Line data Source code
1 : use anyhow::{bail, Result};
2 : use byteorder::{ByteOrder, BE};
3 : use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
4 : use postgres_ffi::Oid;
5 : use postgres_ffi::RepOriginId;
6 : use serde::{Deserialize, Serialize};
7 : use std::{fmt, ops::Range};
8 :
9 : use crate::reltag::{BlockNumber, RelTag, SlruKind};
10 :
11 : /// Key used in the Repository kv-store.
12 : ///
13 : /// The Repository treats this as an opaque struct, but see the code in pgdatadir_mapping.rs
14 : /// for what we actually store in these fields.
15 2380 : #[derive(Debug, Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize)]
16 : pub struct Key {
17 : pub field1: u8,
18 : pub field2: u32,
19 : pub field3: u32,
20 : pub field4: u32,
21 : pub field5: u8,
22 : pub field6: u32,
23 : }
24 :
25 : /// When working with large numbers of Keys in-memory, it is more efficient to handle them as i128 than as
26 : /// a struct of fields.
27 : #[derive(Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd)]
28 : pub struct CompactKey(i128);
29 :
30 : /// The storage key size.
31 : pub const KEY_SIZE: usize = 18;
32 :
33 : /// The metadata key size. 2B fewer than the storage key size because field2 is not fully utilized.
34 : /// See [`Key::to_i128`] for more information on the encoding.
35 : pub const METADATA_KEY_SIZE: usize = 16;
36 :
37 : /// The key prefix start range for the metadata keys. All keys with the first byte >= 0x60 is a metadata key.
38 : pub const METADATA_KEY_BEGIN_PREFIX: u8 = 0x60;
39 : pub const METADATA_KEY_END_PREFIX: u8 = 0x7F;
40 :
41 : /// The (reserved) key prefix of relation sizes.
42 : pub const RELATION_SIZE_PREFIX: u8 = 0x61;
43 :
44 : /// The key prefix of AUX file keys.
45 : pub const AUX_KEY_PREFIX: u8 = 0x62;
46 :
47 : /// The key prefix of ReplOrigin keys.
48 : pub const REPL_ORIGIN_KEY_PREFIX: u8 = 0x63;
49 :
50 : /// Check if the key falls in the range of metadata keys.
51 36 : pub const fn is_metadata_key_slice(key: &[u8]) -> bool {
52 36 : key[0] >= METADATA_KEY_BEGIN_PREFIX && key[0] < METADATA_KEY_END_PREFIX
53 36 : }
54 :
55 : impl Key {
56 : /// Check if the key falls in the range of metadata keys.
57 67 : pub const fn is_metadata_key(&self) -> bool {
58 67 : self.field1 >= METADATA_KEY_BEGIN_PREFIX && self.field1 < METADATA_KEY_END_PREFIX
59 67 : }
60 :
61 : /// Encode a metadata key to a storage key.
62 35 : pub fn from_metadata_key_fixed_size(key: &[u8; METADATA_KEY_SIZE]) -> Self {
63 35 : assert!(is_metadata_key_slice(key), "key not in metadata key range");
64 : // Metadata key space ends at 0x7F so it's fine to directly convert it to i128.
65 35 : Self::from_i128(i128::from_be_bytes(*key))
66 35 : }
67 :
68 : /// Encode a metadata key to a storage key.
69 1 : pub fn from_metadata_key(key: &[u8]) -> Self {
70 1 : Self::from_metadata_key_fixed_size(key.try_into().expect("expect 16 byte metadata key"))
71 1 : }
72 :
73 : /// Get the range of metadata keys.
74 1180 : pub const fn metadata_key_range() -> Range<Self> {
75 1180 : Key {
76 1180 : field1: METADATA_KEY_BEGIN_PREFIX,
77 1180 : field2: 0,
78 1180 : field3: 0,
79 1180 : field4: 0,
80 1180 : field5: 0,
81 1180 : field6: 0,
82 1180 : }..Key {
83 1180 : field1: METADATA_KEY_END_PREFIX,
84 1180 : field2: 0,
85 1180 : field3: 0,
86 1180 : field4: 0,
87 1180 : field5: 0,
88 1180 : field6: 0,
89 1180 : }
90 1180 : }
91 :
92 : /// Get the range of aux keys.
