risingwave_hummock_sdk/
key.rs

1// Copyright 2025 RisingWave Labs
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15use std::borrow::Borrow;
16use std::cmp::Ordering;
17use std::fmt::Debug;
18use std::iter::once;
19use std::ops::Bound::*;
20use std::ops::{Bound, Deref, DerefMut, RangeBounds};
21use std::ptr;
22
23use bytes::{Buf, BufMut, Bytes, BytesMut};
24use risingwave_common::catalog::TableId;
25use risingwave_common::hash::VirtualNode;
26use risingwave_common_estimate_size::EstimateSize;
27
28use crate::{EpochWithGap, HummockEpoch};
29
30pub const EPOCH_LEN: usize = std::mem::size_of::<HummockEpoch>();
31pub const TABLE_PREFIX_LEN: usize = std::mem::size_of::<u32>();
32// Max length for key overlap and diff length. See KeyPrefix::encode.
33pub const MAX_KEY_LEN: usize = u16::MAX as usize;
34
35pub type KeyPayloadType = Bytes;
36pub type TableKeyRange = (
37    Bound<TableKey<KeyPayloadType>>,
38    Bound<TableKey<KeyPayloadType>>,
39);
40pub type UserKeyRange = (
41    Bound<UserKey<KeyPayloadType>>,
42    Bound<UserKey<KeyPayloadType>>,
43);
44pub type UserKeyRangeRef<'a> = (Bound<UserKey<&'a [u8]>>, Bound<UserKey<&'a [u8]>>);
45pub type FullKeyRange = (
46    Bound<FullKey<KeyPayloadType>>,
47    Bound<FullKey<KeyPayloadType>>,
48);
49
50pub fn is_empty_key_range(key_range: &TableKeyRange) -> bool {
51    match key_range {
52        (Included(start), Excluded(end)) => start == end,
53        _ => false,
54    }
55}
56
57/// Returns left inclusive and right exclusive vnode index of the given range.
58///
59/// # Vnode count unawareness
60///
61/// Note that this function is not aware of the vnode count that is actually used in this table.
62/// For example, if the total vnode count is 256, `Unbounded` can be a correct end bound for vnode 255,
63/// but this function will still return `Excluded(256)`.
64///
65/// See also [`vnode`] and [`end_bound_of_vnode`] which hold such invariant.
66pub fn vnode_range(range: &TableKeyRange) -> (usize, usize) {
67    let (left, right) = range;
68    let left = match left {
69        Included(key) | Excluded(key) => key.vnode_part().to_index(),
70        Unbounded => 0,
71    };
72    let right = match right {
73        Included(key) => key.vnode_part().to_index() + 1,
74        Excluded(key) => {
75            let (vnode, inner_key) = key.split_vnode();
76            if inner_key.is_empty() {
77                // When the exclusive end key range contains only a vnode,
78                // the whole vnode is excluded.
79                vnode.to_index()
80            } else {
81                vnode.to_index() + 1
82            }
83        }
84        Unbounded => VirtualNode::MAX_REPRESENTABLE.to_index() + 1,
85    };
86    (left, right)
87}
88
89/// Ensure there is only one vnode involved in table key range and return the vnode.
90///
91/// # Vnode count unawareness
92///
93/// Note that this function is not aware of the vnode count that is actually used in this table.
94/// For example, if the total vnode count is 256, `Unbounded` can be a correct end bound for vnode 255,
95/// but this function will still require `Excluded(256)`.
96///
97/// See also [`vnode_range`] and [`end_bound_of_vnode`] which hold such invariant.
98pub fn vnode(range: &TableKeyRange) -> VirtualNode {
99    let (l, r_exclusive) = vnode_range(range);
100    assert_eq!(r_exclusive - l, 1);
101    VirtualNode::from_index(l)
102}
103
104/// Converts user key to full key by appending `epoch` to the user key.
105pub fn key_with_epoch(mut user_key: Vec<u8>, epoch: HummockEpoch) -> Vec<u8> {
106    let res = epoch.to_be();
107    user_key.reserve(EPOCH_LEN);
108    let buf = user_key.chunk_mut();
109
110    // TODO: check whether this hack improves performance
111    unsafe {
112        ptr::copy_nonoverlapping(
113            &res as *const _ as *const u8,
114            buf.as_mut_ptr() as *mut _,
115            EPOCH_LEN,
116        );
117        user_key.advance_mut(EPOCH_LEN);
118    }
119
120    user_key
121}
122
123/// Splits a full key into its user key part and epoch part.
124#[inline]
125pub fn split_key_epoch(full_key: &[u8]) -> (&[u8], &[u8]) {
126    let pos = full_key
127        .len()
128        .checked_sub(EPOCH_LEN)
129        .unwrap_or_else(|| panic!("bad full key format: {:?}", full_key));
130    full_key.split_at(pos)
131}
132
133/// Extract encoded [`UserKey`] from encoded [`FullKey`] without epoch part
134pub fn user_key(full_key: &[u8]) -> &[u8] {
135    split_key_epoch(full_key).0
136}
137
138/// Extract table key from encoded [`UserKey`] without table id part
139pub fn table_key(user_key: &[u8]) -> &[u8] {
140    &user_key[TABLE_PREFIX_LEN..]
141}
142
143#[inline(always)]
144/// Extract encoded [`UserKey`] from encoded [`FullKey`] but allow empty slice
145pub fn get_user_key(full_key: &[u8]) -> Vec<u8> {
146    if full_key.is_empty() {
147        vec![]
148    } else {
149        user_key(full_key).to_vec()
150    }
151}
152
153/// Extract table id from encoded [`FullKey`]
154#[inline(always)]
155pub fn get_table_id(full_key: &[u8]) -> u32 {
156    let mut buf = full_key;
157    buf.get_u32()
158}
159
160// Copyright 2016 TiKV Project Authors. Licensed under Apache-2.0.
161
162/// Computes the next key of the given key.
163///
164/// If the key has no successor key (e.g. the input is "\xff\xff"), the result
165/// would be an empty vector.
166///
167/// # Examples
168///
169/// ```rust
170/// use risingwave_hummock_sdk::key::next_key;
171/// assert_eq!(next_key(b"123"), b"124");
172/// assert_eq!(next_key(b"12\xff"), b"13");
173/// assert_eq!(next_key(b"\xff\xff"), b"");
174/// assert_eq!(next_key(b"\xff\xfe"), b"\xff\xff");
175/// assert_eq!(next_key(b"T"), b"U");
176/// assert_eq!(next_key(b""), b"");
177/// ```
178pub fn next_key(key: &[u8]) -> Vec<u8> {
179    if let Some((s, e)) = next_key_no_alloc(key) {
180        let mut res = Vec::with_capacity(s.len() + 1);
181        res.extend_from_slice(s);
182        res.push(e);
183        res
184    } else {
185        Vec::new()
186    }
187}
188
189/// Computes the previous key of the given key.
