-
Notifications
You must be signed in to change notification settings - Fork 75
/
retry.rs
313 lines (280 loc) · 9.75 KB
/
retry.rs
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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
use std::{ops::SubAssign, time::Duration};
use async_trait::async_trait;
use futures::Future;
use tracing::warn;
use super::{ClientApi, LeaderStateUpdate, ProposeResponse, RepeatableClientApi};
use crate::{
members::ServerId,
rpc::{ConfChange, CurpError, FetchClusterResponse, Member, ReadState, Redirect},
};
/// Backoff config
#[derive(Debug, Clone)]
enum BackoffConfig {
/// A fixed delay backoff
Fixed,
/// A exponential delay backoff
Exponential {
/// Control the max delay of exponential
max_delay: Duration,
},
}
/// Retry config to control the retry policy
#[derive(Debug, Clone)]
pub(super) struct RetryConfig {
/// Backoff config
backoff: BackoffConfig,
/// Initial delay
delay: Duration,
/// Retry count
count: usize,
}
/// Backoff tool
#[derive(Debug)]
struct Backoff {
/// The retry config
config: RetryConfig,
/// Current delay
cur_delay: Duration,
/// Total RPC count
count: usize,
}
impl RetryConfig {
/// Create a fixed retry config
pub(super) fn new_fixed(delay: Duration, count: usize) -> Self {
assert!(count > 0, "retry count should be larger than 0");
Self {
backoff: BackoffConfig::Fixed,
delay,
count,
}
}
/// Create a exponential retry config
pub(super) fn new_exponential(delay: Duration, max_delay: Duration, count: usize) -> Self {
assert!(count > 0, "retry count should be larger than 0");
Self {
backoff: BackoffConfig::Exponential { max_delay },
delay,
count,
}
}
/// Create a backoff process
fn init_backoff(&self) -> Backoff {
Backoff {
config: self.clone(),
cur_delay: self.delay,
count: self.count,
}
}
}
impl Backoff {
/// Get the next delay duration, None means the end.
fn next_delay(&mut self) -> Option<Duration> {
if self.count == 0 {
return None;
}
self.count.sub_assign(1);
let cur = self.cur_delay;
if let BackoffConfig::Exponential { max_delay } = self.config.backoff {
self.cur_delay = self
.cur_delay
.checked_mul(2)
.unwrap_or(self.cur_delay)
.min(max_delay);
}
Some(cur)
}
}
/// The retry client automatically retry the requests of the inner client api
/// which raises the [`tonic::Status`] error
#[derive(Debug)]
pub(super) struct Retry<Api> {
/// Inner client
inner: Api,
/// Retry config
config: RetryConfig,
}
impl<Api> Retry<Api>
where
Api: RepeatableClientApi<Error = CurpError> + LeaderStateUpdate + Send + Sync + 'static,
{
/// Create a retry client
pub(super) fn new(inner: Api, config: RetryConfig) -> Self {
Self { inner, config }
}
/// Takes a function f and run retry.
async fn retry<'a, R, F>(&'a self, f: impl Fn(&'a Api) -> F) -> Result<R, tonic::Status>
where
F: Future<Output = Result<R, CurpError>>,
{
let mut backoff = self.config.init_backoff();
let mut last_err = None;
while let Some(delay) = backoff.next_delay() {
let err = match f(&self.inner).await {
Ok(res) => return Ok(res),
Err(err) => err,
};
match err {
// some errors that should not retry
CurpError::Duplicated(_)
| CurpError::ShuttingDown(_)
| CurpError::InvalidConfig(_)
| CurpError::NodeNotExists(_)
| CurpError::NodeAlreadyExists(_)
| CurpError::LearnerNotCatchUp(_) => {
return Err(tonic::Status::from(err));
}
// some errors that could have a retry
CurpError::ExpiredClientId(_)
| CurpError::KeyConflict(_)
| CurpError::Internal(_)
| CurpError::LeaderTransfer(_) => {}
// update leader state if we got a rpc transport error
CurpError::RpcTransport(_) => {
if let Err(e) = self.inner.fetch_leader_id(true).await {
warn!("fetch leader failed, error {e:?}");
}
}
// update the cluster state if got WrongClusterVersion
CurpError::WrongClusterVersion(_) => {
// the inner client should automatically update cluster state when fetch_cluster
if let Err(e) = self.inner.fetch_cluster(true).await {
warn!("fetch cluster failed, error {e:?}");
}
}
// update the leader state if got Redirect
CurpError::Redirect(Redirect { leader_id, term }) => {
let _ig = self.inner.update_leader(leader_id, term).await;
}
}
#[cfg(feature = "client-metrics")]
super::metrics::get().client_retry_count.add(1, &[]);
warn!(
"got error: {err:?}, retry on {} seconds later",
delay.as_secs_f32()
);
last_err = Some(err);
tokio::time::sleep(delay).await;
}
Err(tonic::Status::deadline_exceeded(format!(
"request timeout, last error: {:?}",
last_err.unwrap_or_else(|| unreachable!("last error must be set"))
)))
}
}
#[async_trait]
impl<Api> ClientApi for Retry<Api>
where
Api: RepeatableClientApi<Error = CurpError> + LeaderStateUpdate + Send + Sync + 'static,
{
/// The client error
type Error = tonic::Status;
/// Inherit the command type
type Cmd = Api::Cmd;
/// Send propose to the whole cluster, `use_fast_path` set to `false` to fallback into ordered
/// requests (event the requests are commutative).