93 302 : pub fn metadata_aux_key_range() -> Range<Self> {
94 302 : Key {
95 302 : field1: AUX_KEY_PREFIX,
96 302 : field2: 0,
97 302 : field3: 0,
98 302 : field4: 0,
99 302 : field5: 0,
100 302 : field6: 0,
101 302 : }..Key {
102 302 : field1: AUX_KEY_PREFIX + 1,
103 302 : field2: 0,
104 302 : field3: 0,
105 302 : field4: 0,
106 302 : field5: 0,
107 302 : field6: 0,
108 302 : }
109 302 : }
110 :
111 : /// This function checks more extensively what keys we can take on the write path.
112 : /// If a key beginning with 00 does not have a global/default tablespace OID, it
113 : /// will be rejected on the write path.
114 : #[allow(dead_code)]
115 0 : pub fn is_valid_key_on_write_path_strong(&self) -> bool {
116 : use postgres_ffi::pg_constants::{DEFAULTTABLESPACE_OID, GLOBALTABLESPACE_OID};
117 0 : if !self.is_i128_representable() {
118 0 : return false;
119 0 : }
120 0 : if self.field1 == 0
121 0 : && !(self.field2 == GLOBALTABLESPACE_OID
122 0 : || self.field2 == DEFAULTTABLESPACE_OID
123 0 : || self.field2 == 0)
124 : {
125 0 : return false; // User defined tablespaces are not supported
126 0 : }
127 0 : true
128 0 : }
129 :
130 : /// This is a weaker version of `is_valid_key_on_write_path_strong` that simply
131 : /// checks if the key is i128 representable. Note that some keys can be successfully
132 : /// ingested into the pageserver, but will cause errors on generating basebackup.
133 4816642 : pub fn is_valid_key_on_write_path(&self) -> bool {
134 4816642 : self.is_i128_representable()
135 4816642 : }
136 :
137 16472313 : pub fn is_i128_representable(&self) -> bool {
138 16472313 : self.field2 <= 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222
139 16472313 : }
140 :
141 : /// 'field2' is used to store tablespaceid for relations and small enum numbers for other relish.
142 : /// As long as Neon does not support tablespace (because of lack of access to local file system),
143 : /// we can assume that only some predefined namespace OIDs are used which can fit in u16
144 11655671 : pub fn to_i128(&self) -> i128 {
145 11655671 : assert!(self.is_i128_representable(), "invalid key: {self}");
146 11655671 : (((self.field1 & 0x7F) as i128) << 120)
147 11655671 : | (((self.field2 & 0xFFFF) as i128) << 104)
148 11655671 : | ((self.field3 as i128) << 72)
149 11655671 : | ((self.field4 as i128) << 40)
150 11655671 : | ((self.field5 as i128) << 32)
151 11655671 : | self.field6 as i128
152 11655671 : }
153 :
154 5532080 : pub const fn from_i128(x: i128) -> Self {
155 5532080 : Key {
156 5532080 : field1: ((x >> 120) & 0x7F) as u8,
157 5532080 : field2: ((x >> 104) & 0xFFFF) as u32,
158 5532080 : field3: (x >> 72) as u32,
159 5532080 : field4: (x >> 40) as u32,
160 5532080 : field5: (x >> 32) as u8,
161 5532080 : field6: x as u32,
162 5532080 : }
163 5532080 : }
164 :
165 6875751 : pub fn to_compact(&self) -> CompactKey {
166 6875751 : CompactKey(self.to_i128())
167 6875751 : }
168 :
169 5029982 : pub fn from_compact(k: CompactKey) -> Self {
170 5029982 : Self::from_i128(k.0)
171 5029982 : }
172 :
173 8865182 : pub const fn next(&self) -> Key {
174 8865182 : self.add(1)
175 8865182 : }
176 :
177 8875073 : pub const fn add(&self, x: u32) -> Key {
178 8875073 : let mut key = *self;
179 8875073 :
180 8875073 : let r = key.field6.overflowing_add(x);
181 8875073 : key.field6 = r.0;
182 8875073 : if r.1 {
183 827101 : let r = key.field5.overflowing_add(1);
184 827101 : key.field5 = r.0;
185 827101 : if r.1 {
186 0 : let r = key.field4.overflowing_add(1);
187 0 : key.field4 = r.0;
188 0 : if r.1 {
189 0 : let r = key.field3.overflowing_add(1);
190 0 : key.field3 = r.0;
191 0 : if r.1 {
192 0 : let r = key.field2.overflowing_add(1);
193 0 : key.field2 = r.0;
194 0 : if r.1 {
195 0 : let r = key.field1.overflowing_add(1);
196 0 : key.field1 = r.0;
197 0 : assert!(!r.1);
198 0 : }
199 0 : }
200 0 : }
201 827101 : }
202 8047972 : }
203 8875073 : key
204 8875073 : }
205 :
206 : /// Convert a 18B slice to a key. This function should not be used for 16B metadata keys because `field2` is handled differently.
