Line data Source code
1 : use std::convert::Infallible;
2 : use std::net::IpAddr;
3 : use std::net::Ipv6Addr;
4 : use std::net::SocketAddr;
5 : use std::sync::Arc;
6 : use std::thread;
7 :
8 : use crate::catalog::SchemaDumpError;
9 : use crate::catalog::{get_database_schema, get_dbs_and_roles};
10 : use crate::compute::forward_termination_signal;
11 : use crate::compute::{ComputeNode, ComputeState, ParsedSpec};
12 : use crate::installed_extensions;
13 : use compute_api::requests::{ConfigurationRequest, ExtensionInstallRequest, SetRoleGrantsRequest};
14 : use compute_api::responses::{
15 : ComputeStatus, ComputeStatusResponse, ExtensionInstallResult, GenericAPIError,
16 : SetRoleGrantsResponse,
17 : };
18 :
19 : use anyhow::Result;
20 : use hyper::header::CONTENT_TYPE;
21 : use hyper::service::{make_service_fn, service_fn};
22 : use hyper::{Body, Method, Request, Response, Server, StatusCode};
23 : use metrics::Encoder;
24 : use metrics::TextEncoder;
25 : use tokio::task;
26 : use tokio_util::sync::CancellationToken;
27 : use tracing::{debug, error, info, warn};
28 : use tracing_utils::http::OtelName;
29 : use utils::failpoint_support::failpoints_handler;
30 : use utils::http::error::ApiError;
31 : use utils::http::request::must_get_query_param;
32 :
33 0 : fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
34 0 : ComputeStatusResponse {
35 0 : start_time: state.start_time,
36 0 : tenant: state
37 0 : .pspec
38 0 : .as_ref()
39 0 : .map(|pspec| pspec.tenant_id.to_string()),
40 0 : timeline: state
41 0 : .pspec
42 0 : .as_ref()
43 0 : .map(|pspec| pspec.timeline_id.to_string()),
44 0 : status: state.status,
45 0 : last_active: state.last_active,
46 0 : error: state.error.clone(),
47 0 : }
48 0 : }
49 :
50 : // Service function to handle all available routes.
51 0 : async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body> {
52 0 : //
53 0 : // NOTE: The URI path is currently included in traces. That's OK because
54 0 : // it doesn't contain any variable parts or sensitive information. But
55 0 : // please keep that in mind if you change the routing here.
56 0 : //
57 0 : match (req.method(), req.uri().path()) {
58 : // Serialized compute state.
59 0 : (&Method::GET, "/status") => {
60 0 : debug!("serving /status GET request");
61 0 : let state = compute.state.lock().unwrap();
62 0 : let status_response = status_response_from_state(&state);
63 0 : Response::new(Body::from(serde_json::to_string(&status_response).unwrap()))
64 : }
65 :
66 : // Startup metrics in JSON format. Keep /metrics reserved for a possible
67 : // future use for Prometheus metrics format.
