aboutsummaryrefslogtreecommitdiff
path: root/src/block/manager.rs
blob: 848d91411826c97774f61e24fe0f75d2d286bd6c (plain) (blame)
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
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
use std::path::PathBuf;
use std::pin::Pin;
use std::sync::Arc;
use std::time::Duration;

use arc_swap::{ArcSwap, ArcSwapOption};
use async_trait::async_trait;
use bytes::Bytes;
use rand::prelude::*;
use serde::{Deserialize, Serialize};

use futures::Stream;
use futures_util::stream::StreamExt;
use tokio::fs;
use tokio::io::{AsyncReadExt, AsyncWriteExt, BufReader};
use tokio::sync::{mpsc, Mutex, MutexGuard};

use opentelemetry::{
	trace::{FutureExt as OtelFutureExt, TraceContextExt, Tracer},
	Context,
};

use garage_net::stream::{stream_asyncread, ByteStream};

use garage_db as db;

use garage_util::background::{vars, BackgroundRunner};
use garage_util::config::DataDirEnum;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::metrics::RecordDuration;
use garage_util::persister::{Persister, PersisterShared};
use garage_util::time::msec_to_rfc3339;

use garage_rpc::rpc_helper::OrderTag;
use garage_rpc::system::System;
use garage_rpc::*;

use garage_table::replication::{TableReplication, TableShardedReplication};

use crate::block::*;
use crate::layout::*;
use crate::metrics::*;
use crate::rc::*;
use crate::repair::*;
use crate::resync::*;

/// Size under which data will be stored inlined in database instead of as files
pub const INLINE_THRESHOLD: usize = 3072;

// The delay between the moment when the reference counter
// drops to zero, and the moment where we allow ourselves
// to delete the block locally.
pub(crate) const BLOCK_GC_DELAY: Duration = Duration::from_secs(600);

pub type BlockStream =
	Pin<Box<dyn Stream<Item = Result<Bytes, std::io::Error>> + Send + Sync + 'static>>;

/// RPC messages used to share blocks of data between nodes
#[derive(Debug, Serialize, Deserialize)]
pub enum BlockRpc {
	Ok,
	/// Message to ask for a block of data, by hash
	GetBlock(Hash, Option<OrderTag>),
	/// Message to send a block of data, either because requested, of for first delivery of new
	/// block
	PutBlock {
		hash: Hash,
		header: DataBlockHeader,
	},
	/// Ask other node if they should have this block, but don't actually have it
	NeedBlockQuery(Hash),
	/// Response : whether the node do require that block
	NeedBlockReply(bool),
}

impl Rpc for BlockRpc {
	type Response = Result<BlockRpc, Error>;
}

/// The block manager, handling block exchange between nodes, and block storage on local node
pub struct BlockManager {
	/// Replication strategy, allowing to find on which node blocks should be located
	pub replication: TableShardedReplication,

	/// Data layout
	pub(crate) data_layout: ArcSwap<DataLayout>,
	/// Data layout persister
	pub(crate) data_layout_persister: Persister<DataLayout>,

	data_fsync: bool,
	compression_level: Option<i32>,

	mutation_lock: Vec<Mutex<BlockManagerLocked>>,

	pub(crate) rc: BlockRc,
	pub resync: BlockResyncManager,

	pub(crate) system: Arc<System>,
	pub(crate) endpoint: Arc<Endpoint<BlockRpc, Self>>,

	pub(crate) metrics: BlockManagerMetrics,

	pub scrub_persister: PersisterShared<ScrubWorkerPersisted>,
	tx_scrub_command: ArcSwapOption<mpsc::Sender<ScrubWorkerCommand>>,
}

#[derive(Serialize, Deserialize, Clone, Debug)]
pub struct BlockResyncErrorInfo {
	pub hash: Hash,
	pub refcount: u64,
	pub error_count: u64,
	pub last_try: u64,
	pub next_try: u64,
}

// The number of different mutexes used to parallelize write access to data blocks
const MUTEX_COUNT: usize = 256;

