Line data Source code
1 : //! Connection request monitoring contexts
2 :
3 : use std::net::IpAddr;
4 :
5 : use chrono::Utc;
6 : use once_cell::sync::OnceCell;
7 : use pq_proto::StartupMessageParams;
8 : use smol_str::SmolStr;
9 : use tokio::sync::mpsc;
10 : use tracing::field::display;
11 : use tracing::{Span, debug, error, info_span};
12 : use try_lock::TryLock;
13 : use uuid::Uuid;
14 :
15 : use self::parquet::RequestData;
16 : use crate::control_plane::messages::{ColdStartInfo, MetricsAuxInfo};
17 : use crate::error::ErrorKind;
18 : use crate::intern::{BranchIdInt, ProjectIdInt};
19 : use crate::metrics::{
20 : ConnectOutcome, InvalidEndpointsGroup, LatencyTimer, Metrics, Protocol, Waiting,
21 : };
22 : use crate::protocol2::{ConnectionInfo, ConnectionInfoExtra};
23 : use crate::types::{DbName, EndpointId, RoleName};
24 :
25 : pub mod parquet;
26 :
27 : pub(crate) static LOG_CHAN: OnceCell<mpsc::WeakUnboundedSender<RequestData>> = OnceCell::new();
28 : pub(crate) static LOG_CHAN_DISCONNECT: OnceCell<mpsc::WeakUnboundedSender<RequestData>> =
29 : OnceCell::new();
30 :
31 : /// Context data for a single request to connect to a database.
32 : ///
33 : /// This data should **not** be used for connection logic, only for observability and limiting purposes.
34 : /// All connection logic should instead use strongly typed state machines, not a bunch of Options.
35 : pub struct RequestContext(
36 : /// To allow easier use of the ctx object, we have interior mutability.
37 : /// I would typically use a RefCell but that would break the `Send` requirements
38 : /// so we need something with thread-safety. `TryLock` is a cheap alternative
39 : /// that offers similar semantics to a `RefCell` but with synchronisation.
40 : TryLock<RequestContextInner>,
41 : );
42 :
43 : struct RequestContextInner {
44 : pub(crate) conn_info: ConnectionInfo,
45 : pub(crate) session_id: Uuid,
46 : pub(crate) protocol: Protocol,
47 : first_packet: chrono::DateTime<Utc>,
48 : region: &'static str,
49 : pub(crate) span: Span,
50 :
51 : // filled in as they are discovered
52 : project: Option<ProjectIdInt>,
53 : branch: Option<BranchIdInt>,
54 : endpoint_id: Option<EndpointId>,
55 : dbname: Option<DbName>,
56 : user: Option<RoleName>,
57 : application: Option<SmolStr>,
58 : user_agent: Option<SmolStr>,
59 : error_kind: Option<ErrorKind>,
60 : pub(crate) auth_method: Option<AuthMethod>,
61 : jwt_issuer: Option<String>,
62 : success: bool,
63 : pub(crate) cold_start_info: ColdStartInfo,
64 : pg_options: Option<StartupMessageParams>,
65 :
66 : // extra
67 : // This sender is here to keep the request monitoring channel open while requests are taking place.
68 : sender: Option<mpsc::UnboundedSender<RequestData>>,
69 : // This sender is only used to log the length of session in case of success.
70 : disconnect_sender: Option<mpsc::UnboundedSender<RequestData>>,
71 : pub(crate) latency_timer: LatencyTimer,
72 : // Whether proxy decided that it's not a valid endpoint end rejected it before going to cplane.
