1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
use crate::{Spool, SpoolEntry, SpoolId};
use async_trait::async_trait;
use chrono::{DateTime, Utc};
use flume::Sender;
use prometheus::IntGaugeVec;
use rocksdb::perf::get_memory_usage_stats;
use rocksdb::{
    DBCompressionType, ErrorKind, IteratorMode, LogLevel, Options, WriteBatch, WriteOptions, DB,
};
use serde::{Deserialize, Serialize};
use std::path::Path;
use std::sync::{Arc, LazyLock, Weak};
use std::time::Duration;
use tokio::runtime::Handle;

#[derive(Serialize, Deserialize, Debug)]
pub struct RocksSpoolParams {
    pub increase_parallelism: Option<i32>,

    pub optimize_level_style_compaction: Option<usize>,
    pub optimize_universal_style_compaction: Option<usize>,
    #[serde(default)]
    pub paranoid_checks: bool,
    #[serde(default)]
    pub compression_type: DBCompressionTypeDef,

    /// If non-zero, we perform bigger reads when doing compaction. If you’re running RocksDB on
    /// spinning disks, you should set this to at least 2MB. That way RocksDB’s compaction is doing
    /// sequential instead of random reads
    pub compaction_readahead_size: Option<usize>,

    #[serde(default)]
    pub level_compaction_dynamic_level_bytes: bool,

    #[serde(default)]
    pub max_open_files: Option<usize>,

    #[serde(default)]
    pub log_level: LogLevelDef,

    /// See:
    /// <https://docs.rs/rocksdb/latest/rocksdb/struct.Options.html#method.set_memtable_huge_page_size>
    #[serde(default)]
    pub memtable_huge_page_size: Option<usize>,

    #[serde(
        with = "duration_serde",
        default = "RocksSpoolParams::default_log_file_time_to_roll"
    )]
    pub log_file_time_to_roll: Duration,

    #[serde(
        with = "duration_serde",
        default = "RocksSpoolParams::default_obsolete_files_period"
    )]
    pub obsolete_files_period: Duration,
}

impl Default for RocksSpoolParams {
    fn default() -> Self {
        Self {
            increase_parallelism: None,
            optimize_level_style_compaction: None,
            optimize_universal_style_compaction: None,
            paranoid_checks: false,
            compression_type: DBCompressionTypeDef::default(),
            compaction_readahead_size: None,
            level_compaction_dynamic_level_bytes: false,
            max_open_files: None,
            log_level: LogLevelDef::default(),
            memtable_huge_page_size: None,
            log_file_time_to_roll: Self::default_log_file_time_to_roll(),
            obsolete_files_period: Self::default_obsolete_files_period(),
        }
    }
}

impl RocksSpoolParams {
    fn default_log_file_time_to_roll() -> Duration {
        let one_day = Duration::from_secs(86400);
        one_day
    }

    fn default_obsolete_files_period() -> Duration {
        let six_hours = Duration::from_secs(6 * 60 * 60);
        six_hours
    }
}

#[derive(Serialize, Deserialize, Debug)]
pub enum DBCompressionTypeDef {
    None,
    Snappy,
    Zlib,
    Bz2,
    Lz4,
    Lz4hc,
    Zstd,
}

impl Into<DBCompressionType> for DBCompressionTypeDef {
    fn into(self) -> DBCompressionType {
        match self {
            Self::None => DBCompressionType::None,
            Self::Snappy => DBCompressionType::Snappy,
            Self::Zlib => DBCompressionType::Zlib,
            Self::Bz2 => DBCompressionType::Bz2,
            Self::Lz4 => DBCompressionType::Lz4,
            Self::Lz4hc => DBCompressionType::Lz4hc,
            Self::Zstd => DBCompressionType::Zstd,
        }
    }
}

impl Default for DBCompressionTypeDef {
    fn default() -> Self {
        Self::Snappy
    }
}

