Line data Source code
1 : //! Production console backend.
2 :
3 : use super::{
4 : super::messages::{ConsoleError, GetRoleSecret, WakeCompute},
5 : errors::{ApiError, GetAuthInfoError, WakeComputeError},
6 : ApiCaches, ApiLocks, AuthInfo, AuthSecret, CachedAllowedIps, CachedNodeInfo, CachedRoleSecret,
7 : NodeInfo,
8 : };
9 : use crate::{
10 : auth::backend::ComputeUserInfo,
11 : compute,
12 : console::messages::{ColdStartInfo, Reason},
13 : http,
14 : metrics::{CacheOutcome, Metrics},
15 : rate_limiter::WakeComputeRateLimiter,
16 : scram, EndpointCacheKey,
17 : };
18 : use crate::{cache::Cached, context::RequestMonitoring};
19 : use futures::TryFutureExt;
20 : use std::{sync::Arc, time::Duration};
21 : use tokio::time::Instant;
22 : use tokio_postgres::config::SslMode;
23 : use tracing::{debug, error, info, info_span, warn, Instrument};
24 :
25 : pub struct Api {
26 : endpoint: http::Endpoint,
27 : pub caches: &'static ApiCaches,
28 : pub locks: &'static ApiLocks<EndpointCacheKey>,
29 : pub wake_compute_endpoint_rate_limiter: Arc<WakeComputeRateLimiter>,
30 : jwt: String,
31 : }
32 :
33 : impl Api {
34 : /// Construct an API object containing the auth parameters.
35 0 : pub fn new(
36 0 : endpoint: http::Endpoint,
37 0 : caches: &'static ApiCaches,
38 0 : locks: &'static ApiLocks<EndpointCacheKey>,
39 0 : wake_compute_endpoint_rate_limiter: Arc<WakeComputeRateLimiter>,
40 0 : ) -> Self {
41 0 : let jwt: String = match std::env::var("NEON_PROXY_TO_CONTROLPLANE_TOKEN") {
42 0 : Ok(v) => v,
43 0 : Err(_) => "".to_string(),
44 : };
45 0 : Self {
46 0 : endpoint,
47 0 : caches,
48 0 : locks,
49 0 : wake_compute_endpoint_rate_limiter,
50 0 : jwt,
51 0 : }
52 0 : }
53 :
54 0 : pub fn url(&self) -> &str {
55 0 : self.endpoint.url().as_str()
56 0 : }
57 :
58 0 : async fn do_get_auth_info(
59 0 : &self,
60 0 : ctx: &mut RequestMonitoring,
61 0 : user_info: &ComputeUserInfo,
62 0 : ) -> Result<AuthInfo, GetAuthInfoError> {
63 0 : if !self
64 0 : .caches
65 0 : .endpoints_cache
66 0 : .is_valid(ctx, &user_info.endpoint.normalize())
67 0 : .await
68 : {
69 0 : info!("endpoint is not valid, skipping the request");
70 0 : return Ok(AuthInfo::default());
71 0 : }
72 0 : let request_id = ctx.session_id.to_string();
73 0 : let application_name = ctx.console_application_name();
74 0 : async {
75 0 : let request = self
76 0 : .endpoint
77 0 : .get("proxy_get_role_secret")
78 0 : .header("X-Request-ID", &request_id)
79 0 : .header("Authorization", format!("Bearer {}", &self.jwt))
80 0 : .query(&[("session_id", ctx.session_id)])
81 0 : .query(&[
82 0 : ("application_name", application_name.as_str()),
83 0 : ("project", user_info.endpoint.as_str()),
84 0 : ("role", user_info.user.as_str()),
85 0 : ])
86 0 : .build()?;
87 :
88 0 : info!(url = request.url().as_str(), "sending http request");
89 0 : let start = Instant::now();
90 0 : let pause = ctx.latency_timer.pause(crate::metrics::Waiting::Cplane);
91 0 : let response = self.endpoint.execute(request).await?;
92 0 : drop(pause);
93 0 : info!(duration = ?start.elapsed(), "received http response");
94 0 : let body = match parse_body::<GetRoleSecret>(response).await {
95 0 : Ok(body) => body,
96 : // Error 404 is special: it's ok not to have a secret.
