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 : TimelineImportRequest,
26 : };
27 : use pageserver_api::models::{
28 : DetachBehavior, LsnLeaseRequest, TenantConfigPatchRequest, TenantConfigRequest,
29 : TenantLocationConfigRequest, TenantShardSplitRequest, TenantTimeTravelRequest,
30 : TimelineArchivalConfigRequest, TimelineCreateRequest,
31 : };
32 : use pageserver_api::shard::TenantShardId;
33 : use pageserver_api::upcall_api::{
34 : PutTimelineImportStatusRequest, ReAttachRequest, ValidateRequest,
35 : };
36 : use pageserver_client::{BlockUnblock, mgmt_api};
37 : use routerify::Middleware;
38 : use tokio_util::sync::CancellationToken;
39 : use tracing::warn;
40 : use utils::auth::{Scope, SwappableJwtAuth};
41 : use utils::id::{NodeId, TenantId, TimelineId};
42 :
43 : use crate::http;
44 : use crate::metrics::{
45 : HttpRequestLatencyLabelGroup, HttpRequestStatusLabelGroup, METRICS_REGISTRY,
46 : PageserverRequestLabelGroup,
47 : };
48 : use crate::persistence::SafekeeperUpsert;
49 : use crate::reconciler::ReconcileError;
50 : use crate::service::{LeadershipStatus, RECONCILE_TIMEOUT, STARTUP_RECONCILE_TIMEOUT, Service};
51 :
52 : /// State available to HTTP request handlers
53 : pub struct HttpState {
54 : service: Arc<crate::service::Service>,
55 : auth: Option<Arc<SwappableJwtAuth>>,
56 : rate_limiter: governor::DefaultKeyedRateLimiter<TenantId>,
57 : neon_metrics: NeonMetrics,
58 : allowlist_routes: &'static [&'static str],
59 : }
60 :
61 : impl HttpState {
62 0 : pub fn new(
63 0 : service: Arc<crate::service::Service>,
64 0 : auth: Option<Arc<SwappableJwtAuth>>,
65 0 : build_info: BuildInfo,
66 0 : ) -> Self {
67 0 : let quota = governor::Quota::per_second(service.get_config().tenant_rate_limit);
68 0 : Self {
69 0 : service,
70 0 : auth,
71 0 : rate_limiter: governor::RateLimiter::keyed(quota),
72 0 : neon_metrics: NeonMetrics::new(build_info),
73 0 : allowlist_routes: &[
74 0 : "/status",
75 0 : "/live",
76 0 : "/ready",
77 0 : "/metrics",
78 0 : "/profile/cpu",
79 0 : "/profile/heap",
80 0 : ],
81 0 : }
82 0 : }
83 : }
84 :
85 : #[inline(always)]
86 0 : fn get_state(request: &Request<Body>) -> &HttpState {
87 0 : request
88 0 : .data::<Arc<HttpState>>()
89 0 : .expect("unknown state type")
90 0 : .as_ref()
91 0 : }
92 :
93 : /// Rate limits tenant requests.
94 : ///
95 : /// TODO: this should be a request middleware, but requires us to extract the tenant ID from
96 : /// different URLs in a systematic way.
97 : ///
98 : /// TODO: consider returning a 429 response if these start piling up.
99 0 : async fn maybe_rate_limit(request: &Request<Body>, tenant_id: TenantId) {
100 0 : // Check if the tenant should be rate-limited.
101 0 : let rate_limiter = &get_state(request).rate_limiter;
102 0 : if rate_limiter.check_key(&tenant_id).is_ok() {
103 0 : return;
104 0 : }
105 0 :
106 0 : // Measure the rate limiting delay.
107 0 : let _timer = METRICS_REGISTRY
108 0 : .metrics_group
109 0 : .storage_controller_http_request_rate_limited
110 0 : .start_timer();
111 :
112 : // Log rate limited tenants once every 10 seconds.
113 : static LOG_RATE_LIMITER: LazyLock<governor::DefaultKeyedRateLimiter<TenantId>> =
114 0 : LazyLock::new(|| {
115 0 : let quota = governor::Quota::with_period(Duration::from_secs(10)).unwrap();
116 0 : governor::RateLimiter::keyed(quota)
117 0 : });
118 :
119 0 : if LOG_RATE_LIMITER.check_key(&tenant_id).is_ok() {
120 0 : warn!("tenant {tenant_id} is rate limited")
121 0 : }
122 :
123 : // Wait for quota.
124 0 : rate_limiter.until_key_ready(&tenant_id).await;
125 0 : }
126 :
127 : /// Pageserver calls into this on startup, to learn which tenants it should attach
128 0 : async fn handle_re_attach(req: Request<Body>) -> Result<Response<Body>, ApiError> {
129 0 : check_permissions(&req, Scope::GenerationsApi)?;
130 :
131 0 : let mut req = match maybe_forward(req).await {
132 0 : ForwardOutcome::Forwarded(res) => {
133 0 : return res;
134 : }
135 0 : ForwardOutcome::NotForwarded(req) => req,
136 : };
137 :
138 0 : let reattach_req = json_request::<ReAttachRequest>(&mut req).await?;
139 0 : let state = get_state(&req);
140 0 : json_response(StatusCode::OK, state.service.re_attach(reattach_req).await?)
141 0 : }
142 :
143 : /// Pageserver calls into this before doing deletions, to confirm that it still
144 : /// holds the latest generation for the tenants with deletions enqueued
145 0 : async fn handle_validate(req: Request<Body>) -> Result<Response<Body>, ApiError> {
146 0 : check_permissions(&req, Scope::GenerationsApi)?;
147 :
148 0 : let mut req = match maybe_forward(req).await {
149 0 : ForwardOutcome::Forwarded(res) => {
150 0 : return res;
151 : }
152 0 : ForwardOutcome::NotForwarded(req) => req,
153 : };
154 :
155 0 : let validate_req = json_request::<ValidateRequest>(&mut req).await?;
156 0 : let state = get_state(&req);
157 0 : json_response(StatusCode::OK, state.service.validate(validate_req).await?)
158 0 : }
159 :
160 0 : async fn handle_put_timeline_import_status(req: Request<Body>) -> Result<Response<Body>, ApiError> {
161 0 : check_permissions(&req, Scope::GenerationsApi)?;
162 :
163 0 : let mut req = match maybe_forward(req).await {
164 0 : ForwardOutcome::Forwarded(res) => {
165 0 : return res;
166 : }
167 0 : ForwardOutcome::NotForwarded(req) => req,
168 : };
169 :
170 0 : let put_req = json_request::<PutTimelineImportStatusRequest>(&mut req).await?;
171 :
172 0 : let state = get_state(&req);
173 0 : json_response(
174 0 : StatusCode::OK,
175 0 : state
176 0 : .service
177 0 : .handle_timeline_shard_import_progress_upcall(put_req)
178 0 : .await?,
179 : )
180 0 : }
181 :
182 : /// Call into this before attaching a tenant to a pageserver, to acquire a generation number
183 : /// (in the real control plane this is unnecessary, because the same program is managing
184 : /// generation numbers and doing attachments).
185 0 : async fn handle_attach_hook(req: Request<Body>) -> Result<Response<Body>, ApiError> {
186 0 : check_permissions(&req, Scope::Admin)?;
187 :
188 0 : let mut req = match maybe_forward(req).await {
189 0 : ForwardOutcome::Forwarded(res) => {
190 0 : return res;
191 : }
192 0 : ForwardOutcome::NotForwarded(req) => req,
193 : };
194 :
195 0 : let attach_req = json_request::<AttachHookRequest>(&mut req).await?;
196 0 : let state = get_state(&req);
197 0 :
198 0 : json_response(
199 0 : StatusCode::OK,
200 0 : state
201 0 : .service
202 0 : .attach_hook(attach_req)
203 0 : .await
204 0 : .map_err(ApiError::InternalServerError)?,
205 : )
206 0 : }
207 :
208 0 : async fn handle_inspect(req: Request<Body>) -> Result<Response<Body>, ApiError> {
209 0 : check_permissions(&req, Scope::Admin)?;
210 :
211 0 : let mut req = match maybe_forward(req).await {
212 0 : ForwardOutcome::Forwarded(res) => {
213 0 : return res;
214 : }
215 0 : ForwardOutcome::NotForwarded(req) => req,
216 : };
217 :
218 0 : let inspect_req = json_request::<InspectRequest>(&mut req).await?;
219 :
220 0 : let state = get_state(&req);
221 0 :
222 0 : json_response(StatusCode::OK, state.service.inspect(inspect_req))
223 0 : }
224 :
225 0 : async fn handle_tenant_create(
226 0 : service: Arc<Service>,
227 0 : req: Request<Body>,
228 0 : ) -> Result<Response<Body>, ApiError> {
229 0 : check_permissions(&req, Scope::PageServerApi)?;
230 :
231 0 : let mut req = match maybe_forward(req).await {
232 0 : ForwardOutcome::Forwarded(res) => {
233 0 : return res;
234 : }
235 0 : ForwardOutcome::NotForwarded(req) => req,
236 : };
237 :
238 0 : let create_req = json_request::<TenantCreateRequest>(&mut req).await?;
239 :
240 : json_response(
241 : StatusCode::CREATED,
242 0 : service.tenant_create(create_req).await?,
243 : )
244 0 : }
245 :
246 0 : async fn handle_tenant_location_config(
247 0 : service: Arc<Service>,
248 0 : req: Request<Body>,
249 0 : ) -> Result<Response<Body>, ApiError> {
250 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
251 0 : check_permissions(&req, Scope::PageServerApi)?;
252 :
253 0 : let mut req = match maybe_forward(req).await {
254 0 : ForwardOutcome::Forwarded(res) => {
255 0 : return res;
256 : }
257 0 : ForwardOutcome::NotForwarded(req) => req,
258 : };
259 :
260 0 : let config_req = json_request::<TenantLocationConfigRequest>(&mut req).await?;
261 : json_response(
262 : StatusCode::OK,
263 0 : service
264 0 : .tenant_location_config(tenant_shard_id, config_req)
265 0 : .await?,
266 : )
267 0 : }
268 :
269 0 : async fn handle_tenant_config_patch(
270 0 : service: Arc<Service>,
271 0 : req: Request<Body>,
272 0 : ) -> Result<Response<Body>, ApiError> {
273 0 : check_permissions(&req, Scope::PageServerApi)?;
274 :
275 0 : let mut req = match maybe_forward(req).await {
276 0 : ForwardOutcome::Forwarded(res) => {
277 0 : return res;
278 : }
279 0 : ForwardOutcome::NotForwarded(req) => req,
280 : };
281 :
282 0 : let config_req = json_request::<TenantConfigPatchRequest>(&mut req).await?;
283 :
284 : json_response(
285 : StatusCode::OK,
286 0 : service.tenant_config_patch(config_req).await?,
287 : )
288 0 : }
289 :
290 0 : async fn handle_tenant_config_set(
291 0 : service: Arc<Service>,
292 0 : req: Request<Body>,
293 0 : ) -> Result<Response<Body>, ApiError> {
294 0 : check_permissions(&req, Scope::PageServerApi)?;
295 :
296 0 : let mut req = match maybe_forward(req).await {
297 0 : ForwardOutcome::Forwarded(res) => {
298 0 : return res;
299 : }
300 0 : ForwardOutcome::NotForwarded(req) => req,
301 : };
302 :
303 0 : let config_req = json_request::<TenantConfigRequest>(&mut req).await?;
304 :
305 0 : json_response(StatusCode::OK, service.tenant_config_set(config_req).await?)
