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::proto::MetricFamily;
24 : use metrics::Encoder;
25 : use metrics::TextEncoder;
26 : use tokio::task;
27 : use tracing::{debug, error, info, warn};
28 : use tracing_utils::http::OtelName;
29 : use utils::http::request::must_get_query_param;
30 :
31 0 : fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
32 0 : ComputeStatusResponse {
33 0 : start_time: state.start_time,
34 0 : tenant: state
35 0 : .pspec
36 0 : .as_ref()
37 0 : .map(|pspec| pspec.tenant_id.to_string()),
38 0 : timeline: state
39 0 : .pspec
40 0 : .as_ref()
41 0 : .map(|pspec| pspec.timeline_id.to_string()),
42 0 : status: state.status,
43 0 : last_active: state.last_active,
44 0 : error: state.error.clone(),
45 0 : }
46 0 : }
47 :
48 : // Service function to handle all available routes.
49 0 : async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body> {
50 0 : //
51 0 : // NOTE: The URI path is currently included in traces. That's OK because
52 0 : // it doesn't contain any variable parts or sensitive information. But
53 0 : // please keep that in mind if you change the routing here.
54 0 : //
55 0 : match (req.method(), req.uri().path()) {
56 : // Serialized compute state.
57 0 : (&Method::GET, "/status") => {
58 0 : debug!("serving /status GET request");
59 0 : let state = compute.state.lock().unwrap();
60 0 : let status_response = status_response_from_state(&state);
61 0 : Response::new(Body::from(serde_json::to_string(&status_response).unwrap()))
62 : }
63 :
64 : // Startup metrics in JSON format. Keep /metrics reserved for a possible
65 : // future use for Prometheus metrics format.
66 0 : (&Method::GET, "/metrics.json") => {
67 0 : info!("serving /metrics.json GET request");
68 0 : let metrics = compute.state.lock().unwrap().metrics.clone();
69 0 : Response::new(Body::from(serde_json::to_string(&metrics).unwrap()))
70 : }
71 :
72 : // Prometheus metrics
73 0 : (&Method::GET, "/metrics") => {
74 0 : debug!("serving /metrics GET request");
75 :
76 : // When we call TextEncoder::encode() below, it will immediately
77 : // return an error if a metric family has no metrics, so we need to
78 : // preemptively filter out metric families with no metrics.
79 0 : let metrics = installed_extensions::collect()
80 0 : .into_iter()
81 0 : .filter(|m| !m.get_metric().is_empty())
82 0 : .collect::<Vec<MetricFamily>>();
83 0 :
84 0 : let encoder = TextEncoder::new();
85 0 : let mut buffer = vec![];
86 :
87 0 : if let Err(err) = encoder.encode(&metrics, &mut buffer) {
88 0 : let msg = format!("error handling /metrics request: {err}");
89 0 : error!(msg);
90 0 : return render_json_error(&msg, StatusCode::INTERNAL_SERVER_ERROR);
91 0 : }
92 0 :
93 0 : match Response::builder()
94 0 : .status(StatusCode::OK)
95 0 : .header(CONTENT_TYPE, encoder.format_type())
96 0 : .body(Body::from(buffer))
97 : {
98 0 : Ok(response) => response,
99 0 : Err(err) => {
100 0 : let msg = format!("error handling /metrics request: {err}");
101 0 : error!(msg);
102 0 : render_json_error(&msg, StatusCode::INTERNAL_SERVER_ERROR)
103 : }
104 : }
105 : }
106 : // Collect Postgres current usage insights
107 0 : (&Method::GET, "/insights") => {
108 0 : info!("serving /insights GET request");
109 0 : let status = compute.get_status();
110 0 : if status != ComputeStatus::Running {
111 0 : let msg = format!("compute is not running, current status: {:?}", status);
112 0 : error!(msg);
113 0 : return Response::new(Body::from(msg));
114 0 : }
115 :
116 0 : let insights = compute.collect_insights().await;
117 0 : Response::new(Body::from(insights))
118 : }
119 :
120 0 : (&Method::POST, "/check_writability") => {
121 0 : info!("serving /check_writability POST request");
122 0 : let status = compute.get_status();
123 0 : if status != ComputeStatus::Running {
124 0 : let msg = format!(
125 0 : "invalid compute status for check_writability request: {:?}",
126 0 : status
127 0 : );
128 0 : error!(msg);
129 0 : return Response::new(Body::from(msg));
130 0 : }
131 :
132 0 : let res = crate::checker::check_writability(compute).await;
133 0 : match res {
134 0 : Ok(_) => Response::new(Body::from("true")),
135 0 : Err(e) => {
136 0 : error!("check_writability failed: {}", e);
137 0 : Response::new(Body::from(e.to_string()))
138 : }
139 : }
140 : }
141 :
142 0 : (&Method::POST, "/extensions") => {
143 0 : info!("serving /extensions POST request");
144 0 : let status = compute.get_status();
145 0 : if status != ComputeStatus::Running {
146 0 : let msg = format!(
147 0 : "invalid compute status for extensions request: {:?}",
148 0 : status