// This custom struct contains functions that must only be ran
// when the lock is held. We ensure that it is the case by storing
// it INSIDE a Mutex.
struct BlockManagerLocked();

impl BlockManager {
	pub fn new(
		db: &db::Db,
		data_dir: DataDirEnum,
		data_fsync: bool,
		compression_level: Option<i32>,
		replication: TableShardedReplication,
		system: Arc<System>,
	) -> Result<Arc<Self>, Error> {
		// Load or compute layout, i.e. assignment of data blocks to the different data directories
		let data_layout_persister: Persister<DataLayout> =
			Persister::new(&system.metadata_dir, "data_layout");
		let data_layout = match data_layout_persister.load() {
			Ok(mut layout) => {
				layout
					.update(&data_dir)
					.ok_or_message("invalid data_dir config")?;
				layout
			}
			Err(_) => DataLayout::initialize(&data_dir).ok_or_message("invalid data_dir config")?,
		};
		data_layout_persister
			.save(&data_layout)
			.expect("cannot save data_layout");

		// Open metadata tables
		let rc = db
			.open_tree("block_local_rc")
			.expect("Unable to open block_local_rc tree");
		let rc = BlockRc::new(rc);

		let resync = BlockResyncManager::new(db, &system);

		let endpoint = system
			.netapp
			.endpoint("garage_block/manager.rs/Rpc".to_string());

		let metrics = BlockManagerMetrics::new(
			compression_level,
			rc.rc.clone(),
			resync.queue.clone(),
			resync.errors.clone(),
		);

		let scrub_persister = PersisterShared::new(&system.metadata_dir, "scrub_info");

		let block_manager = Arc::new(Self {
			replication,
			data_layout: ArcSwap::new(Arc::new(data_layout)),
			data_layout_persister,
			data_fsync,
			compression_level,
			mutation_lock: vec![(); MUTEX_COUNT]
				.iter()
				.map(|_| Mutex::new(BlockManagerLocked()))
				.collect::<Vec<_>>(),
			rc,
			resync,
			system,
			endpoint,
			metrics,
			scrub_persister,
			tx_scrub_command: ArcSwapOption::new(None),
		});
		block_manager.endpoint.set_handler(block_manager.clone());
		block_manager.scrub_persister.set_with(|_| ()).unwrap();

		Ok(block_manager)
	}

	pub fn spawn_workers(self: &Arc<Self>, bg: &BackgroundRunner) {
		// Spawn a bunch of resync workers
		for index in 0..MAX_RESYNC_WORKERS {
			let worker = ResyncWorker::new(index, self.clone());
			bg.spawn_worker(worker);
		}

		// Spawn scrub worker
		let (scrub_tx, scrub_rx) = mpsc::channel(1);
		self.tx_scrub_command.store(Some(Arc::new(scrub_tx)));
		bg.spawn_worker(ScrubWorker::new(
			self.clone(),
			scrub_rx,
			self.scrub_persister.clone(),
		));
	}

	pub fn register_bg_vars(&self, vars: &mut vars::BgVars) {
		self.resync.register_bg_vars(vars);

		vars.register_rw(
			&self.scrub_persister,
			"scrub-tranquility",
			|p| p.get_with(|x| x.tranquility),
			|p, tranquility| p.set_with(|x| x.tranquility = tranquility),
		);
		vars.register_ro(&self.scrub_persister, "scrub-last-completed", |p| {
			p.get_with(|x| msec_to_rfc3339(x.time_last_complete_scrub))
		});
		vars.register_ro(&self.scrub_persister, "scrub-next-run", |p| {
			p.get_with(|x| msec_to_rfc3339(x.time_next_run_scrub))
		});
		vars.register_ro(&self.scrub_persister, "scrub-corruptions_detected", |p| {
			p.get_with(|x| x.corruptions_detected)
		});
	}

	/// Ask nodes that might have a (possibly compressed) block for it
	/// Return it as a stream with a header
	async fn rpc_get_raw_block_streaming(
		&self,
		hash: &Hash,
		order_tag: Option<OrderTag>,
	) -> Result<(DataBlockHeader, ByteStream), Error> {
		self.rpc_get_raw_block_internal(hash, order_tag, |header, stream| async move {
			Ok((header, stream))
		})
		.await
	}