207 : /// Use [`Key::from_i128`] instead if you want to handle 16B keys (i.e., metadata keys). There are some restrictions on `field2`,
208 : /// and therefore not all 18B slices are valid page server keys.
209 5943125 : pub fn from_slice(b: &[u8]) -> Self {
210 5943125 : Key {
211 5943125 : field1: b[0],
212 5943125 : field2: u32::from_be_bytes(b[1..5].try_into().unwrap()),
213 5943125 : field3: u32::from_be_bytes(b[5..9].try_into().unwrap()),
214 5943125 : field4: u32::from_be_bytes(b[9..13].try_into().unwrap()),
215 5943125 : field5: b[13],
216 5943125 : field6: u32::from_be_bytes(b[14..18].try_into().unwrap()),
217 5943125 : }
218 5943125 : }
219 :
220 : /// Convert a key to a 18B slice. This function should not be used for getting a 16B metadata key because `field2` is handled differently.
221 : /// Use [`Key::to_i128`] instead if you want to get a 16B key (i.e., metadata keys).
222 7257377 : pub fn write_to_byte_slice(&self, buf: &mut [u8]) {
223 7257377 : buf[0] = self.field1;
224 7257377 : BE::write_u32(&mut buf[1..5], self.field2);
225 7257377 : BE::write_u32(&mut buf[5..9], self.field3);
226 7257377 : BE::write_u32(&mut buf[9..13], self.field4);
227 7257377 : buf[13] = self.field5;
228 7257377 : BE::write_u32(&mut buf[14..18], self.field6);
229 7257377 : }
230 : }
231 :
232 : impl fmt::Display for Key {
233 375771 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
234 375771 : write!(
235 375771 : f,
236 375771 : "{:02X}{:08X}{:08X}{:08X}{:02X}{:08X}",
237 375771 : self.field1, self.field2, self.field3, self.field4, self.field5, self.field6
238 375771 : )
239 375771 : }
240 : }
241 :
242 : impl fmt::Display for CompactKey {
243 0 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
244 0 : let k = Key::from_compact(*self);
245 0 : k.fmt(f)
246 0 : }
247 : }
248 :
249 : impl Key {
250 : pub const MIN: Key = Key {
251 : field1: u8::MIN,
252 : field2: u32::MIN,
253 : field3: u32::MIN,
254 : field4: u32::MIN,
255 : field5: u8::MIN,
256 : field6: u32::MIN,
257 : };
258 : pub const MAX: Key = Key {
259 : field1: u8::MAX,
260 : field2: u32::MAX,
261 : field3: u32::MAX,
262 : field4: u32::MAX,
263 : field5: u8::MAX,
264 : field6: u32::MAX,
265 : };
266 :
267 39613 : pub fn from_hex(s: &str) -> Result<Self> {
268 39613 : if s.len() != 36 {
269 4 : bail!("parse error");
270 39609 : }
271 39609 : Ok(Key {
272 39609 : field1: u8::from_str_radix(&s[0..2], 16)?,
273 39609 : field2: u32::from_str_radix(&s[2..10], 16)?,
274 39609 : field3: u32::from_str_radix(&s[10..18], 16)?,
275 39609 : field4: u32::from_str_radix(&s[18..26], 16)?,
276 39609 : field5: u8::from_str_radix(&s[26..28], 16)?,
277 39609 : field6: u32::from_str_radix(&s[28..36], 16)?,
278 : })
279 39613 : }
280 : }
281 :
282 : // Layout of the Key address space
283 : //
284 : // The Key struct, used to address the underlying key-value store, consists of
285 : // 18 bytes, split into six fields. See 'Key' in repository.rs. We need to map
286 : // all the data and metadata keys into those 18 bytes.