306 0 : }
307 :
308 0 : async fn handle_tenant_config_get(
309 0 : service: Arc<Service>,
310 0 : req: Request<Body>,
311 0 : ) -> Result<Response<Body>, ApiError> {
312 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
313 0 : check_permissions(&req, Scope::PageServerApi)?;
314 0 : maybe_rate_limit(&req, tenant_id).await;
315 :
316 0 : match maybe_forward(req).await {
317 0 : ForwardOutcome::Forwarded(res) => {
318 0 : return res;
319 : }
320 0 : ForwardOutcome::NotForwarded(_req) => {}
321 0 : };
322 0 :
323 0 : json_response(StatusCode::OK, service.tenant_config_get(tenant_id)?)
324 0 : }
325 :
326 0 : async fn handle_tenant_time_travel_remote_storage(
327 0 : service: Arc<Service>,
328 0 : req: Request<Body>,
329 0 : ) -> Result<Response<Body>, ApiError> {
330 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
331 0 : check_permissions(&req, Scope::PageServerApi)?;
332 0 : maybe_rate_limit(&req, tenant_id).await;
333 :
334 0 : let mut req = match maybe_forward(req).await {
335 0 : ForwardOutcome::Forwarded(res) => {
336 0 : return res;
337 : }
338 0 : ForwardOutcome::NotForwarded(req) => req,
339 : };
340 :
341 0 : let time_travel_req = json_request::<TenantTimeTravelRequest>(&mut req).await?;
342 :
343 0 : let timestamp_raw = must_get_query_param(&req, "travel_to")?;
344 0 : let _timestamp = humantime::parse_rfc3339(×tamp_raw).map_err(|_e| {
345 0 : ApiError::BadRequest(anyhow::anyhow!(
346 0 : "Invalid time for travel_to: {timestamp_raw:?}"
347 0 : ))
348 0 : })?;
349 :
350 0 : let done_if_after_raw = must_get_query_param(&req, "done_if_after")?;
351 0 : let _done_if_after = humantime::parse_rfc3339(&done_if_after_raw).map_err(|_e| {
352 0 : ApiError::BadRequest(anyhow::anyhow!(
353 0 : "Invalid time for done_if_after: {done_if_after_raw:?}"
354 0 : ))
355 0 : })?;
356 :
357 0 : service
358 0 : .tenant_time_travel_remote_storage(
359 0 : &time_travel_req,
360 0 : tenant_id,
361 0 : timestamp_raw,
362 0 : done_if_after_raw,
363 0 : )
364 0 : .await?;
365 0 : json_response(StatusCode::OK, ())
366 0 : }
367 :
368 0 : fn map_reqwest_hyper_status(status: reqwest::StatusCode) -> Result<hyper::StatusCode, ApiError> {
369 0 : hyper::StatusCode::from_u16(status.as_u16())
370 0 : .context("invalid status code")
371 0 : .map_err(ApiError::InternalServerError)
372 0 : }
373 :
374 0 : async fn handle_tenant_secondary_download(
375 0 : service: Arc<Service>,
376 0 : req: Request<Body>,
377 0 : ) -> Result<Response<Body>, ApiError> {
378 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
379 0 : let wait = parse_query_param(&req, "wait_ms")?.map(Duration::from_millis);
380 0 : maybe_rate_limit(&req, tenant_id).await;
381 :
382 0 : match maybe_forward(req).await {
383 0 : ForwardOutcome::Forwarded(res) => {
384 0 : return res;
385 : }
386 0 : ForwardOutcome::NotForwarded(_req) => {}
387 : };
388 :
389 0 : let (status, progress) = service.tenant_secondary_download(tenant_id, wait).await?;
390 0 : json_response(map_reqwest_hyper_status(status)?, progress)
391 0 : }
392 :
393 0 : async fn handle_tenant_delete(
394 0 : service: Arc<Service>,
395 0 : req: Request<Body>,
396 0 : ) -> Result<Response<Body>, ApiError> {
397 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
398 0 : check_permissions(&req, Scope::PageServerApi)?;
399 0 : maybe_rate_limit(&req, tenant_id).await;
400 :
401 0 : match maybe_forward(req).await {
402 0 : ForwardOutcome::Forwarded(res) => {
403 0 : return res;
404 : }
405 0 : ForwardOutcome::NotForwarded(_req) => {}
406 : };
407 :
408 0 : let status_code = service
409 0 : .tenant_delete(tenant_id)
410 0 : .await
411 0 : .and_then(map_reqwest_hyper_status)?;
412 :
413 0 : if status_code == StatusCode::NOT_FOUND {
414 : // The pageserver uses 404 for successful deletion, but we use 200
415 0 : json_response(StatusCode::OK, ())
416 : } else {
417 0 : json_response(status_code, ())
418 : }
419 0 : }
420 :
421 0 : async fn handle_tenant_timeline_create(
422 0 : service: Arc<Service>,
423 0 : req: Request<Body>,
424 0 : ) -> Result<Response<Body>, ApiError> {
425 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
426 0 : check_permissions(&req, Scope::PageServerApi)?;
427 0 : maybe_rate_limit(&req, tenant_id).await;
428 :
429 0 : let mut req = match maybe_forward(req).await {
430 0 : ForwardOutcome::Forwarded(res) => {
431 0 : return res;
432 : }
433 0 : ForwardOutcome::NotForwarded(req) => req,
434 : };
435 :
436 0 : let create_req = json_request::<TimelineCreateRequest>(&mut req).await?;
437 : json_response(
438 : StatusCode::CREATED,
439 0 : service
440 0 : .tenant_timeline_create(tenant_id, create_req)
441 0 : .await?,
442 : )
443 0 : }
444 :
445 0 : async fn handle_tenant_timeline_delete(
446 0 : service: Arc<Service>,
447 0 : req: Request<Body>,
448 0 : ) -> Result<Response<Body>, ApiError> {
449 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
450 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
451 :
452 0 : check_permissions(&req, Scope::PageServerApi)?;
453 0 : maybe_rate_limit(&req, tenant_id).await;
454 :
455 0 : match maybe_forward(req).await {
456 0 : ForwardOutcome::Forwarded(res) => {
457 0 : return res;
458 : }
459 0 : ForwardOutcome::NotForwarded(_req) => {}
460 : };
461 :
462 : // For timeline deletions, which both implement an "initially return 202, then 404 once
463 : // we're done" semantic, we wrap with a retry loop to expose a simpler API upstream.
464 0 : async fn deletion_wrapper<R, F>(service: Arc<Service>, f: F) -> Result<Response<Body>, ApiError>
465 0 : where
466 0 : R: std::future::Future<Output = Result<StatusCode, ApiError>> + Send + 'static,
467 0 : F: Fn(Arc<Service>) -> R + Send + Sync + 'static,
468 0 : {
469 : // On subsequent retries, wait longer.
470 : // Enable callers with a 25 second request timeout to reliably get a response
471 : const MAX_WAIT: Duration = Duration::from_secs(25);
472 : const MAX_RETRY_PERIOD: Duration = Duration::from_secs(5);
473 :
474 0 : let started_at = Instant::now();
475 0 :
476 0 : // To keep deletion reasonably snappy for small tenants, initially check after 1 second if deletion
477 0 : // completed.
478 0 : let mut retry_period = Duration::from_secs(1);
479 :
480 : loop {
481 0 : let status = f(service.clone()).await?;
482 0 : match status {
483 : StatusCode::ACCEPTED => {
484 0 : tracing::info!("Deletion accepted, waiting to try again...");
485 0 : tokio::time::sleep(retry_period).await;
486 0 : retry_period = MAX_RETRY_PERIOD;
487 : }
488 : StatusCode::CONFLICT => {
489 0 : tracing::info!("Deletion already in progress, waiting to try again...");
490 0 : tokio::time::sleep(retry_period).await;
491 : }
492 : StatusCode::NOT_FOUND => {
493 0 : tracing::info!("Deletion complete");
494 0 : return json_response(StatusCode::OK, ());
495 : }
496 : _ => {
497 0 : tracing::warn!("Unexpected status {status}");
498 0 : return json_response(status, ());
499 : }
500 : }
501 :
502 0 : let now = Instant::now();
503 0 : if now + retry_period > started_at + MAX_WAIT {
504 0 : tracing::info!("Deletion timed out waiting for 404");
505 : // REQUEST_TIMEOUT would be more appropriate, but CONFLICT is already part of
506 : // the pageserver's swagger definition for this endpoint, and has the same desired
507 : // effect of causing the control plane to retry later.