97 : // TODO(anna): retry
98 0 : Err(e) => {
99 0 : if e.get_reason().is_not_found() {
100 0 : return Ok(AuthInfo::default());
101 : } else {
102 0 : return Err(e.into());
103 : }
104 : }
105 : };
106 :
107 0 : let secret = if body.role_secret.is_empty() {
108 0 : None
109 : } else {
110 0 : let secret = scram::ServerSecret::parse(&body.role_secret)
111 0 : .map(AuthSecret::Scram)
112 0 : .ok_or(GetAuthInfoError::BadSecret)?;
113 0 : Some(secret)
114 : };
115 0 : let allowed_ips = body.allowed_ips.unwrap_or_default();
116 0 : Metrics::get()
117 0 : .proxy
118 0 : .allowed_ips_number
119 0 : .observe(allowed_ips.len() as f64);
120 0 : Ok(AuthInfo {
121 0 : secret,
122 0 : allowed_ips,
123 0 : project_id: body.project_id,
124 0 : })
125 0 : }
126 0 : .map_err(crate::error::log_error)
127 0 : .instrument(info_span!("http", id = request_id))
128 0 : .await
129 0 : }
130 :
131 0 : async fn do_wake_compute(
132 0 : &self,
133 0 : ctx: &mut RequestMonitoring,
134 0 : user_info: &ComputeUserInfo,
135 0 : ) -> Result<NodeInfo, WakeComputeError> {
136 0 : let request_id = ctx.session_id.to_string();
137 0 : let application_name = ctx.console_application_name();
138 0 : async {
139 0 : let mut request_builder = self
140 0 : .endpoint
141 0 : .get("proxy_wake_compute")
142 0 : .header("X-Request-ID", &request_id)
143 0 : .header("Authorization", format!("Bearer {}", &self.jwt))
144 0 : .query(&[("session_id", ctx.session_id)])
145 0 : .query(&[
146 0 : ("application_name", application_name.as_str()),
147 0 : ("project", user_info.endpoint.as_str()),
148 0 : ]);
149 0 :
150 0 : let options = user_info.options.to_deep_object();
151 0 : if !options.is_empty() {
152 0 : request_builder = request_builder.query(&options);
153 0 : }
154 :
155 0 : let request = request_builder.build()?;
156 :
157 0 : info!(url = request.url().as_str(), "sending http request");
158 0 : let start = Instant::now();
159 0 : let pause = ctx.latency_timer.pause(crate::metrics::Waiting::Cplane);
160 0 : let response = self.endpoint.execute(request).await?;
161 0 : drop(pause);
162 0 : info!(duration = ?start.elapsed(), "received http response");
163 0 : let body = parse_body::<WakeCompute>(response).await?;
164 :
165 : // Unfortunately, ownership won't let us use `Option::ok_or` here.
166 0 : let (host, port) = match parse_host_port(&body.address) {
167 0 : None => return Err(WakeComputeError::BadComputeAddress(body.address)),
168 0 : Some(x) => x,
169 0 : };
170 0 :
171 0 : // Don't set anything but host and port! This config will be cached.
172 0 : // We'll set username and such later using the startup message.
173 0 : // TODO: add more type safety (in progress).
174 0 : let mut config = compute::ConnCfg::new();
175 0 : config.host(host).port(port).ssl_mode(SslMode::Disable); // TLS is not configured on compute nodes.
176 0 :
177 0 : let node = NodeInfo {
178 0 : config,
179 0 : aux: body.aux,
180 0 : allow_self_signed_compute: false,
181 0 : };
182 0 :
183 0 : Ok(node)
184 0 : }
185 0 : .map_err(crate::error::log_error)
186 0 : .instrument(info_span!("http", id = request_id))
187 0 : .await
188 0 : }
189 : }
190 :
191 : impl super::Api for Api {
192 0 : #[tracing::instrument(skip_all)]
193 : async fn get_role_secret(
194 : &self,
195 : ctx: &mut RequestMonitoring,
196 : user_info: &ComputeUserInfo,
197 : ) -> Result<CachedRoleSecret, GetAuthInfoError> {
198 : let normalized_ep = &user_info.endpoint.normalize();
199 : let user = &user_info.user;
200 : if let Some(role_secret) = self
201 : .caches
202 : .project_info
203 : .get_role_secret(normalized_ep, user)
204 : {
205 : return Ok(role_secret);
206 : }
207 : let auth_info = self.do_get_auth_info(ctx, user_info).await?;
208 : if let Some(project_id) = auth_info.project_id {
209 : let normalized_ep_int = normalized_ep.into();
210 : self.caches.project_info.insert_role_secret(
211 : project_id,
212 : normalized_ep_int,
213 : user.into(),
214 : auth_info.secret.clone(),
215 : );
216 : self.caches.project_info.insert_allowed_ips(
217 : project_id,
218 : normalized_ep_int,
219 : Arc::new(auth_info.allowed_ips),
220 : );
221 : ctx.set_project_id(project_id);
222 : }
223 : // When we just got a secret, we don't need to invalidate it.