287 : //
288 : // Principles for the mapping:
289 : //
290 : // - Things that are often accessed or modified together, should be close to
291 : // each other in the key space. For example, if a relation is extended by one
292 : // block, we create a new key-value pair for the block data, and update the
293 : // relation size entry. Because of that, the RelSize key comes after all the
294 : // RelBlocks of a relation: the RelSize and the last RelBlock are always next
295 : // to each other.
296 : //
297 : // The key space is divided into four major sections, identified by the first
298 : // byte, and the form a hierarchy:
299 : //
300 : // 00 Relation data and metadata
301 : //
302 : // DbDir () -> (dbnode, spcnode)
303 : // Filenodemap
304 : // RelDir -> relnode forknum
305 : // RelBlocks
306 : // RelSize
307 : //
308 : // 01 SLRUs
309 : //
310 : // SlruDir kind
311 : // SlruSegBlocks segno
312 : // SlruSegSize
313 : //
314 : // 02 pg_twophase
315 : //
316 : // 03 misc
317 : // Controlfile
318 : // checkpoint
319 : // pg_version
320 : //
321 : // 04 aux files
322 : //
323 : // Below is a full list of the keyspace allocation:
324 : //
325 : // DbDir:
326 : // 00 00000000 00000000 00000000 00 00000000
327 : //
328 : // Filenodemap:
329 : // 00 SPCNODE DBNODE 00000000 00 00000000
330 : //
331 : // RelDir:
332 : // 00 SPCNODE DBNODE 00000000 00 00000001 (Postgres never uses relfilenode 0)
333 : //
334 : // RelBlock:
335 : // 00 SPCNODE DBNODE RELNODE FORK BLKNUM
336 : //
337 : // RelSize:
338 : // 00 SPCNODE DBNODE RELNODE FORK FFFFFFFF
339 : //
340 : // SlruDir:
341 : // 01 kind 00000000 00000000 00 00000000
342 : //
343 : // SlruSegBlock:
344 : // 01 kind 00000001 SEGNO 00 BLKNUM
345 : //
346 : // SlruSegSize:
347 : // 01 kind 00000001 SEGNO 00 FFFFFFFF
348 : //
349 : // TwoPhaseDir:
350 : // 02 00000000 00000000 00000000 00 00000000
351 : //
352 : // TwoPhaseFile:
353 : //
354 : // 02 00000000 00000000 00XXXXXX XX XXXXXXXX
355 : //
356 : // \______XID_________/
357 : //
358 : // The 64-bit XID is stored a little awkwardly in field6, field5 and
359 : // field4. PostgreSQL v16 and below only stored a 32-bit XID, which
360 : // fit completely in field6, but starting with PostgreSQL v17, a full
361 : // 64-bit XID is used. Most pageserver code that accesses
362 : // TwoPhaseFiles now deals with 64-bit XIDs even on v16, the high bits
363 : // are just unused.