508 0 : return json_response(StatusCode::CONFLICT, ());
509 0 : }
510 : }
511 0 : }
512 :
513 0 : deletion_wrapper(service, move |service| async move {
514 0 : service
515 0 : .tenant_timeline_delete(tenant_id, timeline_id)
516 0 : .await
517 0 : .and_then(map_reqwest_hyper_status)
518 0 : })
519 0 : .await
520 0 : }
521 :
522 0 : async fn handle_tenant_timeline_archival_config(
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 : let mut req = match maybe_forward(req).await {
533 0 : ForwardOutcome::Forwarded(res) => {
534 0 : return res;
535 : }
536 0 : ForwardOutcome::NotForwarded(req) => req,
537 : };
538 :
539 0 : let create_req = json_request::<TimelineArchivalConfigRequest>(&mut req).await?;
540 :
541 0 : service
542 0 : .tenant_timeline_archival_config(tenant_id, timeline_id, create_req)
543 0 : .await?;
544 :
545 0 : json_response(StatusCode::OK, ())
546 0 : }
547 :
548 0 : async fn handle_tenant_timeline_detach_ancestor(
549 0 : service: Arc<Service>,
550 0 : req: Request<Body>,
551 0 : ) -> Result<Response<Body>, ApiError> {
552 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
553 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
554 0 : let behavior: Option<DetachBehavior> = parse_query_param(&req, "detach_behavior")?;
555 :
556 0 : check_permissions(&req, Scope::PageServerApi)?;
557 0 : maybe_rate_limit(&req, tenant_id).await;
558 :
559 0 : match maybe_forward(req).await {
560 0 : ForwardOutcome::Forwarded(res) => {
561 0 : return res;
562 : }
563 0 : ForwardOutcome::NotForwarded(_req) => {}
564 : };
565 :
566 0 : let res = service
567 0 : .tenant_timeline_detach_ancestor(tenant_id, timeline_id, behavior)
568 0 : .await?;
569 :
570 0 : json_response(StatusCode::OK, res)
571 0 : }
572 :
573 0 : async fn handle_tenant_timeline_block_unblock_gc(
574 0 : service: Arc<Service>,
575 0 : req: Request<Body>,
576 0 : dir: BlockUnblock,
577 0 : ) -> Result<Response<Body>, ApiError> {
578 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
579 0 : check_permissions(&req, Scope::PageServerApi)?;
580 0 : maybe_rate_limit(&req, tenant_id).await;
581 :
582 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
583 :
584 0 : service
585 0 : .tenant_timeline_block_unblock_gc(tenant_id, timeline_id, dir)
586 0 : .await?;
587 :
588 0 : json_response(StatusCode::OK, ())
589 0 : }
590 :
591 0 : async fn handle_tenant_timeline_download_heatmap_layers(
592 0 : service: Arc<Service>,
593 0 : req: Request<Body>,
594 0 : ) -> Result<Response<Body>, ApiError> {
595 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
596 :
597 0 : check_permissions(&req, Scope::PageServerApi)?;
598 0 : maybe_rate_limit(&req, tenant_shard_id.tenant_id).await;
599 :
600 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
601 0 : let concurrency: Option<usize> = parse_query_param(&req, "concurrency")?;
602 0 : let recurse = parse_query_param(&req, "recurse")?.unwrap_or(false);
603 0 :
604 0 : service
605 0 : .tenant_timeline_download_heatmap_layers(tenant_shard_id, timeline_id, concurrency, recurse)
606 0 : .await?;
607 :
608 0 : json_response(StatusCode::OK, ())
609 0 : }
610 :
611 0 : async fn handle_tenant_timeline_lsn_lease(
612 0 : service: Arc<Service>,
613 0 : req: Request<Body>,
614 0 : ) -> Result<Response<Body>, ApiError> {
615 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
616 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
617 :
618 0 : check_permissions(&req, Scope::PageServerApi)?;
619 0 : maybe_rate_limit(&req, tenant_id).await;
620 :
621 0 : let mut req = match maybe_forward(req).await {
622 0 : ForwardOutcome::Forwarded(res) => {
623 0 : return res;
624 : }
625 0 : ForwardOutcome::NotForwarded(req) => req,
626 : };
627 :
628 0 : let lsn_lease_request = json_request::<LsnLeaseRequest>(&mut req).await?;
629 :
630 0 : service
631 0 : .tenant_timeline_lsn_lease(tenant_id, timeline_id, lsn_lease_request.lsn)
632 0 : .await?;
633 :
634 0 : json_response(StatusCode::OK, ())
635 0 : }
636 :
637 : // For metric labels where we would like to include the approximate path, but exclude high-cardinality fields like query parameters
638 : // and tenant/timeline IDs. Since we are proxying to arbitrary paths, we don't have routing templates to
639 : // compare to, so we can just filter out our well known ID format with regexes.
640 3 : fn path_without_ids(path: &str) -> String {
641 : static ID_REGEX: std::sync::OnceLock<regex::Regex> = std::sync::OnceLock::new();
642 3 : ID_REGEX
643 3 : .get_or_init(|| regex::Regex::new(r"([0-9a-fA-F]{32}(-[0-9]{4})?|\?.*)").unwrap())
644 3 : .replace_all(path, "")
645 3 : .to_string()
646 3 : }
647 :
648 0 : async fn handle_tenant_timeline_passthrough(
649 0 : service: Arc<Service>,
650 0 : req: Request<Body>,
651 0 : ) -> Result<Response<Body>, ApiError> {
652 0 : let tenant_or_shard_id: TenantShardId = parse_request_param(&req, "tenant_id")?;
653 0 : check_permissions(&req, Scope::PageServerApi)?;
654 0 : maybe_rate_limit(&req, tenant_or_shard_id.tenant_id).await;
655 :
656 0 : let req = match maybe_forward(req).await {
657 0 : ForwardOutcome::Forwarded(res) => {
658 0 : return res;
659 : }
660 0 : ForwardOutcome::NotForwarded(req) => req,
661 : };
662 :
663 0 : let Some(path) = req.uri().path_and_query() else {
664 : // This should never happen, our request router only calls us if there is a path
665 0 : return Err(ApiError::BadRequest(anyhow::anyhow!("Missing path")));
666 : };
667 :
668 0 : let method = match *req.method() {
669 0 : hyper::Method::GET => reqwest::Method::GET,
670 0 : hyper::Method::POST => reqwest::Method::POST,
671 0 : hyper::Method::PUT => reqwest::Method::PUT,
672 0 : hyper::Method::DELETE => reqwest::Method::DELETE,
673 0 : hyper::Method::PATCH => reqwest::Method::PATCH,
674 0 : _ => return Err(ApiError::BadRequest(anyhow::anyhow!("Unsupported method"))),
675 : };
676 :
677 0 : tracing::info!(
678 0 : "Proxying request for tenant {} ({})",
679 : tenant_or_shard_id.tenant_id,
680 : path
681 : );
682 :
683 : // Find the node that holds shard zero
684 0 : let (node, tenant_shard_id) = if tenant_or_shard_id.is_unsharded() {
685 0 : service
686 0 : .tenant_shard0_node(tenant_or_shard_id.tenant_id)
687 0 : .await?
688 : } else {
689 : (
690 0 : service.tenant_shard_node(tenant_or_shard_id).await?,
691 0 : tenant_or_shard_id,
692 : )
693 : };
694 :
695 : // Callers will always pass an unsharded tenant ID. Before proxying, we must
696 : // rewrite this to a shard-aware shard zero ID.
697 0 : let path = format!("{}", path);
698 0 : let tenant_str = tenant_or_shard_id.tenant_id.to_string();
699 0 : let tenant_shard_str = format!("{}", tenant_shard_id);
700 0 : let path = path.replace(&tenant_str, &tenant_shard_str);
701 0 :
702 0 : let latency = &METRICS_REGISTRY
703 0 : .metrics_group
704 0 : .storage_controller_passthrough_request_latency;
705 0 :
706 0 : let path_label = path_without_ids(&path)
707 0 : .split('/')
708 0 : .filter(|token| !token.is_empty())
709 0 : .collect::<Vec<_>>()
710 0 : .join("_");
711 0 : let labels = PageserverRequestLabelGroup {
712 0 : pageserver_id: &node.get_id().to_string(),
713 0 : path: &path_label,
714 0 : method: crate::metrics::Method::Get,
715 0 : };
716 0 :
717 0 : let _timer = latency.start_timer(labels.clone());
718 0 :
719 0 : let client = mgmt_api::Client::new(
720 0 : service.get_http_client().clone(),
721 0 : node.base_url(),
722 0 : service.get_config().pageserver_jwt_token.as_deref(),
723 0 : );
724 0 : let resp = client.op_raw(method, path).await.map_err(|e|
725 : // We return 503 here because if we can't successfully send a request to the pageserver,
726 : // either we aren't available or the pageserver is unavailable.
727 0 : ApiError::ResourceUnavailable(format!("Error sending pageserver API request to {node}: {e}").into()))?;
728 :
729 0 : if !resp.status().is_success() {
730 0 : let error_counter = &METRICS_REGISTRY
731 0 : .metrics_group
732 0 : .storage_controller_passthrough_request_error;
733 0 : error_counter.inc(labels);
734 0 : }
735 :
736 : // Transform 404 into 503 if we raced with a migration
737 0 : if resp.status() == reqwest::StatusCode::NOT_FOUND {
738 : // Look up node again: if we migrated it will be different
739 0 : let new_node = service.tenant_shard_node(tenant_shard_id).await?;
740 0 : if new_node.get_id() != node.get_id() {
741 : // Rather than retry here, send the client a 503 to prompt a retry: this matches
742 : // the pageserver's use of 503, and all clients calling this API should retry on 503.
743 0 : return Err(ApiError::ResourceUnavailable(
744 0 : format!("Pageserver {node} returned 404, was migrated to {new_node}").into(),
745 0 : ));
746 0 : }
747 0 : }
748 :
749 : // We have a reqest::Response, would like a http::Response
750 0 : let mut builder = hyper::Response::builder().status(map_reqwest_hyper_status(resp.status())?);
751 0 : for (k, v) in resp.headers() {
752 0 : builder = builder.header(k.as_str(), v.as_bytes());
753 0 : }
754 :
755 0 : let response = builder
756 0 : .body(Body::wrap_stream(resp.bytes_stream()))
757 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
758 :
759 0 : Ok(response)
760 0 : }
761 :
762 0 : async fn handle_tenant_locate(
763 0 : service: Arc<Service>,
764 0 : req: Request<Body>,
765 0 : ) -> Result<Response<Body>, ApiError> {
766 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
767 :
768 0 : check_permissions(&req, Scope::Admin)?;
769 : // NB: don't rate limit: admin operation.
770 :
771 0 : match maybe_forward(req).await {
772 0 : ForwardOutcome::Forwarded(res) => {
773 0 : return res;
774 : }
775 0 : ForwardOutcome::NotForwarded(_req) => {}
776 0 : };
777 0 :
778 0 : json_response(StatusCode::OK, service.tenant_locate(tenant_id)?)
779 0 : }
780 :
781 0 : async fn handle_tenant_describe(
782 0 : service: Arc<Service>,
783 0 : req: Request<Body>,
784 0 : ) -> Result<Response<Body>, ApiError> {
785 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
786 0 : check_permissions(&req, Scope::Scrubber)?;
787 : // NB: don't rate limit: scrubber operation.
788 :
789 0 : match maybe_forward(req).await {
790 0 : ForwardOutcome::Forwarded(res) => {
791 0 : return res;
792 : }
793 0 : ForwardOutcome::NotForwarded(_req) => {}
794 0 : };
795 0 :
796 0 : json_response(StatusCode::OK, service.tenant_describe(tenant_id)?)
797 0 : }
798 :
799 0 : async fn handle_tenant_list(
800 0 : service: Arc<Service>,
801 0 : req: Request<Body>,
802 0 : ) -> Result<Response<Body>, ApiError> {
803 0 : check_permissions(&req, Scope::Admin)?;
804 :
805 0 : let limit: Option<usize> = parse_query_param(&req, "limit")?;
806 0 : let start_after: Option<TenantId> = parse_query_param(&req, "start_after")?;
807 0 : tracing::info!("start_after: {:?}", start_after);
808 :
809 0 : match maybe_forward(req).await {
810 0 : ForwardOutcome::Forwarded(res) => {
811 0 : return res;
812 : }
813 0 : ForwardOutcome::NotForwarded(_req) => {}
814 0 : };
815 0 :
816 0 : json_response(StatusCode::OK, service.tenant_list(limit, start_after))
817 0 : }
818 :
819 0 : async fn handle_node_register(req: Request<Body>) -> Result<Response<Body>, ApiError> {
820 0 : check_permissions(&req, Scope::Infra)?;
821 :
822 0 : let mut req = match maybe_forward(req).await {
823 0 : ForwardOutcome::Forwarded(res) => {
824 0 : return res;
825 : }
826 0 : ForwardOutcome::NotForwarded(req) => req,
827 : };
828 :
829 0 : let register_req = json_request::<NodeRegisterRequest>(&mut req).await?;
830 0 : let state = get_state(&req);
831 0 : state.service.node_register(register_req).await?;
832 0 : json_response(StatusCode::OK, ())
833 0 : }
834 :
835 0 : async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
836 0 : check_permissions(&req, Scope::Infra)?;
837 :
838 0 : let req = match maybe_forward(req).await {
839 0 : ForwardOutcome::Forwarded(res) => {
840 0 : return res;
841 : }
842 0 : ForwardOutcome::NotForwarded(req) => req,
843 0 : };
844 0 :
845 0 : let state = get_state(&req);
846 0 : let mut nodes = state.service.node_list().await?;
847 0 : nodes.sort_by_key(|n| n.get_id());
848 0 : let api_nodes = nodes.into_iter().map(|n| n.describe()).collect::<Vec<_>>();
849 0 :
850 0 : json_response(StatusCode::OK, api_nodes)
851 0 : }
852 :
853 0 : async fn handle_node_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
854 0 : check_permissions(&req, Scope::Admin)?;
855 :
856 0 : let req = match maybe_forward(req).await {
857 0 : ForwardOutcome::Forwarded(res) => {
858 0 : return res;
859 : }
860 0 : ForwardOutcome::NotForwarded(req) => req,
861 0 : };
862 0 :
863 0 : let state = get_state(&req);
864 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
865 0 : json_response(StatusCode::OK, state.service.node_drop(node_id).await?)