#[derive(Serialize, Deserialize, Debug)]
pub enum LogLevelDef {
    Debug,
    Info,
    Warn,
    Error,
    Fatal,
    Header,
}

impl Default for LogLevelDef {
    fn default() -> Self {
        Self::Info
    }
}

impl Into<LogLevel> for LogLevelDef {
    fn into(self) -> LogLevel {
        match self {
            Self::Debug => LogLevel::Debug,
            Self::Info => LogLevel::Info,
            Self::Warn => LogLevel::Warn,
            Self::Error => LogLevel::Error,
            Self::Fatal => LogLevel::Fatal,
            Self::Header => LogLevel::Header,
        }
    }
}

pub struct RocksSpool {
    db: Arc<DB>,
    runtime: Handle,
}

impl RocksSpool {
    pub fn new(
        path: &Path,
        flush: bool,
        params: Option<RocksSpoolParams>,
        runtime: Handle,
    ) -> anyhow::Result<Self> {
        let mut opts = Options::default();
        opts.set_use_fsync(flush);
        opts.create_if_missing(true);
        // The default is 1000, which is a bit high
        opts.set_keep_log_file_num(10);

        let p = params.unwrap_or_default();
        if let Some(i) = p.increase_parallelism {
            opts.increase_parallelism(i);
        }
        if let Some(i) = p.optimize_level_style_compaction {
            opts.optimize_level_style_compaction(i);
        }
        if let Some(i) = p.optimize_universal_style_compaction {
            opts.optimize_universal_style_compaction(i);
        }
        if let Some(i) = p.compaction_readahead_size {
            opts.set_compaction_readahead_size(i);
        }
        if let Some(i) = p.max_open_files {
            opts.set_max_open_files(i as _);
        }
        if let Some(i) = p.memtable_huge_page_size {
            opts.set_memtable_huge_page_size(i);
        }
        opts.set_paranoid_checks(p.paranoid_checks);
        opts.set_level_compaction_dynamic_level_bytes(p.level_compaction_dynamic_level_bytes);
        opts.set_compression_type(p.compression_type.into());
        opts.set_log_level(p.log_level.into());
        opts.set_log_file_time_to_roll(p.log_file_time_to_roll.as_secs() as usize);
        opts.set_delete_obsolete_files_period_micros(p.obsolete_files_period.as_micros() as u64);

        let db = Arc::new(DB::open(&opts, path)?);

        {
            let weak = Arc::downgrade(&db);
            tokio::spawn(metrics_monitor(weak, format!("{}", path.display())));
        }

        Ok(Self { db, runtime })
    }
}

#[async_trait]
impl Spool for RocksSpool {
    async fn load(&self, id: SpoolId) -> anyhow::Result<Vec<u8>> {
        let db = self.db.clone();
        tokio::task::Builder::new()
            .name("rocksdb load")
            .spawn_blocking_on(
                move || {
                    Ok(db
                        .get(id.as_bytes())?
                        .ok_or_else(|| anyhow::anyhow!("no such key {id}"))?)
                },
                &self.runtime,
            )?
            .await?
    }

    async fn store(
        &self,
        id: SpoolId,
        data: Arc<Box<[u8]>>,
        force_sync: bool,
    ) -> anyhow::Result<()> {
        let mut opts = WriteOptions::default();
        opts.set_sync(force_sync);
        opts.set_no_slowdown(true);
        let mut batch = WriteBatch::default();
        batch.put(id.as_bytes(), &*data);

        match self.db.write_opt(batch, &opts) {
            Ok(()) => Ok(()),
            Err(err) if err.kind() == ErrorKind::Incomplete => {
                let db = self.db.clone();
                tokio::task::Builder::new()
                    .name("rocksdb store")
                    .spawn_blocking_on(
                        move || {
                            opts.set_no_slowdown(false);
                            let mut batch = WriteBatch::default();
                            batch.put(id.as_bytes(), &*data);
                            Ok(db.write_opt(batch, &opts)?)
                        },
                        &self.runtime,
                    )?
                    .await?
            }
            Err(err) => Err(err.into()),
        }
    }

    async fn remove(&self, id: SpoolId) -> anyhow::Result<()> {
        let mut opts = WriteOptions::default();
        opts.set_no_slowdown(true);
        let mut batch = WriteBatch::default();
        batch.delete(id.as_bytes());

        match self.db.write_opt(batch, &opts) {
            Ok(()) => Ok(()),
            Err(err) if err.kind() == ErrorKind::Incomplete => {
                let db = self.db.clone();
                tokio::task::Builder::new()
                    .name("rocksdb remove")
                    .spawn_blocking_on(
                        move || {
                            opts.set_no_slowdown(false);
                            let mut batch = WriteBatch::default();
                            batch.delete(id.as_bytes());
                            Ok(db.write_opt(batch, &opts)?)
                        },
                        &self.runtime,
                    )?
                    .await?
            }
            Err(err) => Err(err.into()),
        }
    }