	/// Ask nodes that might have a (possibly compressed) block for it
	/// Return its entire body
	pub(crate) async fn rpc_get_raw_block(
		&self,
		hash: &Hash,
		order_tag: Option<OrderTag>,
	) -> Result<DataBlock, Error> {
		self.rpc_get_raw_block_internal(hash, order_tag, |header, stream| async move {
			read_stream_to_end(stream)
				.await
				.map(|data| DataBlock::from_parts(header, data))
		})
		.await
	}

	async fn rpc_get_raw_block_internal<F, Fut, T>(
		&self,
		hash: &Hash,
		order_tag: Option<OrderTag>,
		f: F,
	) -> Result<T, Error>
	where
		F: Fn(DataBlockHeader, ByteStream) -> Fut,
		Fut: futures::Future<Output = Result<T, Error>>,
	{
		let who = self.replication.read_nodes(hash);
		let who = self.system.rpc.request_order(&who);

		for node in who.iter() {
			let node_id = NodeID::from(*node);
			let rpc = self.endpoint.call_streaming(
				&node_id,
				BlockRpc::GetBlock(*hash, order_tag),
				PRIO_NORMAL | PRIO_SECONDARY,
			);
			tokio::select! {
				res = rpc => {
					let res = match res {
						Ok(res) => res,
						Err(e) => {
							debug!("Get block {:?}: node {:?} could not be contacted: {}", hash, node, e);
							continue;
						}
					};
					let (header, stream) = match res.into_parts() {
						(Ok(BlockRpc::PutBlock { hash: _, header }), Some(stream)) => (header, stream),
						(Ok(_), _) => {
							debug!("Get block {:?}: node {:?} returned a malformed response", hash, node);
							continue;
						}
						(Err(e), _) => {
							debug!("Get block {:?}: node {:?} returned error: {}", hash, node, e);
							continue;
						}
					};
					match f(header, stream).await {
						Ok(ret) => return Ok(ret),
						Err(e) => {
							debug!("Get block {:?}: error reading stream from node {:?}: {}", hash, node, e);
						}
					}
				}
				// TODO: sleep less long (fail early), initiate a second request earlier
				// if the first one doesn't succeed rapidly
				// TODO: keep first request running when initiating a new one and take the
				// one that finishes earlier
				_ = tokio::time::sleep(self.system.rpc.rpc_timeout()) => {
					debug!("Get block {:?}: node {:?} didn't return block in time, trying next.", hash, node);
				}
			};
		}

		let msg = format!("Get block {:?}: no node returned a valid block", hash);
		debug!("{}", msg);
		Err(Error::Message(msg))
	}

	// ---- Public interface ----

	/// Ask nodes that might have a block for it,
	/// return it as a stream
	pub async fn rpc_get_block_streaming(
		&self,
		hash: &Hash,
		order_tag: Option<OrderTag>,
	) -> Result<BlockStream, Error> {
		let (header, stream) = self.rpc_get_raw_block_streaming(hash, order_tag).await?;
		match header {
			DataBlockHeader::Plain => Ok(stream),
			DataBlockHeader::Compressed => {
				// Too many things, I hate it.
				let reader = stream_asyncread(stream);
				let reader = BufReader::new(reader);
				let reader = async_compression::tokio::bufread::ZstdDecoder::new(reader);
				Ok(Box::pin(tokio_util::io::ReaderStream::new(reader)))
			}
		}
	}

	/// Ask nodes that might have a block for it
	pub async fn rpc_get_block(
		&self,
		hash: &Hash,
		order_tag: Option<OrderTag>,
	) -> Result<Bytes, Error> {
		self.rpc_get_raw_block(hash, order_tag)
			.await?
			.verify_get(*hash)
	}

