be/src/util/s3_rate_limiter_manager.cpp
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 | | #include "util/s3_rate_limiter_manager.h" |
19 | | |
20 | | #include <algorithm> |
21 | | #include <limits> |
22 | | #include <thread> |
23 | | |
24 | | #include "common/config.h" |
25 | | #include "common/logging.h" |
26 | | #include "util/cgroup_util.h" |
27 | | |
28 | | namespace doris { |
29 | | |
30 | | bvar::Adder<int64_t> s3_get_bytes_rate_limit_sleep_ns("s3_get_bytes_rate_limit_sleep_ns"); |
31 | | bvar::Adder<int64_t> s3_get_bytes_rate_limit_sleep_count("s3_get_bytes_rate_limit_sleep_count"); |
32 | | bvar::Adder<int64_t> s3_put_bytes_rate_limit_sleep_ns("s3_put_bytes_rate_limit_sleep_ns"); |
33 | | bvar::Adder<int64_t> s3_put_bytes_rate_limit_sleep_count("s3_put_bytes_rate_limit_sleep_count"); |
34 | | |
35 | | namespace { |
36 | | |
37 | 2 | std::function<void(int64_t)> bytes_rate_limiter_metric_func(S3RateLimitType type) { |
38 | 2 | switch (type) { |
39 | 1 | case S3RateLimitType::GET: |
40 | 1 | return metric_func_factory(s3_get_bytes_rate_limit_sleep_ns, |
41 | 1 | s3_get_bytes_rate_limit_sleep_count); |
42 | 1 | case S3RateLimitType::PUT: |
43 | 1 | return metric_func_factory(s3_put_bytes_rate_limit_sleep_ns, |
44 | 1 | s3_put_bytes_rate_limit_sleep_count); |
45 | 0 | default: |
46 | 0 | return [](int64_t) {}; |
47 | 2 | } |
48 | 2 | } |
49 | | |
50 | | // min(per_core * cores, cap) with overflow protection; cap == 0 means no cap. |
51 | 9 | int64_t cap_multiply(int64_t per_core, int64_t cores, int64_t cap) { |
52 | 9 | cap = cap > 0 ? cap : std::numeric_limits<int64_t>::max(); |
53 | 9 | if (per_core > cap / cores) { |
54 | 4 | return cap; |
55 | 4 | } |
56 | 5 | return per_core * cores; |
57 | 9 | } |
58 | | |
59 | 221 | size_t index_of(S3RateLimitType type) { |
60 | 221 | CHECK(type == S3RateLimitType::GET || type == S3RateLimitType::PUT) << to_string(type); |
61 | 221 | return static_cast<size_t>(type); |
62 | 221 | } |
63 | | |
64 | | } // namespace |
65 | | |
66 | 81 | S3EffectiveRateLimit resolve_s3_rate_limit(S3RateLimitType type, int64_t cores) { |
67 | 81 | const bool is_get = type == S3RateLimitType::GET; |
68 | 81 | const int64_t qps_per_core = is_get ? config::s3_get_qps_per_core : config::s3_put_qps_per_core; |
69 | 81 | const int64_t qps_max = is_get ? config::s3_get_qps_max : config::s3_put_qps_max; |
70 | 81 | const int64_t bytes_per_core = is_get ? config::s3_get_bytes_per_second_per_core |
71 | 81 | : config::s3_put_bytes_per_second_per_core; |
72 | 81 | const int64_t bytes_max = |
73 | 81 | is_get ? config::s3_get_bytes_per_second_max : config::s3_put_bytes_per_second_max; |
74 | 81 | cores = std::max<int64_t>(1, cores); |
75 | | |
76 | 81 | S3EffectiveRateLimit limit; |
77 | 81 | if (qps_per_core < 0) { |
78 | | // Unset: the legacy absolute configs stay in charge, bit-for-bit compatible. |
79 | 74 | limit.qps = is_get ? config::s3_get_token_per_second : config::s3_put_token_per_second; |
80 | 74 | limit.burst = is_get ? config::s3_get_bucket_tokens : config::s3_put_bucket_tokens; |
