risingwave_stream/cache/
mod.rs

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
// Copyright 2024 RisingWave Labs
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

mod managed_lru;
pub use managed_lru::*;
use risingwave_common::bitmap::Bitmap;
use risingwave_common::util::iter_util::ZipEqFast;

/// Returns whether we're unsure about the fressness of the cache after the scaling from the
/// previous partition to the current one, denoted by vnode bitmaps. If the value is `true`, we must
/// evict the cache entries that does not belong to the previous partition before further
/// processing.
///
/// TODO: currently most executors simply clear all the cache entries if `true`, we can make the
/// cache aware of the consistent hashing to avoid unnecessary eviction in the future.
/// Check this [issue](https://github.com/risingwavelabs/risingwave/issues/5567).
///
/// TODO: may encapsulate the logic into `ExecutorCache` when ready.
///
/// # Explanation
/// We use a lazy manner to manipulate the cache. When scaling out, the partition of the existing
/// executors will likely shrink and becomes a subset of the previous one (to ensure the best
/// locality). In this case, this function will return `false` and the cache entries that're not in
/// the current partition anymore are still kept. This achieves the best performance as we won't
/// touch and valiate the cache at all when scaling-out, which is the common case and the critical
/// path.
///
/// This brings a problem when scaling in after a while. Some partitions may be reassigned back to
/// the current executor, while the cache entries of these partitions are still unevicted. So it's
/// possible that these entries have been updated by other executors on other workers, and
/// the content is now stale! The executor must evict these entries which are not in the
/// **previous** partition before further processing.
pub(super) fn cache_may_stale(
    previous_vnode_bitmap: &Bitmap,
    current_vnode_bitmap: &Bitmap,
) -> bool {
    let current_is_subset = previous_vnode_bitmap
        .iter()
        .zip_eq_fast(current_vnode_bitmap.iter())
        .all(|(p, c)| p >= c);

    !current_is_subset
}
#[cfg(test)]
mod tests {
    use super::*;

    #[expect(clippy::bool_assert_comparison)]
    #[test]
    fn test_cache_may_stale() {
        let p123 = Bitmap::from_bytes(&[0b_0000_0111]);
        let p1234 = Bitmap::from_bytes(&[0b_0000_1111]);
        let p1245 = Bitmap::from_bytes(&[0b_0001_1011]);

        assert_eq!(cache_may_stale(&p123, &p123), false); // unchanged
        assert_eq!(cache_may_stale(&p1234, &p123), false); // scale-out
        assert_eq!(cache_may_stale(&p123, &p1234), true); // scale-in
        assert_eq!(cache_may_stale(&p123, &p1245), true); // scale-in
    }
}