    async fn cleanup(&self) -> anyhow::Result<()> {
        Ok(())
    }

    async fn shutdown(&self) -> anyhow::Result<()> {
        let db = self.db.clone();
        tokio::task::spawn_blocking(move || db.cancel_all_background_work(true)).await?;
        Ok(())
    }

    async fn advise_low_memory(&self) -> anyhow::Result<isize> {
        let db = self.db.clone();
        tokio::task::spawn_blocking(move || {
            let usage_before = match get_memory_usage_stats(Some(&[&db]), None) {
                Ok(stats) => {
                    let stats: Stats = stats.into();
                    tracing::debug!("pre-flush: {stats:#?}");
                    stats.total()
                }
                Err(err) => {
                    tracing::error!("error getting stats: {err:#}");
                    0
                }
            };

            if let Err(err) = db.flush() {
                tracing::error!("error flushing memory: {err:#}");
            }

            let usage_after = match get_memory_usage_stats(Some(&[&db]), None) {
                Ok(stats) => {
                    let stats: Stats = stats.into();
                    tracing::debug!("post-flush: {stats:#?}");
                    stats.total()
                }
                Err(err) => {
                    tracing::error!("error getting stats: {err:#}");
                    0
                }
            };

            Ok(usage_before - usage_after)
        })
        .await?
    }

    fn enumerate(
        &self,
        sender: Sender<SpoolEntry>,
        start_time: DateTime<Utc>,
    ) -> anyhow::Result<()> {
        let db = Arc::clone(&self.db);
        tokio::task::Builder::new()
            .name("rocksdb enumerate")
            .spawn_blocking_on(
                move || {
                    let iter = db.iterator(IteratorMode::Start);
                    for entry in iter {
                        let (key, value) = entry?;
                        let id = SpoolId::from_slice(&key)
                            .ok_or_else(|| anyhow::anyhow!("invalid spool id {key:?}"))?;

                        if id.created() >= start_time {
                            // Entries created since we started must have
                            // landed there after we started and are thus
                            // not eligible for discovery via enumeration
                            continue;
                        }

                        sender
                            .send(SpoolEntry::Item {
                                id,
                                data: value.to_vec(),
                            })
                            .map_err(|err| {
                                anyhow::anyhow!("failed to send SpoolEntry for {id}: {err:#}")
                            })?;
                    }
                    Ok::<(), anyhow::Error>(())
                },
                &self.runtime,
            )?;
        Ok(())
    }
}

#[cfg(test)]
mod test {
    use super::*;

    #[tokio::test]
    async fn rocks_spool() -> anyhow::Result<()> {
        let location = tempfile::tempdir()?;
        let spool = RocksSpool::new(&location.path(), false, None, Handle::current())?;

        {
            let id1 = SpoolId::new();

            // Can't load an entry that doesn't exist
            assert_eq!(
                format!("{:#}", spool.load(id1).await.unwrap_err()),
                format!("no such key {id1}")
            );
        }

        // Insert some entries
        let mut ids = vec![];
        for i in 0..100 {
            let id = SpoolId::new();
            spool
                .store(
                    id,
                    Arc::new(format!("I am {i}").as_bytes().to_vec().into_boxed_slice()),
                    false,
                )
                .await?;
            ids.push(id);
        }

        // Verify that we can load those entries
        for (i, &id) in ids.iter().enumerate() {
            let data = spool.load(id).await?;
            let text = String::from_utf8(data)?;
            assert_eq!(text, format!("I am {i}"));
        }

        {
            // Verify that we can enumerate them
            let (tx, rx) = flume::bounded(32);
            spool.enumerate(tx, Utc::now())?;
            let mut count = 0;

            while let Ok(item) = rx.recv_async().await {
                match item {
                    SpoolEntry::Item { id, data } => {
                        let i = ids
                            .iter()
                            .position(|&item| item == id)
                            .ok_or_else(|| anyhow::anyhow!("{id} not found in ids!"))?;

                        let text = String::from_utf8(data)?;
                        assert_eq!(text, format!("I am {i}"));