81 | 74 | limit.count_limit = is_get ? config::s3_get_token_limit : config::s3_put_token_limit; |
82 | 74 | } else if (qps_per_core > 0) { |
83 | 6 | limit.qps = cap_multiply(qps_per_core, cores, qps_max); |
84 | 6 | limit.burst = limit.qps; // burst = 1 second worth of quota |
85 | 6 | } // qps_per_core == 0: QPS limiting disabled, all fields stay 0. |
86 | | |
87 | 81 | if (bytes_per_core > 0) { |
88 | 3 | limit.bytes_per_second = cap_multiply(bytes_per_core, cores, bytes_max); |
89 | 3 | } |
90 | 81 | return limit; |
91 | 81 | } |
92 | | |
93 | 0 | int reset_s3_rate_limiter(S3RateLimitType type, size_t max_speed, size_t max_burst, size_t limit) { |
94 | 0 | if (type == S3RateLimitType::UNKNOWN) { |
95 | 0 | return -1; |
96 | 0 | } |
97 | 0 | return S3RateLimiterManager::instance().qps_limiter(type)->reset(max_speed, max_burst, limit); |
98 | 0 | } |
99 | | |
100 | 38 | int64_t s3_rate_limiter_cpu_cores() { |
101 | 38 | if (int64_t overridden = config::s3_rate_limiter_cpu_cores; overridden > 0) { |
102 | 6 | return overridden; |
103 | 6 | } |
104 | 32 | int physical = static_cast<int>(std::thread::hardware_concurrency()); |
105 | | // Re-read the cgroup quota on every call: serverless BEs can be resized in place, |
106 | | // and the daemon refresh thread picks the change up through here. |
107 | 32 | int limited = CGroupUtil::get_cgroup_limited_cpu_number(physical); |
108 | 32 | return std::max(1, limited); |
109 | 38 | } |
110 | | |
111 | 1 | S3RateLimiterManager::S3RateLimiterManager() { |
112 | 1 | const int64_t cores = s3_rate_limiter_cpu_cores(); |
113 | 2 | for (auto type : {S3RateLimitType::GET, S3RateLimitType::PUT}) { |
114 | 2 | auto limit = resolve_s3_rate_limit(type, cores); |
115 | 2 | _qps_limiters[index_of(type)] = std::make_unique<S3RateLimiterHolder>( |
116 | 2 | limit.qps, limit.burst, limit.count_limit, s3_rate_limiter_metric_func(type)); |
117 | 2 | _bytes_limiters[index_of(type)] = std::make_unique<S3RateLimiterHolder>( |
118 | 2 | limit.bytes_per_second, limit.bytes_per_second, limit.bytes_per_second, |
119 | 2 | bytes_rate_limiter_metric_func(type)); |
120 | 2 | } |
121 | 1 | } |
122 | | |
123 | 71 | S3RateLimiterManager& S3RateLimiterManager::instance() { |
124 | 71 | static S3RateLimiterManager ret; |
125 | 71 | return ret; |
126 | 71 | } |
127 | | |
128 | 119 | S3RateLimiterHolder* S3RateLimiterManager::qps_limiter(S3RateLimitType type) { |
129 | 119 | return _qps_limiters[index_of(type)].get(); |
130 | 119 | } |
131 | | |
132 | 98 | S3RateLimiterHolder* S3RateLimiterManager::bytes_limiter(S3RateLimitType type) { |
133 | 98 | return _bytes_limiters[index_of(type)].get(); |
134 | 98 | } |
135 | | |
136 | 35 | void S3RateLimiterManager::refresh() { |
137 | 35 | std::lock_guard guard(_refresh_lock); |
138 | 35 | const int64_t cores = s3_rate_limiter_cpu_cores(); |
139 | 70 | for (auto type : {S3RateLimitType::GET, S3RateLimitType::PUT}) { |
140 | 70 | const auto limit = resolve_s3_rate_limit(type, cores); |
141 | | |
142 | 70 | auto* qps = qps_limiter(type); |
143 | 70 | if (qps->get_max_speed() != static_cast<size_t>(limit.qps) || |