73 : rejected: Option<bool>,
74 : disconnect_timestamp: Option<chrono::DateTime<Utc>>,
75 : }
76 :
77 : #[derive(Clone, Debug)]
78 : pub(crate) enum AuthMethod {
79 : // aka passwordless, fka link
80 : ConsoleRedirect,
81 : ScramSha256,
82 : ScramSha256Plus,
83 : Cleartext,
84 : Jwt,
85 : }
86 :
87 : impl Clone for RequestContext {
88 0 : fn clone(&self) -> Self {
89 0 : let inner = self.0.try_lock().expect("should not deadlock");
90 0 : let new = RequestContextInner {
91 0 : conn_info: inner.conn_info.clone(),
92 0 : session_id: inner.session_id,
93 0 : protocol: inner.protocol,
94 0 : first_packet: inner.first_packet,
95 0 : region: inner.region,
96 0 : span: info_span!("background_task"),
97 :
98 0 : project: inner.project,
99 0 : branch: inner.branch,
100 0 : endpoint_id: inner.endpoint_id.clone(),
101 0 : dbname: inner.dbname.clone(),
102 0 : user: inner.user.clone(),
103 0 : application: inner.application.clone(),
104 0 : user_agent: inner.user_agent.clone(),
105 0 : error_kind: inner.error_kind,
106 0 : auth_method: inner.auth_method.clone(),
107 0 : jwt_issuer: inner.jwt_issuer.clone(),
108 0 : success: inner.success,
109 0 : rejected: inner.rejected,
110 0 : cold_start_info: inner.cold_start_info,
111 0 : pg_options: inner.pg_options.clone(),
112 0 :
113 0 : sender: None,
114 0 : disconnect_sender: None,
115 0 : latency_timer: LatencyTimer::noop(inner.protocol),
116 0 : disconnect_timestamp: inner.disconnect_timestamp,
117 0 : };
118 0 :
119 0 : Self(TryLock::new(new))
120 0 : }
121 : }
122 :
123 : impl RequestContext {
124 70 : pub fn new(
125 70 : session_id: Uuid,
126 70 : conn_info: ConnectionInfo,
127 70 : protocol: Protocol,
128 70 : region: &'static str,
129 70 : ) -> Self {
130 : // TODO: be careful with long lived spans
131 70 : let span = info_span!(
132 70 : "connect_request",
133 70 : %protocol,
134 70 : ?session_id,
135 70 : %conn_info,
136 70 : ep = tracing::field::Empty,
137 70 : role = tracing::field::Empty,
138 70 : );
139 :
140 70 : let inner = RequestContextInner {
141 70 : conn_info,
142 70 : session_id,
143 70 : protocol,
144 70 : first_packet: Utc::now(),
145 70 : region,
146 70 : span,
147 70 :
148 70 : project: None,
149 70 : branch: None,
150 70 : endpoint_id: None,
151 70 : dbname: None,
152 70 : user: None,
153 70 : application: None,
154 70 : user_agent: None,
155 70 : error_kind: None,
156 70 : auth_method: None,
157 70 : jwt_issuer: None,
158 70 : success: false,
159 70 : rejected: None,
160 70 : cold_start_info: ColdStartInfo::Unknown,
161 70 : pg_options: None,
162 70 :
163 70 : sender: LOG_CHAN.get().and_then(|tx| tx.upgrade()),
164 70 : disconnect_sender: LOG_CHAN_DISCONNECT.get().and_then(|tx| tx.upgrade()),
165 70 : latency_timer: LatencyTimer::new(protocol),
166 70 : disconnect_timestamp: None,
167 70 : };
168 70 :
169 70 : Self(TryLock::new(inner))
170 70 : }
171 :
172 : #[cfg(test)]
173 70 : pub(crate) fn test() -> Self {
174 : use std::net::SocketAddr;
175 70 : let ip = IpAddr::from([127, 0, 0, 1]);
176 70 : let addr = SocketAddr::new(ip, 5432);
177 70 : let conn_info = ConnectionInfo { addr, extra: None };
178 70 : RequestContext::new(Uuid::now_v7(), conn_info, Protocol::Tcp, "test")
179 70 : }
180 :
181 0 : pub(crate) fn console_application_name(&self) -> String {
