Line data Source code
1 : use std::str::FromStr;
2 : use std::sync::{Arc, LazyLock};
3 : use std::time::{Duration, Instant};
4 :
5 : use anyhow::Context;
6 : use control_plane::storage_controller::{AttachHookRequest, InspectRequest};
7 : use futures::Future;
8 : use http_utils::endpoint::{
9 : self, auth_middleware, check_permission_with, profile_cpu_handler, profile_heap_handler,
10 : request_span,
11 : };
12 : use http_utils::error::ApiError;
13 : use http_utils::failpoints::failpoints_handler;
14 : use http_utils::json::{json_request, json_response};
15 : use http_utils::request::{must_get_query_param, parse_query_param, parse_request_param};
16 : use http_utils::{RequestExt, RouterBuilder};
17 : use hyper::header::CONTENT_TYPE;
18 : use hyper::{Body, Request, Response, StatusCode, Uri};
19 : use metrics::{BuildInfo, NeonMetrics};
20 : use pageserver_api::controller_api::{
21 : MetadataHealthListOutdatedRequest, MetadataHealthListOutdatedResponse,
22 : MetadataHealthListUnhealthyResponse, MetadataHealthUpdateRequest, MetadataHealthUpdateResponse,
23 : NodeAvailability, NodeConfigureRequest, NodeRegisterRequest, SafekeeperSchedulingPolicyRequest,
24 : ShardsPreferredAzsRequest, TenantCreateRequest, TenantPolicyRequest, TenantShardMigrateRequest,
25 : };
26 : use pageserver_api::models::{
27 : TenantConfigPatchRequest, TenantConfigRequest, TenantLocationConfigRequest,
28 : TenantShardSplitRequest, TenantTimeTravelRequest, TimelineArchivalConfigRequest,
29 : TimelineCreateRequest,
30 : };
31 : use pageserver_api::shard::TenantShardId;
32 : use pageserver_api::upcall_api::{ReAttachRequest, ValidateRequest};
33 : use pageserver_client::{BlockUnblock, mgmt_api};
34 : use routerify::Middleware;
35 : use tokio_util::sync::CancellationToken;
36 : use tracing::warn;
37 : use utils::auth::{Scope, SwappableJwtAuth};
38 : use utils::id::{NodeId, TenantId, TimelineId};
39 :
40 : use crate::http;
41 : use crate::metrics::{
42 : HttpRequestLatencyLabelGroup, HttpRequestStatusLabelGroup, METRICS_REGISTRY,
43 : PageserverRequestLabelGroup,
44 : };
45 : use crate::persistence::SafekeeperUpsert;
46 : use crate::reconciler::ReconcileError;
47 : use crate::service::{LeadershipStatus, RECONCILE_TIMEOUT, STARTUP_RECONCILE_TIMEOUT, Service};
48 :
49 : /// State available to HTTP request handlers
50 : pub struct HttpState {
51 : service: Arc<crate::service::Service>,
52 : auth: Option<Arc<SwappableJwtAuth>>,
53 : rate_limiter: governor::DefaultKeyedRateLimiter<TenantId>,
54 : neon_metrics: NeonMetrics,
55 : allowlist_routes: &'static [&'static str],
56 : }
57 :
58 : impl HttpState {
59 0 : pub fn new(
60 0 : service: Arc<crate::service::Service>,
61 0 : auth: Option<Arc<SwappableJwtAuth>>,
62 0 : build_info: BuildInfo,
63 0 : ) -> Self {
64 0 : let quota = governor::Quota::per_second(service.get_config().tenant_rate_limit);
65 0 : Self {
66 0 : service,
67 0 : auth,
68 0 : rate_limiter: governor::RateLimiter::keyed(quota),
69 0 : neon_metrics: NeonMetrics::new(build_info),
70 0 : allowlist_routes: &[
71 0 : "/status",
72 0 : "/ready",
73 0 : "/metrics",
74 0 : "/profile/cpu",
75 0 : "/profile/heap",
76 0 : ],
77 0 : }
78 0 : }
79 : }
80 :
81 : #[inline(always)]
82 0 : fn get_state(request: &Request<Body>) -> &HttpState {
83 0 : request
84 0 : .data::<Arc<HttpState>>()
85 0 : .expect("unknown state type")
86 0 : .as_ref()
87 0 : }
88 :
89 : /// Rate limits tenant requests.
90 : ///
91 : /// TODO: this should be a request middleware, but requires us to extract the tenant ID from
92 : /// different URLs in a systematic way.
93 : ///
94 : /// TODO: consider returning a 429 response if these start piling up.
95 0 : async fn maybe_rate_limit(request: &Request<Body>, tenant_id: TenantId) {
96 0 : // Check if the tenant should be rate-limited.
97 0 : let rate_limiter = &get_state(request).rate_limiter;
98 0 : if rate_limiter.check_key(&tenant_id).is_ok() {
99 0 : return;
100 0 : }
101 0 :
102 0 : // Measure the rate limiting delay.
103 0 : let _timer = METRICS_REGISTRY
104 0 : .metrics_group
105 0 : .storage_controller_http_request_rate_limited
106 0 : .start_timer();
107 :
108 : // Log rate limited tenants once every 10 seconds.
109 : static LOG_RATE_LIMITER: LazyLock<governor::DefaultKeyedRateLimiter<TenantId>> =
110 0 : LazyLock::new(|| {
111 0 : let quota = governor::Quota::with_period(Duration::from_secs(10)).unwrap();
112 0 : governor::RateLimiter::keyed(quota)
113 0 : });
114 :
115 0 : if LOG_RATE_LIMITER.check_key(&tenant_id).is_ok() {
116 0 : warn!("tenant {tenant_id} is rate limited")
117 0 : }
118 :
119 : // Wait for quota.
120 0 : rate_limiter.until_key_ready(&tenant_id).await;
121 0 : }
122 :
123 : /// Pageserver calls into this on startup, to learn which tenants it should attach
124 0 : async fn handle_re_attach(req: Request<Body>) -> Result<Response<Body>, ApiError> {
125 0 : check_permissions(&req, Scope::GenerationsApi)?;
126 :
127 0 : let mut req = match maybe_forward(req).await {
128 0 : ForwardOutcome::Forwarded(res) => {
129 0 : return res;
130 : }
131 0 : ForwardOutcome::NotForwarded(req) => req,
132 : };
133 :
134 0 : let reattach_req = json_request::<ReAttachRequest>(&mut req).await?;
135 0 : let state = get_state(&req);
136 0 : json_response(StatusCode::OK, state.service.re_attach(reattach_req).await?)
137 0 : }
138 :
139 : /// Pageserver calls into this before doing deletions, to confirm that it still
140 : /// holds the latest generation for the tenants with deletions enqueued
141 0 : async fn handle_validate(req: Request<Body>) -> Result<Response<Body>, ApiError> {
142 0 : check_permissions(&req, Scope::GenerationsApi)?;
143 :
144 0 : let mut req = match maybe_forward(req).await {
145 0 : ForwardOutcome::Forwarded(res) => {
146 0 : return res;
147 : }
148 0 : ForwardOutcome::NotForwarded(req) => req,
149 : };
150 :
151 0 : let validate_req = json_request::<ValidateRequest>(&mut req).await?;
152 0 : let state = get_state(&req);
153 0 : json_response(StatusCode::OK, state.service.validate(validate_req).await?)
154 0 : }
155 :
156 : /// Call into this before attaching a tenant to a pageserver, to acquire a generation number
157 : /// (in the real control plane this is unnecessary, because the same program is managing
158 : /// generation numbers and doing attachments).
159 0 : async fn handle_attach_hook(req: Request<Body>) -> Result<Response<Body>, ApiError> {
160 0 : check_permissions(&req, Scope::Admin)?;
161 :
162 0 : let mut req = match maybe_forward(req).await {
163 0 : ForwardOutcome::Forwarded(res) => {
164 0 : return res;
165 : }
166 0 : ForwardOutcome::NotForwarded(req) => req,
167 : };
168 :
169 0 : let attach_req = json_request::<AttachHookRequest>(&mut req).await?;
170 0 : let state = get_state(&req);
171 0 :
172 0 : json_response(
173 0 : StatusCode::OK,
174 0 : state
175 0 : .service
176 0 : .attach_hook(attach_req)
177 0 : .await
178 0 : .map_err(ApiError::InternalServerError)?,
179 : )
180 0 : }
181 :
182 0 : async fn handle_inspect(req: Request<Body>) -> Result<Response<Body>, ApiError> {
183 0 : check_permissions(&req, Scope::Admin)?;
184 :
185 0 : let mut req = match maybe_forward(req).await {
186 0 : ForwardOutcome::Forwarded(res) => {
187 0 : return res;
188 : }
189 0 : ForwardOutcome::NotForwarded(req) => req,
190 : };
191 :
192 0 : let inspect_req = json_request::<InspectRequest>(&mut req).await?;
193 :
194 0 : let state = get_state(&req);
195 0 :
196 0 : json_response(StatusCode::OK, state.service.inspect(inspect_req))
197 0 : }
198 :
199 0 : async fn handle_tenant_create(
200 0 : service: Arc<Service>,
201 0 : req: Request<Body>,
202 0 : ) -> Result<Response<Body>, ApiError> {
203 0 : check_permissions(&req, Scope::PageServerApi)?;
204 :
205 0 : let mut req = match maybe_forward(req).await {
206 0 : ForwardOutcome::Forwarded(res) => {
207 0 : return res;
208 : }
209 0 : ForwardOutcome::NotForwarded(req) => req,
210 : };
211 :
212 0 : let create_req = json_request::<TenantCreateRequest>(&mut req).await?;
213 :
214 : json_response(
215 : StatusCode::CREATED,
216 0 : service.tenant_create(create_req).await?,
217 : )
218 0 : }
219 :
220 0 : async fn handle_tenant_location_config(
221 0 : service: Arc<Service>,
222 0 : req: Request<Body>,
223 0 : ) -> Result<Response<Body>, ApiError> {
224 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
225 0 : check_permissions(&req, Scope::PageServerApi)?;
226 :
227 0 : let mut req = match maybe_forward(req).await {
228 0 : ForwardOutcome::Forwarded(res) => {
229 0 : return res;
230 : }
231 0 : ForwardOutcome::NotForwarded(req) => req,
232 : };
233 :
234 0 : let config_req = json_request::<TenantLocationConfigRequest>(&mut req).await?;
235 : json_response(
236 : StatusCode::OK,
237 0 : service
238 0 : .tenant_location_config(tenant_shard_id, config_req)
239 0 : .await?,
240 : )
241 0 : }
242 :
243 0 : async fn handle_tenant_config_patch(
244 0 : service: Arc<Service>,
245 0 : req: Request<Body>,
246 0 : ) -> Result<Response<Body>, ApiError> {
247 0 : check_permissions(&req, Scope::PageServerApi)?;
248 :
249 0 : let mut req = match maybe_forward(req).await {
250 0 : ForwardOutcome::Forwarded(res) => {
251 0 : return res;
252 : }
253 0 : ForwardOutcome::NotForwarded(req) => req,
254 : };
255 :
256 0 : let config_req = json_request::<TenantConfigPatchRequest>(&mut req).await?;
257 :
258 : json_response(
259 : StatusCode::OK,
260 0 : service.tenant_config_patch(config_req).await?,
261 : )
262 0 : }
263 :
264 0 : async fn handle_tenant_config_set(
265 0 : service: Arc<Service>,
266 0 : req: Request<Body>,
267 0 : ) -> Result<Response<Body>, ApiError> {
268 0 : check_permissions(&req, Scope::PageServerApi)?;
269 :
270 0 : let mut req = match maybe_forward(req).await {
271 0 : ForwardOutcome::Forwarded(res) => {
272 0 : return res;
273 : }
274 0 : ForwardOutcome::NotForwarded(req) => req,
275 : };
276 :
277 0 : let config_req = json_request::<TenantConfigRequest>(&mut req).await?;
278 :
279 0 : json_response(StatusCode::OK, service.tenant_config_set(config_req).await?)
