risingwave_storage/hummock/
mod.rs

1// Copyright 2022 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
15//! Hummock is the state store of the streaming system.
16
17use std::ops::Bound;
18use std::sync::Arc;
19
20use bytes::Bytes;
21use risingwave_hummock_sdk::key::{FullKey, TableKey, UserKeyRangeRef};
22use risingwave_hummock_sdk::sstable_info::SstableInfo;
23use risingwave_hummock_sdk::{HummockEpoch, *};
24
25pub mod block_cache;
26pub use block_cache::*;
27
28pub mod sstable;
29pub use sstable::*;
30
31pub mod compactor;
32mod error;
33pub mod hummock_meta_client;
34pub mod iterator;
35pub mod shared_buffer;
36pub mod sstable_store;
37#[cfg(any(test, feature = "test"))]
38pub mod test_utils;
39pub mod utils;
40pub use utils::MemoryLimiter;
41pub mod backup_reader;
42pub mod event_handler;
43pub mod local_version;
44pub mod observer_manager;
45pub mod store;
46pub use store::*;
47mod validator;
48pub mod value;
49pub mod write_limiter;
50
51pub mod recent_filter;
52pub use recent_filter::*;
53
54pub mod block_stream;
55mod time_travel_version_cache;
56
57pub(crate) mod vector;
58
59mod object_id_manager;
60pub use error::*;
61pub use object_id_manager::*;
62pub use risingwave_common::cache::{CacheableEntry, LookupResult, LruCache};
63pub use validator::*;
64use value::*;
65
66use self::iterator::HummockIterator;
67pub use self::sstable_store::*;
68use crate::mem_table::ImmutableMemtable;
69use crate::monitor::StoreLocalStatistic;
70use crate::store::ReadOptions;
71
72pub(in crate::hummock) struct IteratorStatsGuard<'a, TI: HummockIterator> {
73    iter: Option<TI>,
74    parent_stats: &'a mut StoreLocalStatistic,
75}
76
77impl<'a, TI: HummockIterator> IteratorStatsGuard<'a, TI> {
78    pub(in crate::hummock) fn new(iter: TI, parent_stats: &'a mut StoreLocalStatistic) -> Self {
79        Self {
80            iter: Some(iter),
81            parent_stats,
82        }
83    }
84
85    pub(in crate::hummock) fn iter(&self) -> &TI {
86        self.iter.as_ref().expect("iterator must be present")
87    }
88
89    pub(in crate::hummock) fn iter_mut(&mut self) -> &mut TI {
90        self.iter.as_mut().expect("iterator must be present")
91    }
92
93    fn collect(&mut self) {
94        if let Some(iter) = &self.iter {
95            iter.collect_local_statistic(self.parent_stats);
96        }
97    }
98
99    pub(in crate::hummock) fn handoff(mut self) -> TI {
100        self.iter.take().expect("iterator must be present")
101    }
102
103    pub(in crate::hummock) fn collect_and_take(mut self) -> TI {
104        self.collect();
105        self.iter.take().expect("iterator must be present")
106    }
107}
108
109impl<TI: HummockIterator> Drop for IteratorStatsGuard<'_, TI> {
110    fn drop(&mut self) {
111        self.collect();
112    }
113}
114
115pub async fn get_from_sstable_info(
116    sstable_store_ref: SstableStoreRef,
117    sstable_info: &SstableInfo,
118    full_key: FullKey<&[u8]>,
119    read_options: &ReadOptions,
120    dist_key_hash: Option<u64>,
121    local_stats: &mut StoreLocalStatistic,
122) -> HummockResult<Option<impl HummockIterator>> {
123    let sstable = sstable_store_ref.sstable(sstable_info, local_stats).await?;
124
125    // SST filter key is the distribution key, which does not need to be the prefix of pk, and does not
126    // contain `TablePrefix` and `VnodePrefix`.
127    if let Some(hash) = dist_key_hash
128        && !hit_sstable_filter(
129            &sstable,
130            &(
131                Bound::Included(full_key.user_key),
132                Bound::Included(full_key.user_key),
133            ),
134            hash,
135            local_stats,
136        )
137    {
138        return Ok(None);
139    }
140
141    let mut iterator_read_options = SstableIteratorReadOptions::from_read_options(read_options);
142    iterator_read_options.read_table_id = Some(full_key.user_key.table_id);
143
144    let mut iter = IteratorStatsGuard::new(
145        SstableIterator::create(
146            sstable,
147            sstable_store_ref.clone(),
148            Arc::new(iterator_read_options),
149            sstable_info,
150        ),
151        local_stats,
152    );
153    iter.iter_mut().seek(full_key).await?;
154    // Iterator has sought passed the borders.
155    if !iter.iter().is_valid() {
156        return Ok(None);
157    }
158
159    // Iterator gets us the key, we tell if it's the key we want
160    // or key next to it.