182 0 : let this = self.0.try_lock().expect("should not deadlock");
183 0 : format!(
184 0 : "{}/{}",
185 0 : this.application.as_deref().unwrap_or_default(),
186 0 : this.protocol
187 0 : )
188 0 : }
189 :
190 0 : pub(crate) fn set_rejected(&self, rejected: bool) {
191 0 : let mut this = self.0.try_lock().expect("should not deadlock");
192 0 : this.rejected = Some(rejected);
193 0 : }
194 :
195 0 : pub(crate) fn set_cold_start_info(&self, info: ColdStartInfo) {
196 0 : self.0
197 0 : .try_lock()
198 0 : .expect("should not deadlock")
199 0 : .set_cold_start_info(info);
200 0 : }
201 :
202 0 : pub(crate) fn set_db_options(&self, options: StartupMessageParams) {
203 0 : let mut this = self.0.try_lock().expect("should not deadlock");
204 0 : this.set_application(options.get("application_name").map(SmolStr::from));
205 0 : if let Some(user) = options.get("user") {
206 0 : this.set_user(user.into());
207 0 : }
208 0 : if let Some(dbname) = options.get("database") {
209 0 : this.set_dbname(dbname.into());
210 0 : }
211 :
212 0 : this.pg_options = Some(options);
213 0 : }
214 :
215 0 : pub(crate) fn set_project(&self, x: MetricsAuxInfo) {
216 0 : let mut this = self.0.try_lock().expect("should not deadlock");
217 0 : if this.endpoint_id.is_none() {
218 0 : this.set_endpoint_id(x.endpoint_id.as_str().into());
219 0 : }
220 0 : this.branch = Some(x.branch_id);
221 0 : this.project = Some(x.project_id);
222 0 : this.set_cold_start_info(x.cold_start_info);
223 0 : }
224 :
225 0 : pub(crate) fn set_project_id(&self, project_id: ProjectIdInt) {
226 0 : let mut this = self.0.try_lock().expect("should not deadlock");
227 0 : this.project = Some(project_id);
228 0 : }
229 :
230 28 : pub(crate) fn set_endpoint_id(&self, endpoint_id: EndpointId) {
231 28 : self.0
232 28 : .try_lock()
233 28 : .expect("should not deadlock")
234 28 : .set_endpoint_id(endpoint_id);
235 28 : }
236 :
237 0 : pub(crate) fn set_dbname(&self, dbname: DbName) {
238 0 : self.0
239 0 : .try_lock()
240 0 : .expect("should not deadlock")
241 0 : .set_dbname(dbname);
242 0 : }
243 :
244 0 : pub(crate) fn set_user(&self, user: RoleName) {
245 0 : self.0
246 0 : .try_lock()
247 0 : .expect("should not deadlock")
248 0 : .set_user(user);
249 0 : }
250 :
251 0 : pub(crate) fn set_user_agent(&self, user_agent: Option<SmolStr>) {
252 0 : self.0
253 0 : .try_lock()
254 0 : .expect("should not deadlock")
255 0 : .set_user_agent(user_agent);
256 0 : }
257 :
258 15 : pub(crate) fn set_auth_method(&self, auth_method: AuthMethod) {
259 15 : let mut this = self.0.try_lock().expect("should not deadlock");
260 15 : this.auth_method = Some(auth_method);
261 15 : }
262 :
263 12 : pub(crate) fn set_jwt_issuer(&self, jwt_issuer: String) {
264 12 : let mut this = self.0.try_lock().expect("should not deadlock");
265 12 : this.jwt_issuer = Some(jwt_issuer);
266 12 : }
267 :
268 0 : pub fn has_private_peer_addr(&self) -> bool {
269 0 : self.0
270 0 : .try_lock()
271 0 : .expect("should not deadlock")
272 0 : .has_private_peer_addr()
273 0 : }
274 :
275 0 : pub(crate) fn set_error_kind(&self, kind: ErrorKind) {
276 0 : let mut this = self.0.try_lock().expect("should not deadlock");
277 0 : // Do not record errors from the private address to metrics.
