Line data Source code
1 : use std::fmt::Debug;
2 : use std::num::NonZeroUsize;
3 : use std::str::FromStr;
4 : use std::time::Duration;
5 :
6 : use aws_sdk_s3::types::StorageClass;
7 : use camino::Utf8PathBuf;
8 : use serde::{Deserialize, Serialize};
9 :
10 : use crate::{
11 : DEFAULT_MAX_KEYS_PER_LIST_RESPONSE, DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT,
12 : DEFAULT_REMOTE_STORAGE_LOCALFS_CONCURRENCY_LIMIT, DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT,
13 : };
14 :
15 : /// External backup storage configuration, enough for creating a client for that storage.
16 37 : #[derive(Debug, Clone, PartialEq, Eq, Deserialize, Serialize)]
17 : pub struct RemoteStorageConfig {
18 : /// The storage connection configuration.
19 : #[serde(flatten)]
20 : pub storage: RemoteStorageKind,
21 : /// A common timeout enforced for all requests after concurrency limiter permit has been
22 : /// acquired.
23 : #[serde(
24 : with = "humantime_serde",
25 : default = "default_timeout",
26 : skip_serializing_if = "is_default_timeout"
27 : )]
28 : pub timeout: Duration,
29 : /// Alternative timeout used for metadata objects which are expected to be small
30 : #[serde(
31 : with = "humantime_serde",
32 : default = "default_small_timeout",
33 : skip_serializing_if = "is_default_small_timeout"
34 : )]
35 : pub small_timeout: Duration,
36 : }
37 :
38 : impl RemoteStorageKind {
39 0 : pub fn bucket_name(&self) -> Option<&str> {
40 0 : match self {
41 0 : RemoteStorageKind::LocalFs { .. } => None,
42 0 : RemoteStorageKind::AwsS3(config) => Some(&config.bucket_name),
43 0 : RemoteStorageKind::AzureContainer(config) => Some(&config.container_name),
44 : }
45 0 : }
46 : }
47 :
48 : impl RemoteStorageConfig {
49 : /// Helper to fetch the configured concurrency limit.
50 0 : pub fn concurrency_limit(&self) -> usize {
51 0 : match &self.storage {
52 0 : RemoteStorageKind::LocalFs { .. } => DEFAULT_REMOTE_STORAGE_LOCALFS_CONCURRENCY_LIMIT,
53 0 : RemoteStorageKind::AwsS3(c) => c.concurrency_limit.into(),
54 0 : RemoteStorageKind::AzureContainer(c) => c.concurrency_limit.into(),
55 : }
56 0 : }
57 : }
58 :
59 1 : fn default_timeout() -> Duration {
60 1 : RemoteStorageConfig::DEFAULT_TIMEOUT
61 1 : }
62 :
63 10 : fn default_small_timeout() -> Duration {
64 10 : RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
65 10 : }
66 :
67 0 : fn is_default_timeout(d: &Duration) -> bool {
68 0 : *d == RemoteStorageConfig::DEFAULT_TIMEOUT
69 0 : }
70 :
71 0 : fn is_default_small_timeout(d: &Duration) -> bool {
72 0 : *d == RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
73 0 : }
74 :
75 : /// A kind of a remote storage to connect to, with its connection configuration.
76 28 : #[derive(Debug, Clone, PartialEq, Eq, Deserialize, Serialize)]
77 : #[serde(untagged)]
78 : pub enum RemoteStorageKind {
79 : /// Storage based on local file system.
80 : /// Specify a root folder to place all stored files into.
81 : LocalFs { local_path: Utf8PathBuf },
82 : /// AWS S3 based storage, storing all files in the S3 bucket
83 : /// specified by the config
84 : AwsS3(S3Config),
85 : /// Azure Blob based storage, storing all files in the container
86 : /// specified by the config
87 : AzureContainer(AzureConfig),
88 : }
89 :
90 : /// AWS S3 bucket coordinates and access credentials to manage the bucket contents (read and write).
91 25 : #[derive(Clone, PartialEq, Eq, Deserialize, Serialize)]
92 : pub struct S3Config {
93 : /// Name of the bucket to connect to.
94 : pub bucket_name: String,
95 : /// The region where the bucket is located at.
96 : pub bucket_region: String,
97 : /// A "subfolder" in the bucket, to use the same bucket separately by multiple remote storage users at once.
98 : pub prefix_in_bucket: Option<String>,
99 : /// A base URL to send S3 requests to.
100 : /// By default, the endpoint is derived from a region name, assuming it's
101 : /// an AWS S3 region name, erroring on wrong region name.
102 : /// Endpoint provides a way to support other S3 flavors and their regions.
103 : ///
104 : /// Example: `http://127.0.0.1:5000`
105 : pub endpoint: Option<String>,
106 : /// AWS S3 has various limits on its API calls, we need not to exceed those.
107 : /// See [`DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT`] for more details.