68 0 : (&Method::GET, "/metrics.json") => {
69 0 : info!("serving /metrics.json GET request");
70 0 : let metrics = compute.state.lock().unwrap().metrics.clone();
71 0 : Response::new(Body::from(serde_json::to_string(&metrics).unwrap()))
72 : }
73 :
74 : // Prometheus metrics
75 0 : (&Method::GET, "/metrics") => {
76 0 : debug!("serving /metrics GET request");
77 :
78 0 : let mut buffer = vec![];
79 0 : let metrics = installed_extensions::collect();
80 0 : let encoder = TextEncoder::new();
81 0 : encoder.encode(&metrics, &mut buffer).unwrap();
82 0 :
83 0 : match Response::builder()
84 0 : .status(StatusCode::OK)
85 0 : .header(CONTENT_TYPE, encoder.format_type())
86 0 : .body(Body::from(buffer))
87 : {
88 0 : Ok(response) => response,
89 0 : Err(err) => {
90 0 : let msg = format!("error handling /metrics request: {err}");
91 0 : error!(msg);
92 0 : render_json_error(&msg, StatusCode::INTERNAL_SERVER_ERROR)
93 : }
94 : }
95 : }
96 : // Collect Postgres current usage insights
97 0 : (&Method::GET, "/insights") => {
98 0 : info!("serving /insights GET request");
99 0 : let status = compute.get_status();
100 0 : if status != ComputeStatus::Running {
101 0 : let msg = format!("compute is not running, current status: {:?}", status);
102 0 : error!(msg);
103 0 : return Response::new(Body::from(msg));
104 0 : }
105 :
106 0 : let insights = compute.collect_insights().await;
107 0 : Response::new(Body::from(insights))
108 : }
109 :
110 0 : (&Method::POST, "/check_writability") => {
111 0 : info!("serving /check_writability POST request");
112 0 : let status = compute.get_status();
113 0 : if status != ComputeStatus::Running {
114 0 : let msg = format!(
115 0 : "invalid compute status for check_writability request: {:?}",
116 0 : status
117 0 : );
118 0 : error!(msg);
119 0 : return Response::new(Body::from(msg));
120 0 : }
121 :
122 0 : let res = crate::checker::check_writability(compute).await;
123 0 : match res {
124 0 : Ok(_) => Response::new(Body::from("true")),
125 0 : Err(e) => {
126 0 : error!("check_writability failed: {}", e);
127 0 : Response::new(Body::from(e.to_string()))
128 : }
129 : }
130 : }
131 :
132 0 : (&Method::POST, "/extensions") => {
133 0 : info!("serving /extensions POST request");
134 0 : let status = compute.get_status();
135 0 : if status != ComputeStatus::Running {
136 0 : let msg = format!(
137 0 : "invalid compute status for extensions request: {:?}",
138 0 : status
139 0 : );
140 0 : error!(msg);
141 0 : return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
142 0 : }
143 :
144 0 : let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
145 0 : let request = serde_json::from_slice::<ExtensionInstallRequest>(&request).unwrap();
146 0 : let res = compute
147 0 : .install_extension(&request.extension, &request.database, request.version)
148 0 : .await;
149 0 : match res {
150 0 : Ok(version) => render_json(Body::from(
151 0 : serde_json::to_string(&ExtensionInstallResult {
152 0 : extension: request.extension,
153 0 : version,
154 0 : })
155 0 : .unwrap(),
156 0 : )),
157 0 : Err(e) => {
158 0 : error!("install_extension failed: {}", e);
159 0 : render_json_error(&e.to_string(), StatusCode::INTERNAL_SERVER_ERROR)
160 : }
161 : }
162 : }
163 :
164 0 : (&Method::GET, "/info") => {
165 0 : let num_cpus = num_cpus::get_physical();
166 0 : info!("serving /info GET request. num_cpus: {}", num_cpus);
167 0 : Response::new(Body::from(
168 0 : serde_json::json!({
169 0 : "num_cpus": num_cpus,
170 0 : })
171 0 : .to_string(),
172 0 : ))
173 : }
174 :
175 : // Accept spec in JSON format and request compute configuration. If
176 : // anything goes wrong after we set the compute status to `ConfigurationPending`
177 : // and update compute state with new spec, we basically leave compute
178 : // in the potentially wrong state. That said, it's control-plane's
179 : // responsibility to watch compute state after reconfiguration request
180 : // and to clean restart in case of errors.