278 0 : if !this.has_private_peer_addr() {
279 0 : Metrics::get().proxy.errors_total.inc(kind);
280 0 : }
281 0 : if let Some(ep) = &this.endpoint_id {
282 0 : let metric = &Metrics::get().proxy.endpoints_affected_by_errors;
283 0 : let label = metric.with_labels(kind);
284 0 : metric.get_metric(label).measure(ep);
285 0 : }
286 0 : this.error_kind = Some(kind);
287 0 : }
288 :
289 0 : pub fn set_success(&self) {
290 0 : let mut this = self.0.try_lock().expect("should not deadlock");
291 0 : this.success = true;
292 0 : }
293 :
294 0 : pub fn log_connect(self) -> DisconnectLogger {
295 0 : let mut this = self.0.into_inner();
296 0 : this.log_connect();
297 0 :
298 0 : // close current span.
299 0 : this.span = Span::none();
300 0 :
301 0 : DisconnectLogger(this)
302 0 : }
303 :
304 0 : pub(crate) fn protocol(&self) -> Protocol {
305 0 : self.0.try_lock().expect("should not deadlock").protocol
306 0 : }
307 :
308 0 : pub(crate) fn span(&self) -> Span {
309 0 : self.0.try_lock().expect("should not deadlock").span.clone()
310 0 : }
311 :
312 0 : pub(crate) fn session_id(&self) -> Uuid {
313 0 : self.0.try_lock().expect("should not deadlock").session_id
314 0 : }
315 :
316 6 : pub(crate) fn peer_addr(&self) -> IpAddr {
317 6 : self.0
318 6 : .try_lock()
319 6 : .expect("should not deadlock")
320 6 : .conn_info
321 6 : .addr
322 6 : .ip()
323 6 : }
324 :
325 0 : pub(crate) fn extra(&self) -> Option<ConnectionInfoExtra> {
326 0 : self.0
327 0 : .try_lock()
328 0 : .expect("should not deadlock")
329 0 : .conn_info
330 0 : .extra
331 0 : .clone()
332 0 : }
333 :
334 0 : pub(crate) fn cold_start_info(&self) -> ColdStartInfo {
335 0 : self.0
336 0 : .try_lock()
337 0 : .expect("should not deadlock")
338 0 : .cold_start_info
339 0 : }
340 :
341 28 : pub(crate) fn latency_timer_pause(&self, waiting_for: Waiting) -> LatencyTimerPause<'_> {
342 28 : LatencyTimerPause {
343 28 : ctx: self,
344 28 : start: tokio::time::Instant::now(),
345 28 : waiting_for,
346 28 : }
347 28 : }
348 :
349 4 : pub(crate) fn success(&self) {
350 4 : self.0
351 4 : .try_lock()
352 4 : .expect("should not deadlock")
353 4 : .latency_timer
354 4 : .success();
355 4 : }
356 : }
357 :
358 : pub(crate) struct LatencyTimerPause<'a> {
359 : ctx: &'a RequestContext,
360 : start: tokio::time::Instant,
361 : waiting_for: Waiting,
362 : }
363 :
364 : impl Drop for LatencyTimerPause<'_> {
365 28 : fn drop(&mut self) {
366 28 : self.ctx
367 28 : .0
368 28 : .try_lock()
369 28 : .expect("should not deadlock")
370 28 : .latency_timer
371 28 : .unpause(self.start, self.waiting_for);
372 28 : }
373 : }
374 :
375 : impl RequestContextInner {
376 0 : fn set_cold_start_info(&mut self, info: ColdStartInfo) {
377 0 : self.cold_start_info = info;
378 0 : self.latency_timer.cold_start_info(info);
379 0 : }
380 :
381 28 : fn set_endpoint_id(&mut self, endpoint_id: EndpointId) {
382 28 : if self.endpoint_id.is_none() {
383 28 : self.span.record("ep", display(&endpoint_id));
384 28 : let metric = &Metrics::get().proxy.connecting_endpoints;