190///
191/// If the key has no predecessor key (e.g. the input is "\x00\x00"), the result
192/// would be a "\xff\xff" vector.
193///
194/// # Examples
195///
196/// ```rust
197/// use risingwave_hummock_sdk::key::prev_key;
198/// assert_eq!(prev_key(b"123"), b"122");
199/// assert_eq!(prev_key(b"12\x00"), b"11\xff");
200/// assert_eq!(prev_key(b"\x00\x00"), b"\xff\xff");
201/// assert_eq!(prev_key(b"\x00\x01"), b"\x00\x00");
202/// assert_eq!(prev_key(b"T"), b"S");
203/// assert_eq!(prev_key(b""), b"");
204/// ```
205pub fn prev_key(key: &[u8]) -> Vec<u8> {
206    let pos = key.iter().rposition(|b| *b != 0x00);
207    match pos {
208        Some(pos) => {
209            let mut res = Vec::with_capacity(key.len());
210            res.extend_from_slice(&key[0..pos]);
211            res.push(key[pos] - 1);
212            if pos + 1 < key.len() {
213                res.push(b"\xff".to_owned()[0]);
214            }
215            res
216        }
217        None => {
218            vec![0xff; key.len()]
219        }
220    }
221}
222
223fn next_key_no_alloc(key: &[u8]) -> Option<(&[u8], u8)> {
224    let pos = key.iter().rposition(|b| *b != 0xff)?;
225    Some((&key[..pos], key[pos] + 1))
226}
227
228// End Copyright 2016 TiKV Project Authors. Licensed under Apache-2.0.
229
230/// compute the next epoch, and don't change the bytes of the u8 slice.
231/// # Examples
232///
233/// ```rust
234/// use risingwave_hummock_sdk::key::next_epoch;
235/// assert_eq!(next_epoch(b"123"), b"124");
236/// assert_eq!(next_epoch(b"\xff\x00\xff"), b"\xff\x01\x00");
237/// assert_eq!(next_epoch(b"\xff\xff"), b"\x00\x00");
238/// assert_eq!(next_epoch(b"\x00\x00"), b"\x00\x01");
239/// assert_eq!(next_epoch(b"S"), b"T");
240/// assert_eq!(next_epoch(b""), b"");
241/// ```
242pub fn next_epoch(epoch: &[u8]) -> Vec<u8> {
243    let pos = epoch.iter().rposition(|b| *b != 0xff);
244    match pos {
245        Some(mut pos) => {
246            let mut res = Vec::with_capacity(epoch.len());
247            res.extend_from_slice(&epoch[0..pos]);
248            res.push(epoch[pos] + 1);
249            while pos + 1 < epoch.len() {
250                res.push(0x00);
251                pos += 1;
252            }
253            res
254        }
255        None => {
256            vec![0x00; epoch.len()]
257        }
258    }
259}
260
261/// compute the prev epoch, and don't change the bytes of the u8 slice.
262/// # Examples
263///
264/// ```rust
265/// use risingwave_hummock_sdk::key::prev_epoch;
266/// assert_eq!(prev_epoch(b"124"), b"123");
267/// assert_eq!(prev_epoch(b"\xff\x01\x00"), b"\xff\x00\xff");
268/// assert_eq!(prev_epoch(b"\x00\x00"), b"\xff\xff");
269/// assert_eq!(prev_epoch(b"\x00\x01"), b"\x00\x00");
270/// assert_eq!(prev_epoch(b"T"), b"S");
271/// assert_eq!(prev_epoch(b""), b"");
272/// ```
273pub fn prev_epoch(epoch: &[u8]) -> Vec<u8> {
274    let pos = epoch.iter().rposition(|b| *b != 0x00);
275    match pos {
276        Some(mut pos) => {
277            let mut res = Vec::with_capacity(epoch.len());
278            res.extend_from_slice(&epoch[0..pos]);
279            res.push(epoch[pos] - 1);
280            while pos + 1 < epoch.len() {
281                res.push(0xff);
282                pos += 1;
283            }
284            res
285        }
286        None => {
287            vec![0xff; epoch.len()]
288        }
289    }
290}
291
292/// compute the next full key of the given full key
293///
294/// if the `user_key` has no successor key, the result will be a empty vec
295pub fn next_full_key(full_key: &[u8]) -> Vec<u8> {
296    let (user_key, epoch) = split_key_epoch(full_key);
297    let prev_epoch = prev_epoch(epoch);
298    let mut res = Vec::with_capacity(full_key.len());
299    if prev_epoch.cmp(&vec![0xff; prev_epoch.len()]) == Ordering::Equal {
300        let next_user_key = next_key(user_key);
301        if next_user_key.is_empty() {
302            return Vec::new();
303        }
304        res.extend_from_slice(next_user_key.as_slice());
305        res.extend_from_slice(prev_epoch.as_slice());
306        res
307    } else {
308        res.extend_from_slice(user_key);
309        res.extend_from_slice(prev_epoch.as_slice());
310        res
311    }
312}
313
314/// compute the prev full key of the given full key
315///
316/// if the `user_key` has no predecessor key, the result will be a empty vec
317pub fn prev_full_key(full_key: &[u8]) -> Vec<u8> {
318    let (user_key, epoch) = split_key_epoch(full_key);
319    let next_epoch = next_epoch(epoch);
320    let mut res = Vec::with_capacity(full_key.len());
321    if next_epoch.cmp(&vec![0x00; next_epoch.len()]) == Ordering::Equal {
322        let prev_user_key = prev_key(user_key);
323        if prev_user_key.cmp(&vec![0xff; prev_user_key.len()]) == Ordering::Equal {
324            return Vec::new();
325        }
326        res.extend_from_slice(prev_user_key.as_slice());
327        res.extend_from_slice(next_epoch.as_slice());
328        res
329    } else {
330        res.extend_from_slice(user_key);
331        res.extend_from_slice(next_epoch.as_slice());
332        res
333    }
334}
335
336/// [`Unbounded`] if the vnode is the maximum representable value (i.e. [`VirtualNode::MAX_REPRESENTABLE`]),
337/// otherwise [`Excluded`] the next vnode.
338///
339/// Note that this function is not aware of the vnode count that is actually used in this table.