280 0 : }
281 :
282 0 : async fn handle_tenant_config_get(
283 0 : service: Arc<Service>,
284 0 : req: Request<Body>,
285 0 : ) -> Result<Response<Body>, ApiError> {
286 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
287 0 : check_permissions(&req, Scope::PageServerApi)?;
288 0 : maybe_rate_limit(&req, tenant_id).await;
289 :
290 0 : match maybe_forward(req).await {
291 0 : ForwardOutcome::Forwarded(res) => {
292 0 : return res;
293 : }
294 0 : ForwardOutcome::NotForwarded(_req) => {}
295 0 : };
296 0 :
297 0 : json_response(StatusCode::OK, service.tenant_config_get(tenant_id)?)
298 0 : }
299 :
300 0 : async fn handle_tenant_time_travel_remote_storage(
301 0 : service: Arc<Service>,
302 0 : req: Request<Body>,
303 0 : ) -> Result<Response<Body>, ApiError> {
304 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
305 0 : check_permissions(&req, Scope::PageServerApi)?;
306 0 : maybe_rate_limit(&req, tenant_id).await;
307 :
308 0 : let mut req = match maybe_forward(req).await {
309 0 : ForwardOutcome::Forwarded(res) => {
310 0 : return res;
311 : }
312 0 : ForwardOutcome::NotForwarded(req) => req,
313 : };
314 :
315 0 : let time_travel_req = json_request::<TenantTimeTravelRequest>(&mut req).await?;
316 :
317 0 : let timestamp_raw = must_get_query_param(&req, "travel_to")?;
318 0 : let _timestamp = humantime::parse_rfc3339(×tamp_raw).map_err(|_e| {
319 0 : ApiError::BadRequest(anyhow::anyhow!(
320 0 : "Invalid time for travel_to: {timestamp_raw:?}"
321 0 : ))
322 0 : })?;
323 :
324 0 : let done_if_after_raw = must_get_query_param(&req, "done_if_after")?;
325 0 : let _done_if_after = humantime::parse_rfc3339(&done_if_after_raw).map_err(|_e| {
326 0 : ApiError::BadRequest(anyhow::anyhow!(
327 0 : "Invalid time for done_if_after: {done_if_after_raw:?}"
328 0 : ))
329 0 : })?;
330 :
331 0 : service
332 0 : .tenant_time_travel_remote_storage(
333 0 : &time_travel_req,
334 0 : tenant_id,
335 0 : timestamp_raw,
336 0 : done_if_after_raw,
337 0 : )
338 0 : .await?;
339 0 : json_response(StatusCode::OK, ())
340 0 : }
341 :
342 0 : fn map_reqwest_hyper_status(status: reqwest::StatusCode) -> Result<hyper::StatusCode, ApiError> {
343 0 : hyper::StatusCode::from_u16(status.as_u16())
344 0 : .context("invalid status code")
345 0 : .map_err(ApiError::InternalServerError)
346 0 : }
347 :
348 0 : async fn handle_tenant_secondary_download(
349 0 : service: Arc<Service>,
350 0 : req: Request<Body>,
351 0 : ) -> Result<Response<Body>, ApiError> {
352 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
353 0 : let wait = parse_query_param(&req, "wait_ms")?.map(Duration::from_millis);
354 0 : maybe_rate_limit(&req, tenant_id).await;
355 :
356 0 : match maybe_forward(req).await {
357 0 : ForwardOutcome::Forwarded(res) => {
358 0 : return res;
359 : }
360 0 : ForwardOutcome::NotForwarded(_req) => {}
361 : };
362 :
363 0 : let (status, progress) = service.tenant_secondary_download(tenant_id, wait).await?;
364 0 : json_response(map_reqwest_hyper_status(status)?, progress)
365 0 : }
366 :
367 0 : async fn handle_tenant_delete(
368 0 : service: Arc<Service>,
369 0 : req: Request<Body>,
370 0 : ) -> Result<Response<Body>, ApiError> {
371 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
372 0 : check_permissions(&req, Scope::PageServerApi)?;
373 0 : maybe_rate_limit(&req, tenant_id).await;
374 :
375 0 : match maybe_forward(req).await {
376 0 : ForwardOutcome::Forwarded(res) => {
377 0 : return res;
378 : }
379 0 : ForwardOutcome::NotForwarded(_req) => {}
380 : };
381 :
382 0 : let status_code = service
383 0 : .tenant_delete(tenant_id)
384 0 : .await
385 0 : .and_then(map_reqwest_hyper_status)?;
386 :
387 0 : if status_code == StatusCode::NOT_FOUND {
388 : // The pageserver uses 404 for successful deletion, but we use 200
389 0 : json_response(StatusCode::OK, ())
390 : } else {
391 0 : json_response(status_code, ())
392 : }
393 0 : }
394 :
395 0 : async fn handle_tenant_timeline_create(
396 0 : service: Arc<Service>,
397 0 : req: Request<Body>,
398 0 : ) -> Result<Response<Body>, ApiError> {
399 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
400 0 : check_permissions(&req, Scope::PageServerApi)?;
401 0 : maybe_rate_limit(&req, tenant_id).await;
402 :
403 0 : let mut req = match maybe_forward(req).await {
404 0 : ForwardOutcome::Forwarded(res) => {
405 0 : return res;
406 : }
407 0 : ForwardOutcome::NotForwarded(req) => req,
408 : };
409 :
410 0 : let create_req = json_request::<TimelineCreateRequest>(&mut req).await?;
411 : json_response(
412 : StatusCode::CREATED,
413 0 : service
414 0 : .tenant_timeline_create(tenant_id, create_req)
415 0 : .await?,
416 : )
417 0 : }
418 :
419 0 : async fn handle_tenant_timeline_delete(
420 0 : service: Arc<Service>,
421 0 : req: Request<Body>,
422 0 : ) -> Result<Response<Body>, ApiError> {
423 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
424 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
425 :
426 0 : check_permissions(&req, Scope::PageServerApi)?;
427 0 : maybe_rate_limit(&req, tenant_id).await;
428 :
429 0 : match maybe_forward(req).await {
430 0 : ForwardOutcome::Forwarded(res) => {
431 0 : return res;
432 : }
433 0 : ForwardOutcome::NotForwarded(_req) => {}
434 : };
435 :
436 : // For timeline deletions, which both implement an "initially return 202, then 404 once
437 : // we're done" semantic, we wrap with a retry loop to expose a simpler API upstream.
438 0 : async fn deletion_wrapper<R, F>(service: Arc<Service>, f: F) -> Result<Response<Body>, ApiError>
439 0 : where
440 0 : R: std::future::Future<Output = Result<StatusCode, ApiError>> + Send + 'static,
441 0 : F: Fn(Arc<Service>) -> R + Send + Sync + 'static,
442 0 : {
443 : // On subsequent retries, wait longer.
444 : // Enable callers with a 25 second request timeout to reliably get a response
445 : const MAX_WAIT: Duration = Duration::from_secs(25);
446 : const MAX_RETRY_PERIOD: Duration = Duration::from_secs(5);
447 :
448 0 : let started_at = Instant::now();
449 0 :
450 0 : // To keep deletion reasonably snappy for small tenants, initially check after 1 second if deletion
451 0 : // completed.
452 0 : let mut retry_period = Duration::from_secs(1);
453 :
454 : loop {
455 0 : let status = f(service.clone()).await?;
456 0 : match status {
457 : StatusCode::ACCEPTED => {
458 0 : tracing::info!("Deletion accepted, waiting to try again...");
459 0 : tokio::time::sleep(retry_period).await;
460 0 : retry_period = MAX_RETRY_PERIOD;
461 : }
462 : StatusCode::CONFLICT => {
463 0 : tracing::info!("Deletion already in progress, waiting to try again...");
464 0 : tokio::time::sleep(retry_period).await;
465 : }
466 : StatusCode::NOT_FOUND => {
467 0 : tracing::info!("Deletion complete");
468 0 : return json_response(StatusCode::OK, ());
469 : }
470 : _ => {
471 0 : tracing::warn!("Unexpected status {status}");
472 0 : return json_response(status, ());
473 : }
474 : }
475 :
476 0 : let now = Instant::now();
477 0 : if now + retry_period > started_at + MAX_WAIT {
478 0 : tracing::info!("Deletion timed out waiting for 404");
479 : // REQUEST_TIMEOUT would be more appropriate, but CONFLICT is already part of
480 : // the pageserver's swagger definition for this endpoint, and has the same desired
481 : // effect of causing the control plane to retry later.
482 0 : return json_response(StatusCode::CONFLICT, ());
483 0 : }
484 : }
485 0 : }
486 :
487 0 : deletion_wrapper(service, move |service| async move {
488 0 : service
489 0 : .tenant_timeline_delete(tenant_id, timeline_id)
490 0 : .await
491 0 : .and_then(map_reqwest_hyper_status)
492 0 : })
493 0 : .await
494 0 : }
495 :
496 0 : async fn handle_tenant_timeline_archival_config(
497 0 : service: Arc<Service>,
498 0 : req: Request<Body>,
499 0 : ) -> Result<Response<Body>, ApiError> {
500 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
501 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
502 :
503 0 : check_permissions(&req, Scope::PageServerApi)?;
504 0 : maybe_rate_limit(&req, tenant_id).await;
505 :
506 0 : let mut req = match maybe_forward(req).await {
507 0 : ForwardOutcome::Forwarded(res) => {
508 0 : return res;
509 : }
510 0 : ForwardOutcome::NotForwarded(req) => req,
511 : };
512 :
513 0 : let create_req = json_request::<TimelineArchivalConfigRequest>(&mut req).await?;
514 :
515 0 : service
516 0 : .tenant_timeline_archival_config(tenant_id, timeline_id, create_req)
517 0 : .await?;
518 :
519 0 : json_response(StatusCode::OK, ())
520 0 : }
521 :
522 0 : async fn handle_tenant_timeline_detach_ancestor(
523 0 : service: Arc<Service>,
524 0 : req: Request<Body>,
525 0 : ) -> Result<Response<Body>, ApiError> {
526 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
527 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
528 :
529 0 : check_permissions(&req, Scope::PageServerApi)?;
530 0 : maybe_rate_limit(&req, tenant_id).await;
531 :
532 0 : match maybe_forward(req).await {
533 0 : ForwardOutcome::Forwarded(res) => {
534 0 : return res;
535 : }
536 0 : ForwardOutcome::NotForwarded(_req) => {}
537 : };
538 :
539 0 : let res = service
540 0 : .tenant_timeline_detach_ancestor(tenant_id, timeline_id)
541 0 : .await?;
542 :
543 0 : json_response(StatusCode::OK, res)
544 0 : }
545 :
546 0 : async fn handle_tenant_timeline_block_unblock_gc(
547 0 : service: Arc<Service>,
548 0 : req: Request<Body>,
549 0 : dir: BlockUnblock,
550 0 : ) -> Result<Response<Body>, ApiError> {
551 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
552 0 : check_permissions(&req, Scope::PageServerApi)?;
553 0 : maybe_rate_limit(&req, tenant_id).await;
554 :
555 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
556 :
557 0 : service
558 0 : .tenant_timeline_block_unblock_gc(tenant_id, timeline_id, dir)
559 0 : .await?;
560 :
561 0 : json_response(StatusCode::OK, ())
562 0 : }
563 :
564 0 : async fn handle_tenant_timeline_download_heatmap_layers(
565 0 : service: Arc<Service>,
566 0 : req: Request<Body>,
567 0 : ) -> Result<Response<Body>, ApiError> {
568 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
569 :
570 0 : check_permissions(&req, Scope::PageServerApi)?;
571 0 : maybe_rate_limit(&req, tenant_shard_id.tenant_id).await;
572 :
573 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
574 0 : let concurrency: Option<usize> = parse_query_param(&req, "concurrency")?;
575 0 : let recurse = parse_query_param(&req, "recurse")?.unwrap_or(false);
576 0 :
577 0 : service
578 0 : .tenant_timeline_download_heatmap_layers(tenant_shard_id, timeline_id, concurrency, recurse)
579 0 : .await?;
580 :
581 0 : json_response(StatusCode::OK, ())
582 0 : }
583 :
584 : // For metric labels where we would like to include the approximate path, but exclude high-cardinality fields like query parameters
585 : // and tenant/timeline IDs. Since we are proxying to arbitrary paths, we don't have routing templates to
586 : // compare to, so we can just filter out our well known ID format with regexes.