	/// Send block to nodes that should have it
	pub async fn rpc_put_block(&self, hash: Hash, data: Bytes) -> Result<(), Error> {
		let who = self.replication.write_nodes(&hash);

		let (header, bytes) = DataBlock::from_buffer(data, self.compression_level)
			.await
			.into_parts();
		let put_block_rpc =
			Req::new(BlockRpc::PutBlock { hash, header })?.with_stream_from_buffer(bytes);

		self.system
			.rpc
			.try_call_many(
				&self.endpoint,
				&who[..],
				put_block_rpc,
				RequestStrategy::with_priority(PRIO_NORMAL | PRIO_SECONDARY)
					.with_quorum(self.replication.write_quorum()),
			)
			.await?;

		Ok(())
	}

	/// Get number of items in the refcount table
	pub fn rc_len(&self) -> Result<usize, Error> {
		Ok(self.rc.rc.len()?)
	}

	/// Get number of items in the refcount table
	pub fn rc_fast_len(&self) -> Result<Option<usize>, Error> {
		Ok(self.rc.rc.fast_len()?)
	}

	/// Send command to start/stop/manager scrub worker
	pub async fn send_scrub_command(&self, cmd: ScrubWorkerCommand) -> Result<(), Error> {
		let tx = self.tx_scrub_command.load();
		let tx = tx.as_ref().ok_or_message("scrub worker is not running")?;
		tx.send(cmd).await.ok_or_message("send error")?;
		Ok(())
	}

	/// Get the reference count of a block
	pub fn get_block_rc(&self, hash: &Hash) -> Result<u64, Error> {
		Ok(self.rc.get_block_rc(hash)?.as_u64())
	}

	/// List all resync errors
	pub fn list_resync_errors(&self) -> Result<Vec<BlockResyncErrorInfo>, Error> {
		let mut blocks = Vec::with_capacity(self.resync.errors.len());
		for ent in self.resync.errors.iter()? {
			let (hash, cnt) = ent?;
			let cnt = ErrorCounter::decode(&cnt);
			blocks.push(BlockResyncErrorInfo {
				hash: Hash::try_from(&hash).unwrap(),
				refcount: 0,
				error_count: cnt.errors,
				last_try: cnt.last_try,
				next_try: cnt.next_try(),
			});
		}
		for block in blocks.iter_mut() {
			block.refcount = self.get_block_rc(&block.hash)?;
		}
		Ok(blocks)
	}

	//// ----- Managing the reference counter ----

	/// Increment the number of time a block is used, putting it to resynchronization if it is
	/// required, but not known
	pub fn block_incref(
		self: &Arc<Self>,
		tx: &mut db::Transaction,
		hash: Hash,
	) -> db::TxOpResult<()> {
		if self.rc.block_incref(tx, &hash)? {
			// When the reference counter is incremented, there is
			// normally a node that is responsible for sending us the
			// data of the block. However that operation may fail,
			// so in all cases we add the block here to the todo list
			// to check later that it arrived correctly, and if not
			// we will fecth it from someone.
			let this = self.clone();
			tokio::spawn(async move {
				if let Err(e) = this
					.resync
					.put_to_resync(&hash, 2 * this.system.rpc.rpc_timeout())
				{
					error!("Block {:?} could not be put in resync queue: {}.", hash, e);
				}
			});
		}
		Ok(())
	}

	/// Decrement the number of time a block is used
	pub fn block_decref(
		self: &Arc<Self>,
		tx: &mut db::Transaction,
		hash: Hash,
	) -> db::TxOpResult<()> {
		if self.rc.block_decref(tx, &hash)? {
			// When the RC is decremented, it might drop to zero,
			// indicating that we don't need the block.
			// There is a delay before we garbage collect it;
			// make sure that it is handled in the resync loop
			// after that delay has passed.
			let this = self.clone();
			tokio::spawn(async move {
				if let Err(e) = this
					.resync
					.put_to_resync(&hash, BLOCK_GC_DELAY + Duration::from_secs(10))
				{
					error!("Block {:?} could not be put in resync queue: {}.", hash, e);
				}
			});
		}
		Ok(())
	}

