Line data Source code
1 : use std::fmt::{self, Display};
2 :
3 : use measured::FixedCardinalityLabel;
4 : use serde::{Deserialize, Serialize};
5 :
6 : use crate::auth::IpPattern;
7 : use crate::intern::{BranchIdInt, EndpointIdInt, ProjectIdInt, RoleNameInt};
8 : use crate::proxy::retry::CouldRetry;
9 :
10 : /// Generic error response with human-readable description.
11 : /// Note that we can't always present it to user as is.
12 0 : #[derive(Debug, Deserialize, Clone)]
13 : pub(crate) struct ControlPlaneErrorMessage {
14 : pub(crate) error: Box<str>,
15 : #[serde(skip)]
16 : pub(crate) http_status_code: http::StatusCode,
17 : pub(crate) status: Option<Status>,
18 : }
19 :
20 : impl ControlPlaneErrorMessage {
21 3 : pub(crate) fn get_reason(&self) -> Reason {
22 3 : self.status
23 3 : .as_ref()
24 3 : .and_then(|s| s.details.error_info.as_ref())
25 3 : .map_or(Reason::Unknown, |e| e.reason)
26 3 : }
27 :
28 0 : pub(crate) fn get_user_facing_message(&self) -> String {
29 : use super::errors::REQUEST_FAILED;
30 0 : self.status
31 0 : .as_ref()
32 0 : .and_then(|s| s.details.user_facing_message.as_ref())
33 0 : .map_or_else(|| {
34 0 : // Ask @neondatabase/control-plane for review before adding more.
35 0 : match self.http_status_code {
36 : http::StatusCode::NOT_FOUND => {
37 : // Status 404: failed to get a project-related resource.
38 0 : format!("{REQUEST_FAILED}: endpoint cannot be found")
39 : }
40 : http::StatusCode::NOT_ACCEPTABLE => {
41 : // Status 406: endpoint is disabled (we don't allow connections).
42 0 : format!("{REQUEST_FAILED}: endpoint is disabled")
43 : }
44 : http::StatusCode::LOCKED | http::StatusCode::UNPROCESSABLE_ENTITY => {
45 : // Status 423: project might be in maintenance mode (or bad state), or quotas exceeded.
46 0 : format!("{REQUEST_FAILED}: endpoint is temporarily unavailable. Check your quotas and/or contact our support.")
47 : }
48 0 : _ => REQUEST_FAILED.to_owned(),
49 : }
50 0 : }, |m| m.message.clone().into())
51 0 : }
52 : }
53 :
54 : impl Display for ControlPlaneErrorMessage {
55 0 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
56 0 : let msg: &str = self
57 0 : .status
58 0 : .as_ref()
59 0 : .and_then(|s| s.details.user_facing_message.as_ref())
60 0 : .map_or_else(|| self.error.as_ref(), |m| m.message.as_ref());
61 0 : write!(f, "{msg}")
62 0 : }
63 : }
64 :
65 : impl CouldRetry for ControlPlaneErrorMessage {
66 6 : fn could_retry(&self) -> bool {
67 : // If the error message does not have a status,
68 : // the error is unknown and probably should not retry automatically
69 6 : let Some(status) = &self.status else {
70 2 : return false;
71 : };
72 :
73 : // retry if the retry info is set.
74 4 : if status.details.retry_info.is_some() {
75 4 : return true;
76 0 : }
77 0 :
78 0 : // if no retry info set, attempt to use the error code to guess the retry state.
79 0 : let reason = status
80 0 : .details
81 0 : .error_info
82 0 : .map_or(Reason::Unknown, |e| e.reason);
83 0 :
84 0 : reason.can_retry()
85 6 : }
86 : }
87 :
88 0 : #[derive(Debug, Deserialize, Clone)]
89 : #[allow(dead_code)]
90 : pub(crate) struct Status {
91 : pub(crate) code: Box<str>,
92 : pub(crate) message: Box<str>,
93 : pub(crate) details: Details,
94 : }
95 :
96 0 : #[derive(Debug, Deserialize, Clone)]
97 : pub(crate) struct Details {
98 : pub(crate) error_info: Option<ErrorInfo>,
99 : pub(crate) retry_info: Option<RetryInfo>,
100 : pub(crate) user_facing_message: Option<UserFacingMessage>,
101 : }
102 :
103 0 : #[derive(Copy, Clone, Debug, Deserialize)]
104 : pub(crate) struct ErrorInfo {
105 : pub(crate) reason: Reason,
106 : // Schema could also have `metadata` field, but it's not structured. Skip it for now.