587 3 : fn path_without_ids(path: &str) -> String {
588 : static ID_REGEX: std::sync::OnceLock<regex::Regex> = std::sync::OnceLock::new();
589 3 : ID_REGEX
590 3 : .get_or_init(|| regex::Regex::new(r"([0-9a-fA-F]{32}(-[0-9]{4})?|\?.*)").unwrap())
591 3 : .replace_all(path, "")
592 3 : .to_string()
593 3 : }
594 :
595 0 : async fn handle_tenant_timeline_passthrough(
596 0 : service: Arc<Service>,
597 0 : req: Request<Body>,
598 0 : ) -> Result<Response<Body>, ApiError> {
599 0 : let tenant_or_shard_id: TenantShardId = parse_request_param(&req, "tenant_id")?;
600 0 : check_permissions(&req, Scope::PageServerApi)?;
601 0 : maybe_rate_limit(&req, tenant_or_shard_id.tenant_id).await;
602 :
603 0 : let req = match maybe_forward(req).await {
604 0 : ForwardOutcome::Forwarded(res) => {
605 0 : return res;
606 : }
607 0 : ForwardOutcome::NotForwarded(req) => req,
608 : };
609 :
610 0 : let Some(path) = req.uri().path_and_query() else {
611 : // This should never happen, our request router only calls us if there is a path
612 0 : return Err(ApiError::BadRequest(anyhow::anyhow!("Missing path")));
613 : };
614 :
615 0 : tracing::info!(
616 0 : "Proxying request for tenant {} ({})",
617 : tenant_or_shard_id.tenant_id,
618 : path
619 : );
620 :
621 : // Find the node that holds shard zero
622 0 : let (node, tenant_shard_id) = if tenant_or_shard_id.is_unsharded() {
623 0 : service
624 0 : .tenant_shard0_node(tenant_or_shard_id.tenant_id)
625 0 : .await?
626 : } else {
627 : (
628 0 : service.tenant_shard_node(tenant_or_shard_id).await?,
629 0 : tenant_or_shard_id,
630 : )
631 : };
632 :
633 : // Callers will always pass an unsharded tenant ID. Before proxying, we must
634 : // rewrite this to a shard-aware shard zero ID.
635 0 : let path = format!("{}", path);
636 0 : let tenant_str = tenant_or_shard_id.tenant_id.to_string();
637 0 : let tenant_shard_str = format!("{}", tenant_shard_id);
638 0 : let path = path.replace(&tenant_str, &tenant_shard_str);
639 0 :
640 0 : let latency = &METRICS_REGISTRY
641 0 : .metrics_group
642 0 : .storage_controller_passthrough_request_latency;
643 0 :
644 0 : let path_label = path_without_ids(&path)
645 0 : .split('/')
646 0 : .filter(|token| !token.is_empty())
647 0 : .collect::<Vec<_>>()
648 0 : .join("_");
649 0 : let labels = PageserverRequestLabelGroup {
650 0 : pageserver_id: &node.get_id().to_string(),
651 0 : path: &path_label,
652 0 : method: crate::metrics::Method::Get,
653 0 : };
654 0 :
655 0 : let _timer = latency.start_timer(labels.clone());
656 :
657 0 : let client = mgmt_api::Client::new(
658 0 : node.base_url(),
659 0 : service.get_config().pageserver_jwt_token.as_deref(),
660 0 : service.get_config().ssl_ca_cert.clone(),
661 0 : )
662 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
663 0 : let resp = client.get_raw(path).await.map_err(|e|
664 : // We return 503 here because if we can't successfully send a request to the pageserver,
665 : // either we aren't available or the pageserver is unavailable.
666 0 : ApiError::ResourceUnavailable(format!("Error sending pageserver API request to {node}: {e}").into()))?;
667 :
668 0 : if !resp.status().is_success() {
669 0 : let error_counter = &METRICS_REGISTRY
670 0 : .metrics_group
671 0 : .storage_controller_passthrough_request_error;
672 0 : error_counter.inc(labels);
673 0 : }
674 :
675 : // Transform 404 into 503 if we raced with a migration
676 0 : if resp.status() == reqwest::StatusCode::NOT_FOUND {
677 : // Look up node again: if we migrated it will be different
678 0 : let new_node = service.tenant_shard_node(tenant_shard_id).await?;
679 0 : if new_node.get_id() != node.get_id() {
680 : // Rather than retry here, send the client a 503 to prompt a retry: this matches
681 : // the pageserver's use of 503, and all clients calling this API should retry on 503.
682 0 : return Err(ApiError::ResourceUnavailable(
683 0 : format!("Pageserver {node} returned 404, was migrated to {new_node}").into(),
684 0 : ));
685 0 : }
686 0 : }
687 :
688 : // We have a reqest::Response, would like a http::Response
689 0 : let mut builder = hyper::Response::builder().status(map_reqwest_hyper_status(resp.status())?);
690 0 : for (k, v) in resp.headers() {
691 0 : builder = builder.header(k.as_str(), v.as_bytes());
692 0 : }
693 :
694 0 : let response = builder
695 0 : .body(Body::wrap_stream(resp.bytes_stream()))
696 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
697 :
698 0 : Ok(response)
699 0 : }
700 :
701 0 : async fn handle_tenant_locate(
702 0 : service: Arc<Service>,
703 0 : req: Request<Body>,
704 0 : ) -> Result<Response<Body>, ApiError> {
705 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
706 :
707 0 : check_permissions(&req, Scope::Admin)?;
708 : // NB: don't rate limit: admin operation.
709 :
710 0 : match maybe_forward(req).await {
711 0 : ForwardOutcome::Forwarded(res) => {
712 0 : return res;
713 : }
714 0 : ForwardOutcome::NotForwarded(_req) => {}
715 0 : };
716 0 :
717 0 : json_response(StatusCode::OK, service.tenant_locate(tenant_id)?)
718 0 : }
719 :
720 0 : async fn handle_tenant_describe(
721 0 : service: Arc<Service>,
722 0 : req: Request<Body>,
723 0 : ) -> Result<Response<Body>, ApiError> {
724 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
725 0 : check_permissions(&req, Scope::Scrubber)?;
726 : // NB: don't rate limit: scrubber operation.
727 :
728 0 : match maybe_forward(req).await {
729 0 : ForwardOutcome::Forwarded(res) => {
730 0 : return res;
731 : }
732 0 : ForwardOutcome::NotForwarded(_req) => {}
733 0 : };
734 0 :
735 0 : json_response(StatusCode::OK, service.tenant_describe(tenant_id)?)
736 0 : }
737 :
738 0 : async fn handle_tenant_list(
739 0 : service: Arc<Service>,
740 0 : req: Request<Body>,
741 0 : ) -> Result<Response<Body>, ApiError> {
742 0 : check_permissions(&req, Scope::Admin)?;
743 :
744 0 : let limit: Option<usize> = parse_query_param(&req, "limit")?;
745 0 : let start_after: Option<TenantId> = parse_query_param(&req, "start_after")?;
746 0 : tracing::info!("start_after: {:?}", start_after);
747 :
748 0 : match maybe_forward(req).await {
749 0 : ForwardOutcome::Forwarded(res) => {
750 0 : return res;
751 : }
752 0 : ForwardOutcome::NotForwarded(_req) => {}
753 0 : };
754 0 :
755 0 : json_response(StatusCode::OK, service.tenant_list(limit, start_after))
756 0 : }
757 :
758 0 : async fn handle_node_register(req: Request<Body>) -> Result<Response<Body>, ApiError> {
759 0 : check_permissions(&req, Scope::Infra)?;
760 :
761 0 : let mut req = match maybe_forward(req).await {
762 0 : ForwardOutcome::Forwarded(res) => {
763 0 : return res;
764 : }
765 0 : ForwardOutcome::NotForwarded(req) => req,
766 : };
767 :
768 0 : let register_req = json_request::<NodeRegisterRequest>(&mut req).await?;
769 0 : let state = get_state(&req);
770 0 : state.service.node_register(register_req).await?;
771 0 : json_response(StatusCode::OK, ())
772 0 : }
773 :
774 0 : async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
775 0 : check_permissions(&req, Scope::Infra)?;
776 :
777 0 : let req = match maybe_forward(req).await {
778 0 : ForwardOutcome::Forwarded(res) => {
779 0 : return res;
780 : }
781 0 : ForwardOutcome::NotForwarded(req) => req,
782 0 : };
783 0 :
784 0 : let state = get_state(&req);
785 0 : let mut nodes = state.service.node_list().await?;
786 0 : nodes.sort_by_key(|n| n.get_id());
787 0 : let api_nodes = nodes.into_iter().map(|n| n.describe()).collect::<Vec<_>>();
788 0 :
789 0 : json_response(StatusCode::OK, api_nodes)
790 0 : }
791 :
792 0 : async fn handle_node_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
793 0 : check_permissions(&req, Scope::Admin)?;
794 :
795 0 : let req = match maybe_forward(req).await {
796 0 : ForwardOutcome::Forwarded(res) => {
797 0 : return res;
798 : }
799 0 : ForwardOutcome::NotForwarded(req) => req,
800 0 : };
801 0 :
802 0 : let state = get_state(&req);
803 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
804 0 : json_response(StatusCode::OK, state.service.node_drop(node_id).await?)
805 0 : }
806 :
807 0 : async fn handle_node_delete(req: Request<Body>) -> Result<Response<Body>, ApiError> {
808 0 : check_permissions(&req, Scope::Admin)?;
809 :
810 0 : let req = match maybe_forward(req).await {
811 0 : ForwardOutcome::Forwarded(res) => {
812 0 : return res;
813 : }
814 0 : ForwardOutcome::NotForwarded(req) => req,
815 0 : };
816 0 :
817 0 : let state = get_state(&req);
818 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
819 0 : json_response(StatusCode::OK, state.service.node_delete(node_id).await?)
