>,
185        mode: DurabilityMode,
186        offset: &mut u64,
187    ) -> io::Result<()> {
188        if let Err(e) = Self::sync_wal(wal, mode) {
189            if let Ok(meta) = wal.get_ref().metadata() {
190                *offset = meta.len();
191            }
192            return Err(e);
193        }
194        Ok(())
195    }
196
197    /// Replays WAL entries from `start_pos` to `end_pos`, updating the index.
198    fn replay_wal_from(
199        log_path: &Path,
200        mut index: FxHashMap<u64, u64>,
201        start_pos: u64,
202        end_pos: u64,
203    ) -> io::Result<FxHashMap<u64, u64>> {
204        if start_pos >= end_pos {
205            return Ok(index);
206        }
207
208        let file = File::open(log_path)?;
209        let mut reader_buf = BufReader::new(file);
210        reader_buf.seek(SeekFrom::Start(start_pos))?;
211
212        let mut pos = start_pos;
213        while pos < end_pos {
214            let Some(entry) = WalEntry::read(&mut reader_buf, pos)? else {
215                break;
216            };
217            // `None` signals a torn tail (crash mid-append): stop cleanly,
218            // keeping every entry replayed so far.
219            let Some(next_pos) = entry.apply(&mut index, &mut reader_buf, end_pos)? else {
220                break;
221            };
222            pos = next_pos;
223        }
224
225        Ok(index)
226    }
227
228    /// Creates a snapshot of the current index state.
229    ///
230    /// The snapshot captures:
231    /// - Current WAL position
232    /// - All index entries (ID -> offset mappings)
233    /// - CRC32 checksum for integrity
234    ///
235    /// # Errors
236    ///
237    /// Returns an error if file operations fail.
238    pub fn create_snapshot(&mut self) -> io::Result<()> {
239        // Flush WAL before snapshotting to ensure data is on disk for the reader
240        {
241            let mut wal = self.wal.write();
242            wal.flush()?;
243            wal.get_ref().sync_all()?;
244        }
245
246        let index = self.index.read();
247        let wal_pos = *self.write_offset.read();
248
249        snapshot::create_snapshot_file(&self.path, &index, wal_pos)?;
250
251        *self.last_snapshot_wal_pos.write() = wal_pos;
252
253        Ok(())
254    }
255
256    /// Returns whether a new snapshot should be created.
257    ///
258    /// Heuristic: Returns true if WAL has grown by more than the default threshold
259    /// bytes since the last snapshot.
260    #[must_use]
261    pub fn should_create_snapshot(&self) -> bool {
262        snapshot::should_create_snapshot(
263            *self.last_snapshot_wal_pos.read(),
264            *self.write_offset.read(),
265        )
266    }
267
268    /// Attempts to create a snapshot if the WAL has grown past the threshold.
269    ///
270    /// Best-effort: on failure the error is logged but not propagated,
271    /// because the WAL write that triggered the check already succeeded.
272    fn maybe_auto_snapshot(&mut self) {
273        if self.should_create_snapshot() {
274            if let Err(e) = self.create_snapshot() {
275                tracing::warn!(
276                    error = %e,
277                    "Auto-snapshot after WAL growth failed; will retry on next write"
278                );
279            }
280        }
281    }
282
283    /// Stores multiple payloads in a single batch operation.
284    ///
285    /// Optimized for bulk imports: acquires WAL + index + offset locks once,
286    /// writes all records sequentially, and performs a **single** durability
287    /// sync at the end instead of per-point fsync.
288    ///
289    /// # Errors
290    ///
291    /// Returns an error if serialization or WAL write fails. On partial failure,
292    /// entries written before the error are durable (WAL is append-only).
293    pub fn store_batch(&mut self, entries: &[(u64, &serde_json::Value)]) -> io::Result<()> {
294        self.store_batch_inner(entries, true)
295    }
296
297    /// Stores multiple payloads without forcing an fsync at the end.
298    ///
299    /// Identical to [`store_batch`](Self::store_batch) except the final
300    /// `sync_all()` is replaced by a buffer-only `flush()`. WAL entries are
301    /// written and the `BufWriter` is flushed to the OS kernel, but not
302    /// fsynced to disk.
303    ///
304    /// Use this for intermediate batches in a streaming bulk import, where
305    /// only the final batch needs full durability. Call
306    /// [`PayloadStorage::flush()`] after the last batch to force fsync.
307    ///
308    /// # Errors
309    ///
310    /// Returns an error if serialization or WAL write fails.
311    pub fn store_batch_deferred(
312        &mut self,
313        entries: &[(u64, &serde_json::Value)],
314    ) -> io::Result<()> {
315        self.store_batch_inner(entries, false)
316    }
317
318    /// Shared implementation for [`store_batch`] and [`store_batch_deferred`].
319    ///
320    /// When `fsync` is `true`, the configured durability mode is applied.
321    /// When `false`, only a buffer flush is performed (no `sync_all`).
322    fn store_batch_inner(
323        &mut self,
324        entries: &[(u64, &serde_json::Value)],
325        fsync: bool,
326    ) -> io::Result<()> {
327        if entries.is_empty() {
328            return Ok(());
329        }
330
331        {
332            let mut wal = self.wal.write();
333            let mut index = self.index.write();
334            let mut offset = self.write_offset.write();
335            let mut record_buf = Vec::with_capacity(256);
336
337            for &(id, payload) in entries {
338                write_store_record(
339                    &mut wal,
340                    id,
341                    payload,
342                    &mut offset,
343                    &mut index,
344                    &mut record_buf,
345                )?;
346            }
347
348            if fsync {
349                Self::sync_wal_or_resync(&mut wal, self.durability, &mut offset)?;
350            } else {
351                // Buffer-only flush: data reaches OS kernel but is not fsynced.