	// ---- Reading and writing blocks locally ----

	async fn handle_put_block(
		&self,
		hash: Hash,
		header: DataBlockHeader,
		stream: Option<ByteStream>,
	) -> Result<(), Error> {
		let stream = stream.ok_or_message("missing stream")?;
		let bytes = read_stream_to_end(stream).await?;
		let data = DataBlock::from_parts(header, bytes);
		self.write_block(&hash, &data).await
	}

	/// Write a block to disk
	pub(crate) async fn write_block(&self, hash: &Hash, data: &DataBlock) -> Result<(), Error> {
		let tracer = opentelemetry::global::tracer("garage");

		self.lock_mutate(hash)
			.await
			.write_block(hash, data, self)
			.bound_record_duration(&self.metrics.block_write_duration)
			.with_context(Context::current_with_span(
				tracer.start("BlockManagerLocked::write_block"),
			))
			.await?;

		Ok(())
	}

	async fn handle_get_block(&self, hash: &Hash, order_tag: Option<OrderTag>) -> Resp<BlockRpc> {
		let block = match self.read_block(hash).await {
			Ok(data) => data,
			Err(e) => return Resp::new(Err(e)),
		};

		let (header, data) = block.into_parts();

		let resp = Resp::new(Ok(BlockRpc::PutBlock {
			hash: *hash,
			header,
		}))
		.with_stream_from_buffer(data);

		if let Some(order_tag) = order_tag {
			resp.with_order_tag(order_tag)
		} else {
			resp
		}
	}

	/// Read block from disk, verifying it's integrity
	pub(crate) async fn read_block(&self, hash: &Hash) -> Result<DataBlock, Error> {
		let tracer = opentelemetry::global::tracer("garage");
		async {
			match self.find_block(hash).await {
				Some(p) => self.read_block_from(hash, &p).await,
				None => {
					// Not found but maybe we should have had it ??
					self.resync
						.put_to_resync(hash, 2 * self.system.rpc.rpc_timeout())?;
					return Err(Error::Message(format!(
						"block {:?} not found on node",
						hash
					)));
				}
			}
		}
		.bound_record_duration(&self.metrics.block_read_duration)
		.with_context(Context::current_with_span(
			tracer.start("BlockManager::read_block"),
		))
		.await
	}

	pub(crate) async fn read_block_from(
		&self,
		hash: &Hash,
		block_path: &DataBlockPath,
	) -> Result<DataBlock, Error> {
		let (path, compressed) = match block_path {
			DataBlockPath::Plain(p) => (p, false),
			DataBlockPath::Compressed(p) => (p, true),
		};

		let mut f = fs::File::open(&path).await?;
		let mut data = vec![];
		f.read_to_end(&mut data).await?;
		self.metrics.bytes_read.add(data.len() as u64);
		drop(f);

		let data = if compressed {
			DataBlock::Compressed(data.into())
		} else {
			DataBlock::Plain(data.into())
		};

		if data.verify(*hash).is_err() {
			self.metrics.corruption_counter.add(1);

			warn!(
				"Block {:?} is corrupted. Renaming to .corrupted and resyncing.",
				hash
			);
			self.lock_mutate(hash)
				.await
				.move_block_to_corrupted(block_path)
				.await?;
			self.resync.put_to_resync(hash, Duration::from_millis(0))?;

			return Err(Error::CorruptData(*hash));
		}

		Ok(data)
	}

	/// Check if this node should have a block, but don't actually have it
	async fn need_block(&self, hash: &Hash) -> Result<bool, Error> {
		let rc = self.rc.get_block_rc(hash)?;
		let exists = self.find_block(hash).await.is_some();
		Ok(rc.is_nonzero() && !exists)
	}

	/// Delete block if it is not needed anymore
	pub(crate) async fn delete_if_unneeded(&self, hash: &Hash) -> Result<(), Error> {
		self.lock_mutate(hash)
			.await
			.delete_if_unneeded(hash, self)
			.await
	}