181 0 : (&Method::POST, "/configure") => {
182 0 : info!("serving /configure POST request");
183 0 : match handle_configure_request(req, compute).await {
184 0 : Ok(msg) => Response::new(Body::from(msg)),
185 0 : Err((msg, code)) => {
186 0 : error!("error handling /configure request: {msg}");
187 0 : render_json_error(&msg, code)
188 : }
189 : }
190 : }
191 :
192 0 : (&Method::POST, "/terminate") => {
193 0 : info!("serving /terminate POST request");
194 0 : match handle_terminate_request(compute).await {
195 0 : Ok(()) => Response::new(Body::empty()),
196 0 : Err((msg, code)) => {
197 0 : error!("error handling /terminate request: {msg}");
198 0 : render_json_error(&msg, code)
199 : }
200 : }
201 : }
202 :
203 0 : (&Method::GET, "/dbs_and_roles") => {
204 0 : info!("serving /dbs_and_roles GET request",);
205 0 : match get_dbs_and_roles(compute).await {
206 0 : Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
207 : Err(_) => {
208 0 : render_json_error("can't get dbs and roles", StatusCode::INTERNAL_SERVER_ERROR)
209 : }
210 : }
211 : }
212 :
213 0 : (&Method::GET, "/database_schema") => {
214 0 : let database = match must_get_query_param(&req, "database") {
215 0 : Err(e) => return e.into_response(),
216 0 : Ok(database) => database,
217 0 : };
218 0 : info!("serving /database_schema GET request with database: {database}",);
219 0 : match get_database_schema(compute, &database).await {
220 0 : Ok(res) => render_plain(Body::wrap_stream(res)),
221 : Err(SchemaDumpError::DatabaseDoesNotExist) => {
222 0 : render_json_error("database does not exist", StatusCode::NOT_FOUND)
223 : }
224 0 : Err(e) => {
225 0 : error!("can't get schema dump: {}", e);
226 0 : render_json_error("can't get schema dump", StatusCode::INTERNAL_SERVER_ERROR)
227 : }
228 : }
229 : }
230 :
231 0 : (&Method::POST, "/grants") => {
232 0 : info!("serving /grants POST request");
233 0 : let status = compute.get_status();
234 0 : if status != ComputeStatus::Running {
235 0 : let msg = format!(
236 0 : "invalid compute status for set_role_grants request: {:?}",
237 0 : status
238 0 : );
239 0 : error!(msg);
240 0 : return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
241 0 : }
242 :
243 0 : let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
244 0 : let request = serde_json::from_slice::<SetRoleGrantsRequest>(&request).unwrap();
245 :
246 0 : let res = compute
247 0 : .set_role_grants(
248 0 : &request.database,
249 0 : &request.schema,
250 0 : &request.privileges,
251 0 : &request.role,
252 0 : )
253 0 : .await;
254 0 : match res {
255 0 : Ok(()) => render_json(Body::from(
256 0 : serde_json::to_string(&SetRoleGrantsResponse {
257 0 : database: request.database,
258 0 : schema: request.schema,
259 0 : role: request.role,
260 0 : privileges: request.privileges,
261 0 : })
262 0 : .unwrap(),
263 0 : )),
264 0 : Err(e) => render_json_error(
265 0 : &format!("could not grant role privileges to the schema: {e}"),
266 0 : // TODO: can we filter on role/schema not found errors
267 0 : // and return appropriate error code?