866 0 : }
867 :
868 0 : async fn handle_node_delete(req: Request<Body>) -> Result<Response<Body>, ApiError> {
869 0 : check_permissions(&req, Scope::Admin)?;
870 :
871 0 : let req = match maybe_forward(req).await {
872 0 : ForwardOutcome::Forwarded(res) => {
873 0 : return res;
874 : }
875 0 : ForwardOutcome::NotForwarded(req) => req,
876 0 : };
877 0 :
878 0 : let state = get_state(&req);
879 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
880 0 : json_response(StatusCode::OK, state.service.node_delete(node_id).await?)
881 0 : }
882 :
883 0 : async fn handle_node_configure(req: Request<Body>) -> Result<Response<Body>, ApiError> {
884 0 : check_permissions(&req, Scope::Admin)?;
885 :
886 0 : let mut req = match maybe_forward(req).await {
887 0 : ForwardOutcome::Forwarded(res) => {
888 0 : return res;
889 : }
890 0 : ForwardOutcome::NotForwarded(req) => req,
891 : };
892 :
893 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
894 0 : let config_req = json_request::<NodeConfigureRequest>(&mut req).await?;
895 0 : if node_id != config_req.node_id {
896 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
897 0 : "Path and body node_id differ"
898 0 : )));
899 0 : }
900 0 : let state = get_state(&req);
901 0 :
902 0 : json_response(
903 0 : StatusCode::OK,
904 0 : state
905 0 : .service
906 0 : .external_node_configure(
907 0 : config_req.node_id,
908 0 : config_req.availability.map(NodeAvailability::from),
909 0 : config_req.scheduling,
910 0 : )
911 0 : .await?,
912 : )
913 0 : }
914 :
915 0 : async fn handle_node_status(req: Request<Body>) -> Result<Response<Body>, ApiError> {
916 0 : check_permissions(&req, Scope::Infra)?;
917 :
918 0 : let req = match maybe_forward(req).await {
919 0 : ForwardOutcome::Forwarded(res) => {
920 0 : return res;
921 : }
922 0 : ForwardOutcome::NotForwarded(req) => req,
923 0 : };
924 0 :
925 0 : let state = get_state(&req);
926 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
927 :
928 0 : let node_status = state.service.get_node(node_id).await?.describe();
929 0 :
930 0 : json_response(StatusCode::OK, node_status)
931 0 : }
932 :
933 0 : async fn handle_node_shards(req: Request<Body>) -> Result<Response<Body>, ApiError> {
934 0 : check_permissions(&req, Scope::Admin)?;
935 :
936 0 : let state = get_state(&req);
937 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
938 :
939 0 : let node_status = state.service.get_node_shards(node_id).await?;
940 :
941 0 : json_response(StatusCode::OK, node_status)
942 0 : }
943 :
944 0 : async fn handle_get_leader(req: Request<Body>) -> Result<Response<Body>, ApiError> {
945 0 : check_permissions(&req, Scope::Admin)?;
946 :
947 0 : let req = match maybe_forward(req).await {
948 0 : ForwardOutcome::Forwarded(res) => {
949 0 : return res;
950 : }
951 0 : ForwardOutcome::NotForwarded(req) => req,
952 0 : };
953 0 :
954 0 : let state = get_state(&req);
955 0 : let leader = state.service.get_leader().await.map_err(|err| {
956 0 : ApiError::InternalServerError(anyhow::anyhow!(
957 0 : "Failed to read leader from database: {err}"
958 0 : ))
959 0 : })?;
960 :
961 0 : json_response(StatusCode::OK, leader)
962 0 : }
963 :
964 0 : async fn handle_node_drain(req: Request<Body>) -> Result<Response<Body>, ApiError> {
965 0 : check_permissions(&req, Scope::Infra)?;
966 :
967 0 : let req = match maybe_forward(req).await {
968 0 : ForwardOutcome::Forwarded(res) => {
969 0 : return res;
970 : }
971 0 : ForwardOutcome::NotForwarded(req) => req,
972 0 : };
973 0 :
974 0 : let state = get_state(&req);
975 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
976 :
977 0 : state.service.start_node_drain(node_id).await?;
978 :
979 0 : json_response(StatusCode::ACCEPTED, ())
980 0 : }
981 :
982 0 : async fn handle_cancel_node_drain(req: Request<Body>) -> Result<Response<Body>, ApiError> {
983 0 : check_permissions(&req, Scope::Infra)?;
984 :
985 0 : let req = match maybe_forward(req).await {
986 0 : ForwardOutcome::Forwarded(res) => {
987 0 : return res;
988 : }
989 0 : ForwardOutcome::NotForwarded(req) => req,
990 0 : };
991 0 :
992 0 : let state = get_state(&req);
993 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
994 :
995 0 : state.service.cancel_node_drain(node_id).await?;
996 :
997 0 : json_response(StatusCode::ACCEPTED, ())
998 0 : }
999 :
1000 0 : async fn handle_node_fill(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1001 0 : check_permissions(&req, Scope::Infra)?;
1002 :
1003 0 : let req = match maybe_forward(req).await {
1004 0 : ForwardOutcome::Forwarded(res) => {
1005 0 : return res;
1006 : }
1007 0 : ForwardOutcome::NotForwarded(req) => req,
1008 0 : };
1009 0 :
1010 0 : let state = get_state(&req);
1011 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
1012 :
1013 0 : state.service.start_node_fill(node_id).await?;
1014 :
1015 0 : json_response(StatusCode::ACCEPTED, ())
1016 0 : }
1017 :
1018 0 : async fn handle_cancel_node_fill(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1019 0 : check_permissions(&req, Scope::Infra)?;
1020 :
1021 0 : let req = match maybe_forward(req).await {
1022 0 : ForwardOutcome::Forwarded(res) => {
1023 0 : return res;
1024 : }
1025 0 : ForwardOutcome::NotForwarded(req) => req,
1026 0 : };
1027 0 :
1028 0 : let state = get_state(&req);
1029 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
1030 :
1031 0 : state.service.cancel_node_fill(node_id).await?;
1032 :
1033 0 : json_response(StatusCode::ACCEPTED, ())
1034 0 : }
1035 :
1036 0 : async fn handle_safekeeper_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1037 0 : check_permissions(&req, Scope::Infra)?;
1038 :
1039 0 : let req = match maybe_forward(req).await {
1040 0 : ForwardOutcome::Forwarded(res) => {
1041 0 : return res;
1042 : }
1043 0 : ForwardOutcome::NotForwarded(req) => req,
1044 0 : };
1045 0 :
1046 0 : let state = get_state(&req);
1047 0 : let safekeepers = state.service.safekeepers_list().await?;
1048 0 : json_response(StatusCode::OK, safekeepers)
1049 0 : }
1050 :
1051 0 : async fn handle_metadata_health_update(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1052 0 : check_permissions(&req, Scope::Scrubber)?;
1053 :
1054 0 : let mut req = match maybe_forward(req).await {
1055 0 : ForwardOutcome::Forwarded(res) => {
1056 0 : return res;
1057 : }
1058 0 : ForwardOutcome::NotForwarded(req) => req,
1059 : };
1060 :
1061 0 : let update_req = json_request::<MetadataHealthUpdateRequest>(&mut req).await?;
1062 0 : let state = get_state(&req);
1063 0 :
1064 0 : state.service.metadata_health_update(update_req).await?;
1065 :
1066 0 : json_response(StatusCode::OK, MetadataHealthUpdateResponse {})
1067 0 : }
1068 :
1069 0 : async fn handle_metadata_health_list_unhealthy(
1070 0 : req: Request<Body>,
1071 0 : ) -> Result<Response<Body>, ApiError> {
1072 0 : check_permissions(&req, Scope::Admin)?;
1073 :
1074 0 : let req = match maybe_forward(req).await {
1075 0 : ForwardOutcome::Forwarded(res) => {
1076 0 : return res;
1077 : }
1078 0 : ForwardOutcome::NotForwarded(req) => req,
1079 0 : };
1080 0 :
1081 0 : let state = get_state(&req);
1082 0 : let unhealthy_tenant_shards = state.service.metadata_health_list_unhealthy().await?;
1083 :
1084 0 : json_response(
1085 0 : StatusCode::OK,
1086 0 : MetadataHealthListUnhealthyResponse {
1087 0 : unhealthy_tenant_shards,
1088 0 : },
1089 0 : )
1090 0 : }
1091 :
1092 0 : async fn handle_metadata_health_list_outdated(
1093 0 : req: Request<Body>,
1094 0 : ) -> Result<Response<Body>, ApiError> {
1095 0 : check_permissions(&req, Scope::Admin)?;
1096 :
1097 0 : let mut req = match maybe_forward(req).await {
1098 0 : ForwardOutcome::Forwarded(res) => {
1099 0 : return res;
1100 : }
1101 0 : ForwardOutcome::NotForwarded(req) => req,
1102 : };
1103 :
1104 0 : let list_outdated_req = json_request::<MetadataHealthListOutdatedRequest>(&mut req).await?;
1105 0 : let state = get_state(&req);
1106 0 : let health_records = state
1107 0 : .service
1108 0 : .metadata_health_list_outdated(list_outdated_req.not_scrubbed_for)
1109 0 : .await?;
1110 :
1111 0 : json_response(
1112 0 : StatusCode::OK,
1113 0 : MetadataHealthListOutdatedResponse { health_records },
1114 0 : )
1115 0 : }
1116 :
1117 0 : async fn handle_tenant_shard_split(
1118 0 : service: Arc<Service>,
1119 0 : req: Request<Body>,
1120 0 : ) -> Result<Response<Body>, ApiError> {
1121 0 : check_permissions(&req, Scope::Admin)?;
1122 : // NB: don't rate limit: admin operation.