144 | 70 | qps->get_max_burst() != static_cast<size_t>(limit.burst) || |
145 | 70 | qps->get_limit() != static_cast<size_t>(limit.count_limit)) { |
146 | 25 | qps->reset(limit.qps, limit.burst, limit.count_limit); |
147 | 25 | LOG(INFO) << "reset S3 " << to_string(type) << " QPS rate limiter, qps=" << limit.qps |
148 | 25 | << ", burst=" << limit.burst << ", count_limit=" << limit.count_limit |
149 | 25 | << ", cores=" << cores; |
150 | 25 | } |
151 | | |
152 | 70 | auto* bytes = bytes_limiter(type); |
153 | 70 | const auto bytes_per_second = static_cast<size_t>(limit.bytes_per_second); |
154 | 70 | if (bytes->get_max_speed() != bytes_per_second || |
155 | 70 | bytes->get_max_burst() != bytes_per_second || bytes->get_limit() != bytes_per_second) { |
156 | 7 | bytes->reset(bytes_per_second, bytes_per_second, bytes_per_second); |
157 | 7 | LOG(INFO) << "reset S3 " << to_string(type) |
158 | 7 | << " bytes rate limiter, bytes_per_second=" << limit.bytes_per_second |
159 | 7 | << ", cores=" << cores; |
160 | 7 | } |
161 | 70 | } |
162 | 35 | } |
163 | | |
164 | 293k | S3RateLimitGuard::S3RateLimitGuard(S3RateLimitType type, size_t estimated_bytes) { |
165 | 293k | if (!config::enable_s3_rate_limiter) { |
166 | 293k | return; |
167 | 293k | } |
168 | 11 | auto& mgr = S3RateLimiterManager::instance(); |
169 | | |
170 | 11 | auto* qps = mgr.qps_limiter(type); |
171 | 11 | if (qps->is_enabled() && |
172 | 12 | apply_s3_rate_limit(type, qps, config::s3_rate_limiter_log_interval) < 0) { |
173 | 5 | _ok = false; |
174 | 5 | return; |
175 | 5 | } |
176 | | |
177 | 7 | if (estimated_bytes == 0) { |
178 | 7 | return; |
179 | 7 | } |
180 | 18.4E | auto* bytes = mgr.bytes_limiter(type); |
181 | 18.4E | if (!bytes->is_enabled()) { |
182 | 1 | return; |
183 | 1 | } |
184 | | // Clamp the reservation to 1 second worth of bandwidth so a single oversized IO |
185 | | // (e.g. a whole-file read_at) cannot create unbounded upfront debt. The clamped |
186 | | // remainder is intentionally not accounted; effective quotas below the single-IO |
187 | | // upper bound are excluded by the config contract (see config.cpp). |
188 | 18.4E | _reserved = std::min(estimated_bytes, bytes->get_max_speed()); |
189 | 18.4E | if (_reserved > 0) { |
190 | | // Pin the admitted bucket generation for settlement and count release. |
191 | 17 | _charged_bucket = bytes->charge(_reserved); |
192 | 17 | if (_charged_bucket == nullptr) { |
193 | 5 | _ok = false; |
194 | 5 | } |
195 | 17 | } |
196 | 18.4E | } |
197 | | |
198 | 293k | S3RateLimitGuard::~S3RateLimitGuard() { |
199 | 293k | if (_charged_bucket != nullptr) { |
200 | 12 | _charged_bucket->refund_count(_reserved); |
201 | 12 | } |
202 | 293k | } |
203 | | |
204 | 116k | void S3RateLimitGuard::settle(size_t actual_bytes) { |
205 | 116k | if (_settled) { |
206 | 0 | return; |
207 | 0 | } |
208 | 116k | _settled = true; |
209 | 116k | if (_charged_bucket != nullptr && _reserved > actual_bytes) { |
210 | 3 | _charged_bucket->refund_tokens(_reserved - actual_bytes); |
211 | 3 | } |
212 | 116k | } |
213 | | |
214 | | } // namespace doris |