820 0 : }
821 :
822 0 : async fn handle_node_configure(req: Request<Body>) -> Result<Response<Body>, ApiError> {
823 0 : check_permissions(&req, Scope::Admin)?;
824 :
825 0 : let mut req = match maybe_forward(req).await {
826 0 : ForwardOutcome::Forwarded(res) => {
827 0 : return res;
828 : }
829 0 : ForwardOutcome::NotForwarded(req) => req,
830 : };
831 :
832 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
833 0 : let config_req = json_request::<NodeConfigureRequest>(&mut req).await?;
834 0 : if node_id != config_req.node_id {
835 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
836 0 : "Path and body node_id differ"
837 0 : )));
838 0 : }
839 0 : let state = get_state(&req);
840 0 :
841 0 : json_response(
842 0 : StatusCode::OK,
843 0 : state
844 0 : .service
845 0 : .external_node_configure(
846 0 : config_req.node_id,
847 0 : config_req.availability.map(NodeAvailability::from),
848 0 : config_req.scheduling,
849 0 : )
850 0 : .await?,
851 : )
852 0 : }
853 :
854 0 : async fn handle_node_status(req: Request<Body>) -> Result<Response<Body>, ApiError> {
855 0 : check_permissions(&req, Scope::Infra)?;
856 :
857 0 : let req = match maybe_forward(req).await {
858 0 : ForwardOutcome::Forwarded(res) => {
859 0 : return res;
860 : }
861 0 : ForwardOutcome::NotForwarded(req) => req,
862 0 : };
863 0 :
864 0 : let state = get_state(&req);
865 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
866 :
867 0 : let node_status = state.service.get_node(node_id).await?;
868 :
869 0 : json_response(StatusCode::OK, node_status)
870 0 : }
871 :
872 0 : async fn handle_node_shards(req: Request<Body>) -> Result<Response<Body>, ApiError> {
873 0 : check_permissions(&req, Scope::Admin)?;
874 :
875 0 : let state = get_state(&req);
876 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
877 :
878 0 : let node_status = state.service.get_node_shards(node_id).await?;
879 :
880 0 : json_response(StatusCode::OK, node_status)
881 0 : }
882 :
883 0 : async fn handle_get_leader(req: Request<Body>) -> Result<Response<Body>, ApiError> {
884 0 : check_permissions(&req, Scope::Admin)?;
885 :
886 0 : let req = match maybe_forward(req).await {
887 0 : ForwardOutcome::Forwarded(res) => {
888 0 : return res;
889 : }
890 0 : ForwardOutcome::NotForwarded(req) => req,
891 0 : };
892 0 :
893 0 : let state = get_state(&req);
894 0 : let leader = state.service.get_leader().await.map_err(|err| {
895 0 : ApiError::InternalServerError(anyhow::anyhow!(
896 0 : "Failed to read leader from database: {err}"
897 0 : ))
898 0 : })?;
899 :
900 0 : json_response(StatusCode::OK, leader)
901 0 : }
902 :
903 0 : async fn handle_node_drain(req: Request<Body>) -> Result<Response<Body>, ApiError> {
904 0 : check_permissions(&req, Scope::Infra)?;
905 :
906 0 : let req = match maybe_forward(req).await {
907 0 : ForwardOutcome::Forwarded(res) => {
908 0 : return res;
909 : }
910 0 : ForwardOutcome::NotForwarded(req) => req,
911 0 : };
912 0 :
913 0 : let state = get_state(&req);
914 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
915 :
916 0 : state.service.start_node_drain(node_id).await?;
917 :
918 0 : json_response(StatusCode::ACCEPTED, ())
919 0 : }
920 :
921 0 : async fn handle_cancel_node_drain(req: Request<Body>) -> Result<Response<Body>, ApiError> {
922 0 : check_permissions(&req, Scope::Infra)?;
923 :
924 0 : let req = match maybe_forward(req).await {
925 0 : ForwardOutcome::Forwarded(res) => {
926 0 : return res;
927 : }
928 0 : ForwardOutcome::NotForwarded(req) => req,
929 0 : };
930 0 :
931 0 : let state = get_state(&req);
932 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
933 :
934 0 : state.service.cancel_node_drain(node_id).await?;
935 :
936 0 : json_response(StatusCode::ACCEPTED, ())
937 0 : }
938 :
939 0 : async fn handle_node_fill(req: Request<Body>) -> Result<Response<Body>, ApiError> {
940 0 : check_permissions(&req, Scope::Infra)?;
941 :
942 0 : let req = match maybe_forward(req).await {
943 0 : ForwardOutcome::Forwarded(res) => {
944 0 : return res;
945 : }
946 0 : ForwardOutcome::NotForwarded(req) => req,
947 0 : };
948 0 :
949 0 : let state = get_state(&req);
950 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
951 :
952 0 : state.service.start_node_fill(node_id).await?;
953 :
954 0 : json_response(StatusCode::ACCEPTED, ())
955 0 : }
956 :
957 0 : async fn handle_cancel_node_fill(req: Request<Body>) -> Result<Response<Body>, ApiError> {
958 0 : check_permissions(&req, Scope::Infra)?;
959 :
960 0 : let req = match maybe_forward(req).await {
961 0 : ForwardOutcome::Forwarded(res) => {
962 0 : return res;
963 : }
964 0 : ForwardOutcome::NotForwarded(req) => req,
965 0 : };
966 0 :
967 0 : let state = get_state(&req);
968 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
969 :
970 0 : state.service.cancel_node_fill(node_id).await?;
971 :
972 0 : json_response(StatusCode::ACCEPTED, ())
973 0 : }
974 :
975 0 : async fn handle_safekeeper_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
976 0 : check_permissions(&req, Scope::Infra)?;
977 :
978 0 : let req = match maybe_forward(req).await {
979 0 : ForwardOutcome::Forwarded(res) => {
980 0 : return res;
981 : }
982 0 : ForwardOutcome::NotForwarded(req) => req,
983 0 : };
984 0 :
985 0 : let state = get_state(&req);
986 0 : let safekeepers = state.service.safekeepers_list().await?;
987 0 : json_response(StatusCode::OK, safekeepers)
988 0 : }
989 :
990 0 : async fn handle_metadata_health_update(req: Request<Body>) -> Result<Response<Body>, ApiError> {
991 0 : check_permissions(&req, Scope::Scrubber)?;
992 :
993 0 : let mut req = match maybe_forward(req).await {
994 0 : ForwardOutcome::Forwarded(res) => {
995 0 : return res;
996 : }
997 0 : ForwardOutcome::NotForwarded(req) => req,
998 : };
999 :
1000 0 : let update_req = json_request::<MetadataHealthUpdateRequest>(&mut req).await?;
1001 0 : let state = get_state(&req);
1002 0 :
1003 0 : state.service.metadata_health_update(update_req).await?;
1004 :
1005 0 : json_response(StatusCode::OK, MetadataHealthUpdateResponse {})
1006 0 : }
1007 :
1008 0 : async fn handle_metadata_health_list_unhealthy(
1009 0 : req: Request<Body>,
1010 0 : ) -> Result<Response<Body>, ApiError> {
1011 0 : check_permissions(&req, Scope::Admin)?;
1012 :
1013 0 : let req = match maybe_forward(req).await {
1014 0 : ForwardOutcome::Forwarded(res) => {
1015 0 : return res;
1016 : }
1017 0 : ForwardOutcome::NotForwarded(req) => req,
1018 0 : };
1019 0 :
1020 0 : let state = get_state(&req);
1021 0 : let unhealthy_tenant_shards = state.service.metadata_health_list_unhealthy().await?;
1022 :
1023 0 : json_response(
1024 0 : StatusCode::OK,
1025 0 : MetadataHealthListUnhealthyResponse {
1026 0 : unhealthy_tenant_shards,
1027 0 : },
1028 0 : )
1029 0 : }
1030 :
1031 0 : async fn handle_metadata_health_list_outdated(
1032 0 : req: Request<Body>,
1033 0 : ) -> Result<Response<Body>, ApiError> {
1034 0 : check_permissions(&req, Scope::Admin)?;
1035 :
1036 0 : let mut req = match maybe_forward(req).await {
1037 0 : ForwardOutcome::Forwarded(res) => {
1038 0 : return res;
1039 : }
1040 0 : ForwardOutcome::NotForwarded(req) => req,
1041 : };
1042 :
1043 0 : let list_outdated_req = json_request::<MetadataHealthListOutdatedRequest>(&mut req).await?;
1044 0 : let state = get_state(&req);
1045 0 : let health_records = state
1046 0 : .service
1047 0 : .metadata_health_list_outdated(list_outdated_req.not_scrubbed_for)
1048 0 : .await?;
1049 :
1050 0 : json_response(
1051 0 : StatusCode::OK,
1052 0 : MetadataHealthListOutdatedResponse { health_records },
1053 0 : )
1054 0 : }
1055 :
1056 0 : async fn handle_tenant_shard_split(
1057 0 : service: Arc<Service>,
1058 0 : req: Request<Body>,
1059 0 : ) -> Result<Response<Body>, ApiError> {
1060 0 : check_permissions(&req, Scope::Admin)?;
1061 : // NB: don't rate limit: admin operation.
1062 :
1063 0 : let mut req = match maybe_forward(req).await {
1064 0 : ForwardOutcome::Forwarded(res) => {
1065 0 : return res;
1066 : }
1067 0 : ForwardOutcome::NotForwarded(req) => req,
1068 : };
1069 :
1070 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1071 0 : let split_req = json_request::<TenantShardSplitRequest>(&mut req).await?;
1072 :
1073 : json_response(
1074 : StatusCode::OK,
1075 0 : service.tenant_shard_split(tenant_id, split_req).await?,
1076 : )
1077 0 : }
1078 :
1079 0 : async fn handle_tenant_shard_migrate(
1080 0 : service: Arc<Service>,
1081 0 : req: Request<Body>,
1082 0 : ) -> Result<Response<Body>, ApiError> {
1083 0 : check_permissions(&req, Scope::Admin)?;
1084 : // NB: don't rate limit: admin operation.
1085 :
1086 0 : let mut req = match maybe_forward(req).await {
1087 0 : ForwardOutcome::Forwarded(res) => {
1088 0 : return res;
1089 : }
1090 0 : ForwardOutcome::NotForwarded(req) => req,
1091 : };
1092 :
1093 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1094 0 : let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
1095 : json_response(
1096 : StatusCode::OK,
1097 0 : service
1098 0 : .tenant_shard_migrate(tenant_shard_id, migrate_req)
1099 0 : .await?,
1100 : )
1101 0 : }
1102 :
1103 0 : async fn handle_tenant_shard_migrate_secondary(
1104 0 : service: Arc<Service>,
1105 0 : req: Request<Body>,
1106 0 : ) -> Result<Response<Body>, ApiError> {
1107 0 : check_permissions(&req, Scope::Admin)?;
1108 : // NB: don't rate limit: admin operation.
1109 :
1110 0 : let mut req = match maybe_forward(req).await {
1111 0 : ForwardOutcome::Forwarded(res) => {
1112 0 : return res;
1113 : }
1114 0 : ForwardOutcome::NotForwarded(req) => req,
1115 : };
1116 :
1117 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1118 0 : let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
1119 : json_response(
1120 : StatusCode::OK,
1121 0 : service
1122 0 : .tenant_shard_migrate_secondary(tenant_shard_id, migrate_req)
1123 0 : .await?,
1124 : )
1125 0 : }
1126 :
1127 0 : async fn handle_tenant_shard_cancel_reconcile(
1128 0 : service: Arc<Service>,
1129 0 : req: Request<Body>,
1130 0 : ) -> Result<Response<Body>, ApiError> {
1131 0 : check_permissions(&req, Scope::Admin)?;
1132 : // NB: don't rate limit: admin operation.