1123 :
1124 0 : let mut req = match maybe_forward(req).await {
1125 0 : ForwardOutcome::Forwarded(res) => {
1126 0 : return res;
1127 : }
1128 0 : ForwardOutcome::NotForwarded(req) => req,
1129 : };
1130 :
1131 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1132 0 : let split_req = json_request::<TenantShardSplitRequest>(&mut req).await?;
1133 :
1134 : json_response(
1135 : StatusCode::OK,
1136 0 : service.tenant_shard_split(tenant_id, split_req).await?,
1137 : )
1138 0 : }
1139 :
1140 0 : async fn handle_tenant_shard_migrate(
1141 0 : service: Arc<Service>,
1142 0 : req: Request<Body>,
1143 0 : ) -> Result<Response<Body>, ApiError> {
1144 0 : check_permissions(&req, Scope::Admin)?;
1145 : // NB: don't rate limit: admin operation.
1146 :
1147 0 : let mut req = match maybe_forward(req).await {
1148 0 : ForwardOutcome::Forwarded(res) => {
1149 0 : return res;
1150 : }
1151 0 : ForwardOutcome::NotForwarded(req) => req,
1152 : };
1153 :
1154 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1155 0 : let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
1156 : json_response(
1157 : StatusCode::OK,
1158 0 : service
1159 0 : .tenant_shard_migrate(tenant_shard_id, migrate_req)
1160 0 : .await?,
1161 : )
1162 0 : }
1163 :
1164 0 : async fn handle_tenant_shard_migrate_secondary(
1165 0 : service: Arc<Service>,
1166 0 : req: Request<Body>,
1167 0 : ) -> Result<Response<Body>, ApiError> {
1168 0 : check_permissions(&req, Scope::Admin)?;
1169 : // NB: don't rate limit: admin operation.
1170 :
1171 0 : let mut req = match maybe_forward(req).await {
1172 0 : ForwardOutcome::Forwarded(res) => {
1173 0 : return res;
1174 : }
1175 0 : ForwardOutcome::NotForwarded(req) => req,
1176 : };
1177 :
1178 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1179 0 : let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
1180 : json_response(
1181 : StatusCode::OK,
1182 0 : service
1183 0 : .tenant_shard_migrate_secondary(tenant_shard_id, migrate_req)
1184 0 : .await?,
1185 : )
1186 0 : }
1187 :
1188 0 : async fn handle_tenant_shard_cancel_reconcile(
1189 0 : service: Arc<Service>,
1190 0 : req: Request<Body>,
1191 0 : ) -> Result<Response<Body>, ApiError> {
1192 0 : check_permissions(&req, Scope::Admin)?;
1193 : // NB: don't rate limit: admin operation.
1194 :
1195 0 : let req = match maybe_forward(req).await {
1196 0 : ForwardOutcome::Forwarded(res) => {
1197 0 : return res;
1198 : }
1199 0 : ForwardOutcome::NotForwarded(req) => req,
1200 : };
1201 :
1202 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
1203 : json_response(
1204 : StatusCode::OK,
1205 0 : service
1206 0 : .tenant_shard_cancel_reconcile(tenant_shard_id)
1207 0 : .await?,
1208 : )
1209 0 : }
1210 :
1211 0 : async fn handle_tenant_update_policy(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1212 0 : check_permissions(&req, Scope::Admin)?;
1213 : // NB: don't rate limit: admin operation.
1214 :
1215 0 : let mut req = match maybe_forward(req).await {
1216 0 : ForwardOutcome::Forwarded(res) => {
1217 0 : return res;
1218 : }
1219 0 : ForwardOutcome::NotForwarded(req) => req,
1220 : };
1221 :
1222 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1223 0 : let update_req = json_request::<TenantPolicyRequest>(&mut req).await?;
1224 0 : let state = get_state(&req);
1225 0 :
1226 0 : json_response(
1227 0 : StatusCode::OK,
1228 0 : state
1229 0 : .service
1230 0 : .tenant_update_policy(tenant_id, update_req)
1231 0 : .await?,
1232 : )
1233 0 : }
1234 :
1235 0 : async fn handle_update_preferred_azs(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1236 0 : check_permissions(&req, Scope::Admin)?;
1237 :
1238 0 : let mut req = match maybe_forward(req).await {
1239 0 : ForwardOutcome::Forwarded(res) => {
1240 0 : return res;
1241 : }
1242 0 : ForwardOutcome::NotForwarded(req) => req,
1243 : };
1244 :
1245 0 : let azs_req = json_request::<ShardsPreferredAzsRequest>(&mut req).await?;
1246 0 : let state = get_state(&req);
1247 0 :
1248 0 : json_response(
1249 0 : StatusCode::OK,
1250 0 : state.service.update_shards_preferred_azs(azs_req).await?,
1251 : )
1252 0 : }
1253 :
1254 0 : async fn handle_step_down(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1255 0 : check_permissions(&req, Scope::ControllerPeer)?;
1256 :
1257 0 : let req = match maybe_forward(req).await {
1258 0 : ForwardOutcome::Forwarded(res) => {
1259 0 : return res;
1260 : }
1261 0 : ForwardOutcome::NotForwarded(req) => req,
1262 0 : };
1263 0 :
1264 0 : let state = get_state(&req);
1265 0 : let result = state.service.step_down().await;
1266 :
1267 0 : json_response(StatusCode::OK, result)
1268 0 : }
1269 :
1270 0 : async fn handle_tenant_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1271 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1272 0 : check_permissions(&req, Scope::PageServerApi)?;
1273 0 : maybe_rate_limit(&req, tenant_id).await;
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.tenant_drop(tenant_id).await?)
1285 0 : }
1286 :
1287 0 : async fn handle_tenant_import(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1288 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1289 0 : check_permissions(&req, Scope::PageServerApi)?;
1290 0 : maybe_rate_limit(&req, tenant_id).await;
1291 :
1292 0 : let req = match maybe_forward(req).await {
1293 0 : ForwardOutcome::Forwarded(res) => {
1294 0 : return res;
1295 : }
1296 0 : ForwardOutcome::NotForwarded(req) => req,
1297 0 : };
1298 0 :
1299 0 : let state = get_state(&req);
1300 0 :
1301 0 : json_response(
1302 0 : StatusCode::OK,
1303 0 : state.service.tenant_import(tenant_id).await?,
1304 : )
1305 0 : }
1306 :
1307 0 : async fn handle_timeline_import(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1308 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
1309 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
1310 0 : check_permissions(&req, Scope::PageServerApi)?;
1311 0 : maybe_rate_limit(&req, tenant_id).await;
1312 :
1313 0 : let mut req = match maybe_forward(req).await {
1314 0 : ForwardOutcome::Forwarded(res) => {
1315 0 : return res;
1316 : }
1317 0 : ForwardOutcome::NotForwarded(req) => req,
1318 : };
1319 :
1320 0 : let import_req = json_request::<TimelineImportRequest>(&mut req).await?;
1321 :
1322 0 : let state = get_state(&req);
1323 0 :
1324 0 : if import_req.tenant_id != tenant_id || import_req.timeline_id != timeline_id {
1325 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
1326 0 : "tenant id or timeline id mismatch: url={tenant_id}/{timeline_id}, body={}/{}",
1327 0 : import_req.tenant_id,
1328 0 : import_req.timeline_id
1329 0 : )));
1330 0 : }
1331 0 :
1332 0 : json_response(
1333 0 : StatusCode::OK,
1334 0 : state.service.timeline_import(import_req).await?,
1335 : )
1336 0 : }
1337 :
1338 0 : async fn handle_tenants_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1339 0 : check_permissions(&req, Scope::Admin)?;
1340 :
1341 0 : let req = match maybe_forward(req).await {
1342 0 : ForwardOutcome::Forwarded(res) => {
1343 0 : return res;
1344 : }
1345 0 : ForwardOutcome::NotForwarded(req) => req,
1346 0 : };
1347 0 :
1348 0 : let state = get_state(&req);
1349 0 : state.service.tenants_dump()
1350 0 : }
1351 :
1352 0 : async fn handle_scheduler_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1353 0 : check_permissions(&req, Scope::Admin)?;
1354 :
1355 0 : let req = match maybe_forward(req).await {
1356 0 : ForwardOutcome::Forwarded(res) => {
1357 0 : return res;
1358 : }
1359 0 : ForwardOutcome::NotForwarded(req) => req,
1360 0 : };
1361 0 :
1362 0 : let state = get_state(&req);
1363 0 : state.service.scheduler_dump()
1364 0 : }
1365 :
1366 0 : async fn handle_consistency_check(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1367 0 : check_permissions(&req, Scope::Admin)?;
1368 :
1369 0 : let req = match maybe_forward(req).await {
1370 0 : ForwardOutcome::Forwarded(res) => {
1371 0 : return res;
1372 : }
1373 0 : ForwardOutcome::NotForwarded(req) => req,
1374 0 : };
1375 0 :
1376 0 : let state = get_state(&req);
1377 0 :
1378 0 : json_response(StatusCode::OK, state.service.consistency_check().await?)
1379 0 : }
1380 :
1381 0 : async fn handle_reconcile_all(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1382 0 : check_permissions(&req, Scope::Admin)?;
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 : json_response(StatusCode::OK, state.service.reconcile_all_now().await?)
1394 0 : }
1395 :
1396 : /// Status endpoint is just used for checking that our HTTP listener is up
1397 : ///
1398 : /// This serves as our k8s startup probe.
1399 0 : async fn handle_status(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1400 0 : match maybe_forward(req).await {
1401 0 : ForwardOutcome::Forwarded(res) => {
1402 0 : return res;
1403 : }
1404 0 : ForwardOutcome::NotForwarded(_req) => {}
1405 0 : };
1406 0 :
1407 0 : json_response(StatusCode::OK, ())
1408 0 : }
1409 :
1410 : /// Liveness endpoint indicates that this storage controller is in a state
1411 : /// where it can fulfill it's responsibilties. Namely, startup has finished
1412 : /// and it is the current leader.
1413 : ///
1414 : /// This serves as our k8s liveness probe.
1415 0 : async fn handle_live(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1416 0 : let req = match maybe_forward(req).await {
1417 0 : ForwardOutcome::Forwarded(res) => {
1418 0 : return res;
1419 : }
1420 0 : ForwardOutcome::NotForwarded(req) => req,
1421 0 : };
1422 0 :
1423 0 : let state = get_state(&req);
1424 0 : let live = state.service.startup_complete.is_ready()
1425 0 : && state.service.get_leadership_status() == LeadershipStatus::Leader;
1426 :
1427 0 : if live {
1428 0 : json_response(StatusCode::OK, ())
1429 : } else {
1430 0 : json_response(StatusCode::SERVICE_UNAVAILABLE, ())
1431 : }
1432 0 : }
1433 :
1434 : /// Readiness endpoint indicates when we're done doing startup I/O (e.g. reconciling
1435 : /// with remote pageserver nodes). This is intended for use as a kubernetes readiness probe.