224 : Ok(Cached::new_uncached(auth_info.secret))
225 : }
226 :
227 0 : async fn get_allowed_ips_and_secret(
228 0 : &self,
229 0 : ctx: &mut RequestMonitoring,
230 0 : user_info: &ComputeUserInfo,
231 0 : ) -> Result<(CachedAllowedIps, Option<CachedRoleSecret>), GetAuthInfoError> {
232 0 : let normalized_ep = &user_info.endpoint.normalize();
233 0 : if let Some(allowed_ips) = self.caches.project_info.get_allowed_ips(normalized_ep) {
234 0 : Metrics::get()
235 0 : .proxy
236 0 : .allowed_ips_cache_misses
237 0 : .inc(CacheOutcome::Hit);
238 0 : return Ok((allowed_ips, None));
239 0 : }
240 0 : Metrics::get()
241 0 : .proxy
242 0 : .allowed_ips_cache_misses
243 0 : .inc(CacheOutcome::Miss);
244 0 : let auth_info = self.do_get_auth_info(ctx, user_info).await?;
245 0 : let allowed_ips = Arc::new(auth_info.allowed_ips);
246 0 : let user = &user_info.user;
247 0 : if let Some(project_id) = auth_info.project_id {
248 0 : let normalized_ep_int = normalized_ep.into();
249 0 : self.caches.project_info.insert_role_secret(
250 0 : project_id,
251 0 : normalized_ep_int,
252 0 : user.into(),
253 0 : auth_info.secret.clone(),
254 0 : );
255 0 : self.caches.project_info.insert_allowed_ips(
256 0 : project_id,
257 0 : normalized_ep_int,
258 0 : allowed_ips.clone(),
259 0 : );
260 0 : ctx.set_project_id(project_id);
261 0 : }
262 0 : Ok((
263 0 : Cached::new_uncached(allowed_ips),
264 0 : Some(Cached::new_uncached(auth_info.secret)),
265 0 : ))
266 0 : }
267 :
268 0 : #[tracing::instrument(skip_all)]
269 : async fn wake_compute(
270 : &self,
271 : ctx: &mut RequestMonitoring,
272 : user_info: &ComputeUserInfo,
273 : ) -> Result<CachedNodeInfo, WakeComputeError> {
274 : let key = user_info.endpoint_cache_key();
275 :
276 : macro_rules! check_cache {
277 : () => {
278 : if let Some(cached) = self.caches.node_info.get(&key) {
279 : let (cached, info) = cached.take_value();
280 0 : let info = info.map_err(|c| {
281 0 : info!(key = &*key, "found cached wake_compute error");
282 0 : WakeComputeError::ApiError(ApiError::Console(*c))
283 0 : })?;
284 :
285 : debug!(key = &*key, "found cached compute node info");
286 : ctx.set_project(info.aux.clone());
287 0 : return Ok(cached.map(|()| info));
288 : }
289 : };
290 : }
291 :
292 : // Every time we do a wakeup http request, the compute node will stay up
293 : // for some time (highly depends on the console's scale-to-zero policy);
294 : // The connection info remains the same during that period of time,
295 : // which means that we might cache it to reduce the load and latency.