340/// For example, if the total vnode count is 256, `Unbounded` can be a correct end bound for vnode 255,
341/// but this function will still return `Excluded(256)`. See also [`vnode`] and [`vnode_range`] which
342/// rely on such invariant.
343pub fn end_bound_of_vnode(vnode: VirtualNode) -> Bound<Bytes> {
344    if vnode == VirtualNode::MAX_REPRESENTABLE {
345        Unbounded
346    } else {
347        let end_bound_index = vnode.to_index() + 1;
348        Excluded(Bytes::copy_from_slice(
349            &VirtualNode::from_index(end_bound_index).to_be_bytes(),
350        ))
351    }
352}
353
354/// Get the end bound of the given `prefix` when transforming it to a key range.
355pub fn end_bound_of_prefix(prefix: &[u8]) -> Bound<Bytes> {
356    if let Some((s, e)) = next_key_no_alloc(prefix) {
357        let mut buf = BytesMut::with_capacity(s.len() + 1);
358        buf.extend_from_slice(s);
359        buf.put_u8(e);
360        Excluded(buf.freeze())
361    } else {
362        Unbounded
363    }
364}
365
366/// Get the start bound of the given `prefix` when it is excluded from the range.
367pub fn start_bound_of_excluded_prefix(prefix: &[u8]) -> Bound<Bytes> {
368    if let Some((s, e)) = next_key_no_alloc(prefix) {
369        let mut buf = BytesMut::with_capacity(s.len() + 1);
370        buf.extend_from_slice(s);
371        buf.put_u8(e);
372        Included(buf.freeze())
373    } else {
374        panic!("the prefix is the maximum value")
375    }
376}
377
378/// Transform the given `prefix` to a key range.
379pub fn range_of_prefix(prefix: &[u8]) -> (Bound<Bytes>, Bound<Bytes>) {
380    if prefix.is_empty() {
381        (Unbounded, Unbounded)
382    } else {
383        (
384            Included(Bytes::copy_from_slice(prefix)),
385            end_bound_of_prefix(prefix),
386        )
387    }
388}
389
390pub fn prefix_slice_with_vnode(vnode: VirtualNode, slice: &[u8]) -> Bytes {
391    let prefix = vnode.to_be_bytes();
392    let mut buf = BytesMut::with_capacity(prefix.len() + slice.len());
393    buf.extend_from_slice(&prefix);
394    buf.extend_from_slice(slice);
395    buf.freeze()
396}
397
398/// Prepend the `prefix` to the given key `range`.
399pub fn prefixed_range_with_vnode<B: AsRef<[u8]>>(
400    range: impl RangeBounds<B>,
401    vnode: VirtualNode,
402) -> TableKeyRange {
403    let prefixed = |b: &B| -> Bytes { prefix_slice_with_vnode(vnode, b.as_ref()) };
404
405    let start: Bound<Bytes> = match range.start_bound() {
406        Included(b) => Included(prefixed(b)),
407        Excluded(b) => {
408            assert!(!b.as_ref().is_empty());
409            Excluded(prefixed(b))
410        }
411        Unbounded => Included(Bytes::copy_from_slice(&vnode.to_be_bytes())),
412    };
413
414    let end = match range.end_bound() {
415        Included(b) => Included(prefixed(b)),
416        Excluded(b) => {
417            assert!(!b.as_ref().is_empty());
418            Excluded(prefixed(b))
419        }
420        Unbounded => end_bound_of_vnode(vnode),
421    };
422
423    map_table_key_range((start, end))
424}
425
426pub trait SetSlice<S: AsRef<[u8]> + ?Sized> {
427    fn set(&mut self, value: &S);
428}
429
430impl<S: AsRef<[u8]> + ?Sized> SetSlice<S> for Vec<u8> {
431    fn set(&mut self, value: &S) {
432        self.clear();
433        self.extend_from_slice(value.as_ref());
434    }
435}
436
437impl SetSlice<Bytes> for Bytes {
438    fn set(&mut self, value: &Bytes) {
439        *self = value.clone()
440    }
441}
442
443pub trait CopyFromSlice: Send + 'static {
444    fn copy_from_slice(slice: &[u8]) -> Self;
445}
446
447impl CopyFromSlice for Vec<u8> {
448    fn copy_from_slice(slice: &[u8]) -> Self {
449        Vec::from(slice)
450    }
451}
452
453impl CopyFromSlice for Bytes {
454    fn copy_from_slice(slice: &[u8]) -> Self {
455        Bytes::copy_from_slice(slice)
456    }
457}
458
459impl CopyFromSlice for () {
460    fn copy_from_slice(_: &[u8]) -> Self {}
461}
462
463/// [`TableKey`] is an internal concept in storage. It's a wrapper around the key directly from the
464/// user, to make the code clearer and avoid confusion with encoded [`UserKey`] and [`FullKey`].
465///
466/// Its name come from the assumption that Hummock is always accessed by a table-like structure
467/// identified by a [`TableId`].
468#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Default)]
469pub struct TableKey<T: AsRef<[u8]>>(pub T);
470
471impl<T: AsRef<[u8]>> Debug for TableKey<T> {
472    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
473        write!(f, "TableKey {{ {} }}", hex::encode(self.0.as_ref()))
474    }
475}
476
477impl<T: AsRef<[u8]>> Deref for TableKey<T> {
478    type Target = T;
479
480    fn deref(&self) -> &Self::Target {
481        &self.0
482    }
483}
484
485impl<T: AsRef<[u8]>> DerefMut for TableKey<T> {
486    fn deref_mut(&mut self) -> &mut Self::Target {
487        &mut self.0
488    }
489}
490
491impl<T: AsRef<[u8]>> AsRef<[u8]> for TableKey<T> {
492    fn as_ref(&self) -> &[u8] {
493        self.0.as_ref()
494    }
495}
496
497impl<T: AsRef<[u8]>> TableKey<T> {
498    pub fn split_vnode(&self) -> (VirtualNode, &[u8]) {
499        debug_assert!(
500            self.0.as_ref().len() >= VirtualNode::SIZE,
501            "too short table key: {:?}",
502            self.0.as_ref()
503        );
504        let (vnode, inner_key) = self
505            .0
506            .as_ref()
507            .split_first_chunk::<{ VirtualNode::SIZE }>()
508            .unwrap();
509        (VirtualNode::from_be_bytes(*vnode), inner_key)
510    }
511
512    pub fn vnode_part(&self) -> VirtualNode {
513        self.split_vnode().0
514    }
515
516    pub fn key_part(&self) -> &[u8] {
517        self.split_vnode().1
518    }
519
520    pub fn to_ref(&self) -> TableKey<&[u8]> {
521        TableKey(self.0.as_ref())
522    }
523}
524
525impl<T: AsRef<[u8]>> Borrow<[u8]> for TableKey<T> {
526    fn borrow(&self) -> &[u8] {
527        self.0.as_ref()
528    }
529}
530
531impl EstimateSize for TableKey<Bytes> {
532    fn estimated_heap_size(&self) -> usize {
533        self.0.estimated_heap_size()
534    }
535}
536
537impl TableKey<&[u8]> {
538    pub fn copy_into<T: CopyFromSlice + AsRef<[u8]>>(&self) -> TableKey<T> {
539        TableKey(T::copy_from_slice(self.as_ref()))
540    }
541}
542
543#[inline]
544pub fn map_table_key_range(range: (Bound<KeyPayloadType>, Bound<KeyPayloadType>)) -> TableKeyRange {
545    (range.0.map(TableKey), range.1.map(TableKey))
546}
547
548pub fn gen_key_from_bytes(vnode: VirtualNode, payload: &[u8]) -> TableKey<Bytes> {
549    TableKey(Bytes::from(
550        [vnode.to_be_bytes().as_slice(), payload].concat(),
551    ))
552}
553
554pub fn gen_key_from_str(vnode: VirtualNode, payload: &str) -> TableKey<Bytes> {
555    gen_key_from_bytes(vnode, payload.as_bytes())
556}
557
558/// [`UserKey`] is is an internal concept in storage. In the storage interface, user specifies
559/// `table_key` and `table_id` (in `ReadOptions` or `WriteOptions`) as the input. The storage
560/// will group these two values into one struct for convenient filtering.