364 : //
365 : // ControlFile:
366 : // 03 00000000 00000000 00000000 00 00000000
367 : //
368 : // Checkpoint:
369 : // 03 00000000 00000000 00000000 00 00000001
370 : //
371 : // AuxFiles:
372 : // 03 00000000 00000000 00000000 00 00000002
373 : //
374 :
375 : //-- Section 01: relation data and metadata
376 :
377 : pub const DBDIR_KEY: Key = Key {
378 : field1: 0x00,
379 : field2: 0,
380 : field3: 0,
381 : field4: 0,
382 : field5: 0,
383 : field6: 0,
384 : };
385 :
386 : #[inline(always)]
387 0 : pub fn dbdir_key_range(spcnode: Oid, dbnode: Oid) -> Range<Key> {
388 0 : Key {
389 0 : field1: 0x00,
390 0 : field2: spcnode,
391 0 : field3: dbnode,
392 0 : field4: 0,
393 0 : field5: 0,
394 0 : field6: 0,
395 0 : }..Key {
396 0 : field1: 0x00,
397 0 : field2: spcnode,
398 0 : field3: dbnode,
399 0 : field4: 0xffffffff,
400 0 : field5: 0xff,
401 0 : field6: 0xffffffff,
402 0 : }
403 0 : }
404 :
405 : #[inline(always)]
406 16 : pub fn relmap_file_key(spcnode: Oid, dbnode: Oid) -> Key {
407 16 : Key {
408 16 : field1: 0x00,
409 16 : field2: spcnode,
410 16 : field3: dbnode,
411 16 : field4: 0,
412 16 : field5: 0,
413 16 : field6: 0,
414 16 : }
415 16 : }
416 :
417 : #[inline(always)]
418 1948 : pub fn rel_dir_to_key(spcnode: Oid, dbnode: Oid) -> Key {
419 1948 : Key {
420 1948 : field1: 0x00,
421 1948 : field2: spcnode,
422 1948 : field3: dbnode,
423 1948 : field4: 0,
424 1948 : field5: 0,
425 1948 : field6: 1,
426 1948 : }
427 1948 : }
428 :
429 : #[inline(always)]
430 1291898 : pub fn rel_block_to_key(rel: RelTag, blknum: BlockNumber) -> Key {
431 1291898 : Key {
432 1291898 : field1: 0x00,
433 1291898 : field2: rel.spcnode,
434 1291898 : field3: rel.dbnode,
435 1291898 : field4: rel.relnode,
436 1291898 : field5: rel.forknum,
437 1291898 : field6: blknum,
438 1291898 : }
439 1291898 : }
440 :
441 : #[inline(always)]
442 289754 : pub fn rel_size_to_key(rel: RelTag) -> Key {
443 289754 : Key {
444 289754 : field1: 0x00,
445 289754 : field2: rel.spcnode,
446 289754 : field3: rel.dbnode,
447 289754 : field4: rel.relnode,
448 289754 : field5: rel.forknum,
449 289754 : field6: 0xffff_ffff,
450 289754 : }
451 289754 : }
452 :
453 : impl Key {
454 : #[inline(always)]
455 0 : pub fn is_rel_size_key(&self) -> bool {
456 0 : self.field1 == 0 && self.field6 == u32::MAX
457 0 : }
458 : }
459 :
460 : #[inline(always)]
461 2 : pub fn rel_key_range(rel: RelTag) -> Range<Key> {
462 2 : Key {
463 2 : field1: 0x00,
464 2 : field2: rel.spcnode,
465 2 : field3: rel.dbnode,
466 2 : field4: rel.relnode,
467 2 : field5: rel.forknum,
468 2 : field6: 0,
469 2 : }..Key {
470 2 : field1: 0x00,
471 2 : field2: rel.spcnode,
472 2 : field3: rel.dbnode,
473 2 : field4: rel.relnode,
474 2 : field5: rel.forknum + 1,
475 2 : field6: 0,
476 2 : }
477 2 : }
478 :
479 : //-- Section 02: SLRUs
480 :
481 : #[inline(always)]
482 1404 : pub fn slru_dir_to_key(kind: SlruKind) -> Key {
483 1404 : Key {
484 1404 : field1: 0x01,
485 1404 : field2: match kind {
486 468 : SlruKind::Clog => 0x00,
487 468 : SlruKind::MultiXactMembers => 0x01,
488 468 : SlruKind::MultiXactOffsets => 0x02,
489 : },
490 : field3: 0,
491 : field4: 0,
492 : field5: 0,
493 : field6: 0,