268 0 : StatusCode::INTERNAL_SERVER_ERROR,
269 0 : ),
270 : }
271 : }
272 :
273 : // get the list of installed extensions
274 : // currently only used in python tests
275 : // TODO: call it from cplane
276 0 : (&Method::GET, "/installed_extensions") => {
277 0 : info!("serving /installed_extensions GET request");
278 0 : let status = compute.get_status();
279 0 : if status != ComputeStatus::Running {
280 0 : let msg = format!(
281 0 : "invalid compute status for extensions request: {:?}",
282 0 : status
283 0 : );
284 0 : error!(msg);
285 0 : return Response::new(Body::from(msg));
286 0 : }
287 0 :
288 0 : let connstr = compute.connstr.clone();
289 0 : let res = crate::installed_extensions::get_installed_extensions(connstr).await;
290 0 : match res {
291 0 : Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
292 0 : Err(e) => render_json_error(
293 0 : &format!("could not get list of installed extensions: {}", e),
294 0 : StatusCode::INTERNAL_SERVER_ERROR,
295 0 : ),
296 : }
297 : }
298 :
299 0 : (&Method::POST, "/failpoints") if cfg!(feature = "testing") => {
300 0 : match failpoints_handler(req, CancellationToken::new()).await {
301 0 : Ok(r) => r,
302 0 : Err(ApiError::BadRequest(e)) => {
303 0 : render_json_error(&e.to_string(), StatusCode::BAD_REQUEST)
304 : }
305 : Err(_) => {
306 0 : render_json_error("Internal server error", StatusCode::INTERNAL_SERVER_ERROR)
307 : }
308 : }
309 : }
310 :
311 : // download extension files from remote extension storage on demand
312 0 : (&Method::POST, route) if route.starts_with("/extension_server/") => {
313 0 : info!("serving {:?} POST request", route);
314 0 : info!("req.uri {:?}", req.uri());
315 :
316 : // don't even try to download extensions
317 : // if no remote storage is configured
318 0 : if compute.ext_remote_storage.is_none() {
319 0 : info!("no extensions remote storage configured");
320 0 : let mut resp = Response::new(Body::from("no remote storage configured"));
321 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
322 0 : return resp;
323 0 : }
324 0 :
325 0 : let mut is_library = false;
326 0 : if let Some(params) = req.uri().query() {
327 0 : info!("serving {:?} POST request with params: {}", route, params);
328 0 : if params == "is_library=true" {
329 0 : is_library = true;
330 0 : } else {
331 0 : let mut resp = Response::new(Body::from("Wrong request parameters"));
332 0 : *resp.status_mut() = StatusCode::BAD_REQUEST;
333 0 : return resp;
334 : }
335 0 : }
336 0 : let filename = route.split('/').last().unwrap().to_string();
337 0 : info!("serving /extension_server POST request, filename: {filename:?} is_library: {is_library}");
338 :
339 : // get ext_name and path from spec
340 : // don't lock compute_state for too long
341 0 : let ext = {
342 0 : let compute_state = compute.state.lock().unwrap();
343 0 : let pspec = compute_state.pspec.as_ref().expect("spec must be set");
344 0 : let spec = &pspec.spec;
345 0 :
346 0 : // debug only
347 0 : info!("spec: {:?}", spec);
348 :
349 0 : let remote_extensions = match spec.remote_extensions.as_ref() {
350 0 : Some(r) => r,
351 : None => {
352 0 : info!("no remote extensions spec was provided");
353 0 : let mut resp = Response::new(Body::from("no remote storage configured"));
354 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
355 0 : return resp;
356 : }
357 : };
358 :
359 0 : remote_extensions.get_ext(
360 0 : &filename,
361 0 : is_library,
362 0 : &compute.build_tag,
363 0 : &compute.pgversion,
364 0 : )
365 0 : };
366 0 :
367 0 : match ext {
368 0 : Ok((ext_name, ext_path)) => {
369 0 : match compute.download_extension(ext_name, ext_path).await {
370 0 : Ok(_) => Response::new(Body::from("OK")),
371 0 : Err(e) => {
372 0 : error!("extension download failed: {}", e);
373 0 : let mut resp = Response::new(Body::from(e.to_string()));
374 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
375 0 : resp
376 : }
377 : }
378 : }
379 0 : Err(e) => {
380 0 : warn!("extension download failed to find extension: {}", e);
381 0 : let mut resp = Response::new(Body::from("failed to find file"));
382 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
383 0 : resp
384 : }
385 : }
386 : }
387 :
388 : // Return the `404 Not Found` for any other routes.