161    let value = if iter.iter().key().user_key == full_key.user_key {
162        Some(iter.collect_and_take())
163    } else {
164        None
165    };
166
167    Ok(value)
168}
169
170pub fn hit_sstable_filter(
171    sstable_ref: &Sstable,
172    user_key_range: &UserKeyRangeRef<'_>,
173    prefix_hash: u64,
174    local_stats: &mut StoreLocalStatistic,
175) -> bool {
176    local_stats.bloom_filter_check_counts += 1;
177    let may_exist = sstable_ref.may_match_hash(user_key_range, prefix_hash);
178    if !may_exist {
179        local_stats.bloom_filter_true_negative_counts += 1;
180    }
181    may_exist
182}
183
184/// Get `user_value` from `ImmutableMemtable`
185pub fn get_from_batch<'a>(
186    imm: &'a ImmutableMemtable,
187    table_key: TableKey<&[u8]>,
188    read_epoch: HummockEpoch,
189    read_options: &ReadOptions,
190    local_stats: &mut StoreLocalStatistic,
191) -> Option<(HummockValue<&'a Bytes>, EpochWithGap)> {
192    imm.get(table_key, read_epoch, read_options).inspect(|_| {
193        local_stats.get_shared_buffer_hit_counts += 1;
194    })
195}
196
197#[cfg(test)]
198mod tests {
199    use bytes::Bytes;
200    use risingwave_common::catalog::TableId;
201    use risingwave_common::hash::VirtualNode;
202    use risingwave_common::util::epoch::test_epoch;
203    use risingwave_hummock_sdk::EpochWithGap;
204    use risingwave_hummock_sdk::key::{FullKey, TableKey, UserKey};
205
206    use super::{CachePolicy, get_from_sstable_info};
207    use crate::hummock::iterator::test_utils::{iterator_test_key_of, mock_sstable_store};
208    use crate::hummock::test_utils::{
209        default_builder_opt_for_test, gen_test_sstable_info, gen_test_sstable_with_table_ids,
210        test_value_of,
211    };
212    use crate::hummock::value::HummockValue;
213    use crate::monitor::StoreLocalStatistic;
214    use crate::store::ReadOptions;
215
216    #[tokio::test]
217    async fn test_get_collects_stats_when_seek_passes_sst_end() {
218        let sstable_store = mock_sstable_store().await;
219        let sstable_info = gen_test_sstable_info(
220            default_builder_opt_for_test(),
221            1,
222            (0..10).map(|idx| {
223                (
224                    iterator_test_key_of(idx),
225                    HummockValue::put(format!("value_{idx}").into_bytes()),
226                )
227            }),
228            sstable_store.clone(),
229        )
230        .await;
231        let mut stats = StoreLocalStatistic::default();
232        let key = iterator_test_key_of(10);
233        let read_options = ReadOptions::default();
234
235        let result = get_from_sstable_info(
236            sstable_store,
237            &sstable_info,
238            key.to_ref(),
239            &read_options,
240            None,
241            &mut stats,
242        )
243        .await
244        .unwrap();
245
246        assert!(result.is_none());
247        drop(result);
248        assert_eq!(stats.cache_data_block_total, 1);
249    }
250
251    #[tokio::test]
252    async fn test_point_get_reads_only_requested_table_blocks() {
253        let sstable_store = mock_sstable_store().await;
254        let mut builder_options = default_builder_opt_for_test();
255        builder_options.block_capacity = 128;
256
257        let test_user_key = |table_id, key: &str| {
258            UserKey::new(
259                TableId::new(table_id),
260                TableKey(Bytes::from(
261                    [VirtualNode::ZERO.to_be_bytes().as_slice(), key.as_bytes()].concat(),
262                )),
263            )
264        };
265        let test_key = |table_id, key: &str| FullKey {
266            user_key: test_user_key(table_id, key),
267            epoch_with_gap: EpochWithGap::new_from_epoch(test_epoch(1)),
268        };
269        let kv_pairs = (1..=2).flat_map(|table_id| {
270            (0..8).map(move |idx| {
271                (
272                    test_key(table_id, &format!("key_{idx:05}")),
273                    HummockValue::put(Bytes::from(test_value_of(idx))),
274                )
275            })
276        });
277        let (sstable, sstable_info) = gen_test_sstable_with_table_ids(
278            builder_options,
279            10,
280            kv_pairs,
281            sstable_store.clone(),
282            vec![1, 2],
283        )
284        .await;
285        let table_2_block_start = sstable
286            .meta
287            .block_metas
288            .partition_point(|block_meta| block_meta.table_id() < TableId::new(2));
289        assert!(table_2_block_start > 0);
290
291        // The queried table-2 key sorts before the first table-2 block. Without a point-get table
292        // id filter, the SST iterator seeks to the previous table-1 block first.
293        let full_key = test_key(2, "key");
294        let mut local_stats = StoreLocalStatistic::default();
295        let read_options = ReadOptions {
296            cache_policy: CachePolicy::Disable,
297            ..Default::default()
298        };
299
300        {
301            let result = get_from_sstable_info(
302                sstable_store,
303                &sstable_info,
304                full_key.to_ref(),
305                &read_options,
306                None,
307                &mut local_stats,
308            )
309            .await
310            .unwrap();
311            assert!(result.is_none());
312        }
313        assert_eq!(local_stats.cache_data_block_total, 1);
314    }
315}
risingwave_storage/hummock/mod.rs

risingwave_storage/hummock/
mod.rs
