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
|
//! Per-node, in-memory compaction state: snapshots built ahead of the switch, switch
//! progress for reports, and archived documents of previous epochs.
//!
//! Nothing here is replicated or durable. After a restart the node simply has no
//! archive and rebuilds snapshots when needed.
use std::collections::HashMap;
use std::sync::{Mutex, MutexGuard};
use automerge::AutoCommit;
use serde::Serialize;
use tokio::time::Instant;
use super::records::{unix_millis, Retention};
use super::snapshot::Snapshot;
use crate::replica::Epoch;
#[derive(Default)]
pub(crate) struct Compaction {
// A std mutex: every access is short and never spans an await.
inner: Mutex<Inner>,
}
#[derive(Default)]
struct Inner {
/// Built by the coordinator at commit time; consumed by the local switch.
prebuilt: HashMap<String, Snapshot>,
/// When this node first saw a compaction committed.
first_seen: HashMap<String, Instant>,
status: HashMap<String, LocalStatus>,
archive: Vec<Archived>,
}
/// What happened to a compaction on this node.
#[derive(Clone, Debug, Default, Serialize)]
#[serde(rename_all = "camelCase")]
pub(crate) struct LocalStatus {
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) switched_at: Option<u64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) replayed_paths: Option<usize>,
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) failed_paths: Option<usize>,
/// Why the switch did not happen (yet).
#[serde(skip_serializing_if = "Option::is_none")]
pub(crate) error: Option<String>,
}
struct Archived {
epoch: Epoch,
// Kept for later steps (serving or auditing old epochs); not read yet.
#[allow(dead_code)]
doc: AutoCommit,
archived_at: u64,
retention: Retention,
}
/// Summary of an archived epoch.
#[derive(Clone, Debug, Serialize)]
#[serde(rename_all = "camelCase")]
pub(crate) struct ArchivedEpoch {
pub(crate) generation: u64,
pub(crate) id: String,
pub(crate) archived_at: u64,
}
impl Compaction {
fn lock(&self) -> MutexGuard<'_, Inner> {
self.inner
.lock()
.unwrap_or_else(|poisoned| poisoned.into_inner())
}
pub(crate) fn store_prebuilt(&self, id: &str, snapshot: Snapshot) {
self.lock().prebuilt.insert(id.to_owned(), snapshot);
}
pub(crate) fn take_prebuilt(&self, id: &str) -> Option<Snapshot> {
self.lock().prebuilt.remove(id)
}
/// When this node first saw the compaction committed (now, on the first call).
pub(crate) fn first_seen(&self, id: &str) -> Instant {
*self
.lock()
.first_seen
.entry(id.to_owned())
.or_insert_with(Instant::now)
}
pub(crate) fn status(&self, id: &str) -> Option<LocalStatus> {
self.lock().status.get(id).cloned()
}
pub(crate) fn set_status(&self, id: &str, status: LocalStatus) {
self.lock().status.insert(id.to_owned(), status);
}
/// Keeps the previous epoch's document, then drops what has outlived its retention.
pub(crate) fn archive(
&self,
epoch: Epoch,
doc: AutoCommit,
retention: Retention,
current_generation: u64,
) {
let mut inner = self.lock();
inner.archive.push(Archived {
epoch,
doc,
archived_at: unix_millis(),
retention,
});
prune(&mut inner.archive, current_generation);
}
pub(crate) fn prune_archive(&self, current_generation: u64) {
prune(&mut self.lock().archive, current_generation);
}
pub(crate) fn archived_epochs(&self) -> Vec<ArchivedEpoch> {
self.lock()
.archive
.iter()
.map(|archived| ArchivedEpoch {
generation: archived.epoch.generation,
id: archived.epoch.id_hex(),
archived_at: archived.archived_at,
})
.collect()
}
}
/// An archive is kept until every limit of its retention has passed.
fn prune(archive: &mut Vec<Archived>, current_generation: u64) {
let now = unix_millis();
archive.retain(|archived| {
let within_epochs = archived.retention.epochs.is_some_and(|epochs| {
current_generation.saturating_sub(archived.epoch.generation) <= epochs
});
let within_time = archived.retention.seconds.is_some_and(|seconds| {
now.saturating_sub(archived.archived_at) <= seconds.saturating_mul(1000)
});
within_epochs || within_time
});
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn archive_retention() {
let compaction = Compaction::default();
let epoch = |generation| Epoch {
generation,
id: [generation as u8; 32],
};
// Kept by epoch count only.
compaction.archive(
epoch(0),
AutoCommit::new(),
Retention {
epochs: Some(1),
seconds: None,
},
1,
);
assert_eq!(compaction.archived_epochs().len(), 1);
compaction.prune_archive(2);
assert!(compaction.archived_epochs().is_empty());
// Kept by time while the epoch count has passed.
compaction.archive(
epoch(1),
AutoCommit::new(),
Retention {
epochs: Some(0),
seconds: Some(3600),
},
5,
);
assert_eq!(compaction.archived_epochs().len(), 1);
// No limits at all: not kept.
compaction.archive(
epoch(2),
AutoCommit::new(),
Retention {
epochs: None,
seconds: None,
},
3,
);
assert_eq!(compaction.archived_epochs().len(), 1);
}
}
|