1133 :
1134 0 : let req = match maybe_forward(req).await {
1135 0 : ForwardOutcome::Forwarded(res) => {
1136 0 : return res;
1137 : }
1138 0 : ForwardOutcome::NotForwarded(req) => req,
1139 : };
1140 :
1141 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1142 : json_response(
1143 : StatusCode::OK,
1144 0 : service
1145 0 : .tenant_shard_cancel_reconcile(tenant_shard_id)
1146 0 : .await?,
1147 : )
1148 0 : }
1149 :
1150 0 : async fn handle_tenant_update_policy(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1151 0 : check_permissions(&req, Scope::Admin)?;
1152 : // NB: don't rate limit: admin operation.
1153 :
1154 0 : let mut req = match maybe_forward(req).await {
1155 0 : ForwardOutcome::Forwarded(res) => {
1156 0 : return res;
1157 : }
1158 0 : ForwardOutcome::NotForwarded(req) => req,
1159 : };
1160 :
1161 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1162 0 : let update_req = json_request::<TenantPolicyRequest>(&mut req).await?;
1163 0 : let state = get_state(&req);
1164 0 :
1165 0 : json_response(
1166 0 : StatusCode::OK,
1167 0 : state
1168 0 : .service
1169 0 : .tenant_update_policy(tenant_id, update_req)
1170 0 : .await?,
1171 : )
1172 0 : }
1173 :
1174 0 : async fn handle_update_preferred_azs(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1175 0 : check_permissions(&req, Scope::Admin)?;
1176 :
1177 0 : let mut req = match maybe_forward(req).await {
1178 0 : ForwardOutcome::Forwarded(res) => {
1179 0 : return res;
1180 : }
1181 0 : ForwardOutcome::NotForwarded(req) => req,
1182 : };
1183 :
1184 0 : let azs_req = json_request::<ShardsPreferredAzsRequest>(&mut req).await?;
1185 0 : let state = get_state(&req);
1186 0 :
1187 0 : json_response(
1188 0 : StatusCode::OK,
1189 0 : state.service.update_shards_preferred_azs(azs_req).await?,
1190 : )
1191 0 : }
1192 :
1193 0 : async fn handle_step_down(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1194 0 : check_permissions(&req, Scope::ControllerPeer)?;
1195 :
1196 0 : let req = match maybe_forward(req).await {
1197 0 : ForwardOutcome::Forwarded(res) => {
1198 0 : return res;
1199 : }
1200 0 : ForwardOutcome::NotForwarded(req) => req,
1201 0 : };
1202 0 :
1203 0 : let state = get_state(&req);
1204 0 : json_response(StatusCode::OK, state.service.step_down().await)
1205 0 : }
1206 :
1207 0 : async fn handle_tenant_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1208 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1209 0 : check_permissions(&req, Scope::PageServerApi)?;
1210 0 : maybe_rate_limit(&req, tenant_id).await;
1211 :
1212 0 : let req = match maybe_forward(req).await {
1213 0 : ForwardOutcome::Forwarded(res) => {
1214 0 : return res;
1215 : }
1216 0 : ForwardOutcome::NotForwarded(req) => req,
1217 0 : };
1218 0 :
1219 0 : let state = get_state(&req);
1220 0 :
1221 0 : json_response(StatusCode::OK, state.service.tenant_drop(tenant_id).await?)
1222 0 : }
1223 :
1224 0 : async fn handle_tenant_import(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1225 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1226 0 : check_permissions(&req, Scope::PageServerApi)?;
1227 0 : maybe_rate_limit(&req, tenant_id).await;
1228 :
1229 0 : let req = match maybe_forward(req).await {
1230 0 : ForwardOutcome::Forwarded(res) => {
1231 0 : return res;
1232 : }
1233 0 : ForwardOutcome::NotForwarded(req) => req,
1234 0 : };
1235 0 :
1236 0 : let state = get_state(&req);
1237 0 :
1238 0 : json_response(
1239 0 : StatusCode::OK,
1240 0 : state.service.tenant_import(tenant_id).await?,
1241 : )
1242 0 : }
1243 :
1244 0 : async fn handle_tenants_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1245 0 : check_permissions(&req, Scope::Admin)?;
1246 :
1247 0 : let req = match maybe_forward(req).await {
1248 0 : ForwardOutcome::Forwarded(res) => {
1249 0 : return res;
1250 : }
1251 0 : ForwardOutcome::NotForwarded(req) => req,
1252 0 : };
1253 0 :
1254 0 : let state = get_state(&req);
1255 0 : state.service.tenants_dump()
1256 0 : }
1257 :
1258 0 : async fn handle_scheduler_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1259 0 : check_permissions(&req, Scope::Admin)?;
1260 :
1261 0 : let req = match maybe_forward(req).await {
1262 0 : ForwardOutcome::Forwarded(res) => {
1263 0 : return res;
1264 : }
1265 0 : ForwardOutcome::NotForwarded(req) => req,
1266 0 : };
1267 0 :
1268 0 : let state = get_state(&req);
1269 0 : state.service.scheduler_dump()
1270 0 : }
1271 :
1272 0 : async fn handle_consistency_check(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1273 0 : check_permissions(&req, Scope::Admin)?;
1274 :
1275 0 : let req = match maybe_forward(req).await {
1276 0 : ForwardOutcome::Forwarded(res) => {
1277 0 : return res;
1278 : }
1279 0 : ForwardOutcome::NotForwarded(req) => req,
1280 0 : };
1281 0 :
1282 0 : let state = get_state(&req);
1283 0 :
1284 0 : json_response(StatusCode::OK, state.service.consistency_check().await?)
1285 0 : }
1286 :
1287 0 : async fn handle_reconcile_all(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1288 0 : check_permissions(&req, Scope::Admin)?;
1289 :
1290 0 : let req = match maybe_forward(req).await {
1291 0 : ForwardOutcome::Forwarded(res) => {
1292 0 : return res;
1293 : }
1294 0 : ForwardOutcome::NotForwarded(req) => req,
1295 0 : };
1296 0 :
1297 0 : let state = get_state(&req);
1298 0 :
1299 0 : json_response(StatusCode::OK, state.service.reconcile_all_now().await?)
1300 0 : }
1301 :
1302 : /// Status endpoint is just used for checking that our HTTP listener is up
1303 0 : async fn handle_status(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1304 0 : match maybe_forward(req).await {
1305 0 : ForwardOutcome::Forwarded(res) => {
1306 0 : return res;
1307 : }
1308 0 : ForwardOutcome::NotForwarded(_req) => {}
1309 0 : };
1310 0 :
1311 0 : json_response(StatusCode::OK, ())
1312 0 : }
1313 :
1314 : /// Readiness endpoint indicates when we're done doing startup I/O (e.g. reconciling
1315 : /// with remote pageserver nodes). This is intended for use as a kubernetes readiness probe.
1316 0 : async fn handle_ready(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1317 0 : let req = match maybe_forward(req).await {
1318 0 : ForwardOutcome::Forwarded(res) => {
1319 0 : return res;
1320 : }
1321 0 : ForwardOutcome::NotForwarded(req) => req,
1322 0 : };
1323 0 :
1324 0 : let state = get_state(&req);
1325 0 : if state.service.startup_complete.is_ready() {
1326 0 : json_response(StatusCode::OK, ())
1327 : } else {
1328 0 : json_response(StatusCode::SERVICE_UNAVAILABLE, ())
1329 : }
1330 0 : }
1331 :
1332 : impl From<ReconcileError> for ApiError {
1333 0 : fn from(value: ReconcileError) -> Self {
1334 0 : ApiError::Conflict(format!("Reconciliation error: {}", value))
1335 0 : }
1336 : }
1337 :
1338 : /// Return the safekeeper record by instance id, or 404.
1339 : ///
1340 : /// Not used by anything except manual testing.
1341 0 : async fn handle_get_safekeeper(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1342 0 : check_permissions(&req, Scope::Infra)?;
1343 :
1344 0 : let id = parse_request_param::<i64>(&req, "id")?;
1345 :
1346 0 : let req = match maybe_forward(req).await {
1347 0 : ForwardOutcome::Forwarded(res) => {
1348 0 : return res;
1349 : }
1350 0 : ForwardOutcome::NotForwarded(req) => req,
1351 0 : };
1352 0 :
1353 0 : let state = get_state(&req);
1354 :
1355 0 : let res = state.service.get_safekeeper(id).await;
1356 :
1357 0 : match res {
1358 0 : Ok(b) => json_response(StatusCode::OK, b),
1359 : Err(crate::persistence::DatabaseError::Query(diesel::result::Error::NotFound)) => {
1360 0 : Err(ApiError::NotFound("unknown instance id".into()))
1361 : }
1362 0 : Err(other) => Err(other.into()),
1363 : }
1364 0 : }
1365 :
1366 : /// Used as part of deployment scripts.
1367 : ///
1368 : /// Assumes information is only relayed to storage controller after first selecting an unique id on
1369 : /// control plane database, which means we have an id field in the request and payload.
1370 0 : async fn handle_upsert_safekeeper(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
1371 0 : check_permissions(&req, Scope::Infra)?;
1372 :
1373 0 : let body = json_request::<SafekeeperUpsert>(&mut req).await?;
1374 0 : let id = parse_request_param::<i64>(&req, "id")?;
1375 :
1376 0 : if id != body.id {
1377 : // it should be repeated
1378 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
1379 0 : "id mismatch: url={id:?}, body={:?}",
1380 0 : body.id
1381 0 : )));
1382 0 : }
1383 :
1384 0 : let req = match maybe_forward(req).await {
1385 0 : ForwardOutcome::Forwarded(res) => {
1386 0 : return res;
1387 : }
1388 0 : ForwardOutcome::NotForwarded(req) => req,
1389 0 : };
1390 0 :
1391 0 : let state = get_state(&req);
1392 0 :
1393 0 : state.service.upsert_safekeeper(body).await?;
1394 :
1395 0 : Ok(Response::builder()
1396 0 : .status(StatusCode::NO_CONTENT)
1397 0 : .body(Body::empty())
1398 0 : .unwrap())
1399 0 : }
1400 :
1401 : /// Sets the scheduling policy of the specified safekeeper
1402 0 : async fn handle_safekeeper_scheduling_policy(
1403 0 : mut req: Request<Body>,
1404 0 : ) -> Result<Response<Body>, ApiError> {
1405 0 : check_permissions(&req, Scope::Admin)?;
1406 :
1407 0 : let body = json_request::<SafekeeperSchedulingPolicyRequest>(&mut req).await?;
1408 0 : let id = parse_request_param::<i64>(&req, "id")?;
1409 :
1410 0 : let req = match maybe_forward(req).await {
1411 0 : ForwardOutcome::Forwarded(res) => {
1412 0 : return res;
1413 : }
1414 0 : ForwardOutcome::NotForwarded(req) => req,
1415 0 : };
1416 0 :
1417 0 : let state = get_state(&req);
1418 0 :
1419 0 : state
1420 0 : .service
1421 0 : .set_safekeeper_scheduling_policy(id, body.scheduling_policy)
1422 0 : .await?;
1423 :
1424 0 : json_response(StatusCode::OK, ())
1425 0 : }
1426 :
1427 : /// Common wrapper for request handlers that call into Service and will operate on tenants: they must only
1428 : /// be allowed to run if Service has finished its initial reconciliation.