	/// Find the path where a block is currently stored
	pub(crate) async fn find_block(&self, hash: &Hash) -> Option<DataBlockPath> {
		let data_layout = self.data_layout.load_full();
		let dirs = Some(data_layout.primary_block_dir(hash))
			.into_iter()
			.chain(data_layout.secondary_block_dirs(hash));
		let filename = hex::encode(hash.as_ref());

		for dir in dirs {
			let mut path = dir;
			path.push(&filename);

			if self.compression_level.is_none() {
				// If compression is disabled on node - check for the raw block
				// first and then a compressed one (as compression may have been
				// previously enabled).
				if fs::metadata(&path).await.is_ok() {
					return Some(DataBlockPath::Plain(path));
				}
				path.set_extension("zst");
				if fs::metadata(&path).await.is_ok() {
					return Some(DataBlockPath::Compressed(path));
				}
			} else {
				path.set_extension("zst");
				if fs::metadata(&path).await.is_ok() {
					return Some(DataBlockPath::Compressed(path));
				}
				path.set_extension("");
				if fs::metadata(&path).await.is_ok() {
					return Some(DataBlockPath::Plain(path));
				}
			}
		}

		None
	}

	/// Rewrite a block at the primary location for its path and delete the old path.
	/// Returns the number of bytes read/written
	pub(crate) async fn fix_block_location(
		&self,
		hash: &Hash,
		wrong_path: DataBlockPath,
	) -> Result<usize, Error> {
		self.lock_mutate(hash)
			.await
			.fix_block_location(hash, wrong_path, self)
			.await
	}

	async fn lock_mutate(&self, hash: &Hash) -> MutexGuard<'_, BlockManagerLocked> {
		let tracer = opentelemetry::global::tracer("garage");
		let ilock = u16::from_be_bytes([hash.as_slice()[0], hash.as_slice()[1]]) as usize
			% self.mutation_lock.len();
		self.mutation_lock[ilock]
			.lock()
			.with_context(Context::current_with_span(
				tracer.start(format!("Acquire mutation_lock #{}", ilock)),
			))
			.await
	}
}

#[async_trait]
impl StreamingEndpointHandler<BlockRpc> for BlockManager {
	async fn handle(self: &Arc<Self>, mut message: Req<BlockRpc>, _from: NodeID) -> Resp<BlockRpc> {
		match message.msg() {
			BlockRpc::PutBlock { hash, header } => Resp::new(
				self.handle_put_block(*hash, *header, message.take_stream())
					.await
					.map(|()| BlockRpc::Ok),
			),
			BlockRpc::GetBlock(h, order_tag) => self.handle_get_block(h, *order_tag).await,
			BlockRpc::NeedBlockQuery(h) => {
				Resp::new(self.need_block(h).await.map(BlockRpc::NeedBlockReply))
			}
			m => Resp::new(Err(Error::unexpected_rpc_message(m))),
		}
	}
}

impl BlockManagerLocked {
	async fn write_block(
		&self,
		hash: &Hash,
		data: &DataBlock,
		mgr: &BlockManager,
	) -> Result<(), Error> {
		let existing_path = mgr.find_block(hash).await;
		self.write_block_inner(hash, data, mgr, existing_path).await
	}

	async fn write_block_inner(
		&self,
		hash: &Hash,
		data: &DataBlock,
		mgr: &BlockManager,
		existing_path: Option<DataBlockPath>,
	) -> Result<(), Error> {
		let compressed = data.is_compressed();
		let data = data.inner_buffer();

		let directory = mgr.data_layout.load().primary_block_dir(hash);

		let mut tgt_path = directory.clone();
		tgt_path.push(hex::encode(hash));
		if compressed {
			tgt_path.set_extension("zst");
		}

		let to_delete = match (existing_path, compressed) {
			// If the block is stored in the wrong directory,
			// write it again at the correct path and delete the old path
			(Some(DataBlockPath::Plain(p)), false) if p != tgt_path => Some(p),
			(Some(DataBlockPath::Compressed(p)), true) if p != tgt_path => Some(p),

			// If the block is already stored not compressed but we have a compressed
			// copy, write the compressed copy and delete the uncompressed one
			(Some(DataBlockPath::Plain(plain_path)), true) => Some(plain_path),

			// If the block is already stored compressed,
			// keep the stored copy, we have nothing to do
			(Some(DataBlockPath::Compressed(_)), _) => return Ok(()),