389 : _ => {
390 0 : let mut not_found = Response::new(Body::from("404 Not Found"));
391 0 : *not_found.status_mut() = StatusCode::NOT_FOUND;
392 0 : not_found
393 : }
394 : }
395 0 : }
396 :
397 0 : async fn handle_configure_request(
398 0 : req: Request<Body>,
399 0 : compute: &Arc<ComputeNode>,
400 0 : ) -> Result<String, (String, StatusCode)> {
401 0 : if !compute.live_config_allowed {
402 0 : return Err((
403 0 : "live configuration is not allowed for this compute node".to_string(),
404 0 : StatusCode::PRECONDITION_FAILED,
405 0 : ));
406 0 : }
407 :
408 0 : let body_bytes = hyper::body::to_bytes(req.into_body()).await.unwrap();
409 0 : let spec_raw = String::from_utf8(body_bytes.to_vec()).unwrap();
410 0 : if let Ok(request) = serde_json::from_str::<ConfigurationRequest>(&spec_raw) {
411 0 : let spec = request.spec;
412 :
413 0 : let parsed_spec = match ParsedSpec::try_from(spec) {
414 0 : Ok(ps) => ps,
415 0 : Err(msg) => return Err((msg, StatusCode::BAD_REQUEST)),
416 : };
417 :
418 : // XXX: wrap state update under lock in code blocks. Otherwise,
419 : // we will try to `Send` `mut state` into the spawned thread
420 : // bellow, which will cause error:
421 : // ```
422 : // error: future cannot be sent between threads safely
423 : // ```
424 : {
425 0 : let mut state = compute.state.lock().unwrap();
426 0 : if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
427 0 : let msg = format!(
428 0 : "invalid compute status for configuration request: {:?}",
429 0 : state.status.clone()
430 0 : );
431 0 : return Err((msg, StatusCode::PRECONDITION_FAILED));
432 0 : }
433 0 : state.pspec = Some(parsed_spec);
434 0 : state.set_status(ComputeStatus::ConfigurationPending, &compute.state_changed);
435 0 : drop(state);
436 0 : info!("set new spec and notified waiters");
437 : }
438 :
439 : // Spawn a blocking thread to wait for compute to become Running.
440 : // This is needed to do not block the main pool of workers and
441 : // be able to serve other requests while some particular request
442 : // is waiting for compute to finish configuration.
443 0 : let c = compute.clone();
444 0 : task::spawn_blocking(move || {
445 0 : let mut state = c.state.lock().unwrap();
446 0 : while state.status != ComputeStatus::Running {
447 0 : state = c.state_changed.wait(state).unwrap();
448 0 : info!(
449 0 : "waiting for compute to become Running, current status: {:?}",
450 0 : state.status
451 : );
452 :
453 0 : if state.status == ComputeStatus::Failed {
454 0 : let err = state.error.as_ref().map_or("unknown error", |x| x);
455 0 : let msg = format!("compute configuration failed: {:?}", err);
456 0 : return Err((msg, StatusCode::INTERNAL_SERVER_ERROR));
457 0 : }
458 : }
459 :
460 0 : Ok(())
461 0 : })
462 0 : .await
463 0 : .unwrap()?;
464 :
465 : // Return current compute state if everything went well.