561///
562/// The encoded format is | `table_id` | `table_key` |.
563#[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Default)]
564pub struct UserKey<T: AsRef<[u8]>> {
565    // When comparing `UserKey`, we first compare `table_id`, then `table_key`. So the order of
566    // declaration matters.
567    pub table_id: TableId,
568    pub table_key: TableKey<T>,
569}
570
571impl<T: AsRef<[u8]>> Debug for UserKey<T> {
572    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
573        write!(
574            f,
575            "UserKey {{ {}, {:?} }}",
576            self.table_id.table_id, self.table_key
577        )
578    }
579}
580
581impl<T: AsRef<[u8]>> UserKey<T> {
582    pub fn new(table_id: TableId, table_key: TableKey<T>) -> Self {
583        Self {
584            table_id,
585            table_key,
586        }
587    }
588
589    /// Pass the inner type of `table_key` to make the code less verbose.
590    pub fn for_test(table_id: TableId, table_key: T) -> Self {
591        Self {
592            table_id,
593            table_key: TableKey(table_key),
594        }
595    }
596
597    /// Encode in to a buffer.
598    pub fn encode_into(&self, buf: &mut impl BufMut) {
599        buf.put_u32(self.table_id.table_id());
600        buf.put_slice(self.table_key.as_ref());
601    }
602
603    pub fn encode_table_key_into(&self, buf: &mut impl BufMut) {
604        buf.put_slice(self.table_key.as_ref());
605    }
606
607    pub fn encode(&self) -> Vec<u8> {
608        let mut ret = Vec::with_capacity(TABLE_PREFIX_LEN + self.table_key.as_ref().len());
609        self.encode_into(&mut ret);
610        ret
611    }
612
613    pub fn is_empty(&self) -> bool {
614        self.table_key.as_ref().is_empty()
615    }
616
617    /// Get the length of the encoded format.
618    pub fn encoded_len(&self) -> usize {
619        self.table_key.as_ref().len() + TABLE_PREFIX_LEN
620    }
621
622    pub fn get_vnode_id(&self) -> usize {
623        self.table_key.vnode_part().to_index()
624    }
625}
626
627impl<'a> UserKey<&'a [u8]> {
628    /// Construct a [`UserKey`] from a byte slice. Its `table_key` will be a part of the input
629    /// `slice`.
630    pub fn decode(slice: &'a [u8]) -> Self {
631        let table_id: u32 = (&slice[..]).get_u32();
632
633        Self {
634            table_id: TableId::new(table_id),
635            table_key: TableKey(&slice[TABLE_PREFIX_LEN..]),
636        }
637    }
638
639    pub fn to_vec(self) -> UserKey<Vec<u8>> {
640        self.copy_into()
641    }
642
643    pub fn copy_into<T: CopyFromSlice + AsRef<[u8]>>(self) -> UserKey<T> {
644        UserKey {
645            table_id: self.table_id,
646            table_key: TableKey(T::copy_from_slice(self.table_key.0)),
647        }
648    }
649}
650
651impl<T: AsRef<[u8]> + Clone> UserKey<&T> {
652    pub fn cloned(self) -> UserKey<T> {
653        UserKey {
654            table_id: self.table_id,
655            table_key: TableKey(self.table_key.0.clone()),
656        }
657    }
658}
659
660impl<T: AsRef<[u8]>> UserKey<T> {
661    pub fn as_ref(&self) -> UserKey<&[u8]> {
662        UserKey::new(self.table_id, TableKey(self.table_key.as_ref()))
663    }
664}
665
666impl<T: AsRef<[u8]>> UserKey<T> {
667    /// Use this method to override an old `UserKey<Vec<u8>>` with a `UserKey<&[u8]>` to own the
668    /// table key without reallocating a new `UserKey` object.
669    pub fn set<F>(&mut self, other: UserKey<F>)
670    where
671        T: SetSlice<F>,
672        F: AsRef<[u8]>,
673    {
674        self.table_id = other.table_id;
675        self.table_key.0.set(&other.table_key.0);
676    }
677}
678
679impl UserKey<Vec<u8>> {
680    pub fn into_bytes(self) -> UserKey<Bytes> {
681        UserKey {
682            table_id: self.table_id,
683            table_key: TableKey(Bytes::from(self.table_key.0)),
684        }
685    }
686}
687
688/// [`FullKey`] is an internal concept in storage. It associates [`UserKey`] with an epoch.
689///
690/// The encoded format is | `user_key` | `epoch` |.