			// If the block is already stored not compressed,
			// and we don't have a compressed copy either,
			// keep the stored copy, we have nothing to do
			(Some(DataBlockPath::Plain(_)), false) => return Ok(()),

			// If the block isn't stored already, just store what is given to us
			(None, _) => None,
		};
		assert!(to_delete.as_ref() != Some(&tgt_path));

		let mut path_tmp = tgt_path.clone();
		let tmp_extension = format!("tmp{}", hex::encode(thread_rng().gen::<[u8; 4]>()));
		path_tmp.set_extension(tmp_extension);

		fs::create_dir_all(&directory).await?;

		let mut delete_on_drop = DeleteOnDrop(Some(path_tmp.clone()));

		let mut f = fs::File::create(&path_tmp).await?;
		f.write_all(data).await?;
		mgr.metrics.bytes_written.add(data.len() as u64);

		if mgr.data_fsync {
			f.sync_all().await?;
		}

		drop(f);

		fs::rename(path_tmp, tgt_path).await?;

		delete_on_drop.cancel();

		if let Some(to_delete) = to_delete {
			fs::remove_file(to_delete).await?;
		}

		if mgr.data_fsync {
			// We want to ensure that when this function returns, data is properly persisted
			// to disk. The first step is the sync_all above that does an fsync on the data file.
			// Now, we do an fsync on the containing directory, to ensure that the rename
			// is persisted properly. See:
			// http://thedjbway.b0llix.net/qmail/syncdir.html
			let dir = fs::OpenOptions::new()
				.read(true)
				.mode(0)
				.open(directory)
				.await?;
			dir.sync_all().await?;
			drop(dir);
		}

		Ok(())
	}

	async fn move_block_to_corrupted(&self, block_path: &DataBlockPath) -> Result<(), Error> {
		let (path, path2) = match block_path {
			DataBlockPath::Plain(p) => {
				let mut p2 = p.clone();
				p2.set_extension("corrupted");
				(p, p2)
			}
			DataBlockPath::Compressed(p) => {
				let mut p2 = p.clone();
				p2.set_extension("zst.corrupted");
				(p, p2)
			}
		};

		fs::rename(path, path2).await?;
		Ok(())
	}

	async fn delete_if_unneeded(&self, hash: &Hash, mgr: &BlockManager) -> Result<(), Error> {
		let rc = mgr.rc.get_block_rc(hash)?;
		if rc.is_deletable() {
			while let Some(path) = mgr.find_block(hash).await {
				let path = match path {
					DataBlockPath::Plain(p) | DataBlockPath::Compressed(p) => p,
				};
				fs::remove_file(path).await?;
				mgr.metrics.delete_counter.add(1);
			}
		}
		Ok(())
	}

	async fn fix_block_location(
		&self,
		hash: &Hash,
		wrong_path: DataBlockPath,
		mgr: &BlockManager,
	) -> Result<usize, Error> {
		let data = mgr.read_block_from(hash, &wrong_path).await?;
		self.write_block_inner(hash, &data, mgr, Some(wrong_path))
			.await?;
		Ok(data.inner_buffer().len())
	}
}

async fn read_stream_to_end(mut stream: ByteStream) -> Result<Bytes, Error> {
	let mut parts: Vec<Bytes> = vec![];
	while let Some(part) = stream.next().await {
		parts.push(part.ok_or_message("error in stream")?);
	}

	Ok(parts
		.iter()
		.map(|x| &x[..])
		.collect::<Vec<_>>()
		.concat()
		.into())
}

struct DeleteOnDrop(Option<PathBuf>);

impl DeleteOnDrop {
	fn cancel(&mut self) {
		drop(self.0.take());
	}
}

impl Drop for DeleteOnDrop {
	fn drop(&mut self) {
		if let Some(path) = self.0.take() {
			tokio::spawn(async move {
				if let Err(e) = fs::remove_file(&path).await {
					debug!("DeleteOnDrop failed for {}: {}", path.display(), e);
				}
			});
		}
	}
}