352                // Safe for intermediate batches — caller must fsync after the
353                // final batch.
354                wal.flush()?;
355            }
356        }
357
358        self.maybe_auto_snapshot();
359        Ok(())
360    }
361}
362
363impl PayloadStorage for LogPayloadStorage {
364    fn store(&mut self, id: u64, payload: &serde_json::Value) -> io::Result<()> {
365        // Scoped block: lock guards released before auto-snapshot (which acquires locks).
366        {
367            let mut wal = self.wal.write();
368            let mut index = self.index.write();
369            let mut offset = self.write_offset.write();
370            let mut record_buf = Vec::new();
371
372            write_store_record(
373                &mut wal,
374                id,
375                payload,
376                &mut offset,
377                &mut index,
378                &mut record_buf,
379            )?;
380
381            Self::sync_wal_or_resync(&mut wal, self.durability, &mut offset)?;
382        }
383
384        self.maybe_auto_snapshot();
385        Ok(())
386    }
387
388    fn retrieve(&self, id: u64) -> io::Result<Option<serde_json::Value>> {
389        let index = self.index.read();
390        let Some(&offset) = index.get(&id) else {
391            return Ok(None);
392        };
393        drop(index);
394
395        // H-2: Only flush when DurabilityMode::None is configured, because sync_wal()
396        // already flushes the BufWriter after every write in Fsync and FlushOnly modes.
397        // Skipping this avoids acquiring the WAL write lock on every read, which would
398        // serialize all readers behind writers.
399        if self.durability == DurabilityMode::None {
400            self.wal.write().flush()?;
401        }
402
403        let mut reader = self.reader.write(); // Need write lock to seek
404        let file_len = reader.metadata()?.len();
405        let payload_bytes = read_length_prefixed_payload(&mut *reader, offset, file_len)?;
406
407        let payload = serde_json::from_slice(&payload_bytes)
408            .map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))?;
409
410        Ok(Some(payload))
411    }
412
413    fn delete(&mut self, id: u64) -> io::Result<()> {
414        // If the id is not in the index there is nothing for a tombstone to
415        // shadow on WAL replay. Without this guard, callers that issue
416        // per-entry deletes (e.g. `write_deduped_payloads` with all-None
417        // payloads) pay one fsync per never-stored id.
418        //
419        // SAFETY: `&mut self` serializes all writers, so the read-then-write
420        // gap below cannot race a concurrent `store(id)`. If this signature
421        // is ever relaxed to `&self`, replace this with a single write-lock
422        // acquisition or fold the check inside the existing scoped block.
423        if !self.index.read().contains_key(&id) {
424            return Ok(());
425        }
426
427        let crc = compute_delete_crc(id);
428
429        // Scoped block: all lock guards are released before the auto-snapshot
430        // check, which itself acquires locks (see `create_snapshot`).
431        {
432            let mut wal = self.wal.write();
433            let mut index = self.index.write();
434            let mut offset = self.write_offset.write();
435
436            // H-3: Build complete delete record in one buffer to minimize partial-write window.
437            // CRC-protected format: Marker(0xC4) | ID(8) | CRC32(4)
438            let mut record = [0u8; 1 + 8 + 4];
439            record[0] = CRC_DELETE_MARKER;
440            record[1..9].copy_from_slice(&id.to_le_bytes());
441            record[9..13].copy_from_slice(&crc.to_le_bytes());
442            wal.write_all(&record)?;
443
444            // Sync WAL according to durability mode (resync offset on failure).
445            Self::sync_wal_or_resync(&mut wal, self.durability, &mut offset)?;
446
447            *offset += 1 + 8 + 4; // Marker(1) + ID(8) + CRC32(4)
448            index.remove(&id);
449        }
450
451        self.maybe_auto_snapshot();
452        Ok(())
453    }
454
455    fn flush(&mut self) -> io::Result<()> {
456        let mut wal = self.wal.write();
457        Self::sync_wal(&mut wal, self.durability)
458    }
459
460    fn ids(&self) -> Vec<u64> {
461        self.index.read().keys().copied().collect()
462    }
463}
464
465/// Reads a length-prefixed payload at `offset`: a 4-byte LE length followed by
466/// that many payload bytes. The declared length is bounded by the bytes
467/// remaining after the prefix (OOM guard #897/#898) before allocating.
468fn read_length_prefixed_payload<R: Read + Seek>(
469    reader: &mut R,
470    offset: u64,
471    file_len: u64,
472) -> io::Result<Vec<u8>> {
473    reader.seek(SeekFrom::Start(offset))?;
474
475    let mut len_bytes = [0u8; 4];
476    reader.read_exact(&mut len_bytes)?;
477    let declared = u64::from(u32::from_le_bytes(len_bytes));
478
479    // OOM guard (#897/#898): a corrupt length field must not drive an unbounded
480    // allocation. The payload cannot exceed the bytes remaining after the prefix.
481    let pos_after_len = offset.saturating_add(4);
482    let remaining = file_len.saturating_sub(pos_after_len);
483    if declared > remaining {
484        return Err(io::Error::new(
485            io::ErrorKind::InvalidData,
486            "payload length exceeds file size",
487        ));
488    }
489    let len = usize::try_from(declared)
490        .map_err(|_| io::Error::new(io::ErrorKind::InvalidData, "payload length overflow"))?;
491
492    let mut payload_bytes = vec![0u8; len];
493    reader.read_exact(&mut payload_bytes)?;
494    Ok(payload_bytes)
495}