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
macro_rules! segmented_table_ordered_methods {
() => {
fn collect_rows_ordered_by_index(
&self,
column_name: &str,
ascending: bool,
limit: usize,
offset: usize,
) -> Option<RowVec> {
if !self.segment_mgr.has_segments() {
return self.hot.collect_rows_ordered_by_index(
column_name,
ascending,
limit,
offset,
);
}
// Snapshot isolation: the merge path doesn't filter by snapshot_seq.
// Fall back to the full scan + sort path which handles MVCC correctly.
if self.snapshot_seq.is_some() {
return None;
}
// Only optimize when ORDER BY column is the INTEGER PRIMARY KEY.
// For PK, row_id order == value order, so we can merge sorted sources.
let schema = self.hot.schema().clone();
let pk_idx = schema.pk_column_index()?;
let pk_col = &schema.columns[pk_idx];
if pk_col.name_lower != column_name.to_lowercase() {
return None;
}
let needed = limit.saturating_add(offset);
if needed == 0 {
return Some(RowVec::new());
}
// 1. Collect hot rows in PK order. Only materialize `needed` rows
// (not all hot rows). The skip set uses row IDs only (no materialization).
let hot_rows =
match self
.hot
.collect_rows_ordered_by_index(column_name, ascending, needed, 0)
{
Some(rows) => rows,
None => {
// Local changes prevent ordered iteration — fall back to collect + sort.
let mut rows = match self.hot.collect_all_rows(None) {
Ok(r) => r,
Err(_) => return None,
};
if ascending {
rows.sort_unstable_by_key(|&(id, _)| id);
} else {
rows.sort_unstable_by_key(|entry| std::cmp::Reverse(entry.0));
}
rows.truncate(needed);
rows
}
};
// 2. Build skip set from ALL hot row IDs (no Row materialization).
// Hot rows shadow cold rows regardless of whether they're in the merge set.
let mut skip: FxHashSet<i64> =
FxHashSet::with_capacity_and_hasher(10_000, Default::default());
self.hot.collect_hot_row_ids_into(&mut skip);
self.segment_mgr
.insert_pending_tombstones_into(self.txn_id(), &mut skip);
let tombstones_arc = self.segment_mgr.tombstone_set_arc();
// 3. Get volumes lazily (oldest first — segment_id ascending order).
// The merge itself needs only row_ids metadata; materialization goes
// through materialize_volume_row_for_read(), so artifact-backed-backed cold
// volumes can stay metadata-only and read individual column blocks.
let volumes = self.segment_mgr.get_volumes_newest_first_lazy();
// volumes is oldest-first after the reverse inside get_volumes_newest_first_lazy().
// 4. K-way merge using per-source cursors.
// Sources: hot_rows (already sorted) + each volume's row_ids (sorted ascending).
// For DESC, we iterate each source from the end.
// Pre-compute column mappings for each volume.
struct VolSource {
row_ids: crate::volume::writer::RowIds,
cursor: usize,
mapping: crate::volume::writer::ColumnMapping,
volume: Arc<FrozenVolume>,
visible: Option<Arc<Vec<u64>>>,
}
let mut vol_sources: Vec<VolSource> = Vec::with_capacity(volumes.len());
for (_seg_id, cs) in volumes.iter() {
// Filter out empty or fully-skipped volumes early via zone-map on row_id range.
let vol = &cs.volume;
if vol.meta.row_count == 0 {
continue;
}
let mapping = self.segment_mgr.get_cold_segment_mapping(cs, &schema);
vol_sources.push(VolSource {
row_ids: vol.meta.row_ids.clone(),
cursor: if ascending { 0 } else { vol.meta.row_count },
mapping,
volume: Arc::clone(&cs.volume),
visible: cs.visible.clone(),
});
}
let mut result = RowVec::with_capacity(needed.min(1024));
let mut skipped = 0usize;
let mut hot_cursor: usize = 0;
loop {
if result.len() >= limit {
break;
}
// Find the source with the next row_id to emit.
