Line data Source code
1 : //! This module provides a wrapper around a real RemoteStorage implementation that
2 : //! causes the first N attempts at each upload or download operatio to fail. For
3 : //! testing purposes.
4 : use bytes::Bytes;
5 : use futures::stream::Stream;
6 : use std::collections::HashMap;
7 : use std::num::NonZeroU32;
8 : use std::sync::Mutex;
9 : use std::time::SystemTime;
10 : use std::{collections::hash_map::Entry, sync::Arc};
11 : use tokio_util::sync::CancellationToken;
12 :
13 : use crate::{
14 : Download, DownloadError, GenericRemoteStorage, Listing, ListingMode, RemotePath, RemoteStorage,
15 : StorageMetadata, TimeTravelError,
16 : };
17 :
18 : pub struct UnreliableWrapper {
19 : inner: GenericRemoteStorage<Arc<VoidStorage>>,
20 :
21 : // This many attempts of each operation will fail, then we let it succeed.
22 : attempts_to_fail: u64,
23 :
24 : // Tracks how many failed attempts of each operation has been made.
25 : attempts: Mutex<HashMap<RemoteOp, u64>>,
26 : }
27 :
28 : /// Used to identify retries of different unique operation.
29 48 : #[derive(Debug, Hash, Eq, PartialEq)]
30 : enum RemoteOp {
31 : ListPrefixes(Option<RemotePath>),
32 : Upload(RemotePath),
33 : Download(RemotePath),
34 : Delete(RemotePath),
35 : DeleteObjects(Vec<RemotePath>),
36 : TimeTravelRecover(Option<RemotePath>),
37 : }
38 :
39 : impl UnreliableWrapper {
40 4 : pub fn new(inner: crate::GenericRemoteStorage, attempts_to_fail: u64) -> Self {
41 4 : assert!(attempts_to_fail > 0);
42 4 : let inner = match inner {
43 0 : GenericRemoteStorage::AwsS3(s) => GenericRemoteStorage::AwsS3(s),
44 0 : GenericRemoteStorage::AzureBlob(s) => GenericRemoteStorage::AzureBlob(s),
45 4 : GenericRemoteStorage::LocalFs(s) => GenericRemoteStorage::LocalFs(s),
46 : // We could also make this a no-op, as in, extract the inner of the passed generic remote storage
47 0 : GenericRemoteStorage::Unreliable(_s) => {
48 0 : panic!("Can't wrap unreliable wrapper unreliably")
49 : }
50 : };
51 4 : UnreliableWrapper {
52 4 : inner,
53 4 : attempts_to_fail,
54 4 : attempts: Mutex::new(HashMap::new()),
55 4 : }
56 4 : }
57 :
58 : ///
59 : /// Common functionality for all operations.
60 : ///
61 : /// On the first attempts of this operation, return an error. After 'attempts_to_fail'
62 : /// attempts, let the operation go ahead, and clear the counter.
63 : ///
64 48 : fn attempt(&self, op: RemoteOp) -> anyhow::Result<u64> {
65 48 : let mut attempts = self.attempts.lock().unwrap();
66 48 :
67 48 : match attempts.entry(op) {
68 24 : Entry::Occupied(mut e) => {
69 24 : let attempts_before_this = {
70 24 : let p = e.get_mut();
71 24 : *p += 1;
72 24 : *p
73 24 : };
74 24 :
75 24 : if attempts_before_this >= self.attempts_to_fail {
76 : // let it succeed
77 24 : e.remove();
78 24 : Ok(attempts_before_this)
79 : } else {
80 0 : let error =
81 0 : anyhow::anyhow!("simulated failure of remote operation {:?}", e.key());
82 0 : Err(error)
83 : }
84 : }
85 24 : Entry::Vacant(e) => {
86 24 : let error = anyhow::anyhow!("simulated failure of remote operation {:?}", e.key());
87 24 : e.insert(1);
88 24 : Err(error)
89 : }
90 : }
91 48 : }
92 :
93 0 : async fn delete_inner(
94 0 : &self,
95 0 : path: &RemotePath,
96 0 : attempt: bool,
97 0 : cancel: &CancellationToken,
98 0 : ) -> anyhow::Result<()> {
99 0 : if attempt {
100 0 : self.attempt(RemoteOp::Delete(path.clone()))?;
101 0 : }
102 0 : self.inner.delete(path, cancel).await
103 0 : }
104 : }
105 :
106 : // We never construct this, so the type is not important, just has to not be UnreliableWrapper and impl RemoteStorage.