385 28 : let label = metric.with_labels(self.protocol);
386 28 : metric.get_metric(label).measure(&endpoint_id);
387 28 : self.endpoint_id = Some(endpoint_id);
388 28 : }
389 28 : }
390 :
391 0 : fn set_application(&mut self, app: Option<SmolStr>) {
392 0 : if let Some(app) = app {
393 0 : self.application = Some(app);
394 0 : }
395 0 : }
396 :
397 0 : fn set_user_agent(&mut self, user_agent: Option<SmolStr>) {
398 0 : self.user_agent = user_agent;
399 0 : }
400 :
401 0 : fn set_dbname(&mut self, dbname: DbName) {
402 0 : self.dbname = Some(dbname);
403 0 : }
404 :
405 0 : fn set_user(&mut self, user: RoleName) {
406 0 : self.span.record("role", display(&user));
407 0 : self.user = Some(user);
408 0 : }
409 :
410 0 : fn has_private_peer_addr(&self) -> bool {
411 0 : match self.conn_info.addr.ip() {
412 0 : IpAddr::V4(ip) => ip.is_private(),
413 0 : IpAddr::V6(_) => false,
414 : }
415 0 : }
416 :
417 0 : fn log_connect(&mut self) {
418 0 : let outcome = if self.success {
419 0 : ConnectOutcome::Success
420 : } else {
421 0 : ConnectOutcome::Failed
422 : };
423 :
424 : // TODO: get rid of entirely/refactor
425 : // check for false positives
426 : // AND false negatives
427 0 : if let Some(rejected) = self.rejected {
428 0 : let ep = self
429 0 : .endpoint_id
430 0 : .as_ref()
431 0 : .map(|x| x.as_str())
432 0 : .unwrap_or_default();
433 0 : // This makes sense only if cache is disabled
434 0 : debug!(
435 : ?outcome,
436 : ?rejected,
437 : ?ep,
438 0 : "check endpoint is valid with outcome"
439 : );
440 0 : Metrics::get()
441 0 : .proxy
442 0 : .invalid_endpoints_total
443 0 : .inc(InvalidEndpointsGroup {
444 0 : protocol: self.protocol,
445 0 : rejected: rejected.into(),
446 0 : outcome,
447 0 : });
448 0 : }
449 :
450 0 : if let Some(tx) = self.sender.take() {
451 : // If type changes, this error handling needs to be updated.
452 0 : let tx: mpsc::UnboundedSender<RequestData> = tx;
453 0 : if let Err(e) = tx.send(RequestData::from(&*self)) {
454 0 : error!("log_connect channel send failed: {e}");
455 0 : }
456 0 : }
457 0 : }
458 :
459 0 : fn log_disconnect(&mut self) {
460 0 : // If we are here, it's guaranteed that the user successfully connected to the endpoint.
461 0 : // Here we log the length of the session.
462 0 : self.disconnect_timestamp = Some(Utc::now());
463 0 : if let Some(tx) = self.disconnect_sender.take() {
464 : // If type changes, this error handling needs to be updated.
465 0 : let tx: mpsc::UnboundedSender<RequestData> = tx;
466 0 : if let Err(e) = tx.send(RequestData::from(&*self)) {
467 0 : error!("log_disconnect channel send failed: {e}");
468 0 : }
469 0 : }
470 0 : }
471 : }
472 :
473 : impl Drop for RequestContextInner {
474 70 : fn drop(&mut self) {
475 70 : if self.sender.is_some() {
476 0 : self.log_connect();
477 70 : }
478 70 : }
479 : }
480 :
481 : pub struct DisconnectLogger(RequestContextInner);
482 :
483 : impl Drop for DisconnectLogger {
484 0 : fn drop(&mut self) {
485 0 : self.0.log_disconnect();
486 0 : }
487 : }
|