107 : }
108 :
109 0 : #[derive(Clone, Copy, Debug, Deserialize, Default)]
110 : pub(crate) enum Reason {
111 : /// RoleProtected indicates that the role is protected and the attempted operation is not permitted on protected roles.
112 : #[serde(rename = "ROLE_PROTECTED")]
113 : RoleProtected,
114 : /// ResourceNotFound indicates that a resource (project, endpoint, branch, etc.) wasn't found,
115 : /// usually due to the provided ID not being correct or because the subject doesn't have enough permissions to
116 : /// access the requested resource.
117 : /// Prefer a more specific reason if possible, e.g., ProjectNotFound, EndpointNotFound, etc.
118 : #[serde(rename = "RESOURCE_NOT_FOUND")]
119 : ResourceNotFound,
120 : /// ProjectNotFound indicates that the project wasn't found, usually due to the provided ID not being correct,
121 : /// or that the subject doesn't have enough permissions to access the requested project.
122 : #[serde(rename = "PROJECT_NOT_FOUND")]
123 : ProjectNotFound,
124 : /// EndpointNotFound indicates that the endpoint wasn't found, usually due to the provided ID not being correct,
125 : /// or that the subject doesn't have enough permissions to access the requested endpoint.
126 : #[serde(rename = "ENDPOINT_NOT_FOUND")]
127 : EndpointNotFound,
128 : /// BranchNotFound indicates that the branch wasn't found, usually due to the provided ID not being correct,
129 : /// or that the subject doesn't have enough permissions to access the requested branch.
130 : #[serde(rename = "BRANCH_NOT_FOUND")]
131 : BranchNotFound,
132 : /// RateLimitExceeded indicates that the rate limit for the operation has been exceeded.
133 : #[serde(rename = "RATE_LIMIT_EXCEEDED")]
134 : RateLimitExceeded,
135 : /// NonDefaultBranchComputeTimeExceeded indicates that the compute time quota of non-default branches has been
136 : /// exceeded.
137 : #[serde(rename = "NON_PRIMARY_BRANCH_COMPUTE_TIME_EXCEEDED")]
138 : NonDefaultBranchComputeTimeExceeded,
139 : /// ActiveTimeQuotaExceeded indicates that the active time quota was exceeded.
140 : #[serde(rename = "ACTIVE_TIME_QUOTA_EXCEEDED")]
141 : ActiveTimeQuotaExceeded,
142 : /// ComputeTimeQuotaExceeded indicates that the compute time quota was exceeded.
143 : #[serde(rename = "COMPUTE_TIME_QUOTA_EXCEEDED")]
144 : ComputeTimeQuotaExceeded,
145 : /// WrittenDataQuotaExceeded indicates that the written data quota was exceeded.
146 : #[serde(rename = "WRITTEN_DATA_QUOTA_EXCEEDED")]
147 : WrittenDataQuotaExceeded,
148 : /// DataTransferQuotaExceeded indicates that the data transfer quota was exceeded.
149 : #[serde(rename = "DATA_TRANSFER_QUOTA_EXCEEDED")]
150 : DataTransferQuotaExceeded,
151 : /// LogicalSizeQuotaExceeded indicates that the logical size quota was exceeded.
152 : #[serde(rename = "LOGICAL_SIZE_QUOTA_EXCEEDED")]
153 : LogicalSizeQuotaExceeded,
154 : /// RunningOperations indicates that the project already has some running operations
155 : /// and scheduling of new ones is prohibited.
156 : #[serde(rename = "RUNNING_OPERATIONS")]
157 : RunningOperations,
158 : /// ConcurrencyLimitReached indicates that the concurrency limit for an action was reached.
159 : #[serde(rename = "CONCURRENCY_LIMIT_REACHED")]
160 : ConcurrencyLimitReached,
161 : /// LockAlreadyTaken indicates that the we attempted to take a lock that was already taken.