296 : check_cache!();
297 :
298 : let permit = self.locks.get_permit(&key).await?;
299 :
300 : // after getting back a permit - it's possible the cache was filled
301 : // double check
302 : if permit.should_check_cache() {
303 : check_cache!();
304 : }
305 :
306 : // check rate limit
307 : if !self
308 : .wake_compute_endpoint_rate_limiter
309 : .check(user_info.endpoint.normalize_intern(), 1)
310 : {
311 : return Err(WakeComputeError::TooManyConnections);
312 : }
313 :
314 : let node = permit.release_result(self.do_wake_compute(ctx, user_info).await);
315 : match node {
316 : Ok(node) => {
317 : ctx.set_project(node.aux.clone());
318 : debug!(key = &*key, "created a cache entry for woken compute node");
319 :
320 : let mut stored_node = node.clone();
321 : // store the cached node as 'warm_cached'
322 : stored_node.aux.cold_start_info = ColdStartInfo::WarmCached;
323 :
324 : let (_, cached) = self.caches.node_info.insert_unit(key, Ok(stored_node));
325 :
326 0 : Ok(cached.map(|()| node))
327 : }
328 : Err(err) => match err {
329 : WakeComputeError::ApiError(ApiError::Console(err)) => {
330 : let Some(status) = &err.status else {
331 : return Err(WakeComputeError::ApiError(ApiError::Console(err)));
332 : };
333 :
334 : let reason = status
335 : .details
336 : .error_info
337 0 : .map_or(Reason::Unknown, |x| x.reason);
338 :
339 : // if we can retry this error, do not cache it.
340 : if reason.can_retry() {
341 : return Err(WakeComputeError::ApiError(ApiError::Console(err)));
342 : }
343 :
344 : // at this point, we should only have quota errors.
345 : debug!(
346 : key = &*key,
347 : "created a cache entry for the wake compute error"
348 : );
349 :
350 : self.caches.node_info.insert_ttl(
351 : key,
352 : Err(Box::new(err.clone())),
353 : Duration::from_secs(30),
354 : );
355 :
356 : Err(WakeComputeError::ApiError(ApiError::Console(err)))
357 : }
358 : err => return Err(err),
359 : },
360 : }
361 : }
362 : }
363 :
364 : /// Parse http response body, taking status code into account.
365 0 : async fn parse_body<T: for<'a> serde::Deserialize<'a>>(
366 0 : response: http::Response,
367 0 : ) -> Result<T, ApiError> {
368 0 : let status = response.status();
369 0 : if status.is_success() {
370 : // We shouldn't log raw body because it may contain secrets.
371 0 : info!("request succeeded, processing the body");
372 0 : return Ok(response.json().await?);
373 0 : }
374 0 : let s = response.bytes().await?;
375 : // Log plaintext to be able to detect, whether there are some cases not covered by the error struct.
376 0 : info!("response_error plaintext: {:?}", s);
377 :
378 : // Don't throw an error here because it's not as important
379 : // as the fact that the request itself has failed.
380 0 : let mut body = serde_json::from_slice(&s).unwrap_or_else(|e| {
381 0 : warn!("failed to parse error body: {e}");
382 0 : ConsoleError {
383 0 : error: "reason unclear (malformed error message)".into(),
384 0 : http_status_code: status,
385 0 : status: None,
386 0 : }
387 0 : });
388 0 : body.http_status_code = status;
389 0 :
390 0 : error!("console responded with an error ({status}): {body:?}");
391 0 : Err(ApiError::Console(body))
392 0 : }
393 :
394 6 : fn parse_host_port(input: &str) -> Option<(&str, u16)> {
395 6 : let (host, port) = input.rsplit_once(':')?;
396 6 : let ipv6_brackets: &[_] = &['[', ']'];
397 6 : Some((host.trim_matches(ipv6_brackets), port.parse().ok()?))
398 6 : }
399 :
400 : #[cfg(test)]
401 : mod tests {
402 : use super::*;
403 :
404 : #[test]
405 2 : fn test_parse_host_port_v4() {
406 2 : let (host, port) = parse_host_port("127.0.0.1:5432").expect("failed to parse");
407 2 : assert_eq!(host, "127.0.0.1");
408 2 : assert_eq!(port, 5432);
409 2 : }
410 :
411 : #[test]
412 2 : fn test_parse_host_port_v6() {
413 2 : let (host, port) = parse_host_port("[2001:db8::1]:5432").expect("failed to parse");
414 2 : assert_eq!(host, "2001:db8::1");
415 2 : assert_eq!(port, 5432);
416 2 : }
417 :
418 : #[test]
419 2 : fn test_parse_host_port_url() {
420 2 : let (host, port) = parse_host_port("compute-foo-bar-1234.default.svc.cluster.local:5432")
421 2 : .expect("failed to parse");
422 2 : assert_eq!(host, "compute-foo-bar-1234.default.svc.cluster.local");
423 2 : assert_eq!(port, 5432);
424 2 : }
425 : }
|