1429 0 : async fn tenant_service_handler<R, H>(
1430 0 : request: Request<Body>,
1431 0 : handler: H,
1432 0 : request_name: RequestName,
1433 0 : ) -> R::Output
1434 0 : where
1435 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
1436 0 : H: FnOnce(Arc<Service>, Request<Body>) -> R + Send + Sync + 'static,
1437 0 : {
1438 0 : let state = get_state(&request);
1439 0 : let service = state.service.clone();
1440 0 :
1441 0 : let startup_complete = service.startup_complete.clone();
1442 0 : if tokio::time::timeout(STARTUP_RECONCILE_TIMEOUT, startup_complete.wait())
1443 0 : .await
1444 0 : .is_err()
1445 : {
1446 : // This shouldn't happen: it is the responsibilty of [`Service::startup_reconcile`] to use appropriate
1447 : // timeouts around its remote calls, to bound its runtime.
1448 0 : return Err(ApiError::Timeout(
1449 0 : "Timed out waiting for service readiness".into(),
1450 0 : ));
1451 0 : }
1452 0 :
1453 0 : named_request_span(
1454 0 : request,
1455 0 : |request| async move { handler(service, request).await },
1456 0 : request_name,
1457 0 : )
1458 0 : .await
1459 0 : }
1460 :
1461 : /// Check if the required scope is held in the request's token, or if the request has
1462 : /// a token with 'admin' scope then always permit it.
1463 0 : fn check_permissions(request: &Request<Body>, required_scope: Scope) -> Result<(), ApiError> {
1464 0 : check_permission_with(request, |claims| {
1465 0 : match crate::auth::check_permission(claims, required_scope) {
1466 0 : Err(e) => match crate::auth::check_permission(claims, Scope::Admin) {
1467 0 : Ok(()) => Ok(()),
1468 0 : Err(_) => Err(e),
1469 : },
1470 0 : Ok(()) => Ok(()),
1471 : }
1472 0 : })
1473 0 : }
1474 :
1475 : #[derive(Clone, Debug)]
1476 : struct RequestMeta {
1477 : method: hyper::http::Method,
1478 : at: Instant,
1479 : }
1480 :
1481 0 : pub fn prologue_leadership_status_check_middleware<
1482 0 : B: hyper::body::HttpBody + Send + Sync + 'static,
1483 0 : >() -> Middleware<B, ApiError> {
1484 0 : Middleware::pre(move |req| async move {
1485 0 : let state = get_state(&req);
1486 0 : let leadership_status = state.service.get_leadership_status();
1487 :
1488 : enum AllowedRoutes {
1489 : All,
1490 : Some(&'static [&'static str]),
1491 : }
1492 :
1493 0 : let allowed_routes = match leadership_status {
1494 0 : LeadershipStatus::Leader => AllowedRoutes::All,
1495 0 : LeadershipStatus::SteppedDown => AllowedRoutes::All,
1496 0 : LeadershipStatus::Candidate => AllowedRoutes::Some(&[
1497 0 : "/ready",
1498 0 : "/status",
1499 0 : "/metrics",
1500 0 : "/profile/cpu",
1501 0 : "/profile/heap",
1502 0 : ]),
1503 : };
1504 :
1505 0 : match allowed_routes {
1506 0 : AllowedRoutes::All => Ok(req),
1507 0 : AllowedRoutes::Some(allowed) if allowed.contains(&req.uri().path()) => Ok(req),
1508 : _ => {
1509 0 : tracing::info!(
1510 0 : "Request {} not allowed due to current leadership state",
1511 0 : req.uri()
1512 : );
1513 :
1514 0 : Err(ApiError::ResourceUnavailable(
1515 0 : format!("Current leadership status is {leadership_status}").into(),
1516 0 : ))
1517 : }
1518 : }
1519 0 : })
1520 0 : }
1521 :
1522 0 : fn prologue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>()
1523 0 : -> Middleware<B, ApiError> {
1524 0 : Middleware::pre(move |req| async move {
1525 0 : let meta = RequestMeta {
1526 0 : method: req.method().clone(),
1527 0 : at: Instant::now(),
1528 0 : };
1529 0 :
1530 0 : req.set_context(meta);
1531 0 :
1532 0 : Ok(req)
1533 0 : })
1534 0 : }
1535 :
1536 0 : fn epilogue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>()
1537 0 : -> Middleware<B, ApiError> {
1538 0 : Middleware::post_with_info(move |resp, req_info| async move {
1539 0 : let request_name = match req_info.context::<RequestName>() {
1540 0 : Some(name) => name,
1541 : None => {
1542 0 : return Ok(resp);
1543 : }
1544 : };
1545 :
1546 0 : if let Some(meta) = req_info.context::<RequestMeta>() {
1547 0 : let status = &crate::metrics::METRICS_REGISTRY
1548 0 : .metrics_group
1549 0 : .storage_controller_http_request_status;
1550 0 : let latency = &crate::metrics::METRICS_REGISTRY
1551 0 : .metrics_group
1552 0 : .storage_controller_http_request_latency;
1553 0 :
1554 0 : status.inc(HttpRequestStatusLabelGroup {
1555 0 : path: request_name.0,
1556 0 : method: meta.method.clone().into(),
1557 0 : status: crate::metrics::StatusCode(resp.status()),
1558 0 : });
1559 0 :
1560 0 : latency.observe(
1561 0 : HttpRequestLatencyLabelGroup {
1562 0 : path: request_name.0,
1563 0 : method: meta.method.into(),
1564 0 : },
1565 0 : meta.at.elapsed().as_secs_f64(),
1566 0 : );
1567 0 : }
1568 0 : Ok(resp)
1569 0 : })
1570 0 : }
1571 :
1572 0 : pub async fn measured_metrics_handler(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1573 : pub const TEXT_FORMAT: &str = "text/plain; version=0.0.4";
1574 :
1575 0 : let req = match maybe_forward(req).await {
1576 0 : ForwardOutcome::Forwarded(res) => {
1577 0 : return res;
1578 : }
1579 0 : ForwardOutcome::NotForwarded(req) => req,
1580 0 : };
1581 0 :
1582 0 : let state = get_state(&req);
1583 0 : let payload = crate::metrics::METRICS_REGISTRY.encode(&state.neon_metrics);
1584 0 : let response = Response::builder()
1585 0 : .status(200)
1586 0 : .header(CONTENT_TYPE, TEXT_FORMAT)
1587 0 : .body(payload.into())
1588 0 : .unwrap();
1589 0 :
1590 0 : Ok(response)
1591 0 : }
1592 :
1593 : #[derive(Clone)]
1594 : struct RequestName(&'static str);
1595 :
1596 0 : async fn named_request_span<R, H>(
1597 0 : request: Request<Body>,
1598 0 : handler: H,
1599 0 : name: RequestName,
1600 0 : ) -> R::Output
1601 0 : where
1602 0 : R: Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
1603 0 : H: FnOnce(Request<Body>) -> R + Send + Sync + 'static,
1604 0 : {
1605 0 : request.set_context(name);
1606 0 : request_span(request, handler).await
1607 0 : }
1608 :
1609 : enum ForwardOutcome {
1610 : Forwarded(Result<Response<Body>, ApiError>),
1611 : NotForwarded(Request<Body>),
1612 : }
1613 :
1614 : /// Potentially forward the request to the current storage controler leader.
1615 : /// More specifically we forward when:
1616 : /// 1. Request is not one of:
1617 : /// ["/control/v1/step_down", "/status", "/ready", "/metrics", "/profile/cpu", "/profile/heap"]
1618 : /// 2. Current instance is in [`LeadershipStatus::SteppedDown`] state
1619 : /// 3. There is a leader in the database to forward to
1620 : /// 4. Leader from step (3) is not the current instance
1621 : ///
1622 : /// Why forward?
1623 : /// It turns out that we can't rely on external orchestration to promptly route trafic to the
1624 : /// new leader. This is downtime inducing. Forwarding provides a safe way out.
1625 : ///
1626 : /// Why is it safe?
1627 : /// If a storcon instance is persisted in the database, then we know that it is the current leader.
1628 : /// There's one exception: time between handling step-down request and the new leader updating the
1629 : /// database.
1630 : ///
1631 : /// Let's treat the happy case first. The stepped down node does not produce any side effects,
1632 : /// since all request handling happens on the leader.
1633 : ///
1634 : /// As for the edge case, we are guaranteed to always have a maximum of two running instances.
1635 : /// Hence, if we are in the edge case scenario the leader persisted in the database is the
1636 : /// stepped down instance that received the request. Condition (4) above covers this scenario.
1637 0 : async fn maybe_forward(req: Request<Body>) -> ForwardOutcome {
1638 : const NOT_FOR_FORWARD: &[&str] = &[
1639 : "/control/v1/step_down",
1640 : "/status",
1641 : "/ready",
1642 : "/metrics",
1643 : "/profile/cpu",
1644 : "/profile/heap",
1645 : ];
1646 :
1647 0 : let uri = req.uri();
1648 0 : let uri_for_forward = !NOT_FOR_FORWARD.contains(&uri.path());
1649 0 :
1650 0 : // Fast return before trying to take any Service locks, if we will never forward anyway
1651 0 : if !uri_for_forward {
1652 0 : return ForwardOutcome::NotForwarded(req);
1653 0 : }
1654 0 :
1655 0 : let state = get_state(&req);
1656 0 : let leadership_status = state.service.get_leadership_status();
1657 0 :
1658 0 : if leadership_status != LeadershipStatus::SteppedDown {
1659 0 : return ForwardOutcome::NotForwarded(req);
1660 0 : }
1661 :
1662 0 : let leader = state.service.get_leader().await;
1663 0 : let leader = {
1664 0 : match leader {
1665 0 : Ok(Some(leader)) => leader,
1666 : Ok(None) => {
1667 0 : return ForwardOutcome::Forwarded(Err(ApiError::ResourceUnavailable(
1668 0 : "No leader to forward to while in stepped down state".into(),
1669 0 : )));
1670 : }
1671 0 : Err(err) => {
1672 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(
1673 0 : anyhow::anyhow!(
1674 0 : "Failed to get leader for forwarding while in stepped down state: {err}"
1675 0 : ),
1676 0 : )));
1677 : }
1678 : }
1679 : };
1680 :
1681 0 : let cfg = state.service.get_config();
1682 0 : if let Some(ref self_addr) = cfg.address_for_peers {
1683 0 : let leader_addr = match Uri::from_str(leader.address.as_str()) {
1684 0 : Ok(uri) => uri,
1685 0 : Err(err) => {
1686 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(
1687 0 : anyhow::anyhow!(
1688 0 : "Failed to parse leader uri for forwarding while in stepped down state: {err}"
1689 0 : ),
1690 0 : )));
1691 : }
1692 : };
1693 :
1694 0 : if *self_addr == leader_addr {
1695 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1696 0 : "Leader is stepped down instance"
1697 0 : ))));
1698 0 : }
1699 0 : }
1700 :
1701 0 : tracing::info!("Forwarding {} to leader at {}", uri, leader.address);
1702 :
1703 : // Use [`RECONCILE_TIMEOUT`] as the max amount of time a request should block for and
1704 : // include some leeway to get the timeout for proxied requests.