async fn propose(
&self,
cmd: &Self::Cmd,
token: Option<&String>,
use_fast_path: bool,
) -> Result<ProposeResponse<Self::Cmd>, tonic::Status> {
let propose_id = self
.retry::<_, _>(RepeatableClientApi::gen_propose_id)
.await?;
self.retry::<_, _>(|client| {
RepeatableClientApi::propose(client, propose_id, cmd, token, use_fast_path)
})
.await
}
/// Send propose configuration changes to the cluster
async fn propose_conf_change(
&self,
changes: Vec<ConfChange>,
) -> Result<Vec<Member>, tonic::Status> {
let propose_id = self
.retry::<_, _>(RepeatableClientApi::gen_propose_id)
.await?;
self.retry::<_, _>(|client| {
let changes_c = changes.clone();
RepeatableClientApi::propose_conf_change(client, propose_id, changes_c)
})
.await
}
/// Send propose to shutdown cluster
async fn propose_shutdown(&self) -> Result<(), tonic::Status> {
let propose_id = self
.retry::<_, _>(RepeatableClientApi::gen_propose_id)
.await?;
self.retry::<_, _>(|client| RepeatableClientApi::propose_shutdown(client, propose_id))
.await
}
/// Send propose to publish a node id and name
async fn propose_publish(
&self,
node_id: ServerId,
node_name: String,
node_client_urls: Vec<String>,
) -> Result<(), Self::Error> {
let propose_id = self
.retry::<_, _>(RepeatableClientApi::gen_propose_id)
.await?;
self.retry::<_, _>(|client| {
let name_c = node_name.clone();
let node_client_urls_c = node_client_urls.clone();
RepeatableClientApi::propose_publish(
client,
propose_id,
node_id,
name_c,
node_client_urls_c,
)
})
.await
}
/// Send move leader request
async fn move_leader(&self, node_id: u64) -> Result<(), Self::Error> {
self.retry::<_, _>(|client| client.move_leader(node_id))
.await
}
/// Send fetch read state from leader
async fn fetch_read_state(&self, cmd: &Self::Cmd) -> Result<ReadState, tonic::Status> {
self.retry::<_, _>(|client| client.fetch_read_state(cmd))
.await
}
/// Send fetch cluster requests to all servers (That's because initially, we didn't
/// know who the leader is.)
/// Note: The fetched cluster may still be outdated if `linearizable` is false
async fn fetch_cluster(
&self,
linearizable: bool,
) -> Result<FetchClusterResponse, tonic::Status> {
self.retry::<_, _>(|client| client.fetch_cluster(linearizable))
.await
}
}
/// Tests for backoff
#[cfg(test)]
mod tests {
use std::time::Duration;
use super::RetryConfig;
#[test]
fn test_fixed_backoff_works() {
let config = RetryConfig::new_fixed(Duration::from_secs(1), 3);
let mut backoff = config.init_backoff();
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(1)));
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(1)));
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(1)));
assert_eq!(backoff.next_delay(), None);
}
#[test]
fn test_exponential_backoff_works() {
let config =
RetryConfig::new_exponential(Duration::from_secs(1), Duration::from_secs(5), 4);
let mut backoff = config.init_backoff();
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(1)));
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(2)));
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(4)));
assert_eq!(backoff.next_delay(), Some(Duration::from_secs(5))); // 8 > 5
assert_eq!(backoff.next_delay(), None);
}
}