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
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you 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.

use std::num::NonZeroU32;
use std::sync::Arc;
use std::thread;

use governor::clock::Clock;
use governor::clock::DefaultClock;
use governor::middleware::NoOpMiddleware;
use governor::state::InMemoryState;
use governor::state::NotKeyed;
use governor::Quota;
use governor::RateLimiter;

use crate::raw::*;
use crate::*;

/// Add a bandwidth rate limiter to the underlying services.
///
/// # Throttle
///
/// There are several algorithms when it come to rate limiting techniques.
/// This throttle layer uses Generic Cell Rate Algorithm (GCRA) provided by
/// [Governor](https://docs.rs/governor/latest/governor/index.html).
/// By setting the `bandwidth` and `burst`, we can control the byte flow rate of underlying services.
///
/// # Note
///
/// When setting the ThrottleLayer, always consider the largest possible operation size as the burst size,
/// as **the burst size should be larger than any possible byte length to allow it to pass through**.
///
/// Read more about [Quota](https://docs.rs/governor/latest/governor/struct.Quota.html#examples)
///
/// # Examples
///
/// This example limits bandwidth to 10 KiB/s and burst size to 10 MiB.
///
/// ```no_build
/// use anyhow::Result;
/// use opendal::layers::ThrottleLayer;
/// use opendal::services;
/// use opendal::Operator;
/// use opendal::Scheme;
///
/// let _ = Operator::new(services::Memory::default())
///     .expect("must init")
///     .layer(ThrottleLayer::new(10 * 1024, 10000 * 1024))
///     .finish();
/// ```
#[derive(Clone)]
pub struct ThrottleLayer {
    bandwidth: NonZeroU32,
    burst: NonZeroU32,
}

impl ThrottleLayer {
    /// Create a new `ThrottleLayer` with given bandwidth and burst.
    ///
    /// - bandwidth: the maximum number of bytes allowed to pass through per second.
    /// - burst: the maximum number of bytes allowed to pass through at once.
    pub fn new(bandwidth: u32, burst: u32) -> Self {
        assert!(bandwidth > 0);
        assert!(burst > 0);
        Self {
            bandwidth: NonZeroU32::new(bandwidth).unwrap(),
            burst: NonZeroU32::new(burst).unwrap(),
        }
    }
}

impl<A: Access> Layer<A> for ThrottleLayer {
    type LayeredAccess = ThrottleAccessor<A>;

    fn layer(&self, accessor: A) -> Self::LayeredAccess {
        let rate_limiter = Arc::new(RateLimiter::direct(
            Quota::per_second(self.bandwidth).allow_burst(self.burst),
        ));
        ThrottleAccessor {
            inner: accessor,
            rate_limiter,
        }
    }
}

/// Share an atomic RateLimiter instance across all threads in one operator.
/// If want to add more observability in the future, replace the default NoOpMiddleware with other middleware types.
/// Read more about [Middleware](https://docs.rs/governor/latest/governor/middleware/index.html)
type SharedRateLimiter = Arc<RateLimiter<NotKeyed, InMemoryState, DefaultClock, NoOpMiddleware>>;

#[derive(Debug, Clone)]
pub struct ThrottleAccessor<A: Access> {
    inner: A,
    rate_limiter: SharedRateLimiter,
}

impl<A: Access> LayeredAccess for ThrottleAccessor<A> {
    type Inner = A;
    type Reader = ThrottleWrapper<A::Reader>;
    type BlockingReader = ThrottleWrapper<A::BlockingReader>;
    type Writer = ThrottleWrapper<A::Writer>;
    type BlockingWriter = ThrottleWrapper<A::BlockingWriter>;
    type Lister = A::Lister;
    type BlockingLister = A::BlockingLister;

    fn inner(&self) -> &Self::Inner {
        &self.inner
    }

    async fn read(&self, path: &str, args: OpRead) -> Result<(RpRead, Self::Reader)> {
        let limiter = self.rate_limiter.clone();

        self.inner
            .read(path, args)
            .await
            .map(|(rp, r)| (rp, ThrottleWrapper::new(r, limiter)))
    }

    async fn write(&self, path: &str, args: OpWrite) -> Result<(RpWrite, Self::Writer)> {
        let limiter = self.rate_limiter.clone();