                        spool.remove(id).await?;
                        // Can't load an entry that we just removed
                        assert_eq!(
                            format!("{:#}", spool.load(id).await.unwrap_err()),
                            format!("no such key {id}")
                        );
                        count += 1;
                    }
                    SpoolEntry::Corrupt { id, error } => {
                        anyhow::bail!("Corrupt: {id}: {error}");
                    }
                }
            }

            assert_eq!(count, 100);
        }

        // Now that we've removed the files, try enumerating again.
        // We expect to receive no entries.
        // Do it a couple of times to verify that none of the cleanup
        // stuff that happens in enumerate breaks the directory
        // structure
        for _ in 0..2 {
            // Verify that we can enumerate them
            let (tx, rx) = flume::bounded(32);
            spool.enumerate(tx, Utc::now())?;
            let mut unexpected = vec![];

            while let Ok(item) = rx.recv_async().await {
                match item {
                    SpoolEntry::Item { id, .. } | SpoolEntry::Corrupt { id, .. } => {
                        unexpected.push(id)
                    }
                }
            }

            assert_eq!(unexpected.len(), 0);
        }

        Ok(())
    }
}

/// The rocksdb type doesn't impl Debug, so we get to do it
#[allow(unused)]
#[derive(Debug)]
struct Stats {
    pub mem_table_total: u64,
    pub mem_table_unflushed: u64,
    pub mem_table_readers_total: u64,
    pub cache_total: u64,
}

impl Stats {
    fn total(&self) -> isize {
        (self.mem_table_total + self.mem_table_readers_total + self.cache_total) as isize
    }
}

impl From<rocksdb::perf::MemoryUsageStats> for Stats {
    fn from(s: rocksdb::perf::MemoryUsageStats) -> Self {
        Self {
            mem_table_total: s.mem_table_total,
            mem_table_unflushed: s.mem_table_unflushed,
            mem_table_readers_total: s.mem_table_readers_total,
            cache_total: s.cache_total,
        }
    }
}

static MEM_TABLE_TOTAL: LazyLock<IntGaugeVec> = LazyLock::new(|| {
    prometheus::register_int_gauge_vec!(
        "rocks_spool_mem_table_total",
        "Approximate memory usage of all the mem-tables",
        &["path"]
    )
    .unwrap()
});
static MEM_TABLE_UNFLUSHED: LazyLock<IntGaugeVec> = LazyLock::new(|| {
    prometheus::register_int_gauge_vec!(
        "rocks_spool_mem_table_unflushed",
        "Approximate memory usage of un-flushed mem-tables",
        &["path"]
    )
    .unwrap()
});
static MEM_TABLE_READERS_TOTAL: LazyLock<IntGaugeVec> = LazyLock::new(|| {
    prometheus::register_int_gauge_vec!(
        "rocks_spool_mem_table_readers_total",
        "Approximate memory usage of all the table readers",
        &["path"]
    )
    .unwrap()
});
static CACHE_TOTAL: LazyLock<IntGaugeVec> = LazyLock::new(|| {
    prometheus::register_int_gauge_vec!(
        "rocks_spool_cache_total",
        "Approximate memory usage by cache",
        &["path"]
    )
    .unwrap()
});

async fn metrics_monitor(db: Weak<DB>, path: String) {
    let mem_table_total = MEM_TABLE_TOTAL
        .get_metric_with_label_values(&[path.as_str()])
        .unwrap();
    let mem_table_unflushed = MEM_TABLE_UNFLUSHED
        .get_metric_with_label_values(&[path.as_str()])
        .unwrap();
    let mem_table_readers_total = MEM_TABLE_READERS_TOTAL
        .get_metric_with_label_values(&[path.as_str()])
        .unwrap();
    let cache_total = CACHE_TOTAL
        .get_metric_with_label_values(&[path.as_str()])
        .unwrap();

    loop {
        match db.upgrade() {
            Some(db) => {
                match get_memory_usage_stats(Some(&[&db]), None) {
                    Ok(stats) => {
                        mem_table_total.set(stats.mem_table_total as i64);
                        mem_table_unflushed.set(stats.mem_table_unflushed as i64);
                        mem_table_readers_total.set(stats.mem_table_readers_total as i64);
                        cache_total.set(stats.cache_total as i64);
                    }
                    Err(err) => {
                        tracing::error!("error getting stats: {err:#}");
                    }
                };
            }
            None => {
                // Dead
                return;
            }
        }
        tokio::time::sleep(tokio::time::Duration::from_secs(5)).await;
    }
}