162 : #[serde(rename = "LOCK_ALREADY_TAKEN")]
163 : LockAlreadyTaken,
164 : /// ActiveEndpointsLimitExceeded indicates that the limit of concurrently active endpoints was exceeded.
165 : #[serde(rename = "ACTIVE_ENDPOINTS_LIMIT_EXCEEDED")]
166 : ActiveEndpointsLimitExceeded,
167 : #[default]
168 : #[serde(other)]
169 : Unknown,
170 : }
171 :
172 : impl Reason {
173 0 : pub(crate) fn is_not_found(self) -> bool {
174 0 : matches!(
175 0 : self,
176 : Reason::ResourceNotFound
177 : | Reason::ProjectNotFound
178 : | Reason::EndpointNotFound
179 : | Reason::BranchNotFound
180 : )
181 0 : }
182 :
183 0 : pub(crate) fn can_retry(self) -> bool {
184 0 : match self {
185 : // do not retry role protected errors
186 : // not a transitive error
187 0 : Reason::RoleProtected => false,
188 : // on retry, it will still not be found
189 : Reason::ResourceNotFound
190 : | Reason::ProjectNotFound
191 : | Reason::EndpointNotFound
192 0 : | Reason::BranchNotFound => false,
193 : // we were asked to go away
194 : Reason::RateLimitExceeded
195 : | Reason::NonDefaultBranchComputeTimeExceeded
196 : | Reason::ActiveTimeQuotaExceeded
197 : | Reason::ComputeTimeQuotaExceeded
198 : | Reason::WrittenDataQuotaExceeded
199 : | Reason::DataTransferQuotaExceeded
200 : | Reason::LogicalSizeQuotaExceeded
201 0 : | Reason::ActiveEndpointsLimitExceeded => false,
202 : // transitive error. control plane is currently busy
203 : // but might be ready soon
204 : Reason::RunningOperations
205 : | Reason::ConcurrencyLimitReached
206 0 : | Reason::LockAlreadyTaken => true,
207 : // unknown error. better not retry it.
208 0 : Reason::Unknown => false,
209 : }
210 0 : }
211 : }
212 :
213 0 : #[derive(Copy, Clone, Debug, Deserialize)]
214 : #[allow(dead_code)]
215 : pub(crate) struct RetryInfo {
216 : pub(crate) retry_delay_ms: u64,
217 : }
218 :
219 0 : #[derive(Debug, Deserialize, Clone)]
220 : pub(crate) struct UserFacingMessage {
221 : pub(crate) message: Box<str>,
222 : }
223 :
224 : /// Response which holds client's auth secret, e.g. [`crate::scram::ServerSecret`].
225 : /// Returned by the `/get_endpoint_access_control` API method.
226 9 : #[derive(Deserialize)]
227 : pub(crate) struct GetEndpointAccessControl {
228 : pub(crate) role_secret: Box<str>,
229 : pub(crate) allowed_ips: Option<Vec<IpPattern>>,
230 : pub(crate) project_id: Option<ProjectIdInt>,
231 : pub(crate) allowed_vpc_endpoint_ids: Option<Vec<EndpointIdInt>>,
232 : }
233 :
234 : /// Response which holds compute node's `host:port` pair.
235 : /// Returned by the `/proxy_wake_compute` API method.
236 3 : #[derive(Debug, Deserialize)]
237 : pub(crate) struct WakeCompute {
238 : pub(crate) address: Box<str>,
239 : pub(crate) aux: MetricsAuxInfo,
240 : }
241 :
242 : /// Async response which concludes the console redirect auth flow.
243 : /// Also known as `kickResponse` in the console.
244 4 : #[derive(Debug, Deserialize)]
245 : pub(crate) struct KickSession<'a> {
246 : /// Session ID is assigned by the proxy.
247 : pub(crate) session_id: &'a str,
248 :
249 : /// Compute node connection params.
250 : #[serde(deserialize_with = "KickSession::parse_db_info")]
251 : pub(crate) result: DatabaseInfo,
252 : }
253 :
254 : impl KickSession<'_> {
255 1 : fn parse_db_info<'de, D>(des: D) -> Result<DatabaseInfo, D::Error>
256 1 : where
257 1 : D: serde::Deserializer<'de>,
258 1 : {
259 2 : #[derive(Deserialize)]
260 : enum Wrapper {
261 : // Currently, console only reports `Success`.