107 : type VoidStorage = crate::LocalFs;
108 :
109 : impl RemoteStorage for UnreliableWrapper {
110 0 : async fn list_prefixes(
111 0 : &self,
112 0 : prefix: Option<&RemotePath>,
113 0 : cancel: &CancellationToken,
114 0 : ) -> Result<Vec<RemotePath>, DownloadError> {
115 0 : self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
116 0 : .map_err(DownloadError::Other)?;
117 0 : self.inner.list_prefixes(prefix, cancel).await
118 0 : }
119 :
120 0 : async fn list_files(
121 0 : &self,
122 0 : folder: Option<&RemotePath>,
123 0 : max_keys: Option<NonZeroU32>,
124 0 : cancel: &CancellationToken,
125 0 : ) -> Result<Vec<RemotePath>, DownloadError> {
126 0 : self.attempt(RemoteOp::ListPrefixes(folder.cloned()))
127 0 : .map_err(DownloadError::Other)?;
128 0 : self.inner.list_files(folder, max_keys, cancel).await
129 0 : }
130 :
131 0 : async fn list(
132 0 : &self,
133 0 : prefix: Option<&RemotePath>,
134 0 : mode: ListingMode,
135 0 : max_keys: Option<NonZeroU32>,
136 0 : cancel: &CancellationToken,
137 0 : ) -> Result<Listing, DownloadError> {
138 0 : self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
139 0 : .map_err(DownloadError::Other)?;
140 0 : self.inner.list(prefix, mode, max_keys, cancel).await
141 0 : }
142 :
143 48 : async fn upload(
144 48 : &self,
145 48 : data: impl Stream<Item = std::io::Result<Bytes>> + Send + Sync + 'static,
146 48 : // S3 PUT request requires the content length to be specified,
147 48 : // otherwise it starts to fail with the concurrent connection count increasing.
148 48 : data_size_bytes: usize,
149 48 : to: &RemotePath,
150 48 : metadata: Option<StorageMetadata>,
151 48 : cancel: &CancellationToken,
152 48 : ) -> anyhow::Result<()> {
153 48 : self.attempt(RemoteOp::Upload(to.clone()))?;
154 24 : self.inner
155 24 : .upload(data, data_size_bytes, to, metadata, cancel)
156 76 : .await
157 48 : }
158 :
159 0 : async fn download(
160 0 : &self,
161 0 : from: &RemotePath,
162 0 : cancel: &CancellationToken,
163 0 : ) -> Result<Download, DownloadError> {
164 0 : self.attempt(RemoteOp::Download(from.clone()))
165 0 : .map_err(DownloadError::Other)?;
166 0 : self.inner.download(from, cancel).await
167 0 : }
168 :
169 0 : async fn download_byte_range(
170 0 : &self,
171 0 : from: &RemotePath,
172 0 : start_inclusive: u64,
173 0 : end_exclusive: Option<u64>,
174 0 : cancel: &CancellationToken,
175 0 : ) -> Result<Download, DownloadError> {
176 0 : // Note: We treat any download_byte_range as an "attempt" of the same
177 0 : // operation. We don't pay attention to the ranges. That's good enough
178 0 : // for now.
179 0 : self.attempt(RemoteOp::Download(from.clone()))
180 0 : .map_err(DownloadError::Other)?;
181 0 : self.inner
182 0 : .download_byte_range(from, start_inclusive, end_exclusive, cancel)
183 0 : .await
184 0 : }
185 :
186 0 : async fn delete(&self, path: &RemotePath, cancel: &CancellationToken) -> anyhow::Result<()> {
187 0 : self.delete_inner(path, true, cancel).await
188 0 : }
189 :
190 0 : async fn delete_objects<'a>(
191 0 : &self,
192 0 : paths: &'a [RemotePath],
193 0 : cancel: &CancellationToken,
194 0 : ) -> anyhow::Result<()> {
195 0 : self.attempt(RemoteOp::DeleteObjects(paths.to_vec()))?;
196 0 : let mut error_counter = 0;
197 0 : for path in paths {
198 : // Dont record attempt because it was already recorded above
199 0 : if (self.delete_inner(path, false, cancel).await).is_err() {
200 0 : error_counter += 1;
201 0 : }
202 : }
203 0 : if error_counter > 0 {
204 0 : return Err(anyhow::anyhow!(
205 0 : "failed to delete {} objects",
206 0 : error_counter
207 0 : ));
208 0 : }
209 0 : Ok(())
210 0 : }
211 :
212 0 : async fn copy(
213 0 : &self,
214 0 : from: &RemotePath,
215 0 : to: &RemotePath,
216 0 : cancel: &CancellationToken,
217 0 : ) -> anyhow::Result<()> {
218 0 : // copy is equivalent to download + upload
219 0 : self.attempt(RemoteOp::Download(from.clone()))?;
220 0 : self.attempt(RemoteOp::Upload(to.clone()))?;
221 0 : self.inner.copy_object(from, to, cancel).await
222 0 : }
223 :
224 0 : async fn time_travel_recover(
225 0 : &self,
226 0 : prefix: Option<&RemotePath>,
227 0 : timestamp: SystemTime,
228 0 : done_if_after: SystemTime,
229 0 : cancel: &CancellationToken,
230 0 : ) -> Result<(), TimeTravelError> {
231 0 : self.attempt(RemoteOp::TimeTravelRecover(prefix.map(|p| p.to_owned())))
232 0 : .map_err(TimeTravelError::Other)?;
233 0 : self.inner
234 0 : .time_travel_recover(prefix, timestamp, done_if_after, cancel)
235 0 : .await
236 0 : }
237 : }
|