1436 0 : async fn handle_ready(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1437 0 : let req = match maybe_forward(req).await {
1438 0 : ForwardOutcome::Forwarded(res) => {
1439 0 : return res;
1440 : }
1441 0 : ForwardOutcome::NotForwarded(req) => req,
1442 0 : };
1443 0 :
1444 0 : let state = get_state(&req);
1445 0 : if state.service.startup_complete.is_ready() {
1446 0 : json_response(StatusCode::OK, ())
1447 : } else {
1448 0 : json_response(StatusCode::SERVICE_UNAVAILABLE, ())
1449 : }
1450 0 : }
1451 :
1452 : impl From<ReconcileError> for ApiError {
1453 0 : fn from(value: ReconcileError) -> Self {
1454 0 : ApiError::Conflict(format!("Reconciliation error: {}", value))
1455 0 : }
1456 : }
1457 :
1458 : /// Return the safekeeper record by instance id, or 404.
1459 : ///
1460 : /// Not used by anything except manual testing.
1461 0 : async fn handle_get_safekeeper(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1462 0 : check_permissions(&req, Scope::Infra)?;
1463 :
1464 0 : let id = parse_request_param::<i64>(&req, "id")?;
1465 :
1466 0 : let req = match maybe_forward(req).await {
1467 0 : ForwardOutcome::Forwarded(res) => {
1468 0 : return res;
1469 : }
1470 0 : ForwardOutcome::NotForwarded(req) => req,
1471 0 : };
1472 0 :
1473 0 : let state = get_state(&req);
1474 :
1475 0 : let res = state.service.get_safekeeper(id).await;
1476 :
1477 0 : match res {
1478 0 : Ok(b) => json_response(StatusCode::OK, b),
1479 : Err(crate::persistence::DatabaseError::Query(diesel::result::Error::NotFound)) => {
1480 0 : Err(ApiError::NotFound("unknown instance id".into()))
1481 : }
1482 0 : Err(other) => Err(other.into()),
1483 : }
1484 0 : }
1485 :
1486 : /// Used as part of deployment scripts.
1487 : ///
1488 : /// Assumes information is only relayed to storage controller after first selecting an unique id on
1489 : /// control plane database, which means we have an id field in the request and payload.
1490 0 : async fn handle_upsert_safekeeper(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
1491 0 : check_permissions(&req, Scope::Infra)?;
1492 :
1493 0 : let body = json_request::<SafekeeperUpsert>(&mut req).await?;
1494 0 : let id = parse_request_param::<i64>(&req, "id")?;
1495 :
1496 0 : if id != body.id {
1497 : // it should be repeated
1498 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
1499 0 : "id mismatch: url={id:?}, body={:?}",
1500 0 : body.id
1501 0 : )));
1502 0 : }
1503 0 :
1504 0 : if id <= 0 {
1505 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
1506 0 : "id not allowed to be zero or negative: {id}"
1507 0 : )));
1508 0 : }
1509 :
1510 0 : let req = match maybe_forward(req).await {
1511 0 : ForwardOutcome::Forwarded(res) => {
1512 0 : return res;
1513 : }
1514 0 : ForwardOutcome::NotForwarded(req) => req,
1515 0 : };
1516 0 :
1517 0 : let state = get_state(&req);
1518 0 :
1519 0 : state.service.upsert_safekeeper(body).await?;
1520 :
1521 0 : Ok(Response::builder()
1522 0 : .status(StatusCode::NO_CONTENT)
1523 0 : .body(Body::empty())
1524 0 : .unwrap())
1525 0 : }
1526 :
1527 : /// Sets the scheduling policy of the specified safekeeper
1528 0 : async fn handle_safekeeper_scheduling_policy(
1529 0 : mut req: Request<Body>,
1530 0 : ) -> Result<Response<Body>, ApiError> {
1531 0 : check_permissions(&req, Scope::Admin)?;
1532 :
1533 0 : let body = json_request::<SafekeeperSchedulingPolicyRequest>(&mut req).await?;
1534 0 : let id = parse_request_param::<i64>(&req, "id")?;
1535 :
1536 0 : let req = match maybe_forward(req).await {
1537 0 : ForwardOutcome::Forwarded(res) => {
1538 0 : return res;
1539 : }
1540 0 : ForwardOutcome::NotForwarded(req) => req,
1541 0 : };
1542 0 :
1543 0 : let state = get_state(&req);
1544 0 :
1545 0 : state
1546 0 : .service
1547 0 : .set_safekeeper_scheduling_policy(id, body.scheduling_policy)
1548 0 : .await?;
1549 :
1550 0 : json_response(StatusCode::OK, ())
1551 0 : }
1552 :
1553 : /// Common wrapper for request handlers that call into Service and will operate on tenants: they must only
1554 : /// be allowed to run if Service has finished its initial reconciliation.
1555 0 : async fn tenant_service_handler<R, H>(
1556 0 : request: Request<Body>,
1557 0 : handler: H,
1558 0 : request_name: RequestName,
1559 0 : ) -> R::Output
1560 0 : where
1561 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
1562 0 : H: FnOnce(Arc<Service>, Request<Body>) -> R + Send + Sync + 'static,
1563 0 : {
1564 0 : let state = get_state(&request);
1565 0 : let service = state.service.clone();
1566 0 :
1567 0 : let startup_complete = service.startup_complete.clone();
1568 0 : if tokio::time::timeout(STARTUP_RECONCILE_TIMEOUT, startup_complete.wait())
1569 0 : .await
1570 0 : .is_err()
1571 : {
1572 : // This shouldn't happen: it is the responsibilty of [`Service::startup_reconcile`] to use appropriate
1573 : // timeouts around its remote calls, to bound its runtime.
1574 0 : return Err(ApiError::Timeout(
1575 0 : "Timed out waiting for service readiness".into(),
1576 0 : ));
1577 0 : }
1578 0 :
1579 0 : named_request_span(
1580 0 : request,
1581 0 : |request| async move { handler(service, request).await },
1582 0 : request_name,
1583 0 : )
1584 0 : .await
1585 0 : }
1586 :
1587 : /// Check if the required scope is held in the request's token, or if the request has
1588 : /// a token with 'admin' scope then always permit it.
1589 0 : fn check_permissions(request: &Request<Body>, required_scope: Scope) -> Result<(), ApiError> {
1590 0 : check_permission_with(request, |claims| {
1591 0 : match crate::auth::check_permission(claims, required_scope) {
1592 0 : Err(e) => match crate::auth::check_permission(claims, Scope::Admin) {
1593 0 : Ok(()) => Ok(()),
1594 0 : Err(_) => Err(e),
1595 : },
1596 0 : Ok(()) => Ok(()),
1597 : }
1598 0 : })
1599 0 : }
1600 :
1601 : #[derive(Clone, Debug)]
1602 : struct RequestMeta {
1603 : method: hyper::http::Method,
1604 : at: Instant,
1605 : }
1606 :
1607 0 : pub fn prologue_leadership_status_check_middleware<
1608 0 : B: hyper::body::HttpBody + Send + Sync + 'static,
1609 0 : >() -> Middleware<B, ApiError> {
1610 0 : Middleware::pre(move |req| async move {
1611 0 : let state = get_state(&req);
1612 0 : let leadership_status = state.service.get_leadership_status();
1613 :
1614 : enum AllowedRoutes {
1615 : All,
1616 : Some(&'static [&'static str]),
1617 : }
1618 :
1619 0 : let allowed_routes = match leadership_status {
1620 0 : LeadershipStatus::Leader => AllowedRoutes::All,
1621 0 : LeadershipStatus::SteppedDown => AllowedRoutes::All,
1622 0 : LeadershipStatus::Candidate => AllowedRoutes::Some(&[
1623 0 : "/ready",
1624 0 : "/status",
1625 0 : "/metrics",
1626 0 : "/profile/cpu",
1627 0 : "/profile/heap",
1628 0 : ]),
1629 : };
1630 :
1631 0 : match allowed_routes {
1632 0 : AllowedRoutes::All => Ok(req),
1633 0 : AllowedRoutes::Some(allowed) if allowed.contains(&req.uri().path()) => Ok(req),
1634 : _ => {
1635 0 : tracing::info!(
1636 0 : "Request {} not allowed due to current leadership state",
1637 0 : req.uri()
1638 : );
1639 :
1640 0 : Err(ApiError::ResourceUnavailable(
1641 0 : format!("Current leadership status is {leadership_status}").into(),
1642 0 : ))
1643 : }
1644 : }
1645 0 : })
1646 0 : }
1647 :
1648 0 : fn prologue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>()
1649 0 : -> Middleware<B, ApiError> {
1650 0 : Middleware::pre(move |req| async move {
1651 0 : let meta = RequestMeta {
1652 0 : method: req.method().clone(),
1653 0 : at: Instant::now(),
1654 0 : };
1655 0 :
1656 0 : req.set_context(meta);
1657 0 :
1658 0 : Ok(req)
1659 0 : })
1660 0 : }
1661 :
1662 0 : fn epilogue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>()
1663 0 : -> Middleware<B, ApiError> {
1664 0 : Middleware::post_with_info(move |resp, req_info| async move {
1665 0 : let request_name = match req_info.context::<RequestName>() {
1666 0 : Some(name) => name,
1667 : None => {
1668 0 : return Ok(resp);
1669 : }
1670 : };
1671 :
1672 0 : if let Some(meta) = req_info.context::<RequestMeta>() {
1673 0 : let status = &crate::metrics::METRICS_REGISTRY
1674 0 : .metrics_group
1675 0 : .storage_controller_http_request_status;
1676 0 : let latency = &crate::metrics::METRICS_REGISTRY
1677 0 : .metrics_group
1678 0 : .storage_controller_http_request_latency;
1679 0 :
1680 0 : status.inc(HttpRequestStatusLabelGroup {
1681 0 : path: request_name.0,
1682 0 : method: meta.method.clone().into(),
1683 0 : status: crate::metrics::StatusCode(resp.status()),
1684 0 : });
1685 0 :
1686 0 : latency.observe(
1687 0 : HttpRequestLatencyLabelGroup {
1688 0 : path: request_name.0,
1689 0 : method: meta.method.into(),
1690 0 : },
1691 0 : meta.at.elapsed().as_secs_f64(),
1692 0 : );
1693 0 : }
1694 0 : Ok(resp)
1695 0 : })
1696 0 : }
1697 :
1698 0 : pub async fn measured_metrics_handler(req: Request<Body>) -> Result<Response<Body>, ApiError> {
1699 : pub const TEXT_FORMAT: &str = "text/plain; version=0.0.4";
1700 :
1701 0 : let req = match maybe_forward(req).await {
1702 0 : ForwardOutcome::Forwarded(res) => {
1703 0 : return res;
1704 : }
1705 0 : ForwardOutcome::NotForwarded(req) => req,
1706 0 : };
1707 0 :
1708 0 : let state = get_state(&req);
1709 0 : let payload = crate::metrics::METRICS_REGISTRY.encode(&state.neon_metrics);
1710 0 : let response = Response::builder()
1711 0 : .status(200)
1712 0 : .header(CONTENT_TYPE, TEXT_FORMAT)
1713 0 : .body(payload.into())
1714 0 : .unwrap();
1715 0 :
1716 0 : Ok(response)
1717 0 : }
1718 :
1719 : #[derive(Clone)]
1720 : struct RequestName(&'static str);
1721 :
1722 0 : async fn named_request_span<R, H>(
1723 0 : request: Request<Body>,
1724 0 : handler: H,
1725 0 : name: RequestName,
1726 0 : ) -> R::Output
1727 0 : where
1728 0 : R: Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
1729 0 : H: FnOnce(Request<Body>) -> R + Send + Sync + 'static,
1730 0 : {
1731 0 : request.set_context(name);
1732 0 : request_span(request, handler).await
1733 0 : }
1734 :
1735 : enum ForwardOutcome {
1736 : Forwarded(Result<Response<Body>, ApiError>),
1737 : NotForwarded(Request<Body>),
1738 : }
1739 :
1740 : /// Potentially forward the request to the current storage controler leader.