1705 : const PROXIED_REQUEST_TIMEOUT: Duration = Duration::from_secs(RECONCILE_TIMEOUT.as_secs() + 10);
1706 0 : let client = reqwest::ClientBuilder::new()
1707 0 : .timeout(PROXIED_REQUEST_TIMEOUT)
1708 0 : .build();
1709 0 : let client = match client {
1710 0 : Ok(client) => client,
1711 0 : Err(err) => {
1712 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1713 0 : "Failed to build leader client for forwarding while in stepped down state: {err}"
1714 0 : ))));
1715 : }
1716 : };
1717 :
1718 0 : let request: reqwest::Request = match convert_request(req, &client, leader.address).await {
1719 0 : Ok(r) => r,
1720 0 : Err(err) => {
1721 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1722 0 : "Failed to convert request for forwarding while in stepped down state: {err}"
1723 0 : ))));
1724 : }
1725 : };
1726 :
1727 0 : let response = match client.execute(request).await {
1728 0 : Ok(r) => r,
1729 0 : Err(err) => {
1730 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1731 0 : "Failed to forward while in stepped down state: {err}"
1732 0 : ))));
1733 : }
1734 : };
1735 :
1736 0 : ForwardOutcome::Forwarded(convert_response(response).await)
1737 0 : }
1738 :
1739 : /// Convert a [`reqwest::Response`] to a [hyper::Response`] by passing through
1740 : /// a stable representation (string, bytes or integer)
1741 : ///
1742 : /// Ideally, we would not have to do this since both types use the http crate
1743 : /// under the hood. However, they use different versions of the crate and keeping
1744 : /// second order dependencies in sync is difficult.
1745 0 : async fn convert_response(resp: reqwest::Response) -> Result<hyper::Response<Body>, ApiError> {
1746 : use std::str::FromStr;
1747 :
1748 0 : let mut builder = hyper::Response::builder().status(resp.status().as_u16());
1749 0 : for (key, value) in resp.headers().into_iter() {
1750 0 : let key = hyper::header::HeaderName::from_str(key.as_str()).map_err(|err| {
1751 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1752 0 : })?;
1753 :
1754 0 : let value = hyper::header::HeaderValue::from_bytes(value.as_bytes()).map_err(|err| {
1755 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1756 0 : })?;
1757 :
1758 0 : builder = builder.header(key, value);
1759 : }
1760 :
1761 0 : let body = http::Body::wrap_stream(resp.bytes_stream());
1762 0 :
1763 0 : builder.body(body).map_err(|err| {
1764 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1765 0 : })
1766 0 : }
1767 :
1768 : /// Convert a [`reqwest::Request`] to a [hyper::Request`] by passing through
1769 : /// a stable representation (string, bytes or integer)
1770 : ///
1771 : /// See [`convert_response`] for why we are doing it this way.
1772 0 : async fn convert_request(
1773 0 : req: hyper::Request<Body>,
1774 0 : client: &reqwest::Client,
1775 0 : to_address: String,
1776 0 : ) -> Result<reqwest::Request, ApiError> {
1777 : use std::str::FromStr;
1778 :
1779 0 : let (parts, body) = req.into_parts();
1780 0 : let method = reqwest::Method::from_str(parts.method.as_str()).map_err(|err| {
1781 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1782 0 : })?;
1783 :
1784 0 : let path_and_query = parts.uri.path_and_query().ok_or_else(|| {
1785 0 : ApiError::InternalServerError(anyhow::anyhow!(
1786 0 : "Request conversion failed: no path and query"
1787 0 : ))
1788 0 : })?;
1789 :
1790 0 : let uri = reqwest::Url::from_str(
1791 0 : format!(
1792 0 : "{}{}",
1793 0 : to_address.trim_end_matches("/"),
1794 0 : path_and_query.as_str()
1795 0 : )
1796 0 : .as_str(),
1797 0 : )
1798 0 : .map_err(|err| {
1799 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1800 0 : })?;
1801 :
1802 0 : let mut headers = reqwest::header::HeaderMap::new();
1803 0 : for (key, value) in parts.headers.into_iter() {
1804 0 : let key = match key {
1805 0 : Some(k) => k,
1806 : None => {
1807 0 : continue;
1808 : }
1809 : };
1810 :
1811 0 : let key = reqwest::header::HeaderName::from_str(key.as_str()).map_err(|err| {
1812 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1813 0 : })?;
1814 :
1815 0 : let value = reqwest::header::HeaderValue::from_bytes(value.as_bytes()).map_err(|err| {
1816 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1817 0 : })?;
1818 :
1819 0 : headers.insert(key, value);
1820 : }
1821 :
1822 0 : let body = hyper::body::to_bytes(body).await.map_err(|err| {
1823 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1824 0 : })?;
1825 :
1826 0 : client
1827 0 : .request(method, uri)
1828 0 : .headers(headers)
1829 0 : .body(body)
1830 0 : .build()
1831 0 : .map_err(|err| {
1832 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1833 0 : })
1834 0 : }
1835 :
1836 0 : pub fn make_router(
1837 0 : service: Arc<Service>,
1838 0 : auth: Option<Arc<SwappableJwtAuth>>,
1839 0 : build_info: BuildInfo,
1840 0 : ) -> RouterBuilder<hyper::Body, ApiError> {
1841 0 : let mut router = endpoint::make_router()
1842 0 : .middleware(prologue_leadership_status_check_middleware())
1843 0 : .middleware(prologue_metrics_middleware())
1844 0 : .middleware(epilogue_metrics_middleware());
1845 0 : if auth.is_some() {
1846 0 : router = router.middleware(auth_middleware(|request| {
1847 0 : let state = get_state(request);
1848 0 : if state.allowlist_routes.contains(&request.uri().path()) {
1849 0 : None
1850 : } else {
1851 0 : state.auth.as_deref()
1852 : }
1853 0 : }));
1854 0 : }
1855 :
1856 0 : router
1857 0 : .data(Arc::new(HttpState::new(service, auth, build_info)))
1858 0 : .get("/metrics", |r| {
1859 0 : named_request_span(r, measured_metrics_handler, RequestName("metrics"))
1860 0 : })
1861 0 : // Non-prefixed generic endpoints (status, metrics, profiling)
1862 0 : .get("/status", |r| {
1863 0 : named_request_span(r, handle_status, RequestName("status"))
1864 0 : })
1865 0 : .get("/ready", |r| {
1866 0 : named_request_span(r, handle_ready, RequestName("ready"))
1867 0 : })
1868 0 : .get("/profile/cpu", |r| {
1869 0 : named_request_span(r, profile_cpu_handler, RequestName("profile_cpu"))
1870 0 : })
1871 0 : .get("/profile/heap", |r| {
1872 0 : named_request_span(r, profile_heap_handler, RequestName("profile_heap"))
1873 0 : })
1874 0 : // Upcalls for the pageserver: point the pageserver's `control_plane_api` config to this prefix
1875 0 : .post("/upcall/v1/re-attach", |r| {
1876 0 : named_request_span(r, handle_re_attach, RequestName("upcall_v1_reattach"))
1877 0 : })
1878 0 : .post("/upcall/v1/validate", |r| {
1879 0 : named_request_span(r, handle_validate, RequestName("upcall_v1_validate"))
1880 0 : })
1881 0 : // Test/dev/debug endpoints
1882 0 : .post("/debug/v1/attach-hook", |r| {
1883 0 : named_request_span(r, handle_attach_hook, RequestName("debug_v1_attach_hook"))
1884 0 : })
1885 0 : .post("/debug/v1/inspect", |r| {
1886 0 : named_request_span(r, handle_inspect, RequestName("debug_v1_inspect"))
1887 0 : })
1888 0 : .post("/debug/v1/tenant/:tenant_id/drop", |r| {
1889 0 : named_request_span(r, handle_tenant_drop, RequestName("debug_v1_tenant_drop"))
1890 0 : })
1891 0 : .post("/debug/v1/node/:node_id/drop", |r| {
1892 0 : named_request_span(r, handle_node_drop, RequestName("debug_v1_node_drop"))
1893 0 : })
1894 0 : .post("/debug/v1/tenant/:tenant_id/import", |r| {
1895 0 : named_request_span(
1896 0 : r,
1897 0 : handle_tenant_import,
1898 0 : RequestName("debug_v1_tenant_import"),
1899 0 : )
1900 0 : })
1901 0 : .get("/debug/v1/tenant", |r| {
1902 0 : named_request_span(r, handle_tenants_dump, RequestName("debug_v1_tenant"))
1903 0 : })
1904 0 : .get("/debug/v1/tenant/:tenant_id/locate", |r| {
1905 0 : tenant_service_handler(
1906 0 : r,
1907 0 : handle_tenant_locate,
1908 0 : RequestName("debug_v1_tenant_locate"),
1909 0 : )
1910 0 : })
1911 0 : .get("/debug/v1/scheduler", |r| {
1912 0 : named_request_span(r, handle_scheduler_dump, RequestName("debug_v1_scheduler"))
1913 0 : })
1914 0 : .post("/debug/v1/consistency_check", |r| {
1915 0 : named_request_span(
1916 0 : r,
1917 0 : handle_consistency_check,
1918 0 : RequestName("debug_v1_consistency_check"),
1919 0 : )
1920 0 : })
1921 0 : .post("/debug/v1/reconcile_all", |r| {
1922 0 : request_span(r, handle_reconcile_all)
1923 0 : })
1924 0 : .put("/debug/v1/failpoints", |r| {
1925 0 : request_span(r, |r| failpoints_handler(r, CancellationToken::new()))
1926 0 : })
1927 0 : // Node operations
1928 0 : .post("/control/v1/node", |r| {
1929 0 : named_request_span(r, handle_node_register, RequestName("control_v1_node"))
1930 0 : })
1931 0 : .delete("/control/v1/node/:node_id", |r| {
1932 0 : named_request_span(r, handle_node_delete, RequestName("control_v1_node_delete"))
1933 0 : })
1934 0 : .get("/control/v1/node", |r| {
1935 0 : named_request_span(r, handle_node_list, RequestName("control_v1_node"))
1936 0 : })
1937 0 : .put("/control/v1/node/:node_id/config", |r| {
1938 0 : named_request_span(
1939 0 : r,
1940 0 : handle_node_configure,
1941 0 : RequestName("control_v1_node_config"),
1942 0 : )
1943 0 : })
1944 0 : .get("/control/v1/node/:node_id", |r| {
1945 0 : named_request_span(r, handle_node_status, RequestName("control_v1_node_status"))
1946 0 : })
1947 0 : .get("/control/v1/node/:node_id/shards", |r| {
1948 0 : named_request_span(
1949 0 : r,
1950 0 : handle_node_shards,
1951 0 : RequestName("control_v1_node_describe"),
1952 0 : )
1953 0 : })
1954 0 : .get("/control/v1/leader", |r| {
1955 0 : named_request_span(r, handle_get_leader, RequestName("control_v1_get_leader"))
1956 0 : })
1957 0 : .put("/control/v1/node/:node_id/drain", |r| {