        self.inner
            .write(path, args)
            .await
            .map(|(rp, w)| (rp, ThrottleWrapper::new(w, limiter)))
    }

    async fn list(&self, path: &str, args: OpList) -> Result<(RpList, Self::Lister)> {
        self.inner.list(path, args).await
    }

    fn blocking_read(&self, path: &str, args: OpRead) -> Result<(RpRead, Self::BlockingReader)> {
        let limiter = self.rate_limiter.clone();

        self.inner
            .blocking_read(path, args)
            .map(|(rp, r)| (rp, ThrottleWrapper::new(r, limiter)))
    }

    fn blocking_write(&self, path: &str, args: OpWrite) -> Result<(RpWrite, Self::BlockingWriter)> {
        let limiter = self.rate_limiter.clone();

        self.inner
            .blocking_write(path, args)
            .map(|(rp, w)| (rp, ThrottleWrapper::new(w, limiter)))
    }

    fn blocking_list(&self, path: &str, args: OpList) -> Result<(RpList, Self::BlockingLister)> {
        self.inner.blocking_list(path, args)
    }
}

pub struct ThrottleWrapper<R> {
    inner: R,
    limiter: SharedRateLimiter,
}

impl<R> ThrottleWrapper<R> {
    pub fn new(inner: R, rate_limiter: SharedRateLimiter) -> Self {
        Self {
            inner,
            limiter: rate_limiter,
        }
    }
}

impl<R: oio::Read> oio::Read for ThrottleWrapper<R> {
    async fn read_at(&self, offset: u64, limit: usize) -> Result<Buffer> {
        // TODO: How can we handle buffer reads with a limiter?
        self.inner.read_at(offset, limit).await
    }
}

impl<R: oio::BlockingRead> oio::BlockingRead for ThrottleWrapper<R> {
    fn read_at(&self, offset: u64, limit: usize) -> Result<Buffer> {
        // TODO: How can we handle buffer reads with a limiter?
        self.inner.read_at(offset, limit)
    }
}

impl<R: oio::Write> oio::Write for ThrottleWrapper<R> {
    async fn write(&mut self, bs: Buffer) -> Result<usize> {
        let buf_length = NonZeroU32::new(bs.len() as u32).unwrap();

        loop {
            match self.limiter.check_n(buf_length) {
                Ok(res) => match res {
                    Ok(_) => return self.inner.write(bs).await,
                    // the query is valid but the Decider can not accommodate them.
                    Err(not_until) => {
                        let _ = not_until.wait_time_from(DefaultClock::default().now());
                        // TODO: Should lock the limiter and wait for the wait_time, or should let other small requests go first?

                        // FIXME: we should sleep here.
                        // tokio::time::sleep(wait_time).await;
                    }
                },
                // the query was invalid as the rate limit parameters can "never" accommodate the number of cells queried for.
                Err(_) => return Err(Error::new(
                    ErrorKind::RateLimited,
                    "InsufficientCapacity due to burst size being smaller than the request size",
                )),
            }
        }
    }

    async fn abort(&mut self) -> Result<()> {
        self.inner.abort().await
    }

    async fn close(&mut self) -> Result<()> {
        self.inner.close().await
    }
}

impl<R: oio::BlockingWrite> oio::BlockingWrite for ThrottleWrapper<R> {
    fn write(&mut self, bs: Buffer) -> Result<usize> {
        let buf_length = NonZeroU32::new(bs.len() as u32).unwrap();

        loop {
            match self.limiter.check_n(buf_length) {
                Ok(res) => match res {
                    Ok(_) => return self.inner.write(bs),
                    // the query is valid but the Decider can not accommodate them.
                    Err(not_until) => {
                        let wait_time = not_until.wait_time_from(DefaultClock::default().now());
                        thread::sleep(wait_time);
                    }
                },
                // the query was invalid as the rate limit parameters can "never" accommodate the number of cells queried for.
                Err(_) => return Err(Error::new(
                    ErrorKind::RateLimited,
                    "InsufficientCapacity due to burst size being smaller than the request size",
                )),
            }
        }
    }

    fn close(&mut self) -> Result<()> {
        self.inner.close()
    }
}