149 0 : );
150 0 : error!(msg);
151 0 : return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
152 0 : }
153 :
154 0 : let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
155 0 : let request = serde_json::from_slice::<ExtensionInstallRequest>(&request).unwrap();
156 0 : let res = compute
157 0 : .install_extension(&request.extension, &request.database, request.version)
158 0 : .await;
159 0 : match res {
160 0 : Ok(version) => render_json(Body::from(
161 0 : serde_json::to_string(&ExtensionInstallResult {
162 0 : extension: request.extension,
163 0 : version,
164 0 : })
165 0 : .unwrap(),
166 0 : )),
167 0 : Err(e) => {
168 0 : error!("install_extension failed: {}", e);
169 0 : render_json_error(&e.to_string(), StatusCode::INTERNAL_SERVER_ERROR)
170 : }
171 : }
172 : }
173 :
174 0 : (&Method::GET, "/info") => {
175 0 : let num_cpus = num_cpus::get_physical();
176 0 : info!("serving /info GET request. num_cpus: {}", num_cpus);
177 0 : Response::new(Body::from(
178 0 : serde_json::json!({
179 0 : "num_cpus": num_cpus,
180 0 : })
181 0 : .to_string(),
182 0 : ))
183 : }
184 :
185 : // Accept spec in JSON format and request compute configuration. If
186 : // anything goes wrong after we set the compute status to `ConfigurationPending`
187 : // and update compute state with new spec, we basically leave compute
188 : // in the potentially wrong state. That said, it's control-plane's
189 : // responsibility to watch compute state after reconfiguration request
190 : // and to clean restart in case of errors.
191 0 : (&Method::POST, "/configure") => {
192 0 : info!("serving /configure POST request");
193 0 : match handle_configure_request(req, compute).await {
194 0 : Ok(msg) => Response::new(Body::from(msg)),
195 0 : Err((msg, code)) => {
196 0 : error!("error handling /configure request: {msg}");
197 0 : render_json_error(&msg, code)
198 : }
199 : }
200 : }
201 :
202 0 : (&Method::POST, "/terminate") => {
203 0 : info!("serving /terminate POST request");
204 0 : match handle_terminate_request(compute).await {
205 0 : Ok(()) => Response::new(Body::empty()),
206 0 : Err((msg, code)) => {
207 0 : error!("error handling /terminate request: {msg}");
208 0 : render_json_error(&msg, code)
209 : }
210 : }
211 : }
212 :
213 0 : (&Method::GET, "/dbs_and_roles") => {
214 0 : info!("serving /dbs_and_roles GET request",);
215 0 : match get_dbs_and_roles(compute).await {
216 0 : Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
217 : Err(_) => {
218 0 : render_json_error("can't get dbs and roles", StatusCode::INTERNAL_SERVER_ERROR)
219 : }
220 : }
221 : }
222 :
223 0 : (&Method::GET, "/database_schema") => {
224 0 : let database = match must_get_query_param(&req, "database") {
225 0 : Err(e) => return e.into_response(),
226 0 : Ok(database) => database,
227 0 : };
228 0 : info!("serving /database_schema GET request with database: {database}",);
229 0 : match get_database_schema(compute, &database).await {
230 0 : Ok(res) => render_plain(Body::wrap_stream(res)),
231 : Err(SchemaDumpError::DatabaseDoesNotExist) => {
232 0 : render_json_error("database does not exist", StatusCode::NOT_FOUND)
233 : }
234 0 : Err(e) => {
235 0 : error!("can't get schema dump: {}", e);
236 0 : render_json_error("can't get schema dump", StatusCode::INTERNAL_SERVER_ERROR)
237 : }
238 : }
239 : }
240 :
241 0 : (&Method::POST, "/grants") => {
242 0 : info!("serving /grants POST request");
243 0 : let status = compute.get_status();
244 0 : if status != ComputeStatus::Running {
245 0 : let msg = format!(
246 0 : "invalid compute status for set_role_grants request: {:?}",
247 0 : status
248 0 : );
249 0 : error!(msg);
250 0 : return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
251 0 : }
252 :
253 0 : let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
254 0 : let request = serde_json::from_slice::<SetRoleGrantsRequest>(&request).unwrap();
255 :
256 0 : let res = compute
257 0 : .set_role_grants(
258 0 : &request.database,
259 0 : &request.schema,
260 0 : &request.privileges,
261 0 : &request.role,
262 0 : )
263 0 : .await;
264 0 : match res {
265 0 : Ok(()) => render_json(Body::from(
266 0 : serde_json::to_string(&SetRoleGrantsResponse {
267 0 : database: request.database,
268 0 : schema: request.schema,
269 0 : role: request.role,
270 0 : privileges: request.privileges,
271 0 : })
272 0 : .unwrap(),
273 0 : )),
274 0 : Err(e) => render_json_error(
275 0 : &format!("could not grant role privileges to the schema: {e}"),
276 0 : // TODO: can we filter on role/schema not found errors
277 0 : // and return appropriate error code?
