mito2/cache/
index.rs

1// Copyright 2023 Greptime Team
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
15pub mod bloom_filter_index;
16pub mod inverted_index;
17pub mod result_cache;
18
19use std::future::Future;
20use std::hash::Hash;
21use std::ops::Range;
22use std::sync::Arc;
23
24use bytes::Bytes;
25use object_store::Buffer;
26
27use crate::metrics::{CACHE_BYTES, CACHE_HIT, CACHE_MISS};
28
29/// Metrics for index metadata.
30const INDEX_METADATA_TYPE: &str = "index_metadata";
31/// Metrics for index content.
32const INDEX_CONTENT_TYPE: &str = "index_content";
33
34/// Metrics collected from IndexCache operations.
35#[derive(Debug, Default, Clone)]
36pub struct IndexCacheMetrics {
37    /// Number of cache hits.
38    pub cache_hit: usize,
39    /// Number of cache misses.
40    pub cache_miss: usize,
41    /// Number of pages accessed.
42    pub num_pages: usize,
43    /// Total bytes from pages.
44    pub page_bytes: u64,
45}
46
47impl IndexCacheMetrics {
48    /// Merges another set of metrics into this one.
49    pub fn merge(&mut self, other: &Self) {
50        self.cache_hit += other.cache_hit;
51        self.cache_miss += other.cache_miss;
52        self.num_pages += other.num_pages;
53        self.page_bytes += other.page_bytes;
54    }
55}
56
57#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
58pub struct PageKey {
59    page_id: u64,
60}
61
62impl PageKey {
63    /// Converts an offset to a page ID based on the page size.
64    fn calculate_page_id(offset: u64, page_size: u64) -> u64 {
65        offset / page_size
66    }
67
68    /// Calculates the total number of pages that a given size spans, starting from a specific offset.
69    fn calculate_page_count(offset: u64, size: u32, page_size: u64) -> u32 {
70        let start_page = Self::calculate_page_id(offset, page_size);
71        let end_page = Self::calculate_page_id(offset + (size as u64) - 1, page_size);
72        (end_page + 1 - start_page) as u32
73    }
74
75    /// Calculates the byte range for data retrieval based on the specified offset and size.
76    ///
77    /// This function determines the starting and ending byte positions required for reading data.
78    /// For example, with an offset of 5000 and a size of 5000, using a PAGE_SIZE of 4096,
79    /// the resulting byte range will be 904..5904. This indicates that:
80    /// - The reader will first access fixed-size pages [4096, 8192) and [8192, 12288).
81    /// - To read the range [5000..10000), it only needs to fetch bytes within the range [904, 5904) across two pages.
82    fn calculate_range(offset: u64, size: u32, page_size: u64) -> Range<usize> {
83        let start = (offset % page_size) as usize;
84        let end = start + size as usize;
85        start..end
86    }
87
88    /// Generates a iterator of `IndexKey` for the pages that a given offset and size span.
89    fn generate_page_keys(offset: u64, size: u32, page_size: u64) -> impl Iterator<Item = Self> {
90        let start_page = Self::calculate_page_id(offset, page_size);
91        let total_pages = Self::calculate_page_count(offset, size, page_size);
92        (0..total_pages).map(move |i| Self {
93            page_id: start_page + i as u64,
94        })
95    }
96}
97
98/// Cache for index metadata and content.
99pub struct IndexCache<K, M> {
100    /// Cache for index metadata
101    index_metadata: moka::sync::Cache<K, Arc<M>>,
102    /// Cache for index content.
103    index: moka::sync::Cache<(K, PageKey), Bytes>,
104    // Page size for index content.
105    page_size: u64,
106
107    /// Weighter for metadata.
108    weight_of_metadata: fn(&K, &Arc<M>) -> u32,
109    /// Weighter for content.
110    weight_of_content: fn(&(K, PageKey), &Bytes) -> u32,
111}
112
113impl<K, M> IndexCache<K, M>
114where
115    K: Hash + Eq + Send + Sync + 'static,
116    M: Send + Sync + 'static,
117{
118    pub fn new_with_weighter(
119        index_metadata_cap: u64,
120        index_content_cap: u64,
121        page_size: u64,
122        index_type: &'static str,
123        weight_of_metadata: fn(&K, &Arc<M>) -> u32,
124        weight_of_content: fn(&(K, PageKey), &Bytes) -> u32,
125    ) -> Self {
126        common_telemetry::debug!(
127            "Building IndexCache with metadata size: {index_metadata_cap}, content size: {index_content_cap}, page size: {page_size}, index type: {index_type}"
128        );
129        let index_metadata = moka::sync::CacheBuilder::new(index_metadata_cap)
130            .name(&format!("index_metadata_{}", index_type))
131            .weigher(weight_of_metadata)
132            .eviction_listener(move |k, v, _cause| {
133                let size = weight_of_metadata(&k, &v);
134                CACHE_BYTES
135                    .with_label_values(&[INDEX_METADATA_TYPE])
136                    .sub(size.into());
137            })
138            .support_invalidation_closures()
139            .build();
140        let index_cache = moka::sync::CacheBuilder::new(index_content_cap)
141            .name(&format!("index_content_{}", index_type))
142            .weigher(weight_of_content)
143            .eviction_listener(move |k, v, _cause| {
144                let size = weight_of_content(&k, &v);
145                CACHE_BYTES
146                    .with_label_values(&[INDEX_CONTENT_TYPE])
147                    .sub(size.into());
148            })
149            .support_invalidation_closures()
150            .build();
151        Self {
152            index_metadata,
153            index: index_cache,
154            page_size,
155            weight_of_content,
156            weight_of_metadata,
157        }
158    }
159}
160
161impl<K, M> IndexCache<K, M>
162where
163    K: Hash + Eq + Clone + Copy + Send + Sync + 'static,
164    M: Send + Sync + 'static,
165{
166    pub fn get_metadata(&self, key: K) -> Option<Arc<M>> {
167        self.index_metadata.get(&key)
168    }
169
170    pub fn put_metadata(&self, key: K, metadata: Arc<M>) {
171        CACHE_BYTES
172            .with_label_values(&[INDEX_METADATA_TYPE])
173            .add((self.weight_of_metadata)(&key, &metadata).into());
174        self.index_metadata.insert(key, metadata)
175    }
176
177    /// Gets given range of index data from cache, and loads from source if the file
178    /// is not already cached.