108 : #[serde(default = "default_remote_storage_s3_concurrency_limit")]
109 : pub concurrency_limit: NonZeroUsize,
110 : #[serde(default = "default_max_keys_per_list_response")]
111 : pub max_keys_per_list_response: Option<i32>,
112 : #[serde(
113 : deserialize_with = "deserialize_storage_class",
114 : serialize_with = "serialize_storage_class",
115 : default
116 : )]
117 : pub upload_storage_class: Option<StorageClass>,
118 : }
119 :
120 7 : fn default_remote_storage_s3_concurrency_limit() -> NonZeroUsize {
121 7 : DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT
122 7 : .try_into()
123 7 : .unwrap()
124 7 : }
125 :
126 7 : fn default_max_keys_per_list_response() -> Option<i32> {
127 7 : DEFAULT_MAX_KEYS_PER_LIST_RESPONSE
128 7 : }
129 :
130 0 : fn default_azure_conn_pool_size() -> usize {
131 0 : // By default, the Azure SDK does no connection pooling, due to historic reports of hard-to-reproduce issues
132 0 : // (https://github.com/hyperium/hyper/issues/2312)
133 0 : //
134 0 : // However, using connection pooling is important to avoid exhausting client ports when
135 0 : // doing huge numbers of requests (https://github.com/neondatabase/cloud/issues/20971)
136 0 : //
137 0 : // We therefore enable a modest pool size by default: this may be configured to zero if
138 0 : // issues like the alleged upstream hyper issue appear.
139 0 : 8
140 0 : }
141 :
142 : impl Debug for S3Config {
143 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
144 0 : f.debug_struct("S3Config")
145 0 : .field("bucket_name", &self.bucket_name)
146 0 : .field("bucket_region", &self.bucket_region)
147 0 : .field("prefix_in_bucket", &self.prefix_in_bucket)
148 0 : .field("concurrency_limit", &self.concurrency_limit)
149 0 : .field(
150 0 : "max_keys_per_list_response",
151 0 : &self.max_keys_per_list_response,
152 0 : )
153 0 : .finish()
154 0 : }
155 : }
156 :
157 : /// Azure bucket coordinates and access credentials to manage the bucket contents (read and write).
158 12 : #[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
159 : pub struct AzureConfig {
160 : /// Name of the container to connect to.
161 : pub container_name: String,
162 : /// Name of the storage account the container is inside of
163 : pub storage_account: Option<String>,
164 : /// The region where the bucket is located at.
165 : pub container_region: String,
166 : /// A "subfolder" in the container, to use the same container separately by multiple remote storage users at once.
167 : pub prefix_in_container: Option<String>,
168 : /// Azure has various limits on its API calls, we need not to exceed those.
169 : /// See [`DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT`] for more details.
170 : #[serde(default = "default_remote_storage_azure_concurrency_limit")]
171 : pub concurrency_limit: NonZeroUsize,
172 : #[serde(default = "default_max_keys_per_list_response")]
173 : pub max_keys_per_list_response: Option<i32>,
174 : #[serde(default = "default_azure_conn_pool_size")]
175 : pub conn_pool_size: usize,
176 : }
177 :
178 6 : fn default_remote_storage_azure_concurrency_limit() -> NonZeroUsize {
179 6 : NonZeroUsize::new(DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT).unwrap()
180 6 : }
181 :
182 : impl Debug for AzureConfig {
183 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
184 0 : f.debug_struct("AzureConfig")
185 0 : .field("bucket_name", &self.container_name)
186 0 : .field("storage_account", &self.storage_account)
187 0 : .field("bucket_region", &self.container_region)
188 0 : .field("prefix_in_container", &self.prefix_in_container)
189 0 : .field("concurrency_limit", &self.concurrency_limit)
190 0 : .field(
191 0 : "max_keys_per_list_response",
192 0 : &self.max_keys_per_list_response,
193 0 : )
194 0 : .finish()
195 0 : }
196 : }
197 :
198 15 : fn deserialize_storage_class<'de, D: serde::Deserializer<'de>>(
199 15 : deserializer: D,
200 15 : ) -> Result<Option<StorageClass>, D::Error> {
201 15 : Option::<String>::deserialize(deserializer).and_then(|s| {
202 15 : if let Some(s) = s {
203 : use serde::de::Error;
204 12 : let storage_class = StorageClass::from_str(&s).expect("infallible");
205 : #[allow(deprecated)]
206 12 : if matches!(storage_class, StorageClass::Unknown(_)) {
207 0 : return Err(D::Error::custom(format!(
208 0 : "Specified storage class unknown to SDK: '{s}'. Allowed values: {:?}",
209 0 : StorageClass::values()
210 0 : )));
211 12 : }
212 12 : Ok(Some(storage_class))
213 : } else {
214 3 : Ok(None)
215 : }
216 15 : })
217 15 : }
218 :
219 9 : fn serialize_storage_class<S: serde::Serializer>(
220 9 : val: &Option<StorageClass>,
221 9 : serializer: S,
222 9 : ) -> Result<S::Ok, S::Error> {
223 9 : let val = val.as_ref().map(StorageClass::as_str);
224 9 : Option::<&str>::serialize(&val, serializer)
225 9 : }
226 :
227 : impl RemoteStorageConfig {
228 : pub const DEFAULT_TIMEOUT: Duration = std::time::Duration::from_secs(120);
229 : pub const DEFAULT_SMALL_TIMEOUT: Duration = std::time::Duration::from_secs(30);
230 :
231 10 : pub fn from_toml(toml: &toml_edit::Item) -> anyhow::Result<RemoteStorageConfig> {
232 10 : Ok(utils::toml_edit_ext::deserialize_item(toml)?)