1741 : /// More specifically we forward when:
1742 : /// 1. Request is not one of:
1743 : /// ["/control/v1/step_down", "/status", "/ready", "/metrics", "/profile/cpu", "/profile/heap"]
1744 : /// 2. Current instance is in [`LeadershipStatus::SteppedDown`] state
1745 : /// 3. There is a leader in the database to forward to
1746 : /// 4. Leader from step (3) is not the current instance
1747 : ///
1748 : /// Why forward?
1749 : /// It turns out that we can't rely on external orchestration to promptly route trafic to the
1750 : /// new leader. This is downtime inducing. Forwarding provides a safe way out.
1751 : ///
1752 : /// Why is it safe?
1753 : /// If a storcon instance is persisted in the database, then we know that it is the current leader.
1754 : /// There's one exception: time between handling step-down request and the new leader updating the
1755 : /// database.
1756 : ///
1757 : /// Let's treat the happy case first. The stepped down node does not produce any side effects,
1758 : /// since all request handling happens on the leader.
1759 : ///
1760 : /// As for the edge case, we are guaranteed to always have a maximum of two running instances.
1761 : /// Hence, if we are in the edge case scenario the leader persisted in the database is the
1762 : /// stepped down instance that received the request. Condition (4) above covers this scenario.
1763 0 : async fn maybe_forward(req: Request<Body>) -> ForwardOutcome {
1764 : const NOT_FOR_FORWARD: &[&str] = &[
1765 : "/control/v1/step_down",
1766 : "/status",
1767 : "/live",
1768 : "/ready",
1769 : "/metrics",
1770 : "/profile/cpu",
1771 : "/profile/heap",
1772 : ];
1773 :
1774 0 : let uri = req.uri();
1775 0 : let uri_for_forward = !NOT_FOR_FORWARD.contains(&uri.path());
1776 0 :
1777 0 : // Fast return before trying to take any Service locks, if we will never forward anyway
1778 0 : if !uri_for_forward {
1779 0 : return ForwardOutcome::NotForwarded(req);
1780 0 : }
1781 0 :
1782 0 : let state = get_state(&req);
1783 0 : let leadership_status = state.service.get_leadership_status();
1784 0 :
1785 0 : if leadership_status != LeadershipStatus::SteppedDown {
1786 0 : return ForwardOutcome::NotForwarded(req);
1787 0 : }
1788 :
1789 0 : let leader = state.service.get_leader().await;
1790 0 : let leader = {
1791 0 : match leader {
1792 0 : Ok(Some(leader)) => leader,
1793 : Ok(None) => {
1794 0 : return ForwardOutcome::Forwarded(Err(ApiError::ResourceUnavailable(
1795 0 : "No leader to forward to while in stepped down state".into(),
1796 0 : )));
1797 : }
1798 0 : Err(err) => {
1799 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(
1800 0 : anyhow::anyhow!(
1801 0 : "Failed to get leader for forwarding while in stepped down state: {err}"
1802 0 : ),
1803 0 : )));
1804 : }
1805 : }
1806 : };
1807 :
1808 0 : let cfg = state.service.get_config();
1809 0 : if let Some(ref self_addr) = cfg.address_for_peers {
1810 0 : let leader_addr = match Uri::from_str(leader.address.as_str()) {
1811 0 : Ok(uri) => uri,
1812 0 : Err(err) => {
1813 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(
1814 0 : anyhow::anyhow!(
1815 0 : "Failed to parse leader uri for forwarding while in stepped down state: {err}"
1816 0 : ),
1817 0 : )));
1818 : }
1819 : };
1820 :
1821 0 : if *self_addr == leader_addr {
1822 0 : return ForwardOutcome::Forwarded(Err(ApiError::ResourceUnavailable(
1823 0 : "Leader is stepped down instance".into(),
1824 0 : )));
1825 0 : }
1826 0 : }
1827 :
1828 0 : tracing::info!("Forwarding {} to leader at {}", uri, leader.address);
1829 :
1830 : // Use [`RECONCILE_TIMEOUT`] as the max amount of time a request should block for and
1831 : // include some leeway to get the timeout for proxied requests.
1832 : const PROXIED_REQUEST_TIMEOUT: Duration = Duration::from_secs(RECONCILE_TIMEOUT.as_secs() + 10);
1833 :
1834 0 : let client = state.service.get_http_client().clone();
1835 :
1836 0 : let request: reqwest::Request = match convert_request(
1837 0 : req,
1838 0 : &client,
1839 0 : leader.address,
1840 0 : PROXIED_REQUEST_TIMEOUT,
1841 0 : )
1842 0 : .await
1843 : {
1844 0 : Ok(r) => r,
1845 0 : Err(err) => {
1846 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1847 0 : "Failed to convert request for forwarding while in stepped down state: {err}"
1848 0 : ))));
1849 : }
1850 : };
1851 :
1852 0 : let response = match client.execute(request).await {
1853 0 : Ok(r) => r,
1854 0 : Err(err) => {
1855 0 : return ForwardOutcome::Forwarded(Err(ApiError::InternalServerError(anyhow::anyhow!(
1856 0 : "Failed to forward while in stepped down state: {err}"
1857 0 : ))));
1858 : }
1859 : };
1860 :
1861 0 : ForwardOutcome::Forwarded(convert_response(response).await)
1862 0 : }
1863 :
1864 : /// Convert a [`reqwest::Response`] to a [hyper::Response`] by passing through
1865 : /// a stable representation (string, bytes or integer)
1866 : ///
1867 : /// Ideally, we would not have to do this since both types use the http crate
1868 : /// under the hood. However, they use different versions of the crate and keeping
1869 : /// second order dependencies in sync is difficult.
1870 0 : async fn convert_response(resp: reqwest::Response) -> Result<hyper::Response<Body>, ApiError> {
1871 : use std::str::FromStr;
1872 :
1873 0 : let mut builder = hyper::Response::builder().status(resp.status().as_u16());
1874 0 : for (key, value) in resp.headers().into_iter() {
1875 0 : let key = hyper::header::HeaderName::from_str(key.as_str()).map_err(|err| {
1876 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1877 0 : })?;
1878 :
1879 0 : let value = hyper::header::HeaderValue::from_bytes(value.as_bytes()).map_err(|err| {
1880 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1881 0 : })?;
1882 :
1883 0 : builder = builder.header(key, value);
1884 : }
1885 :
1886 0 : let body = http::Body::wrap_stream(resp.bytes_stream());
1887 0 :
1888 0 : builder.body(body).map_err(|err| {
1889 0 : ApiError::InternalServerError(anyhow::anyhow!("Response conversion failed: {err}"))
1890 0 : })
1891 0 : }
1892 :
1893 : /// Convert a [`reqwest::Request`] to a [hyper::Request`] by passing through
1894 : /// a stable representation (string, bytes or integer)
1895 : ///
1896 : /// See [`convert_response`] for why we are doing it this way.
1897 0 : async fn convert_request(
1898 0 : req: hyper::Request<Body>,
1899 0 : client: &reqwest::Client,
1900 0 : to_address: String,
1901 0 : timeout: Duration,
1902 0 : ) -> Result<reqwest::Request, ApiError> {
1903 : use std::str::FromStr;
1904 :
1905 0 : let (parts, body) = req.into_parts();
1906 0 : let method = reqwest::Method::from_str(parts.method.as_str()).map_err(|err| {
1907 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1908 0 : })?;
1909 :
1910 0 : let path_and_query = parts.uri.path_and_query().ok_or_else(|| {
1911 0 : ApiError::InternalServerError(anyhow::anyhow!(
1912 0 : "Request conversion failed: no path and query"
1913 0 : ))
1914 0 : })?;
1915 :
1916 0 : let uri = reqwest::Url::from_str(
1917 0 : format!(
1918 0 : "{}{}",
1919 0 : to_address.trim_end_matches("/"),
1920 0 : path_and_query.as_str()
1921 0 : )
1922 0 : .as_str(),
1923 0 : )
1924 0 : .map_err(|err| {
1925 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1926 0 : })?;
1927 :
1928 0 : let mut headers = reqwest::header::HeaderMap::new();
1929 0 : for (key, value) in parts.headers.into_iter() {
1930 0 : let key = match key {
1931 0 : Some(k) => k,
1932 : None => {
1933 0 : continue;
1934 : }
1935 : };
1936 :
1937 0 : let key = reqwest::header::HeaderName::from_str(key.as_str()).map_err(|err| {
1938 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1939 0 : })?;
1940 :
1941 0 : let value = reqwest::header::HeaderValue::from_bytes(value.as_bytes()).map_err(|err| {
1942 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1943 0 : })?;
1944 :
1945 0 : headers.insert(key, value);
1946 : }
1947 :
1948 0 : let body = hyper::body::to_bytes(body).await.map_err(|err| {
1949 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1950 0 : })?;
1951 :
1952 0 : client
1953 0 : .request(method, uri)
1954 0 : .headers(headers)
1955 0 : .body(body)
1956 0 : .timeout(timeout)
1957 0 : .build()
1958 0 : .map_err(|err| {
1959 0 : ApiError::InternalServerError(anyhow::anyhow!("Request conversion failed: {err}"))
1960 0 : })
1961 0 : }
1962 :
1963 0 : pub fn make_router(
1964 0 : service: Arc<Service>,
1965 0 : auth: Option<Arc<SwappableJwtAuth>>,
1966 0 : build_info: BuildInfo,
1967 0 : ) -> RouterBuilder<hyper::Body, ApiError> {
1968 0 : let mut router = endpoint::make_router()
1969 0 : .middleware(prologue_leadership_status_check_middleware())
1970 0 : .middleware(prologue_metrics_middleware())
1971 0 : .middleware(epilogue_metrics_middleware());
1972 0 : if auth.is_some() {
1973 0 : router = router.middleware(auth_middleware(|request| {
1974 0 : let state = get_state(request);