1958 0 : named_request_span(r, handle_node_drain, RequestName("control_v1_node_drain"))
1959 0 : })
1960 0 : .delete("/control/v1/node/:node_id/drain", |r| {
1961 0 : named_request_span(
1962 0 : r,
1963 0 : handle_cancel_node_drain,
1964 0 : RequestName("control_v1_cancel_node_drain"),
1965 0 : )
1966 0 : })
1967 0 : .put("/control/v1/node/:node_id/fill", |r| {
1968 0 : named_request_span(r, handle_node_fill, RequestName("control_v1_node_fill"))
1969 0 : })
1970 0 : .delete("/control/v1/node/:node_id/fill", |r| {
1971 0 : named_request_span(
1972 0 : r,
1973 0 : handle_cancel_node_fill,
1974 0 : RequestName("control_v1_cancel_node_fill"),
1975 0 : )
1976 0 : })
1977 0 : // Metadata health operations
1978 0 : .post("/control/v1/metadata_health/update", |r| {
1979 0 : named_request_span(
1980 0 : r,
1981 0 : handle_metadata_health_update,
1982 0 : RequestName("control_v1_metadata_health_update"),
1983 0 : )
1984 0 : })
1985 0 : .get("/control/v1/metadata_health/unhealthy", |r| {
1986 0 : named_request_span(
1987 0 : r,
1988 0 : handle_metadata_health_list_unhealthy,
1989 0 : RequestName("control_v1_metadata_health_list_unhealthy"),
1990 0 : )
1991 0 : })
1992 0 : .post("/control/v1/metadata_health/outdated", |r| {
1993 0 : named_request_span(
1994 0 : r,
1995 0 : handle_metadata_health_list_outdated,
1996 0 : RequestName("control_v1_metadata_health_list_outdated"),
1997 0 : )
1998 0 : })
1999 0 : // Safekeepers
2000 0 : .get("/control/v1/safekeeper", |r| {
2001 0 : named_request_span(
2002 0 : r,
2003 0 : handle_safekeeper_list,
2004 0 : RequestName("control_v1_safekeeper_list"),
2005 0 : )
2006 0 : })
2007 0 : .get("/control/v1/safekeeper/:id", |r| {
2008 0 : named_request_span(r, handle_get_safekeeper, RequestName("v1_safekeeper"))
2009 0 : })
2010 0 : .post("/control/v1/safekeeper/:id", |r| {
2011 0 : // id is in the body
2012 0 : named_request_span(
2013 0 : r,
2014 0 : handle_upsert_safekeeper,
2015 0 : RequestName("v1_safekeeper_post"),
2016 0 : )
2017 0 : })
2018 0 : .post("/control/v1/safekeeper/:id/scheduling_policy", |r| {
2019 0 : named_request_span(
2020 0 : r,
2021 0 : handle_safekeeper_scheduling_policy,
2022 0 : RequestName("v1_safekeeper_status"),
2023 0 : )
2024 0 : })
2025 0 : // Tenant Shard operations
2026 0 : .put("/control/v1/tenant/:tenant_shard_id/migrate", |r| {
2027 0 : tenant_service_handler(
2028 0 : r,
2029 0 : handle_tenant_shard_migrate,
2030 0 : RequestName("control_v1_tenant_migrate"),
2031 0 : )
2032 0 : })
2033 0 : .put(
2034 0 : "/control/v1/tenant/:tenant_shard_id/migrate_secondary",
2035 0 : |r| {
2036 0 : tenant_service_handler(
2037 0 : r,
2038 0 : handle_tenant_shard_migrate_secondary,
2039 0 : RequestName("control_v1_tenant_migrate_secondary"),
2040 0 : )
2041 0 : },
2042 0 : )
2043 0 : .put(
2044 0 : "/control/v1/tenant/:tenant_shard_id/cancel_reconcile",
2045 0 : |r| {
2046 0 : tenant_service_handler(
2047 0 : r,
2048 0 : handle_tenant_shard_cancel_reconcile,
2049 0 : RequestName("control_v1_tenant_cancel_reconcile"),
2050 0 : )
2051 0 : },
2052 0 : )
2053 0 : .put("/control/v1/tenant/:tenant_id/shard_split", |r| {
2054 0 : tenant_service_handler(
2055 0 : r,
2056 0 : handle_tenant_shard_split,
2057 0 : RequestName("control_v1_tenant_shard_split"),
2058 0 : )
2059 0 : })
2060 0 : .get("/control/v1/tenant/:tenant_id", |r| {
2061 0 : tenant_service_handler(
2062 0 : r,
2063 0 : handle_tenant_describe,
2064 0 : RequestName("control_v1_tenant_describe"),
2065 0 : )
2066 0 : })
2067 0 : .get("/control/v1/tenant", |r| {
2068 0 : tenant_service_handler(r, handle_tenant_list, RequestName("control_v1_tenant_list"))
2069 0 : })
2070 0 : .put("/control/v1/tenant/:tenant_id/policy", |r| {
2071 0 : named_request_span(
2072 0 : r,
2073 0 : handle_tenant_update_policy,
2074 0 : RequestName("control_v1_tenant_policy"),
2075 0 : )
2076 0 : })
2077 0 : .put("/control/v1/preferred_azs", |r| {
2078 0 : named_request_span(
2079 0 : r,
2080 0 : handle_update_preferred_azs,
2081 0 : RequestName("control_v1_preferred_azs"),
2082 0 : )
2083 0 : })
2084 0 : .put("/control/v1/step_down", |r| {
2085 0 : named_request_span(r, handle_step_down, RequestName("control_v1_step_down"))
2086 0 : })
2087 0 : // Tenant operations
2088 0 : // The ^/v1/ endpoints act as a "Virtual Pageserver", enabling shard-naive clients to call into
2089 0 : // this service to manage tenants that actually consist of many tenant shards, as if they are a single entity.
2090 0 : .post("/v1/tenant", |r| {
2091 0 : tenant_service_handler(r, handle_tenant_create, RequestName("v1_tenant"))
2092 0 : })
2093 0 : .delete("/v1/tenant/:tenant_id", |r| {
2094 0 : tenant_service_handler(r, handle_tenant_delete, RequestName("v1_tenant"))
2095 0 : })
2096 0 : .patch("/v1/tenant/config", |r| {
2097 0 : tenant_service_handler(
2098 0 : r,
2099 0 : handle_tenant_config_patch,
2100 0 : RequestName("v1_tenant_config"),
2101 0 : )
2102 0 : })
2103 0 : .put("/v1/tenant/config", |r| {
2104 0 : tenant_service_handler(r, handle_tenant_config_set, RequestName("v1_tenant_config"))
2105 0 : })
2106 0 : .get("/v1/tenant/:tenant_id/config", |r| {
2107 0 : tenant_service_handler(r, handle_tenant_config_get, RequestName("v1_tenant_config"))
2108 0 : })
2109 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2110 0 : tenant_service_handler(
2111 0 : r,
2112 0 : handle_tenant_location_config,
2113 0 : RequestName("v1_tenant_location_config"),
2114 0 : )
2115 0 : })
2116 0 : .put("/v1/tenant/:tenant_id/time_travel_remote_storage", |r| {
2117 0 : tenant_service_handler(
2118 0 : r,
2119 0 : handle_tenant_time_travel_remote_storage,
2120 0 : RequestName("v1_tenant_time_travel_remote_storage"),
2121 0 : )
2122 0 : })
2123 0 : .post("/v1/tenant/:tenant_id/secondary/download", |r| {
2124 0 : tenant_service_handler(
2125 0 : r,
2126 0 : handle_tenant_secondary_download,
2127 0 : RequestName("v1_tenant_secondary_download"),
2128 0 : )
2129 0 : })
2130 0 : // Timeline operations
2131 0 : .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
2132 0 : tenant_service_handler(
2133 0 : r,
2134 0 : handle_tenant_timeline_delete,
2135 0 : RequestName("v1_tenant_timeline"),
2136 0 : )
2137 0 : })
2138 0 : .post("/v1/tenant/:tenant_id/timeline", |r| {
2139 0 : tenant_service_handler(
2140 0 : r,
2141 0 : handle_tenant_timeline_create,
2142 0 : RequestName("v1_tenant_timeline"),
2143 0 : )
2144 0 : })
2145 0 : .put(
2146 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/archival_config",
2147 0 : |r| {
2148 0 : tenant_service_handler(
2149 0 : r,
2150 0 : handle_tenant_timeline_archival_config,
2151 0 : RequestName("v1_tenant_timeline_archival_config"),
2152 0 : )
2153 0 : },
2154 0 : )
2155 0 : .put(
2156 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/detach_ancestor",
2157 0 : |r| {
2158 0 : tenant_service_handler(
2159 0 : r,
2160 0 : handle_tenant_timeline_detach_ancestor,
2161 0 : RequestName("v1_tenant_timeline_detach_ancestor"),
2162 0 : )
2163 0 : },
2164 0 : )
2165 0 : .post(
2166 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/block_gc",
2167 0 : |r| {
2168 0 : tenant_service_handler(
2169 0 : r,
2170 0 : |s, r| handle_tenant_timeline_block_unblock_gc(s, r, BlockUnblock::Block),
2171 0 : RequestName("v1_tenant_timeline_block_unblock_gc"),
2172 0 : )
2173 0 : },
2174 0 : )
2175 0 : .post(
2176 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/unblock_gc",
2177 0 : |r| {
2178 0 : tenant_service_handler(
2179 0 : r,
2180 0 : |s, r| handle_tenant_timeline_block_unblock_gc(s, r, BlockUnblock::Unblock),
2181 0 : RequestName("v1_tenant_timeline_block_unblock_gc"),
2182 0 : )
2183 0 : },
2184 0 : )
2185 0 : .post(
2186 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
2187 0 : |r| {
2188 0 : tenant_service_handler(
2189 0 : r,
2190 0 : handle_tenant_timeline_download_heatmap_layers,
2191 0 : RequestName("v1_tenant_timeline_download_heatmap_layers"),
2192 0 : )
2193 0 : },
2194 0 : )
2195 0 : // Tenant detail GET passthrough to shard zero:
2196 0 : .get("/v1/tenant/:tenant_id", |r| {
2197 0 : tenant_service_handler(
2198 0 : r,
2199 0 : handle_tenant_timeline_passthrough,
2200 0 : RequestName("v1_tenant_passthrough"),
2201 0 : )
2202 0 : })
2203 0 : // The `*` in the URL is a wildcard: any tenant/timeline GET APIs on the pageserver
2204 0 : // are implicitly exposed here. This must be last in the list to avoid
2205 0 : // taking precedence over other GET methods we might implement by hand.
2206 0 : .get("/v1/tenant/:tenant_id/*", |r| {
2207 0 : tenant_service_handler(
2208 0 : r,
2209 0 : handle_tenant_timeline_passthrough,
2210 0 : RequestName("v1_tenant_passthrough"),
2211 0 : )
2212 0 : })
2213 0 : }
2214 :
2215 : #[cfg(test)]
2216 : mod test {
2217 :
2218 : use super::path_without_ids;
2219 :
2220 : #[test]
2221 1 : fn test_path_without_ids() {
2222 1 : assert_eq!(
2223 1 : path_without_ids(
2224 1 : "/v1/tenant/1a2b3344556677881122334455667788/timeline/AA223344556677881122334455667788"
2225 1 : ),
2226 1 : "/v1/tenant//timeline/"
2227 1 : );
2228 1 : assert_eq!(
2229 1 : path_without_ids(
2230 1 : "/v1/tenant/1a2b3344556677881122334455667788-0108/timeline/AA223344556677881122334455667788"
2231 1 : ),
2232 1 : "/v1/tenant//timeline/"
2233 1 : );
2234 1 : assert_eq!(
2235 1 : path_without_ids(
2236 1 : "/v1/tenant/1a2b3344556677881122334455667788-0108/timeline/AA223344556677881122334455667788?parameter=foo"
2237 1 : ),
2238 1 : "/v1/tenant//timeline/"
2239 1 : );
2240 1 : }
2241 : }
|