233 10 : }
234 :
235 9 : pub fn from_toml_str(input: &str) -> anyhow::Result<RemoteStorageConfig> {
236 9 : let toml_document = toml_edit::DocumentMut::from_str(input)?;
237 9 : if let Some(item) = toml_document.get("remote_storage") {
238 0 : return Self::from_toml(item);
239 9 : }
240 9 : Self::from_toml(toml_document.as_item())
241 9 : }
242 : }
243 :
244 : #[cfg(test)]
245 : mod tests {
246 : use super::*;
247 :
248 9 : fn parse(input: &str) -> anyhow::Result<RemoteStorageConfig> {
249 9 : RemoteStorageConfig::from_toml_str(input)
250 9 : }
251 :
252 : #[test]
253 3 : fn parse_localfs_config_with_timeout() {
254 3 : let input = "local_path = '.'
255 3 : timeout = '5s'";
256 3 :
257 3 : let config = parse(input).unwrap();
258 3 :
259 3 : assert_eq!(
260 3 : config,
261 3 : RemoteStorageConfig {
262 3 : storage: RemoteStorageKind::LocalFs {
263 3 : local_path: Utf8PathBuf::from(".")
264 3 : },
265 3 : timeout: Duration::from_secs(5),
266 3 : small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
267 3 : }
268 3 : );
269 3 : }
270 :
271 : #[test]
272 3 : fn test_s3_parsing() {
273 3 : let toml = "\
274 3 : bucket_name = 'foo-bar'
275 3 : bucket_region = 'eu-central-1'
276 3 : upload_storage_class = 'INTELLIGENT_TIERING'
277 3 : timeout = '7s'
278 3 : ";
279 3 :
280 3 : let config = parse(toml).unwrap();
281 3 :
282 3 : assert_eq!(
283 3 : config,
284 3 : RemoteStorageConfig {
285 3 : storage: RemoteStorageKind::AwsS3(S3Config {
286 3 : bucket_name: "foo-bar".into(),
287 3 : bucket_region: "eu-central-1".into(),
288 3 : prefix_in_bucket: None,
289 3 : endpoint: None,
290 3 : concurrency_limit: default_remote_storage_s3_concurrency_limit(),
291 3 : max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
292 3 : upload_storage_class: Some(StorageClass::IntelligentTiering),
293 3 : }),
294 3 : timeout: Duration::from_secs(7),
295 3 : small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
296 3 : }
297 3 : );
298 3 : }
299 :
300 : #[test]
301 3 : fn test_storage_class_serde_roundtrip() {
302 3 : let classes = [
303 3 : None,
304 3 : Some(StorageClass::Standard),
305 3 : Some(StorageClass::IntelligentTiering),
306 3 : ];
307 12 : for class in classes {
308 9 : #[derive(Serialize, Deserialize)]
309 9 : struct Wrapper {
310 : #[serde(
311 : deserialize_with = "deserialize_storage_class",
312 : serialize_with = "serialize_storage_class"
313 : )]
314 : class: Option<StorageClass>,
315 : }
316 9 : let wrapped = Wrapper {
317 9 : class: class.clone(),
318 9 : };
319 9 : let serialized = serde_json::to_string(&wrapped).unwrap();
320 9 : let deserialized: Wrapper = serde_json::from_str(&serialized).unwrap();
321 9 : assert_eq!(class, deserialized.class);
322 : }
323 3 : }
324 :
325 : #[test]
326 3 : fn test_azure_parsing() {
327 3 : let toml = "\
328 3 : container_name = 'foo-bar'
329 3 : container_region = 'westeurope'
330 3 : upload_storage_class = 'INTELLIGENT_TIERING'
331 3 : timeout = '7s'
332 3 : conn_pool_size = 8
333 3 : ";
334 3 :
335 3 : let config = parse(toml).unwrap();
336 3 :
337 3 : assert_eq!(
338 3 : config,
339 3 : RemoteStorageConfig {
340 3 : storage: RemoteStorageKind::AzureContainer(AzureConfig {
341 3 : container_name: "foo-bar".into(),
342 3 : storage_account: None,
343 3 : container_region: "westeurope".into(),
344 3 : prefix_in_container: None,
345 3 : concurrency_limit: default_remote_storage_azure_concurrency_limit(),
346 3 : max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
347 3 : conn_pool_size: 8,
348 3 : }),
349 3 : timeout: Duration::from_secs(7),
350 3 : small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
351 3 : }
352 3 : );
353 3 : }
354 : }
|