691#[derive(Clone, Copy, PartialEq, Eq, Hash, Default)]
692pub struct FullKey<T: AsRef<[u8]>> {
693    pub user_key: UserKey<T>,
694    pub epoch_with_gap: EpochWithGap,
695}
696
697impl<T: AsRef<[u8]>> Debug for FullKey<T> {
698    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
699        write!(
700            f,
701            "FullKey {{ {:?}, epoch: {}, epoch_with_gap: {}, spill_offset: {}}}",
702            self.user_key,
703            self.epoch_with_gap.pure_epoch(),
704            self.epoch_with_gap.as_u64(),
705            self.epoch_with_gap.as_u64() - self.epoch_with_gap.pure_epoch(),
706        )
707    }
708}
709
710impl<T: AsRef<[u8]>> FullKey<T> {
711    pub fn new(table_id: TableId, table_key: TableKey<T>, epoch: HummockEpoch) -> Self {
712        Self {
713            user_key: UserKey::new(table_id, table_key),
714            epoch_with_gap: EpochWithGap::new(epoch, 0),
715        }
716    }
717
718    pub fn new_with_gap_epoch(
719        table_id: TableId,
720        table_key: TableKey<T>,
721        epoch_with_gap: EpochWithGap,
722    ) -> Self {
723        Self {
724            user_key: UserKey::new(table_id, table_key),
725            epoch_with_gap,
726        }
727    }
728
729    pub fn from_user_key(user_key: UserKey<T>, epoch: HummockEpoch) -> Self {
730        Self {
731            user_key,
732            epoch_with_gap: EpochWithGap::new_from_epoch(epoch),
733        }
734    }
735
736    /// Pass the inner type of `table_key` to make the code less verbose.
737    pub fn for_test(table_id: TableId, table_key: T, epoch: HummockEpoch) -> Self {
738        Self {
739            user_key: UserKey::for_test(table_id, table_key),
740            epoch_with_gap: EpochWithGap::new(epoch, 0),
741        }
742    }
743
744    /// Encode in to a buffer.
745    pub fn encode_into(&self, buf: &mut impl BufMut) {
746        self.user_key.encode_into(buf);
747        buf.put_u64(self.epoch_with_gap.as_u64());
748    }
749
750    pub fn encode(&self) -> Vec<u8> {
751        let mut buf = Vec::with_capacity(
752            TABLE_PREFIX_LEN + self.user_key.table_key.as_ref().len() + EPOCH_LEN,
753        );
754        self.encode_into(&mut buf);
755        buf
756    }
757
758    // Encode in to a buffer.
759    pub fn encode_into_without_table_id(&self, buf: &mut impl BufMut) {
760        self.user_key.encode_table_key_into(buf);
761        buf.put_u64(self.epoch_with_gap.as_u64());
762    }
763
764    pub fn encode_reverse_epoch(&self) -> Vec<u8> {
765        let mut buf = Vec::with_capacity(
766            TABLE_PREFIX_LEN + self.user_key.table_key.as_ref().len() + EPOCH_LEN,
767        );
768        self.user_key.encode_into(&mut buf);
769        buf.put_u64(u64::MAX - self.epoch_with_gap.as_u64());
770        buf
771    }
772
773    pub fn is_empty(&self) -> bool {
774        self.user_key.is_empty()
775    }
776
777    /// Get the length of the encoded format.
778    pub fn encoded_len(&self) -> usize {
779        self.user_key.encoded_len() + EPOCH_LEN
780    }
781}
782
783impl<'a> FullKey<&'a [u8]> {
784    /// Construct a [`FullKey`] from a byte slice.
785    pub fn decode(slice: &'a [u8]) -> Self {
786        let epoch_pos = slice.len() - EPOCH_LEN;
787        let epoch = (&slice[epoch_pos..]).get_u64();
788
789        Self {
790            user_key: UserKey::decode(&slice[..epoch_pos]),
791            epoch_with_gap: EpochWithGap::from_u64(epoch),
792        }
793    }
794
795    /// Construct a [`FullKey`] from a byte slice without `table_id` encoded.
796    pub fn from_slice_without_table_id(
797        table_id: TableId,
798        slice_without_table_id: &'a [u8],
799    ) -> Self {
800        let epoch_pos = slice_without_table_id.len() - EPOCH_LEN;
801        let epoch = (&slice_without_table_id[epoch_pos..]).get_u64();
802
803        Self {
804            user_key: UserKey::new(table_id, TableKey(&slice_without_table_id[..epoch_pos])),
805            epoch_with_gap: EpochWithGap::from_u64(epoch),
806        }
807    }
808
809    /// Construct a [`FullKey`] from a byte slice.
810    pub fn decode_reverse_epoch(slice: &'a [u8]) -> Self {
811        let epoch_pos = slice.len() - EPOCH_LEN;
812        let epoch = (&slice[epoch_pos..]).get_u64();
813
814        Self {
815            user_key: UserKey::decode(&slice[..epoch_pos]),
816            epoch_with_gap: EpochWithGap::from_u64(u64::MAX - epoch),
817        }
818    }
819
820    pub fn to_vec(self) -> FullKey<Vec<u8>> {
821        self.copy_into()
822    }
823
824    pub fn copy_into<T: CopyFromSlice + AsRef<[u8]>>(self) -> FullKey<T> {
825        FullKey {
826            user_key: self.user_key.copy_into(),
827            epoch_with_gap: self.epoch_with_gap,
828        }
829    }
830}
831
832impl FullKey<Vec<u8>> {
833    /// Calling this method may accidentally cause memory allocation when converting `Vec` into
834    /// `Bytes`
835    pub fn into_bytes(self) -> FullKey<Bytes> {
836        FullKey {
837            epoch_with_gap: self.epoch_with_gap,
838            user_key: self.user_key.into_bytes(),
839        }
840    }
841}
842
843impl<T: AsRef<[u8]>> FullKey<T> {
844    pub fn to_ref(&self) -> FullKey<&[u8]> {
845        FullKey {
846            user_key: self.user_key.as_ref(),
847            epoch_with_gap: self.epoch_with_gap,
848        }
849    }
850}
851
852impl<T: AsRef<[u8]>> FullKey<T> {
853    /// Use this method to override an old `FullKey<Vec<u8>>` with a `FullKey<&[u8]>` to own the
854    /// table key without reallocating a new `FullKey` object.
855    pub fn set<F>(&mut self, other: FullKey<F>)
856    where
857        T: SetSlice<F>,
858        F: AsRef<[u8]>,
859    {
860        self.user_key.set(other.user_key);
861        self.epoch_with_gap = other.epoch_with_gap;
862    }
863}
864
865impl<T: AsRef<[u8]> + Ord + Eq> Ord for FullKey<T> {
866    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
867        // When `user_key` is the same, greater epoch comes first.
