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
// SPDX-License-Identifier: BUSL-1.1
use nodedb_wal::record::RecordType;
use super::core::WalManager;
use crate::types::{DatabaseId, Lsn, TenantId, VShardId};
impl WalManager {
/// Internal: append a record of the given type to the WAL.
fn append_record(
&self,
record_type: RecordType,
tenant_id: TenantId,
vshard_id: VShardId,
database_id: DatabaseId,
payload: &[u8],
) -> crate::Result<Lsn> {
let mut wal = self.wal.lock().unwrap_or_else(|p| p.into_inner());
let lsn = wal
.append(
record_type as u32,
tenant_id.as_u64(),
vshard_id.as_u32(),
database_id.as_u64(),
payload,
)
.map_err(crate::Error::Wal)?;
Ok(Lsn::new(lsn))
}
pub fn append_put(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::Put, tid, vs, db, p)
}
pub fn append_delete(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::Delete, tid, vs, db, p)
}
pub fn append_vector_put(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::VectorPut, tid, vs, db, p)
}
pub fn append_vector_delete(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::VectorDelete, tid, vs, db, p)
}
pub fn append_vector_params(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::VectorParams, tid, vs, db, p)
}
pub fn append_transaction(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::Transaction, tid, vs, db, p)
}
pub fn append_crdt_delta(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
delta: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::CrdtDelta, tid, vs, db, delta)
}
/// Append a checkpoint marker. Serializes the LSN before writing.
pub fn append_checkpoint(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
checkpoint_lsn: u64,
) -> crate::Result<Lsn> {
let payload =
zerompk::to_msgpack_vec(&checkpoint_lsn).map_err(|e| crate::Error::Serialization {
format: "msgpack".into(),
detail: format!("checkpoint: {e}"),
})?;
self.append_record(RecordType::Checkpoint, tid, vs, db, &payload)
}
pub fn append_timeseries_batch(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::TimeseriesBatch, tid, vs, db, p)
}
pub fn append_log_batch(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::LogBatch, tid, vs, db, p)
}
pub fn append_array_put(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::ArrayPut, tid, vs, db, p)
}
pub fn append_array_delete(
&self,
tid: TenantId,
vs: VShardId,
db: DatabaseId,
p: &[u8],
) -> crate::Result<Lsn> {
self.append_record(RecordType::ArrayDelete, tid, vs, db, p)
}
/// Append a `CollectionTombstoned` record. Any subsequent replay
/// that extracts this record will filter prior writes for
/// `(tid, collection)` whose LSN is less than `purge_lsn`.
///
/// `vshard_id` of `0` is conventional — tombstones are tenant-level
/// metadata, not sharded user data. Replay filters on
/// `(tenant_id, collection)` pair alone.
/// Append a `TemporalPurge` audit record. Emitted by the
/// Control Plane's bitemporal-retention scheduler after a successful
/// dispatch of `MetaOp::TemporalPurge*` to the Data Plane, providing
/// a durable audit trail distinct from regular `Delete` records.
///
/// `vshard_id` is `0` — bitemporal audit-purge is collection-scoped
/// metadata, not sharded user data.
pub fn append_temporal_purge(
&self,
tid: TenantId,
engine: nodedb_wal::TemporalPurgeEngine,
collection: &str,
cutoff_system_ms: i64,
purged_count: u64,
) -> crate::Result<Lsn> {
let payload = nodedb_wal::TemporalPurgePayload::new(
engine,
collection,
cutoff_system_ms,
purged_count,
)
.to_bytes()
.map_err(crate::Error::Wal)?;
self.append_record(
RecordType::TemporalPurge,
tid,
VShardId::new(0),
DatabaseId::DEFAULT,
&payload,
)
}
/// Append a `SurrogateAlloc` high-watermark record. Emitted by
/// `SurrogateRegistry::flush` (every 1024 allocations or 200 ms,
/// whichever first) so the global surrogate counter is
/// crash-recoverable independent of the redb `_system.surrogate_hwm`
/// row. `vshard_id` is `0` — the surrogate hwm is a node-global
/// allocator counter, not sharded user data.
pub fn append_surrogate_alloc(&self, hi: u32) -> crate::Result<Lsn> {
let payload = nodedb_wal::record::SurrogateAllocPayload::new(hi).to_bytes();
self.append_record(
RecordType::SurrogateAlloc,
TenantId::new(0),
VShardId::new(0),
DatabaseId::DEFAULT,
&payload,
)
}
/// Append a `SurrogateBind` record. Emitted by
/// `SurrogateAssigner::assign` immediately after the catalog
/// two-table txn that writes `_system.surrogate_pk{,_rev}`, so the
/// binding survives a crash before the next hwm checkpoint. The
/// record is durably persisted before `assign` returns.
/// `vshard_id` is `0` — surrogate bindings are node-global metadata.
pub fn append_surrogate_bind(
&self,
surrogate: u32,
collection: &str,
pk_bytes: &[u8],
) -> crate::Result<Lsn> {
let payload =
nodedb_wal::record::SurrogateBindPayload::new(surrogate, collection, pk_bytes.to_vec())
.to_bytes()
.map_err(crate::Error::Wal)?;
self.append_record(
RecordType::SurrogateBind,
TenantId::new(0),
VShardId::new(0),
DatabaseId::DEFAULT,
&payload,
)
}
/// Append a `CalvinApplied` record after a Calvin executor successfully
/// commits a `MetaOp::CalvinExecute` batch.
///
/// The scheduler's restart path scans these records to compute
/// `last_applied_epoch` for a given vshard without re-reading the full
/// Raft sequencer log. `vshard_id` is stored in the payload (not in the
/// WAL record header's `vshard_id` field) so it can be decoded during
/// a scan of all records regardless of which vshard they were written on.
pub fn append_calvin_applied(
&self,
vshard_id: crate::types::VShardId,
epoch: u64,
position: u32,
) -> crate::Result<crate::types::Lsn> {
let payload =
nodedb_wal::CalvinAppliedPayload::new(epoch, position, vshard_id.as_u32()).to_bytes();
self.append_record(
RecordType::CalvinApplied,
TenantId::new(0),
vshard_id,
DatabaseId::DEFAULT,
&payload,
)
}
pub fn append_collection_tombstone(
&self,
tid: TenantId,
collection: &str,
purge_lsn: u64,
) -> crate::Result<Lsn> {
let payload = nodedb_wal::CollectionTombstonePayload::new(collection, purge_lsn)
.to_bytes()
.map_err(crate::Error::Wal)?;
self.append_record(
RecordType::CollectionTombstoned,
tid,
VShardId::new(0),
DatabaseId::DEFAULT,
&payload,
)
}
}