be/src/io/cache/inflight_write_buffer_index.h
Line | Count | Source |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | |
18 | | #pragma once |
19 | | |
20 | | #include <bvar/bvar.h> |
21 | | |
22 | | #include <atomic> |
23 | | #include <cstddef> |
24 | | #include <cstdint> |
25 | | #include <functional> |
26 | | #include <memory> |
27 | | #include <mutex> |
28 | | #include <string> |
29 | | #include <unordered_map> |
30 | | #include <vector> |
31 | | |
32 | | #include "io/cache/async_cache_write_manager.h" |
33 | | #include "io/cache/file_cache_common.h" |
34 | | |
35 | | namespace doris::io { |
36 | | |
37 | | /// Metadata for one block whose remote payload is waiting for asynchronous persistence. |
38 | | /// `buffer` is non-null. `buffer_offset` and `buffer_size` describe its valid file interval, while |
39 | | /// `buffer->size()` is the allocation capacity accounted by the index's memory gauge. |
40 | | struct InflightWriteBufferEntry { |
41 | | AsyncCacheWriteBufferPtr buffer; |
42 | | size_t buffer_offset {0}; |
43 | | size_t buffer_size {0}; |
44 | | int64_t submit_ts_us {0}; |
45 | | |
46 | | InflightWriteBufferEntry(AsyncCacheWriteBufferPtr buffer_, size_t offset_, size_t size_, |
47 | | int64_t submit_ts_us_) |
48 | 66 | : buffer(std::move(buffer_)), |
49 | 66 | buffer_offset(offset_), |
50 | 66 | buffer_size(size_), |
51 | 66 | submit_ts_us(submit_ts_us_) {} |
52 | | }; |
53 | | |
54 | | /// Sharded index from (cache key, aligned block offset) to an accepted async-write payload. |
55 | | /// Readers use it before probing disk cache so concurrent misses can reuse remote bytes in memory. |
56 | | class InflightWriteBufferIndex { |
57 | | public: |
58 | | /// A batch-lookup result retains the requested offset even when `entry` is null. |
59 | | struct LookupResult { |
60 | | size_t block_offset {0}; |
61 | | std::shared_ptr<InflightWriteBufferEntry> entry; |
62 | | }; |
63 | | |
64 | | /// @param shard_count Positive number of independently locked hash shards. |
65 | | /// @param metric_prefix Prefix that makes per-cache-disk bvar names unique. |
66 | | explicit InflightWriteBufferIndex(size_t shard_count, std::string metric_prefix = {}); |
67 | | |
68 | | /// Claim `(cache_hash, block_offset)` for `entry` if no entry already exists. File contents are |
69 | | /// immutable for a cache hash, so every entry at the same offset contains interchangeable data |
70 | | /// even when its persistence task belongs to an invalidated write epoch. |
71 | | /// @return null when insertion succeeds; otherwise the existing owner. |
72 | | std::shared_ptr<InflightWriteBufferEntry> insert_if_absent( |
73 | | const UInt128Wrapper& cache_hash, size_t block_offset, |
74 | | std::shared_ptr<InflightWriteBufferEntry> entry); |
75 | | |
76 | | /// Find the immutable payload for `(cache_hash, block_offset)` regardless of the persistence |
77 | | /// task's write epoch. |
78 | | std::shared_ptr<InflightWriteBufferEntry> lookup(const UInt128Wrapper& cache_hash, |
79 | | size_t block_offset); |
80 | | |
81 | | /// Look up aligned `block_offsets` in input order. |
82 | | std::vector<LookupResult> lookup_all(const UInt128Wrapper& cache_hash, |
83 | | const std::vector<size_t>& block_offsets); |
84 | | |
85 | | /// Remove the key only if it still points to `expected`, preventing an old task callback from |
86 | | /// deleting a replacement entry. |
87 | | /// @return true if this exact owner was removed. |
88 | | bool remove_if(const UInt128Wrapper& cache_hash, size_t block_offset, |
89 | | const std::shared_ptr<InflightWriteBufferEntry>& expected); |
90 | | |
91 | | /// Return the number of indexed block payloads. |
92 | 34 | size_t count() const { return _count.load(std::memory_order_relaxed); } |
93 | | |
94 | | /// Return the buffer-capacity bytes retained by all current index entries. |
95 | 625 | size_t buffer_bytes() const { return _buffer_bytes.load(std::memory_order_relaxed); } |
96 | | |
97 | | /// Record removal of an inserted entry because queue submission hit backpressure. |
98 | 1 | void record_backpressure_rollback() { *_rollback_on_backpressure_metric << 1; } |
99 | | |
100 | | private: |
101 | | struct Key { |
102 | | UInt128Wrapper cache_hash; |
103 | | size_t block_offset {0}; |
104 | | |
105 | 79 | bool operator==(const Key& other) const { |
106 | 79 | return cache_hash == other.cache_hash && block_offset == other.block_offset; |
107 | 79 | } |
108 | | }; |
109 | | |
110 | | struct KeyHash { |
111 | 325 | size_t operator()(const Key& key) const { |
112 | 325 | return doris::io::KeyHash()(key.cache_hash) ^ std::hash<size_t>()(key.block_offset); |
113 | 325 | } |
114 | | }; |
115 | | |
116 | | struct Shard { |
117 | | mutable std::mutex mutex; |
118 | | std::unordered_map<Key, std::shared_ptr<InflightWriteBufferEntry>, KeyHash> entries; |
119 | | }; |
120 | | |
121 | 147 | size_t _shard_index(const Key& key) const { return KeyHash()(key) % _shards.size(); } |
122 | | |
123 | | std::vector<std::unique_ptr<Shard>> _shards; |
124 | | std::atomic<size_t> _count {0}; |
125 | | std::atomic<size_t> _buffer_bytes {0}; |
126 | | |
127 | | std::shared_ptr<bvar::PassiveStatus<size_t>> _buffer_bytes_metric; |
128 | | std::shared_ptr<bvar::Adder<uint64_t>> _lookup_metric; |
129 | | std::shared_ptr<bvar::Adder<uint64_t>> _hit_metric; |
130 | | std::shared_ptr<bvar::Adder<uint64_t>> _miss_metric; |
131 | | std::shared_ptr<bvar::Adder<uint64_t>> _insert_metric; |
132 | | std::shared_ptr<bvar::Adder<uint64_t>> _insert_existing_metric; |
133 | | std::shared_ptr<bvar::Adder<uint64_t>> _remove_success_metric; |
134 | | std::shared_ptr<bvar::Adder<uint64_t>> _remove_failed_metric; |
135 | | std::shared_ptr<bvar::Adder<uint64_t>> _rollback_on_backpressure_metric; |
136 | | std::shared_ptr<bvar::LatencyRecorder> _lock_wait_latency_metric; |
137 | | std::shared_ptr<bvar::LatencyRecorder> _lock_hold_latency_metric; |
138 | | }; |
139 | | |
140 | | } // namespace doris::io |