494 : }
495 1404 : }
496 :
497 : #[inline(always)]
498 0 : pub fn slru_dir_kind(key: &Key) -> Option<Result<SlruKind, u32>> {
499 0 : if key.field1 == 0x01
500 0 : && key.field3 == 0
501 0 : && key.field4 == 0
502 0 : && key.field5 == 0
503 0 : && key.field6 == 0
504 : {
505 0 : match key.field2 {
506 0 : 0 => Some(Ok(SlruKind::Clog)),
507 0 : 1 => Some(Ok(SlruKind::MultiXactMembers)),
508 0 : 2 => Some(Ok(SlruKind::MultiXactOffsets)),
509 0 : x => Some(Err(x)),
510 : }
511 : } else {
512 0 : None
513 : }
514 0 : }
515 :
516 : #[inline(always)]
517 14 : pub fn slru_block_to_key(kind: SlruKind, segno: u32, blknum: BlockNumber) -> Key {
518 14 : Key {
519 14 : field1: 0x01,
520 14 : field2: match kind {
521 10 : SlruKind::Clog => 0x00,
522 2 : SlruKind::MultiXactMembers => 0x01,
523 2 : SlruKind::MultiXactOffsets => 0x02,
524 : },
525 : field3: 1,
526 14 : field4: segno,
527 14 : field5: 0,
528 14 : field6: blknum,
529 14 : }
530 14 : }
531 :
532 : #[inline(always)]
533 6 : pub fn slru_segment_size_to_key(kind: SlruKind, segno: u32) -> Key {
534 6 : Key {
535 6 : field1: 0x01,
536 6 : field2: match kind {
537 2 : SlruKind::Clog => 0x00,
538 2 : SlruKind::MultiXactMembers => 0x01,
539 2 : SlruKind::MultiXactOffsets => 0x02,
540 : },
541 : field3: 1,
542 6 : field4: segno,
543 6 : field5: 0,
544 6 : field6: 0xffff_ffff,
545 6 : }
546 6 : }
547 :
548 : impl Key {
549 0 : pub fn is_slru_segment_size_key(&self) -> bool {
550 0 : self.field1 == 0x01
551 0 : && self.field2 < 0x03
552 0 : && self.field3 == 0x01
553 0 : && self.field5 == 0
554 0 : && self.field6 == u32::MAX
555 0 : }
556 : }
557 :
558 : #[inline(always)]
559 0 : pub fn slru_segment_key_range(kind: SlruKind, segno: u32) -> Range<Key> {
560 0 : let field2 = match kind {
561 0 : SlruKind::Clog => 0x00,
562 0 : SlruKind::MultiXactMembers => 0x01,
563 0 : SlruKind::MultiXactOffsets => 0x02,
564 : };
565 :
566 0 : Key {
567 0 : field1: 0x01,
568 0 : field2,
569 0 : field3: 1,
570 0 : field4: segno,
571 0 : field5: 0,
572 0 : field6: 0,
573 0 : }..Key {
574 0 : field1: 0x01,
575 0 : field2,
576 0 : field3: 1,
577 0 : field4: segno,
578 0 : field5: 1,
579 0 : field6: 0,
580 0 : }
581 0 : }
582 :
583 : //-- Section 03: pg_twophase
584 :
585 : pub const TWOPHASEDIR_KEY: Key = Key {
586 : field1: 0x02,
587 : field2: 0,
588 : field3: 0,
589 : field4: 0,
590 : field5: 0,
591 : field6: 0,
592 : };
593 :
594 : #[inline(always)]
595 0 : pub fn twophase_file_key(xid: u64) -> Key {
596 0 : Key {
597 0 : field1: 0x02,
598 0 : field2: 0,
599 0 : field3: 0,
600 0 : field4: ((xid & 0xFFFFFF0000000000) >> 40) as u32,
601 0 : field5: ((xid & 0x000000FF00000000) >> 32) as u8,
602 0 : field6: (xid & 0x00000000FFFFFFFF) as u32,
603 0 : }
604 0 : }
605 :
606 : #[inline(always)]
607 0 : pub fn twophase_key_range(xid: u64) -> Range<Key> {
608 0 : // 64-bit XIDs really should not overflow
609 0 : let (next_xid, overflowed) = xid.overflowing_add(1);
610 0 :
611 0 : Key {
612 0 : field1: 0x02,
613 0 : field2: 0,
614 0 : field3: 0,
615 0 : field4: ((xid & 0xFFFFFF0000000000) >> 40) as u32,
616 0 : field5: ((xid & 0x000000FF00000000) >> 32) as u8,
617 0 : field6: (xid & 0x00000000FFFFFFFF) as u32,