278 0 : StatusCode::INTERNAL_SERVER_ERROR,
279 0 : ),
280 : }
281 : }
282 :
283 : // get the list of installed extensions
284 : // currently only used in python tests
285 : // TODO: call it from cplane
286 0 : (&Method::GET, "/installed_extensions") => {
287 0 : info!("serving /installed_extensions GET request");
288 0 : let status = compute.get_status();
289 0 : if status != ComputeStatus::Running {
290 0 : let msg = format!(
291 0 : "invalid compute status for extensions request: {:?}",
292 0 : status
293 0 : );
294 0 : error!(msg);
295 0 : return Response::new(Body::from(msg));
296 0 : }
297 0 :
298 0 : let conf = compute.get_conn_conf(None);
299 0 : let res =
300 0 : task::spawn_blocking(move || installed_extensions::get_installed_extensions(conf))
301 0 : .await
302 0 : .unwrap();
303 0 :
304 0 : match res {
305 0 : Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
306 0 : Err(e) => render_json_error(
307 0 : &format!("could not get list of installed extensions: {}", e),
308 0 : StatusCode::INTERNAL_SERVER_ERROR,
309 0 : ),
310 : }
311 : }
312 :
313 : // download extension files from remote extension storage on demand
314 0 : (&Method::POST, route) if route.starts_with("/extension_server/") => {
315 0 : info!("serving {:?} POST request", route);
316 0 : info!("req.uri {:?}", req.uri());
317 :
318 : // don't even try to download extensions
319 : // if no remote storage is configured
320 0 : if compute.ext_remote_storage.is_none() {
321 0 : info!("no extensions remote storage configured");
322 0 : let mut resp = Response::new(Body::from("no remote storage configured"));
323 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
324 0 : return resp;
325 0 : }
326 0 :
327 0 : let mut is_library = false;
328 0 : if let Some(params) = req.uri().query() {
329 0 : info!("serving {:?} POST request with params: {}", route, params);
330 0 : if params == "is_library=true" {
331 0 : is_library = true;
332 0 : } else {
333 0 : let mut resp = Response::new(Body::from("Wrong request parameters"));
334 0 : *resp.status_mut() = StatusCode::BAD_REQUEST;
335 0 : return resp;
336 : }
337 0 : }
338 0 : let filename = route.split('/').last().unwrap().to_string();
339 0 : info!("serving /extension_server POST request, filename: {filename:?} is_library: {is_library}");
340 :
341 : // get ext_name and path from spec
342 : // don't lock compute_state for too long
343 0 : let ext = {
344 0 : let compute_state = compute.state.lock().unwrap();
345 0 : let pspec = compute_state.pspec.as_ref().expect("spec must be set");
346 0 : let spec = &pspec.spec;
347 0 :
348 0 : // debug only
349 0 : info!("spec: {:?}", spec);
350 :
351 0 : let remote_extensions = match spec.remote_extensions.as_ref() {
352 0 : Some(r) => r,
353 : None => {
354 0 : info!("no remote extensions spec was provided");
355 0 : let mut resp = Response::new(Body::from("no remote storage configured"));
356 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
357 0 : return resp;
358 : }
359 : };
360 :
361 0 : remote_extensions.get_ext(
362 0 : &filename,
363 0 : is_library,
364 0 : &compute.build_tag,
365 0 : &compute.pgversion,
366 0 : )
367 0 : };
368 0 :
369 0 : match ext {
370 0 : Ok((ext_name, ext_path)) => {
371 0 : match compute.download_extension(ext_name, ext_path).await {
372 0 : Ok(_) => Response::new(Body::from("OK")),
373 0 : Err(e) => {
374 0 : error!("extension download failed: {}", e);
375 0 : let mut resp = Response::new(Body::from(e.to_string()));
376 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
377 0 : resp
378 : }
379 : }
380 : }
381 0 : Err(e) => {
382 0 : warn!("extension download failed to find extension: {}", e);
383 0 : let mut resp = Response::new(Body::from("failed to find file"));
384 0 : *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
385 0 : resp
386 : }
387 : }
388 : }
389 :
390 : // Return the `404 Not Found` for any other routes.