466 0 : let state = compute.state.lock().unwrap().clone();
467 0 : let status_response = status_response_from_state(&state);
468 0 : Ok(serde_json::to_string(&status_response).unwrap())
469 : } else {
470 0 : Err(("invalid spec".to_string(), StatusCode::BAD_REQUEST))
471 : }
472 0 : }
473 :
474 0 : fn render_json_error(e: &str, status: StatusCode) -> Response<Body> {
475 0 : let error = GenericAPIError {
476 0 : error: e.to_string(),
477 0 : };
478 0 : Response::builder()
479 0 : .status(status)
480 0 : .header(CONTENT_TYPE, "application/json")
481 0 : .body(Body::from(serde_json::to_string(&error).unwrap()))
482 0 : .unwrap()
483 0 : }
484 :
485 0 : fn render_json(body: Body) -> Response<Body> {
486 0 : Response::builder()
487 0 : .header(CONTENT_TYPE, "application/json")
488 0 : .body(body)
489 0 : .unwrap()
490 0 : }
491 :
492 0 : fn render_plain(body: Body) -> Response<Body> {
493 0 : Response::builder()
494 0 : .header(CONTENT_TYPE, "text/plain")
495 0 : .body(body)
496 0 : .unwrap()
497 0 : }
498 :
499 0 : async fn handle_terminate_request(compute: &Arc<ComputeNode>) -> Result<(), (String, StatusCode)> {
500 0 : {
501 0 : let mut state = compute.state.lock().unwrap();
502 0 : if state.status == ComputeStatus::Terminated {
503 0 : return Ok(());
504 0 : }
505 0 : if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
506 0 : let msg = format!(
507 0 : "invalid compute status for termination request: {}",
508 0 : state.status
509 0 : );
510 0 : return Err((msg, StatusCode::PRECONDITION_FAILED));
511 0 : }
512 0 : state.set_status(ComputeStatus::TerminationPending, &compute.state_changed);
513 0 : drop(state);
514 0 : }
515 0 :
516 0 : forward_termination_signal();
517 0 : info!("sent signal and notified waiters");
518 :
519 : // Spawn a blocking thread to wait for compute to become Terminated.
520 : // This is needed to do not block the main pool of workers and
521 : // be able to serve other requests while some particular request
522 : // is waiting for compute to finish configuration.
523 0 : let c = compute.clone();
524 0 : task::spawn_blocking(move || {
525 0 : let mut state = c.state.lock().unwrap();
526 0 : while state.status != ComputeStatus::Terminated {
527 0 : state = c.state_changed.wait(state).unwrap();
528 0 : info!(
529 0 : "waiting for compute to become {}, current status: {:?}",
530 0 : ComputeStatus::Terminated,
531 0 : state.status
532 : );
533 : }
534 :
535 0 : Ok(())
536 0 : })
537 0 : .await
538 0 : .unwrap()?;
539 0 : info!("terminated Postgres");
540 0 : Ok(())
541 0 : }
542 :
543 : // Main Hyper HTTP server function that runs it and blocks waiting on it forever.
544 : #[tokio::main]
545 0 : async fn serve(port: u16, state: Arc<ComputeNode>) {
546 0 : // this usually binds to both IPv4 and IPv6 on linux
547 0 : // see e.g. https://github.com/rust-lang/rust/pull/34440
548 0 : let addr = SocketAddr::new(IpAddr::from(Ipv6Addr::UNSPECIFIED), port);
549 0 :
550 0 : let make_service = make_service_fn(move |_conn| {
551 0 : let state = state.clone();
552 0 : async move {
553 0 : Ok::<_, Infallible>(service_fn(move |req: Request<Body>| {
554 0 : let state = state.clone();
555 0 : async move {
556 0 : Ok::<_, Infallible>(
557 0 : // NOTE: We include the URI path in the string. It
558 0 : // doesn't contain any variable parts or sensitive
559 0 : // information in this API.
560 0 : tracing_utils::http::tracing_handler(
561 0 : req,
562 0 : |req| routes(req, &state),
563 0 : OtelName::UriPath,
564 0 : )
565 0 : .await,
566 0 : )
567 0 : }
568 0 : }))
569 0 : }
570 0 : });
571 0 :
572 0 : info!("starting HTTP server on {}", addr);
573 0 :
574 0 : let server = Server::bind(&addr).serve(make_service);
575 0 :
576 0 : // Run this server forever
577 0 : if let Err(e) = server.await {
578 0 : error!("server error: {}", e);
579 0 : }
580 0 : }
581 :
582 : /// Launch a separate Hyper HTTP API server thread and return its `JoinHandle`.
583 0 : pub fn launch_http_server(port: u16, state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
584 0 : let state = Arc::clone(state);
585 0 :
586 0 : Ok(thread::Builder::new()
587 0 : .name("http-endpoint".into())
588 0 : .spawn(move || serve(port, state))?)
589 0 : }
|