868        self.user_key
869            .cmp(&other.user_key)
870            .then_with(|| other.epoch_with_gap.cmp(&self.epoch_with_gap))
871    }
872}
873
874impl<T: AsRef<[u8]> + Ord + Eq> PartialOrd for FullKey<T> {
875    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
876        Some(self.cmp(other))
877    }
878}
879
880pub mod range_delete_backward_compatibility_serde_struct {
881    use bytes::{Buf, BufMut};
882    use risingwave_common::catalog::TableId;
883    use serde::{Deserialize, Serialize};
884
885    #[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize)]
886    pub struct TableKey(Vec<u8>);
887
888    #[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize)]
889    pub struct UserKey {
890        // When comparing `UserKey`, we first compare `table_id`, then `table_key`. So the order of
891        // declaration matters.
892        pub table_id: TableId,
893        pub table_key: TableKey,
894    }
895
896    impl UserKey {
897        pub fn decode_length_prefixed(buf: &mut &[u8]) -> Self {
898            let table_id = buf.get_u32();
899            let len = buf.get_u32() as usize;
900            let data = buf[..len].to_vec();
901            buf.advance(len);
902            UserKey {
903                table_id: TableId::new(table_id),
904                table_key: TableKey(data),
905            }
906        }
907
908        pub fn encode_length_prefixed(&self, mut buf: impl BufMut) {
909            buf.put_u32(self.table_id.table_id());
910            buf.put_u32(self.table_key.0.as_slice().len() as u32);
911            buf.put_slice(self.table_key.0.as_slice());
912        }
913    }
914
915    #[derive(Clone, Debug, PartialEq, Eq, Deserialize, Serialize)]
916    pub struct PointRange {
917        // When comparing `PointRange`, we first compare `left_user_key`, then
918        // `is_exclude_left_key`. Therefore the order of declaration matters.
919        pub left_user_key: UserKey,
920        /// `PointRange` represents the left user key itself if `is_exclude_left_key==false`
921        /// while represents the right δ Neighborhood of the left user key if
922        /// `is_exclude_left_key==true`.
923        pub is_exclude_left_key: bool,
924    }
925}
926
927pub trait EmptySliceRef {
928    fn empty_slice_ref<'a>() -> &'a Self;
929}
930
931static EMPTY_BYTES: Bytes = Bytes::new();
932impl EmptySliceRef for Bytes {
933    fn empty_slice_ref<'a>() -> &'a Self {
934        &EMPTY_BYTES
935    }
936}
937
938static EMPTY_VEC: Vec<u8> = Vec::new();
939impl EmptySliceRef for Vec<u8> {
940    fn empty_slice_ref<'a>() -> &'a Self {
941        &EMPTY_VEC
942    }
943}
944
945const EMPTY_SLICE: &[u8] = b"";
946impl EmptySliceRef for &[u8] {
947    fn empty_slice_ref<'b>() -> &'b Self {
948        &EMPTY_SLICE
949    }
950}
951
952/// Bound table key range with table id to generate a new user key range.
953pub fn bound_table_key_range<T: AsRef<[u8]> + EmptySliceRef>(
954    table_id: TableId,
955    table_key_range: &impl RangeBounds<TableKey<T>>,
956) -> (Bound<UserKey<&T>>, Bound<UserKey<&T>>) {
957    let start = match table_key_range.start_bound() {
958        Included(b) => Included(UserKey::new(table_id, TableKey(&b.0))),
959        Excluded(b) => Excluded(UserKey::new(table_id, TableKey(&b.0))),
960        Unbounded => Included(UserKey::new(table_id, TableKey(T::empty_slice_ref()))),
961    };
962
963    let end = match table_key_range.end_bound() {
964        Included(b) => Included(UserKey::new(table_id, TableKey(&b.0))),
965        Excluded(b) => Excluded(UserKey::new(table_id, TableKey(&b.0))),
966        Unbounded => {
967            if let Some(next_table_id) = table_id.table_id().checked_add(1) {
968                Excluded(UserKey::new(
969                    next_table_id.into(),
970                    TableKey(T::empty_slice_ref()),
971                ))
972            } else {
973                Unbounded
974            }
975        }
976    };
977
978    (start, end)
979}
980
981/// TODO: Temporary bypass full key check. Remove this field after #15099 is resolved.
982pub struct FullKeyTracker<T: AsRef<[u8]> + Ord + Eq, const SKIP_DEDUP: bool = false> {
983    pub latest_full_key: FullKey<T>,
984    last_observed_epoch_with_gap: EpochWithGap,
985}
986
987impl<T: AsRef<[u8]> + Ord + Eq, const SKIP_DEDUP: bool> FullKeyTracker<T, SKIP_DEDUP> {
988    pub fn new(init_full_key: FullKey<T>) -> Self {
989        let epoch_with_gap = init_full_key.epoch_with_gap;
990        Self {
991            latest_full_key: init_full_key,
992            last_observed_epoch_with_gap: epoch_with_gap,
993        }
994    }
995
996    /// Check and observe a new full key during iteration
997    ///
998    /// # Examples:
999    /// ```
1000    /// use bytes::Bytes;
1001    /// use risingwave_common::catalog::TableId;
1002    /// use risingwave_common::util::epoch::EPOCH_AVAILABLE_BITS;
1003    /// use risingwave_hummock_sdk::EpochWithGap;
1004    /// use risingwave_hummock_sdk::key::{FullKey, FullKeyTracker, TableKey};
1005    ///
1006    /// let table_id = TableId { table_id: 1 };
1007    /// let full_key1 = FullKey::new(table_id, TableKey(Bytes::from("c")), 5 << EPOCH_AVAILABLE_BITS);
1008    /// let mut a: FullKeyTracker<_> = FullKeyTracker::<Bytes>::new(full_key1.clone());
1009    ///
1010    /// // Panic on non-decreasing epoch observed for the same user key.
1011    /// // let full_key_with_larger_epoch = FullKey::new(table_id, TableKey(Bytes::from("c")), 6 << EPOCH_AVAILABLE_BITS);
1012    /// // a.observe(full_key_with_larger_epoch);
1013    ///
1014    /// // Panic on non-increasing user key observed.
1015    /// // let full_key_with_smaller_user_key = FullKey::new(table_id, TableKey(Bytes::from("b")), 3 << EPOCH_AVAILABLE_BITS);
1016    /// // a.observe(full_key_with_smaller_user_key);
1017    ///
1018    /// let full_key2 = FullKey::new(table_id, TableKey(Bytes::from("c")), 3 << EPOCH_AVAILABLE_BITS);
1019    /// assert_eq!(a.observe(full_key2.clone()), false);
1020    /// assert_eq!(a.latest_user_key(), &full_key2.user_key);
1021    ///
1022    /// let full_key3 = FullKey::new(table_id, TableKey(Bytes::from("f")), 4 << EPOCH_AVAILABLE_BITS);
1023    /// assert_eq!(a.observe(full_key3.clone()), true);
1024    /// assert_eq!(a.latest_user_key(), &full_key3.user_key);
1025    /// ```
1026    ///
1027    /// Return:
1028    /// - If the provided `key` contains a new user key, return true.