618 0 : }..Key {
619 0 : field1: 0x02,
620 0 : field2: 0,
621 0 : field3: u32::from(overflowed),
622 0 : field4: ((next_xid & 0xFFFFFF0000000000) >> 40) as u32,
623 0 : field5: ((next_xid & 0x000000FF00000000) >> 32) as u8,
624 0 : field6: (next_xid & 0x00000000FFFFFFFF) as u32,
625 0 : }
626 0 : }
627 :
628 : //-- Section 03: Control file
629 : pub const CONTROLFILE_KEY: Key = Key {
630 : field1: 0x03,
631 : field2: 0,
632 : field3: 0,
633 : field4: 0,
634 : field5: 0,
635 : field6: 0,
636 : };
637 :
638 : pub const CHECKPOINT_KEY: Key = Key {
639 : field1: 0x03,
640 : field2: 0,
641 : field3: 0,
642 : field4: 0,
643 : field5: 0,
644 : field6: 1,
645 : };
646 :
647 : pub const AUX_FILES_KEY: Key = Key {
648 : field1: 0x03,
649 : field2: 0,
650 : field3: 0,
651 : field4: 0,
652 : field5: 0,
653 : field6: 2,
654 : };
655 :
656 : #[inline(always)]
657 0 : pub fn repl_origin_key(origin_id: RepOriginId) -> Key {
658 0 : Key {
659 0 : field1: REPL_ORIGIN_KEY_PREFIX,
660 0 : field2: 0,
661 0 : field3: 0,
662 0 : field4: 0,
663 0 : field5: 0,
664 0 : field6: origin_id as u32,
665 0 : }
666 0 : }
667 :
668 : /// Get the range of replorigin keys.
669 290 : pub fn repl_origin_key_range() -> Range<Key> {
670 290 : Key {
671 290 : field1: REPL_ORIGIN_KEY_PREFIX,
672 290 : field2: 0,
673 290 : field3: 0,
674 290 : field4: 0,
675 290 : field5: 0,
676 290 : field6: 0,
677 290 : }..Key {
678 290 : field1: REPL_ORIGIN_KEY_PREFIX,
679 290 : field2: 0,
680 290 : field3: 0,
681 290 : field4: 0,
682 290 : field5: 0,
683 290 : field6: 0x10000,
684 290 : }
685 290 : }
686 :
687 : // Reverse mappings for a few Keys.
688 : // These are needed by WAL redo manager.
689 :
690 : /// Non inherited range for vectored get.
691 : pub const NON_INHERITED_RANGE: Range<Key> = AUX_FILES_KEY..AUX_FILES_KEY.next();
692 : /// Sparse keyspace range for vectored get. Missing key error will be ignored for this range.
693 : pub const NON_INHERITED_SPARSE_RANGE: Range<Key> = Key::metadata_key_range();
694 :
695 : impl Key {
696 : // AUX_FILES currently stores only data for logical replication (slots etc), and
697 : // we don't preserve these on a branch because safekeepers can't follow timeline
698 : // switch (and generally it likely should be optional), so ignore these.
699 : #[inline(always)]
700 0 : pub fn is_inherited_key(self) -> bool {
701 0 : !NON_INHERITED_RANGE.contains(&self) && !NON_INHERITED_SPARSE_RANGE.contains(&self)
702 0 : }
703 :
704 : #[inline(always)]
705 0 : pub fn is_rel_fsm_block_key(self) -> bool {
706 0 : self.field1 == 0x00
707 0 : && self.field4 != 0
708 0 : && self.field5 == FSM_FORKNUM
709 0 : && self.field6 != 0xffffffff
710 0 : }
711 :
712 : #[inline(always)]
713 0 : pub fn is_rel_vm_block_key(self) -> bool {
714 0 : self.field1 == 0x00
715 0 : && self.field4 != 0
716 0 : && self.field5 == VISIBILITYMAP_FORKNUM
717 0 : && self.field6 != 0xffffffff
718 0 : }
719 :
720 : #[inline(always)]
721 0 : pub fn to_slru_block(self) -> anyhow::Result<(SlruKind, u32, BlockNumber)> {
722 0 : Ok(match self.field1 {
723 : 0x01 => {
724 0 : let kind = match self.field2 {
725 0 : 0x00 => SlruKind::Clog,
726 0 : 0x01 => SlruKind::MultiXactMembers,
