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
|
use core::ops::Bound;
use std::sync::Arc;
use std::time::Duration;
use async_trait::async_trait;
use tokio::fs;
use tokio::sync::watch;
use garage_util::background::*;
use garage_util::data::*;
use garage_util::error::*;
use garage_util::tranquilizer::Tranquilizer;
use crate::manager::*;
pub struct RepairWorker {
manager: Arc<BlockManager>,
next_start: Option<Hash>,
block_iter: Option<BlockStoreIterator>,
}
impl RepairWorker {
pub fn new(manager: Arc<BlockManager>) -> Self {
Self {
manager,
next_start: None,
block_iter: None,
}
}
}
#[async_trait]
impl Worker for RepairWorker {
fn name(&self) -> String {
"Block repair worker".into()
}
async fn work(
&mut self,
_must_exit: &mut watch::Receiver<bool>,
) -> Result<WorkerStatus, Error> {
match self.block_iter.as_mut() {
None => {
// Phase 1: Repair blocks from RC table.
// We have to do this complicated two-step process where we first read a bunch
// of hashes from the RC table, and then insert them in the to-resync queue,
// because of SQLite. Basically, as long as we have an iterator on a DB table,
// we can't do anything else on the DB. The naive approach (which we had previously)
// of just iterating on the RC table and inserting items one to one in the resync
// queue can't work here, it would just provoke a deadlock in the SQLite adapter code.
// This is mostly because the Rust bindings for SQLite assume a worst-case scenario
// where SQLite is not compiled in thread-safe mode, so we have to wrap everything
// in a mutex (see db/sqlite_adapter.rs and discussion in PR #322).
let mut batch_of_hashes = vec![];
let start_bound = match self.next_start.as_ref() {
None => Bound::Unbounded,
Some(x) => Bound::Excluded(x.as_slice()),
};
for entry in self
.manager
.rc
.rc
.range::<&[u8], _>((start_bound, Bound::Unbounded))?
{
let (hash, _) = entry?;
let hash = Hash::try_from(&hash[..]).unwrap();
batch_of_hashes.push(hash);
if batch_of_hashes.len() >= 1000 {
break;
}
}
if batch_of_hashes.is_empty() {
// move on to phase 2
self.block_iter = Some(BlockStoreIterator::new(&self.manager).await?);
return Ok(WorkerStatus::Busy);
}
for hash in batch_of_hashes.into_iter() {
self.manager.put_to_resync(&hash, Duration::from_secs(0))?;
self.next_start = Some(hash)
}
Ok(WorkerStatus::Busy)
}
Some(bi) => {
// Phase 2: Repair blocks actually on disk
// Lists all blocks on disk and adds them to the resync queue.
// This allows us to find blocks we are storing but don't actually need,
// so that we can offload them if necessary and then delete them locally.
if let Some(hash) = bi.next().await? {
self.manager.put_to_resync(&hash, Duration::from_secs(0))?;
Ok(WorkerStatus::Busy)
} else {
Ok(WorkerStatus::Done)
}
}
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerStatus {
unreachable!()
}
}
// ----
pub struct ScrubWorker {
manager: Arc<BlockManager>,
iterator: BlockStoreIterator,
tranquilizer: Tranquilizer,
tranquility: u32,
}
impl ScrubWorker {
pub async fn new(manager: Arc<BlockManager>, tranquility: u32) -> Result<Self, Error> {
let iterator = BlockStoreIterator::new(&manager).await?;
Ok(Self {
manager,
iterator,
tranquilizer: Tranquilizer::new(30),
tranquility,
})
}
}
#[async_trait]
impl Worker for ScrubWorker {
fn name(&self) -> String {
"Block scrub worker".into()
}
async fn work(
&mut self,
_must_exit: &mut watch::Receiver<bool>,
) -> Result<WorkerStatus, Error> {
self.tranquilizer.reset();
if let Some(hash) = self.iterator.next().await? {
let _ = self.manager.read_block(&hash).await;
self.tranquilizer.tranquilize(self.tranquility).await;
Ok(WorkerStatus::Busy)
} else {
Ok(WorkerStatus::Done)
}
}
async fn wait_for_work(&mut self, _must_exit: &watch::Receiver<bool>) -> WorkerStatus {
unreachable!()
}
}
// ----
struct BlockStoreIterator {
path: Vec<fs::ReadDir>,
}
impl BlockStoreIterator {
async fn new(manager: &BlockManager) -> Result<Self, Error> {
let root_dir = manager.data_dir.clone();
let read_root_dir = fs::read_dir(&root_dir).await?;
Ok(Self {
path: vec![read_root_dir],
})
}
async fn next(&mut self) -> Result<Option<Hash>, Error> {
loop {
if let Some(reader) = self.path.last_mut() {
if let Some(data_dir_ent) = reader.next_entry().await? {
let name = data_dir_ent.file_name();
let name = if let Ok(n) = name.into_string() {
n
} else {
continue;
};
let ent_type = data_dir_ent.file_type().await?;
let name = name.strip_suffix(".zst").unwrap_or(&name);
if name.len() == 2 && hex::decode(&name).is_ok() && ent_type.is_dir() {
let read_child_dir = fs::read_dir(&data_dir_ent.path()).await?;
self.path.push(read_child_dir);
continue;
} else if name.len() == 64 {
let hash_bytes = if let Ok(h) = hex::decode(&name) {
h
} else {
continue;
};
let mut hash = [0u8; 32];
hash.copy_from_slice(&hash_bytes[..]);
return Ok(Some(hash.into()));
}
} else {
self.path.pop();
continue;
}
} else {
return Ok(None);
}
}
}
}
|