391 : _ => {
392 0 : let mut not_found = Response::new(Body::from("404 Not Found"));
393 0 : *not_found.status_mut() = StatusCode::NOT_FOUND;
394 0 : not_found
395 : }
396 : }
397 0 : }
398 :
399 0 : async fn handle_configure_request(
400 0 : req: Request<Body>,
401 0 : compute: &Arc<ComputeNode>,
402 0 : ) -> Result<String, (String, StatusCode)> {
403 0 : if !compute.live_config_allowed {
404 0 : return Err((
405 0 : "live configuration is not allowed for this compute node".to_string(),
406 0 : StatusCode::PRECONDITION_FAILED,
407 0 : ));
408 0 : }
409 :
410 0 : let body_bytes = hyper::body::to_bytes(req.into_body()).await.unwrap();
411 0 : let spec_raw = String::from_utf8(body_bytes.to_vec()).unwrap();
412 0 : if let Ok(request) = serde_json::from_str::<ConfigurationRequest>(&spec_raw) {
413 0 : let spec = request.spec;
414 :
415 0 : let parsed_spec = match ParsedSpec::try_from(spec) {
416 0 : Ok(ps) => ps,
417 0 : Err(msg) => return Err((msg, StatusCode::BAD_REQUEST)),
418 : };
419 :
420 : // XXX: wrap state update under lock in code blocks. Otherwise,
421 : // we will try to `Send` `mut state` into the spawned thread
422 : // bellow, which will cause error:
423 : // ```
424 : // error: future cannot be sent between threads safely
425 : // ```
426 : {
427 0 : let mut state = compute.state.lock().unwrap();
428 0 : if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
429 0 : let msg = format!(
430 0 : "invalid compute status for configuration request: {:?}",
431 0 : state.status.clone()
432 0 : );
433 0 : return Err((msg, StatusCode::PRECONDITION_FAILED));
434 0 : }
435 0 : state.pspec = Some(parsed_spec);
436 0 : state.set_status(ComputeStatus::ConfigurationPending, &compute.state_changed);
437 0 : drop(state);
438 0 : info!("set new spec and notified waiters");
439 : }
440 :
441 : // Spawn a blocking thread to wait for compute to become Running.
442 : // This is needed to do not block the main pool of workers and
443 : // be able to serve other requests while some particular request
444 : // is waiting for compute to finish configuration.
445 0 : let c = compute.clone();
446 0 : task::spawn_blocking(move || {
447 0 : let mut state = c.state.lock().unwrap();
448 0 : while state.status != ComputeStatus::Running {
449 0 : state = c.state_changed.wait(state).unwrap();
450 0 : info!(
451 0 : "waiting for compute to become Running, current status: {:?}",
452 0 : state.status
453 : );
454 :
455 0 : if state.status == ComputeStatus::Failed {
456 0 : let err = state.error.as_ref().map_or("unknown error", |x| x);
457 0 : let msg = format!("compute configuration failed: {:?}", err);
458 0 : return Err((msg, StatusCode::INTERNAL_SERVER_ERROR));
459 0 : }
460 : }
461 :
462 0 : Ok(())
463 0 : })
464 0 : .await
465 0 : .unwrap()?;
466 :
467 : // Return current compute state if everything went well.