727 0 : 0x02 => SlruKind::MultiXactOffsets,
728 0 : _ => anyhow::bail!("unrecognized slru kind 0x{:02x}", self.field2),
729 : };
730 0 : let segno = self.field4;
731 0 : let blknum = self.field6;
732 0 :
733 0 : (kind, segno, blknum)
734 : }
735 0 : _ => anyhow::bail!("unexpected value kind 0x{:02x}", self.field1),
736 : })
737 0 : }
738 :
739 : #[inline(always)]
740 572564 : pub fn is_slru_block_key(self) -> bool {
741 572564 : self.field1 == 0x01 // SLRU-related
742 560 : && self.field3 == 0x00000001 // but not SlruDir
743 20 : && self.field6 != 0xffffffff // and not SlruSegSize
744 572564 : }
745 :
746 : #[inline(always)]
747 6695341 : pub fn is_rel_block_key(&self) -> bool {
748 6695341 : self.field1 == 0x00 && self.field4 != 0 && self.field6 != 0xffffffff
749 6695341 : }
750 :
751 : #[inline(always)]
752 200 : pub fn is_rel_dir_key(&self) -> bool {
753 200 : self.field1 == 0x00
754 200 : && self.field2 != 0
755 0 : && self.field3 != 0
756 0 : && self.field4 == 0
757 0 : && self.field5 == 0
758 0 : && self.field6 == 1
759 200 : }
760 :
761 : /// Guaranteed to return `Ok()` if [`Self::is_rel_block_key`] returns `true` for `key`.
762 : #[inline(always)]
763 145648 : pub fn to_rel_block(self) -> anyhow::Result<(RelTag, BlockNumber)> {
764 145648 : Ok(match self.field1 {
765 145648 : 0x00 => (
766 145648 : RelTag {
767 145648 : spcnode: self.field2,
768 145648 : dbnode: self.field3,
769 145648 : relnode: self.field4,
770 145648 : forknum: self.field5,
771 145648 : },
772 145648 : self.field6,
773 145648 : ),
774 0 : _ => anyhow::bail!("unexpected value kind 0x{:02x}", self.field1),
775 : })
776 145648 : }
777 : }
778 :
779 : impl std::str::FromStr for Key {
780 : type Err = anyhow::Error;
781 :
782 9 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
783 9 : Self::from_hex(s)
784 9 : }
785 : }
786 :
787 : #[cfg(test)]
788 : mod tests {
789 : use std::str::FromStr;
790 :
791 : use crate::key::is_metadata_key_slice;
792 : use crate::key::Key;
793 :
794 : use rand::Rng;
795 : use rand::SeedableRng;
796 :
797 : use super::AUX_KEY_PREFIX;
798 :
799 : #[test]
800 1 : fn display_fromstr_bijection() {
801 1 : let mut rng = rand::rngs::StdRng::seed_from_u64(42);
802 1 :
803 1 : let key = Key {
804 1 : field1: rng.gen(),
805 1 : field2: rng.gen(),
806 1 : field3: rng.gen(),
807 1 : field4: rng.gen(),
808 1 : field5: rng.gen(),
809 1 : field6: rng.gen(),
810 1 : };
811 1 :
812 1 : assert_eq!(key, Key::from_str(&format!("{key}")).unwrap());
813 1 : }
814 :
815 : #[test]
816 1 : fn test_metadata_keys() {
817 1 : let mut metadata_key = vec![AUX_KEY_PREFIX];
818 1 : metadata_key.extend_from_slice(&[0xFF; 15]);
819 1 : let encoded_key = Key::from_metadata_key(&metadata_key);
820 1 : let output_key = encoded_key.to_i128().to_be_bytes();
821 1 : assert_eq!(metadata_key, output_key);
822 1 : assert!(encoded_key.is_metadata_key());
823 1 : assert!(is_metadata_key_slice(&metadata_key));
824 1 : }
825 :
826 : #[test]
827 1 : fn test_possible_largest_key() {
828 1 : Key::from_i128(0x7FFF_FFFF_FFFF_FFFF_FFFF_FFFF_FFFF_FFFF);
829 1 : // TODO: put this key into the system and see if anything breaks.
830 1 : }
831 : }
|