1029    /// - Otherwise: return false
1030    pub fn observe<F>(&mut self, key: FullKey<F>) -> bool
1031    where
1032        T: SetSlice<F>,
1033        F: AsRef<[u8]>,
1034    {
1035        self.observe_multi_version(key.user_key, once(key.epoch_with_gap))
1036    }
1037
1038    /// `epochs` comes from greater to smaller
1039    pub fn observe_multi_version<F>(
1040        &mut self,
1041        user_key: UserKey<F>,
1042        mut epochs: impl Iterator<Item = EpochWithGap>,
1043    ) -> bool
1044    where
1045        T: SetSlice<F>,
1046        F: AsRef<[u8]>,
1047    {
1048        let max_epoch_with_gap = epochs.next().expect("non-empty");
1049        let min_epoch_with_gap = epochs.fold(
1050            max_epoch_with_gap,
1051            |prev_epoch_with_gap, curr_epoch_with_gap| {
1052                assert!(
1053                    prev_epoch_with_gap > curr_epoch_with_gap,
1054                    "epoch list not sorted. prev: {:?}, curr: {:?}, user_key: {:?}",
1055                    prev_epoch_with_gap,
1056                    curr_epoch_with_gap,
1057                    user_key
1058                );
1059                curr_epoch_with_gap
1060            },
1061        );
1062        match self
1063            .latest_full_key
1064            .user_key
1065            .as_ref()
1066            .cmp(&user_key.as_ref())
1067        {
1068            Ordering::Less => {
1069                // Observe a new user key
1070
1071                // Reset epochs
1072                self.last_observed_epoch_with_gap = min_epoch_with_gap;
1073
1074                // Take the previous key and set latest key
1075                self.latest_full_key.set(FullKey {
1076                    user_key,
1077                    epoch_with_gap: min_epoch_with_gap,
1078                });
1079                true
1080            }
1081            Ordering::Equal => {
1082                if max_epoch_with_gap > self.last_observed_epoch_with_gap
1083                    || (!SKIP_DEDUP && max_epoch_with_gap == self.last_observed_epoch_with_gap)
1084                {
1085                    // Epoch from the same user key should be monotonically decreasing
1086                    panic!(
1087                        "key {:?} epoch {:?} >= prev epoch {:?}",
1088                        user_key, max_epoch_with_gap, self.last_observed_epoch_with_gap
1089                    );
1090                }
1091                self.last_observed_epoch_with_gap = min_epoch_with_gap;
1092                false
1093            }
1094            Ordering::Greater => {
1095                // User key should be monotonically increasing
1096                panic!(
1097                    "key {:?} <= prev key {:?}",
1098                    user_key,
1099                    FullKey {
1100                        user_key: self.latest_full_key.user_key.as_ref(),
1101                        epoch_with_gap: self.last_observed_epoch_with_gap
1102                    }
1103                );
1104            }
1105        }
1106    }
1107
1108    pub fn latest_user_key(&self) -> &UserKey<T> {
1109        &self.latest_full_key.user_key
1110    }
1111}
1112
1113#[cfg(test)]
1114mod tests {
1115    use risingwave_common::util::epoch::test_epoch;
1116
1117    use super::*;
1118
1119    #[test]
1120    fn test_encode_decode() {
1121        let epoch = test_epoch(1);
1122        let table_key = b"abc".to_vec();
1123        let key = FullKey::for_test(TableId::new(0), &table_key[..], 0);
1124        let buf = key.encode();
1125        assert_eq!(FullKey::decode(&buf), key);
1126        let key = FullKey::for_test(TableId::new(1), &table_key[..], epoch);
1127        let buf = key.encode();
1128        assert_eq!(FullKey::decode(&buf), key);
1129        let mut table_key = vec![1];
1130        let a = FullKey::for_test(TableId::new(1), table_key.clone(), epoch);
1131        table_key[0] = 2;
1132        let b = FullKey::for_test(TableId::new(1), table_key.clone(), epoch);
1133        table_key[0] = 129;
1134        let c = FullKey::for_test(TableId::new(1), table_key, epoch);
1135        assert!(a.lt(&b));
1136        assert!(b.lt(&c));
1137    }
1138
1139    #[test]
1140    fn test_key_cmp() {
1141        let epoch = test_epoch(1);
1142        let epoch2 = test_epoch(2);
1143        // 1 compared with 256 under little-endian encoding would return wrong result.
1144        let key1 = FullKey::for_test(TableId::new(0), b"0".to_vec(), epoch);
1145        let key2 = FullKey::for_test(TableId::new(1), b"0".to_vec(), epoch);
1146        let key3 = FullKey::for_test(TableId::new(1), b"1".to_vec(), epoch2);
1147        let key4 = FullKey::for_test(TableId::new(1), b"1".to_vec(), epoch);
1148
1149        assert_eq!(key1.cmp(&key1), Ordering::Equal);
1150        assert_eq!(key1.cmp(&key2), Ordering::Less);
1151        assert_eq!(key2.cmp(&key3), Ordering::Less);
1152        assert_eq!(key3.cmp(&key4), Ordering::Less);
1153    }
1154
1155    #[test]
1156    fn test_prev_key() {
1157        assert_eq!(prev_key(b"123"), b"122");
1158        assert_eq!(prev_key(b"12\x00"), b"11\xff");
1159        assert_eq!(prev_key(b"\x00\x00"), b"\xff\xff");
1160        assert_eq!(prev_key(b"\x00\x01"), b"\x00\x00");
1161        assert_eq!(prev_key(b"T"), b"S");
1162        assert_eq!(prev_key(b""), b"");
1163    }
1164
1165    #[test]
1166    fn test_bound_table_key_range() {
1167        assert_eq!(
1168            bound_table_key_range(
1169                TableId::default(),
1170                &(
1171                    Included(TableKey(b"a".to_vec())),
1172                    Included(TableKey(b"b".to_vec()))
1173                )
1174            ),
1175            (
1176                Included(UserKey::for_test(TableId::default(), &b"a".to_vec())),
1177                Included(UserKey::for_test(TableId::default(), &b"b".to_vec()),)
1178            )
1179        );
1180        assert_eq!(
1181            bound_table_key_range(
1182                TableId::from(1),
1183                &(Included(TableKey(b"a".to_vec())), Unbounded)
1184            ),
1185            (
1186                Included(UserKey::for_test(TableId::from(1), &b"a".to_vec())),