468 0 : let state = compute.state.lock().unwrap().clone();
469 0 : let status_response = status_response_from_state(&state);
470 0 : Ok(serde_json::to_string(&status_response).unwrap())
471 : } else {
472 0 : Err(("invalid spec".to_string(), StatusCode::BAD_REQUEST))
473 : }
474 0 : }
475 :
476 0 : fn render_json_error(e: &str, status: StatusCode) -> Response<Body> {
477 0 : let error = GenericAPIError {
478 0 : error: e.to_string(),
479 0 : };
480 0 : Response::builder()
481 0 : .status(status)
482 0 : .header(CONTENT_TYPE, "application/json")
483 0 : .body(Body::from(serde_json::to_string(&error).unwrap()))
484 0 : .unwrap()
485 0 : }
486 :
487 0 : fn render_json(body: Body) -> Response<Body> {
488 0 : Response::builder()
489 0 : .header(CONTENT_TYPE, "application/json")
490 0 : .body(body)
491 0 : .unwrap()
492 0 : }
493 :
494 0 : fn render_plain(body: Body) -> Response<Body> {
495 0 : Response::builder()
496 0 : .header(CONTENT_TYPE, "text/plain")
497 0 : .body(body)
498 0 : .unwrap()
499 0 : }
500 :
501 0 : async fn handle_terminate_request(compute: &Arc<ComputeNode>) -> Result<(), (String, StatusCode)> {
502 0 : {
503 0 : let mut state = compute.state.lock().unwrap();
504 0 : if state.status == ComputeStatus::Terminated {
505 0 : return Ok(());
506 0 : }
507 0 : if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
508 0 : let msg = format!(
509 0 : "invalid compute status for termination request: {}",
510 0 : state.status
511 0 : );
512 0 : return Err((msg, StatusCode::PRECONDITION_FAILED));
513 0 : }
514 0 : state.set_status(ComputeStatus::TerminationPending, &compute.state_changed);
515 0 : drop(state);
516 0 : }
517 0 :
518 0 : forward_termination_signal();
519 0 : info!("sent signal and notified waiters");
520 :
521 : // Spawn a blocking thread to wait for compute to become Terminated.
522 : // This is needed to do not block the main pool of workers and
523 : // be able to serve other requests while some particular request
524 : // is waiting for compute to finish configuration.
525 0 : let c = compute.clone();
526 0 : task::spawn_blocking(move || {
527 0 : let mut state = c.state.lock().unwrap();
528 0 : while state.status != ComputeStatus::Terminated {
529 0 : state = c.state_changed.wait(state).unwrap();
530 0 : info!(
531 0 : "waiting for compute to become {}, current status: {:?}",
532 0 : ComputeStatus::Terminated,
533 0 : state.status
534 : );
535 : }
536 :
537 0 : Ok(())
538 0 : })
539 0 : .await
540 0 : .unwrap()?;
541 0 : info!("terminated Postgres");
542 0 : Ok(())
543 0 : }
544 :
545 : // Main Hyper HTTP server function that runs it and blocks waiting on it forever.
546 : #[tokio::main]
547 0 : async fn serve(port: u16, state: Arc<ComputeNode>) {
548 0 : // this usually binds to both IPv4 and IPv6 on linux
549 0 : // see e.g. https://github.com/rust-lang/rust/pull/34440
550 0 : let addr = SocketAddr::new(IpAddr::from(Ipv6Addr::UNSPECIFIED), port);
551 0 :
552 0 : let make_service = make_service_fn(move |_conn| {
553 0 : let state = state.clone();
554 0 : async move {
555 0 : Ok::<_, Infallible>(service_fn(move |req: Request<Body>| {
556 0 : let state = state.clone();
557 0 : async move {
558 0 : Ok::<_, Infallible>(
559 0 : // NOTE: We include the URI path in the string. It
560 0 : // doesn't contain any variable parts or sensitive
561 0 : // information in this API.
562 0 : tracing_utils::http::tracing_handler(
563 0 : req,
564 0 : |req| routes(req, &state),
565 0 : OtelName::UriPath,
566 0 : )
567 0 : .await,
568 0 : )
569 0 : }
570 0 : }))
571 0 : }
572 0 : });
573 0 :
574 0 : info!("starting HTTP server on {}", addr);
575 0 :
576 0 : let server = Server::bind(&addr).serve(make_service);
577 0 :
578 0 : // Run this server forever
579 0 : if let Err(e) = server.await {
580 0 : error!("server error: {}", e);
581 0 : }
582 0 : }
583 :
584 : /// Launch a separate Hyper HTTP API server thread and return its `JoinHandle`.
585 0 : pub fn launch_http_server(port: u16, state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
586 0 : let state = Arc::clone(state);
587 0 :
588 0 : Ok(thread::Builder::new()
589 0 : .name("http-endpoint".into())
590 0 : .spawn(move || serve(port, state))?)
591 0 : }
|