// For ASC: smallest row_id. For DESC: largest row_id.
let mut best_row_id: Option<i64> = None;
// 0 = hot, 1..=num_vol = vol_sources[idx-1]
let mut best_source: usize = usize::MAX;
// Check hot source
if hot_cursor < hot_rows.len() {
let (rid, _) = &hot_rows[hot_cursor];
best_row_id = Some(*rid);
best_source = 0;
}
// Check each volume source
for (vi, vs) in vol_sources.iter().enumerate() {
let rid = if ascending {
if vs.cursor >= vs.row_ids.len() {
continue;
}
vs.row_ids.at(vs.cursor)
} else {
if vs.cursor == 0 {
continue;
}
vs.row_ids.at(vs.cursor - 1)
};
let dominated = match best_row_id {
None => false,
Some(best) => {
if ascending {
best <= rid
} else {
best >= rid
}
}
};
if !dominated {
best_row_id = Some(rid);
best_source = vi + 1;
}
}
// No more rows from any source.
if best_source == usize::MAX {
break;
}
if best_source == 0 {
// Hot source — row is already materialized and visible.
let (rid, row) = hot_rows[hot_cursor].clone();
hot_cursor += 1;
// Hot rows don't need tombstone/skip checks — they ARE the authoritative version.
if skipped < offset {
skipped += 1;
} else {
result.push((rid, row));
}
} else {
// Volume source
let vs = &mut vol_sources[best_source - 1];
let idx = if ascending {
let i = vs.cursor;
vs.cursor += 1;
i
} else {
vs.cursor -= 1;
vs.cursor
};
let rid = vs.row_ids.at(idx);
// Visibility check: inter-volume dedup bitmap
if let Some(ref bits) = vs.visible {
if (bits[idx >> 6] >> (idx & 63)) & 1 == 0 {
continue;
}
}
// Skip if hot shadows this row or if tombstoned
if skip.contains(&rid) {
continue;
}
if self.is_row_tombstoned(&tombstones_arc, rid) {
continue;
}
let row =
match self.materialize_volume_row_for_read(&vs.volume, idx, &vs.mapping) {
Ok(row) => row,
Err(_) => return None,
};
if skipped < offset {
skipped += 1;
} else {
result.push((rid, row));
}
}
}
Some(result)
}
fn collect_rows_composite_ordered_range(
&self,
where_expr: &dyn Expression,
order_column: &str,
ascending: bool,
limit: usize,
offset: usize,
) -> Option<Result<RowVec>> {
if !self.segment_mgr.has_segments() {
return self.hot.collect_rows_composite_ordered_range(
where_expr,
order_column,
ascending,
limit,
offset,
);
}
let plan = self.plan_cold_composite_ordered(where_expr)?;
if !plan
.columns
.last()
.is_some_and(|column| column.eq_ignore_ascii_case(order_column))
{
return None;
}
let comparisons = where_expr.collect_comparisons();
if comparisons
.iter()
.any(|(column, _, _)| !plan.covered_columns.contains(*column))
|| !where_expr.collect_null_check_infos().is_empty()
{
return None;
}
let target = limit.saturating_add(offset);
let mut cold_rows = match self
.collect_cold_composite_ordered_rows(&plan, where_expr, ascending, target)
{
Ok(Some(rows)) => rows,
Ok(None) => return None,
Err(error) => return Some(Err(error)),
};
let hot_rows = match self.hot.collect_rows_composite_ordered_range(
where_expr,
order_column,
ascending,
target,
0,
) {
Some(Ok(rows)) => rows,
Some(Err(error)) => return Some(Err(error)),
None if self.hot.row_count_hint() == 0 => RowVec::new(),
None => return None,
};
cold_rows.extend(hot_rows);
let order_idx = *plan
.column_indices
.last()
.expect("ordered plan always has a range column");
cold_rows.sort_unstable_by(|left, right| {
let ordering = left
.1
.get(order_idx)
.cmp(&right.1.get(order_idx))
.then_with(|| left.0.cmp(&right.0));
if ascending {
ordering
} else {
ordering.reverse()
}
});
cold_rows.dedup_by_key(|entry| entry.0);
Some(Ok(cold_rows.into_iter().skip(offset).take(limit).collect()))
}
fn collect_rows_pk_keyset(
&self,
start_after: Option<i64>,
start_from: Option<i64>,
ascending: bool,
limit: usize,
) -> Option<RowVec> {
if !self.segment_mgr.has_segments() {
return self
.hot
.collect_rows_pk_keyset(start_after, start_from, ascending, limit);
}
// Hot PK index doesn't cover cold data — can't use keyset pagination
None
}
};
}
pub(super) use segmented_table_ordered_methods;