262 : // `Failure(String)` used to be here... RIP.
263 : Success(DatabaseInfo),
264 : }
265 :
266 1 : Wrapper::deserialize(des).map(|x| match x {
267 1 : Wrapper::Success(info) => info,
268 1 : })
269 1 : }
270 : }
271 :
272 : /// Compute node connection params.
273 36 : #[derive(Deserialize)]
274 : pub(crate) struct DatabaseInfo {
275 : pub(crate) host: Box<str>,
276 : pub(crate) port: u16,
277 : pub(crate) dbname: Box<str>,
278 : pub(crate) user: Box<str>,
279 : /// Console always provides a password, but it might
280 : /// be inconvenient for debug with local PG instance.
281 : pub(crate) password: Option<Box<str>>,
282 : pub(crate) aux: MetricsAuxInfo,
283 : #[serde(default)]
284 : pub(crate) allowed_ips: Option<Vec<IpPattern>>,
285 : }
286 :
287 : // Manually implement debug to omit sensitive info.
288 : impl fmt::Debug for DatabaseInfo {
289 0 : fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
290 0 : f.debug_struct("DatabaseInfo")
291 0 : .field("host", &self.host)
292 0 : .field("port", &self.port)
293 0 : .field("dbname", &self.dbname)
294 0 : .field("user", &self.user)
295 0 : .field("allowed_ips", &self.allowed_ips)
296 0 : .finish_non_exhaustive()
297 0 : }
298 : }
299 :
300 : /// Various labels for prometheus metrics.
301 : /// Also known as `ProxyMetricsAuxInfo` in the console.
302 30 : #[derive(Debug, Deserialize, Clone)]
303 : pub(crate) struct MetricsAuxInfo {
304 : pub(crate) endpoint_id: EndpointIdInt,
305 : pub(crate) project_id: ProjectIdInt,
306 : pub(crate) branch_id: BranchIdInt,
307 : #[serde(default)]
308 : pub(crate) cold_start_info: ColdStartInfo,
309 : }
310 :
311 12 : #[derive(Debug, Default, Serialize, Deserialize, Clone, Copy, FixedCardinalityLabel)]
312 : #[serde(rename_all = "snake_case")]
313 : pub enum ColdStartInfo {
314 : #[default]
315 : Unknown,
316 : /// Compute was already running
317 : Warm,
318 : #[serde(rename = "pool_hit")]
319 : #[label(rename = "pool_hit")]
320 : /// Compute was not running but there was an available VM
321 : VmPoolHit,
322 : #[serde(rename = "pool_miss")]
323 : #[label(rename = "pool_miss")]
324 : /// Compute was not running and there were no VMs available
325 : VmPoolMiss,
326 :
327 : // not provided by control plane
328 : /// Connection available from HTTP pool
329 : HttpPoolHit,
330 : /// Cached connection info
331 : WarmCached,
332 : }
333 :
334 : impl ColdStartInfo {
335 0 : pub(crate) fn as_str(self) -> &'static str {
336 0 : match self {
337 0 : ColdStartInfo::Unknown => "unknown",
338 0 : ColdStartInfo::Warm => "warm",
339 0 : ColdStartInfo::VmPoolHit => "pool_hit",
340 0 : ColdStartInfo::VmPoolMiss => "pool_miss",
341 0 : ColdStartInfo::HttpPoolHit => "http_pool_hit",
342 0 : ColdStartInfo::WarmCached => "warm_cached",
343 : }
344 0 : }
345 : }
346 :
347 0 : #[derive(Debug, Deserialize, Clone)]
348 : pub struct EndpointJwksResponse {
349 : pub jwks: Vec<JwksSettings>,
350 : }
351 :
352 0 : #[derive(Debug, Deserialize, Clone)]
353 : pub struct JwksSettings {
354 : pub id: String,
355 : pub jwks_url: url::Url,
356 : pub provider_name: String,
357 : pub jwt_audience: Option<String>,
358 : pub role_names: Vec<RoleNameInt>,
359 : }
360 :
361 : #[cfg(test)]
362 : mod tests {
363 : use serde_json::json;
364 :
365 : use super::*;
366 :
367 6 : fn dummy_aux() -> serde_json::Value {
368 6 : json!({
369 6 : "endpoint_id": "endpoint",
370 6 : "project_id": "project",
371 6 : "branch_id": "branch",
372 6 : "cold_start_info": "unknown",
373 6 : })
374 6 : }
375 :
376 : #[test]
377 1 : fn parse_kick_session() -> anyhow::Result<()> {
378 1 : // This is what the console's kickResponse looks like.