179    async fn get_or_load<F, Fut, E>(
180        &self,
181        key: K,
182        file_size: u64,
183        offset: u64,
184        size: u32,
185        load: F,
186    ) -> Result<(Vec<u8>, IndexCacheMetrics), E>
187    where
188        F: Fn(Vec<Range<u64>>) -> Fut,
189        Fut: Future<Output = Result<Vec<Bytes>, E>>,
190        E: std::error::Error,
191    {
192        let mut metrics = IndexCacheMetrics::default();
193        let page_keys =
194            PageKey::generate_page_keys(offset, size, self.page_size).collect::<Vec<_>>();
195        // Size is 0, return empty data.
196        if page_keys.is_empty() {
197            return Ok((Vec::new(), metrics));
198        }
199        metrics.num_pages = page_keys.len();
200        let mut data = Vec::with_capacity(page_keys.len());
201        data.resize(page_keys.len(), Bytes::new());
202        let mut cache_miss_range = vec![];
203        let mut cache_miss_idx = vec![];
204        let last_index = page_keys.len() - 1;
205        // TODO: Avoid copy as much as possible.
206        for (i, page_key) in page_keys.iter().enumerate() {
207            match self.get_page(key, *page_key) {
208                Some(page) => {
209                    CACHE_HIT.with_label_values(&[INDEX_CONTENT_TYPE]).inc();
210                    metrics.cache_hit += 1;
211                    metrics.page_bytes += page.len() as u64;
212                    data[i] = page;
213                }
214                None => {
215                    CACHE_MISS.with_label_values(&[INDEX_CONTENT_TYPE]).inc();
216                    metrics.cache_miss += 1;
217                    let base_offset = page_key.page_id * self.page_size;
218                    let pruned_size = if i == last_index {
219                        prune_size(page_keys.iter(), file_size, self.page_size)
220                    } else {
221                        self.page_size
222                    };
223                    cache_miss_range.push(base_offset..base_offset + pruned_size);
224                    cache_miss_idx.push(i);
225                }
226            }
227        }
228        if !cache_miss_range.is_empty() {
229            let pages = load(cache_miss_range).await?;
230            for (i, page) in cache_miss_idx.into_iter().zip(pages.into_iter()) {
231                let page_key = page_keys[i];
232                metrics.page_bytes += page.len() as u64;
233                data[i] = page.clone();
234                self.put_page(key, page_key, page.clone());
235            }
236        }
237        let buffer = Buffer::from_iter(data.into_iter());
238        Ok((
239            buffer
240                .slice(PageKey::calculate_range(offset, size, self.page_size))
241                .to_vec(),
242            metrics,
243        ))
244    }
245
246    fn get_page(&self, key: K, page_key: PageKey) -> Option<Bytes> {
247        self.index.get(&(key, page_key))
248    }
249
250    fn put_page(&self, key: K, page_key: PageKey, value: Bytes) {
251        // Clones the value to ensure it doesn't reference a larger buffer.
252        let value = Bytes::from(value.to_vec());
253        CACHE_BYTES
254            .with_label_values(&[INDEX_CONTENT_TYPE])
255            .add((self.weight_of_content)(&(key, page_key), &value).into());
256        self.index.insert((key, page_key), value);
257    }
258
259    /// Invalidates all cache entries whose keys satisfy `predicate`.
260    pub fn invalidate_if<F>(&self, predicate: F)
261    where
262        F: Fn(&K) -> bool + Send + Sync + 'static,
263    {
264        let predicate = Arc::new(predicate);
265        let metadata_predicate = Arc::clone(&predicate);
266
267        self.index_metadata
268            .invalidate_entries_if(move |key, _| metadata_predicate(key))
269            .expect("cache should support invalidation closures");
270
271        self.index
272            .invalidate_entries_if(move |(key, _), _| predicate(key))
273            .expect("cache should support invalidation closures");
274    }
275}
276
277/// Prunes the size of the last page based on the indexes.
278/// We have following cases:
279/// 1. The rest file size is less than the page size, read to the end of the file.
280/// 2. Otherwise, read the page size.
281fn prune_size<'a>(
282    indexes: impl Iterator<Item = &'a PageKey>,
283    file_size: u64,
284    page_size: u64,
285) -> u64 {
286    let last_page_start = indexes.last().map(|i| i.page_id * page_size).unwrap_or(0);
287    page_size.min(file_size - last_page_start)
288}
mito2/cache/index.rs

mito2/cache/
index.rs