1187                Excluded(UserKey::for_test(TableId::from(2), &b"".to_vec()),)
1188            )
1189        );
1190        assert_eq!(
1191            bound_table_key_range(
1192                TableId::from(u32::MAX),
1193                &(Included(TableKey(b"a".to_vec())), Unbounded)
1194            ),
1195            (
1196                Included(UserKey::for_test(TableId::from(u32::MAX), &b"a".to_vec())),
1197                Unbounded,
1198            )
1199        );
1200    }
1201
1202    #[test]
1203    fn test_next_full_key() {
1204        let user_key = b"aaa".to_vec();
1205        let epoch: HummockEpoch = 3;
1206        let mut full_key = key_with_epoch(user_key, epoch);
1207        full_key = next_full_key(full_key.as_slice());
1208        assert_eq!(full_key, key_with_epoch(b"aaa".to_vec(), 2));
1209        full_key = next_full_key(full_key.as_slice());
1210        assert_eq!(full_key, key_with_epoch(b"aaa".to_vec(), 1));
1211        full_key = next_full_key(full_key.as_slice());
1212        assert_eq!(full_key, key_with_epoch(b"aaa".to_vec(), 0));
1213        full_key = next_full_key(full_key.as_slice());
1214        assert_eq!(
1215            full_key,
1216            key_with_epoch("aab".as_bytes().to_vec(), HummockEpoch::MAX)
1217        );
1218        assert_eq!(
1219            next_full_key(&key_with_epoch(b"\xff".to_vec(), 0)),
1220            Vec::<u8>::new()
1221        );
1222    }
1223
1224    #[test]
1225    fn test_prev_full_key() {
1226        let user_key = b"aab";
1227        let epoch: HummockEpoch = HummockEpoch::MAX - 3;
1228        let mut full_key = key_with_epoch(user_key.to_vec(), epoch);
1229        full_key = prev_full_key(full_key.as_slice());
1230        assert_eq!(
1231            full_key,
1232            key_with_epoch(b"aab".to_vec(), HummockEpoch::MAX - 2)
1233        );
1234        full_key = prev_full_key(full_key.as_slice());
1235        assert_eq!(
1236            full_key,
1237            key_with_epoch(b"aab".to_vec(), HummockEpoch::MAX - 1)
1238        );
1239        full_key = prev_full_key(full_key.as_slice());
1240        assert_eq!(full_key, key_with_epoch(b"aab".to_vec(), HummockEpoch::MAX));
1241        full_key = prev_full_key(full_key.as_slice());
1242        assert_eq!(full_key, key_with_epoch(b"aaa".to_vec(), 0));
1243
1244        assert_eq!(
1245            prev_full_key(&key_with_epoch(b"\x00".to_vec(), HummockEpoch::MAX)),
1246            Vec::<u8>::new()
1247        );
1248    }
1249
1250    #[test]
1251    fn test_uesr_key_order() {
1252        let a = UserKey::new(TableId::new(1), TableKey(b"aaa".to_vec()));
1253        let b = UserKey::new(TableId::new(2), TableKey(b"aaa".to_vec()));
1254        let c = UserKey::new(TableId::new(2), TableKey(b"bbb".to_vec()));
1255        assert!(a.lt(&b));
1256        assert!(b.lt(&c));
1257        let a = a.encode();
1258        let b = b.encode();
1259        let c = c.encode();
1260        assert!(a.lt(&b));
1261        assert!(b.lt(&c));
1262    }
1263
1264    #[test]
1265    fn test_prefixed_range_with_vnode() {
1266        let concat = |vnode: usize, b: &[u8]| -> Bytes {
1267            prefix_slice_with_vnode(VirtualNode::from_index(vnode), b)
1268        };
1269        assert_eq!(
1270            prefixed_range_with_vnode(
1271                (Included(Bytes::from("1")), Included(Bytes::from("2"))),
1272                VirtualNode::from_index(233),
1273            ),
1274            (
1275                Included(TableKey(concat(233, b"1"))),
1276                Included(TableKey(concat(233, b"2")))
1277            )
1278        );
1279        assert_eq!(
1280            prefixed_range_with_vnode(
1281                (Excluded(Bytes::from("1")), Excluded(Bytes::from("2"))),
1282                VirtualNode::from_index(233),
1283            ),
1284            (
1285                Excluded(TableKey(concat(233, b"1"))),
1286                Excluded(TableKey(concat(233, b"2")))
1287            )
1288        );
1289        assert_eq!(
1290            prefixed_range_with_vnode(
1291                (Bound::<Bytes>::Unbounded, Bound::<Bytes>::Unbounded),
1292                VirtualNode::from_index(233),
1293            ),
1294            (
1295                Included(TableKey(concat(233, b""))),
1296                Excluded(TableKey(concat(234, b"")))
1297            )
1298        );
1299        let max_vnode = VirtualNode::MAX_REPRESENTABLE.to_index();
1300        assert_eq!(
1301            prefixed_range_with_vnode(
1302                (Bound::<Bytes>::Unbounded, Bound::<Bytes>::Unbounded),
1303                VirtualNode::from_index(max_vnode),
1304            ),
1305            (Included(TableKey(concat(max_vnode, b""))), Unbounded)
1306        );
1307        let second_max_vnode = max_vnode - 1;
1308        assert_eq!(
1309            prefixed_range_with_vnode(
1310                (Bound::<Bytes>::Unbounded, Bound::<Bytes>::Unbounded),
1311                VirtualNode::from_index(second_max_vnode),
1312            ),
1313            (
1314                Included(TableKey(concat(second_max_vnode, b""))),
1315                Excluded(TableKey(concat(max_vnode, b"")))
1316            )
1317        );
1318    }
1319
1320    #[test]
1321    fn test_single_vnode_range() {
1322        let left_bound = vec![
1323            Included(b"0".as_slice()),
1324            Excluded(b"0".as_slice()),
1325            Unbounded,
1326        ];
1327        let right_bound = vec![
1328            Included(b"1".as_slice()),
1329            Excluded(b"1".as_slice()),
1330            Unbounded,
1331        ];
1332        for vnode in 0..VirtualNode::MAX_COUNT {
1333            for left in &left_bound {
1334                for right in &right_bound {
1335                    assert_eq!(
1336                        (vnode, vnode + 1),
1337                        vnode_range(&prefixed_range_with_vnode::<&[u8]>(
1338                            (*left, *right),
1339                            VirtualNode::from_index(vnode)
1340                        ))
1341                    )
1342                }
1343            }
1344        }
1345    }
1346}
risingwave_hummock_sdk/key.rs

risingwave_hummock_sdk/
key.rs