379 1 : let json = json!({
380 1 : "session_id": "deadbeef",
381 1 : "result": {
382 1 : "Success": {
383 1 : "host": "localhost",
384 1 : "port": 5432,
385 1 : "dbname": "postgres",
386 1 : "user": "john_doe",
387 1 : "password": "password",
388 1 : "aux": dummy_aux(),
389 1 : }
390 1 : }
391 1 : });
392 1 : serde_json::from_str::<KickSession<'_>>(&json.to_string())?;
393 :
394 1 : Ok(())
395 1 : }
396 :
397 : #[test]
398 1 : fn parse_db_info() -> anyhow::Result<()> {
399 1 : // with password
400 1 : serde_json::from_value::<DatabaseInfo>(json!({
401 1 : "host": "localhost",
402 1 : "port": 5432,
403 1 : "dbname": "postgres",
404 1 : "user": "john_doe",
405 1 : "password": "password",
406 1 : "aux": dummy_aux(),
407 1 : }))?;
408 :
409 : // without password
410 1 : serde_json::from_value::<DatabaseInfo>(json!({
411 1 : "host": "localhost",
412 1 : "port": 5432,
413 1 : "dbname": "postgres",
414 1 : "user": "john_doe",
415 1 : "aux": dummy_aux(),
416 1 : }))?;
417 :
418 : // new field (forward compatibility)
419 1 : serde_json::from_value::<DatabaseInfo>(json!({
420 1 : "host": "localhost",
421 1 : "port": 5432,
422 1 : "dbname": "postgres",
423 1 : "user": "john_doe",
424 1 : "project": "hello_world",
425 1 : "N.E.W": "forward compatibility check",
426 1 : "aux": dummy_aux(),
427 1 : }))?;
428 :
429 : // with allowed_ips
430 1 : let dbinfo = serde_json::from_value::<DatabaseInfo>(json!({
431 1 : "host": "localhost",
432 1 : "port": 5432,
433 1 : "dbname": "postgres",
434 1 : "user": "john_doe",
435 1 : "password": "password",
436 1 : "aux": dummy_aux(),
437 1 : "allowed_ips": ["127.0.0.1"],
438 1 : }))?;
439 :
440 1 : assert_eq!(
441 1 : dbinfo.allowed_ips,
442 1 : Some(vec![IpPattern::Single("127.0.0.1".parse()?)])
443 : );
444 :
445 1 : Ok(())
446 1 : }
447 :
448 : #[test]
449 1 : fn parse_wake_compute() -> anyhow::Result<()> {
450 1 : let json = json!({
451 1 : "address": "0.0.0.0",
452 1 : "aux": dummy_aux(),
453 1 : });
454 1 : serde_json::from_str::<WakeCompute>(&json.to_string())?;
455 1 : Ok(())
456 1 : }
457 :
458 : #[test]
459 1 : fn parse_get_role_secret() -> anyhow::Result<()> {
460 1 : // Empty `allowed_ips` field.
461 1 : let json = json!({
462 1 : "role_secret": "secret",
463 1 : });
464 1 : serde_json::from_str::<GetEndpointAccessControl>(&json.to_string())?;
465 1 : let json = json!({
466 1 : "role_secret": "secret",
467 1 : "allowed_ips": ["8.8.8.8"],
468 1 : });
469 1 : serde_json::from_str::<GetEndpointAccessControl>(&json.to_string())?;
470 1 : let json = json!({
471 1 : "role_secret": "secret",
472 1 : "allowed_ips": ["8.8.8.8"],
473 1 : "project_id": "project",
474 1 : });
475 1 : serde_json::from_str::<GetEndpointAccessControl>(&json.to_string())?;
476 :
477 1 : Ok(())
478 1 : }
479 : }
|