1975 0 : if state.allowlist_routes.contains(&request.uri().path()) {
1976 0 : None
1977 : } else {
1978 0 : state.auth.as_deref()
1979 : }
1980 0 : }));
1981 0 : }
1982 :
1983 0 : router
1984 0 : .data(Arc::new(HttpState::new(service, auth, build_info)))
1985 0 : .get("/metrics", |r| {
1986 0 : named_request_span(r, measured_metrics_handler, RequestName("metrics"))
1987 0 : })
1988 0 : // Non-prefixed generic endpoints (status, metrics, profiling)
1989 0 : .get("/status", |r| {
1990 0 : named_request_span(r, handle_status, RequestName("status"))
1991 0 : })
1992 0 : .get("/live", |r| {
1993 0 : named_request_span(r, handle_live, RequestName("live"))
1994 0 : })
1995 0 : .get("/ready", |r| {
1996 0 : named_request_span(r, handle_ready, RequestName("ready"))
1997 0 : })
1998 0 : .get("/profile/cpu", |r| {
1999 0 : named_request_span(r, profile_cpu_handler, RequestName("profile_cpu"))
2000 0 : })
2001 0 : .get("/profile/heap", |r| {
2002 0 : named_request_span(r, profile_heap_handler, RequestName("profile_heap"))
2003 0 : })
2004 0 : // Upcalls for the pageserver: point the pageserver's `control_plane_api` config to this prefix
2005 0 : .post("/upcall/v1/re-attach", |r| {
2006 0 : named_request_span(r, handle_re_attach, RequestName("upcall_v1_reattach"))
2007 0 : })
2008 0 : .post("/upcall/v1/validate", |r| {
2009 0 : named_request_span(r, handle_validate, RequestName("upcall_v1_validate"))
2010 0 : })
2011 0 : .post("/upcall/v1/timeline_import_status", |r| {
2012 0 : named_request_span(
2013 0 : r,
2014 0 : handle_put_timeline_import_status,
2015 0 : RequestName("upcall_v1_timeline_import_status"),
2016 0 : )
2017 0 : })
2018 0 : // Test/dev/debug endpoints
2019 0 : .post("/debug/v1/attach-hook", |r| {
2020 0 : named_request_span(r, handle_attach_hook, RequestName("debug_v1_attach_hook"))
2021 0 : })
2022 0 : .post("/debug/v1/inspect", |r| {
2023 0 : named_request_span(r, handle_inspect, RequestName("debug_v1_inspect"))
2024 0 : })
2025 0 : .post("/debug/v1/tenant/:tenant_id/drop", |r| {
2026 0 : named_request_span(r, handle_tenant_drop, RequestName("debug_v1_tenant_drop"))
2027 0 : })
2028 0 : .post("/debug/v1/node/:node_id/drop", |r| {
2029 0 : named_request_span(r, handle_node_drop, RequestName("debug_v1_node_drop"))
2030 0 : })
2031 0 : .post("/debug/v1/tenant/:tenant_id/import", |r| {
2032 0 : named_request_span(
2033 0 : r,
2034 0 : handle_tenant_import,
2035 0 : RequestName("debug_v1_tenant_import"),
2036 0 : )
2037 0 : })
2038 0 : .get("/debug/v1/tenant", |r| {
2039 0 : named_request_span(r, handle_tenants_dump, RequestName("debug_v1_tenant"))
2040 0 : })
2041 0 : .get("/debug/v1/tenant/:tenant_id/locate", |r| {
2042 0 : tenant_service_handler(
2043 0 : r,
2044 0 : handle_tenant_locate,
2045 0 : RequestName("debug_v1_tenant_locate"),
2046 0 : )
2047 0 : })
2048 0 : .post(
2049 0 : "/debug/v1/tenant/:tenant_id/timeline/:timeline_id/import",
2050 0 : |r| {
2051 0 : named_request_span(
2052 0 : r,
2053 0 : handle_timeline_import,
2054 0 : RequestName("debug_v1_timeline_import"),
2055 0 : )
2056 0 : },
2057 0 : )
2058 0 : .get("/debug/v1/scheduler", |r| {
2059 0 : named_request_span(r, handle_scheduler_dump, RequestName("debug_v1_scheduler"))
2060 0 : })
2061 0 : .post("/debug/v1/consistency_check", |r| {
2062 0 : named_request_span(
2063 0 : r,
2064 0 : handle_consistency_check,
2065 0 : RequestName("debug_v1_consistency_check"),
2066 0 : )
2067 0 : })
2068 0 : .post("/debug/v1/reconcile_all", |r| {
2069 0 : request_span(r, handle_reconcile_all)
2070 0 : })
2071 0 : .put("/debug/v1/failpoints", |r| {
2072 0 : request_span(r, |r| failpoints_handler(r, CancellationToken::new()))
2073 0 : })
2074 0 : // Node operations
2075 0 : .post("/control/v1/node", |r| {
2076 0 : named_request_span(r, handle_node_register, RequestName("control_v1_node"))
2077 0 : })
2078 0 : .delete("/control/v1/node/:node_id", |r| {
2079 0 : named_request_span(r, handle_node_delete, RequestName("control_v1_node_delete"))
2080 0 : })
2081 0 : .get("/control/v1/node", |r| {
2082 0 : named_request_span(r, handle_node_list, RequestName("control_v1_node"))
2083 0 : })
2084 0 : .put("/control/v1/node/:node_id/config", |r| {
2085 0 : named_request_span(
2086 0 : r,
2087 0 : handle_node_configure,
2088 0 : RequestName("control_v1_node_config"),
2089 0 : )
2090 0 : })
2091 0 : .get("/control/v1/node/:node_id", |r| {
2092 0 : named_request_span(r, handle_node_status, RequestName("control_v1_node_status"))
2093 0 : })
2094 0 : .get("/control/v1/node/:node_id/shards", |r| {
2095 0 : named_request_span(
2096 0 : r,
2097 0 : handle_node_shards,
2098 0 : RequestName("control_v1_node_describe"),
2099 0 : )
2100 0 : })
2101 0 : .get("/control/v1/leader", |r| {
2102 0 : named_request_span(r, handle_get_leader, RequestName("control_v1_get_leader"))
2103 0 : })
2104 0 : .put("/control/v1/node/:node_id/drain", |r| {
2105 0 : named_request_span(r, handle_node_drain, RequestName("control_v1_node_drain"))
2106 0 : })
2107 0 : .delete("/control/v1/node/:node_id/drain", |r| {
2108 0 : named_request_span(
2109 0 : r,
2110 0 : handle_cancel_node_drain,
2111 0 : RequestName("control_v1_cancel_node_drain"),
2112 0 : )
2113 0 : })
2114 0 : .put("/control/v1/node/:node_id/fill", |r| {
2115 0 : named_request_span(r, handle_node_fill, RequestName("control_v1_node_fill"))
2116 0 : })
2117 0 : .delete("/control/v1/node/:node_id/fill", |r| {
2118 0 : named_request_span(
2119 0 : r,
2120 0 : handle_cancel_node_fill,
2121 0 : RequestName("control_v1_cancel_node_fill"),
2122 0 : )
2123 0 : })
2124 0 : // Metadata health operations
2125 0 : .post("/control/v1/metadata_health/update", |r| {
2126 0 : named_request_span(
2127 0 : r,
2128 0 : handle_metadata_health_update,
2129 0 : RequestName("control_v1_metadata_health_update"),
2130 0 : )
2131 0 : })
2132 0 : .get("/control/v1/metadata_health/unhealthy", |r| {
2133 0 : named_request_span(
2134 0 : r,
2135 0 : handle_metadata_health_list_unhealthy,
2136 0 : RequestName("control_v1_metadata_health_list_unhealthy"),
2137 0 : )
2138 0 : })
2139 0 : .post("/control/v1/metadata_health/outdated", |r| {
2140 0 : named_request_span(
2141 0 : r,
2142 0 : handle_metadata_health_list_outdated,
2143 0 : RequestName("control_v1_metadata_health_list_outdated"),
2144 0 : )
2145 0 : })
2146 0 : // Safekeepers
2147 0 : .get("/control/v1/safekeeper", |r| {
2148 0 : named_request_span(
2149 0 : r,
2150 0 : handle_safekeeper_list,
2151 0 : RequestName("control_v1_safekeeper_list"),
2152 0 : )
2153 0 : })
2154 0 : .get("/control/v1/safekeeper/:id", |r| {
2155 0 : named_request_span(r, handle_get_safekeeper, RequestName("v1_safekeeper"))
2156 0 : })
2157 0 : .post("/control/v1/safekeeper/:id", |r| {
2158 0 : // id is in the body
2159 0 : named_request_span(
2160 0 : r,
2161 0 : handle_upsert_safekeeper,
2162 0 : RequestName("v1_safekeeper_post"),
2163 0 : )
2164 0 : })
2165 0 : .post("/control/v1/safekeeper/:id/scheduling_policy", |r| {
2166 0 : named_request_span(
2167 0 : r,
2168 0 : handle_safekeeper_scheduling_policy,
2169 0 : RequestName("v1_safekeeper_status"),
2170 0 : )
2171 0 : })
2172 0 : // Tenant Shard operations
2173 0 : .put("/control/v1/tenant/:tenant_shard_id/migrate", |r| {
2174 0 : tenant_service_handler(
2175 0 : r,
2176 0 : handle_tenant_shard_migrate,
2177 0 : RequestName("control_v1_tenant_migrate"),
2178 0 : )
2179 0 : })
2180 0 : .put(
2181 0 : "/control/v1/tenant/:tenant_shard_id/migrate_secondary",
2182 0 : |r| {
2183 0 : tenant_service_handler(
2184 0 : r,
2185 0 : handle_tenant_shard_migrate_secondary,
2186 0 : RequestName("control_v1_tenant_migrate_secondary"),
2187 0 : )
2188 0 : },
2189 0 : )
2190 0 : .put(
2191 0 : "/control/v1/tenant/:tenant_shard_id/cancel_reconcile",
2192 0 : |r| {
2193 0 : tenant_service_handler(
2194 0 : r,
2195 0 : handle_tenant_shard_cancel_reconcile,
2196 0 : RequestName("control_v1_tenant_cancel_reconcile"),
2197 0 : )
2198 0 : },
2199 0 : )
2200 0 : .put("/control/v1/tenant/:tenant_id/shard_split", |r| {
2201 0 : tenant_service_handler(
2202 0 : r,
2203 0 : handle_tenant_shard_split,
2204 0 : RequestName("control_v1_tenant_shard_split"),
2205 0 : )
2206 0 : })
2207 0 : .get("/control/v1/tenant/:tenant_id", |r| {
2208 0 : tenant_service_handler(
2209 0 : r,
2210 0 : handle_tenant_describe,
2211 0 : RequestName("control_v1_tenant_describe"),
2212 0 : )
2213 0 : })
2214 0 : .get("/control/v1/tenant", |r| {
2215 0 : tenant_service_handler(r, handle_tenant_list, RequestName("control_v1_tenant_list"))
2216 0 : })
2217 0 : .put("/control/v1/tenant/:tenant_id/policy", |r| {
2218 0 : named_request_span(
2219 0 : r,
2220 0 : handle_tenant_update_policy,
2221 0 : RequestName("control_v1_tenant_policy"),
2222 0 : )
2223 0 : })
2224 0 : .put("/control/v1/preferred_azs", |r| {
2225 0 : named_request_span(
2226 0 : r,
2227 0 : handle_update_preferred_azs,
2228 0 : RequestName("control_v1_preferred_azs"),
2229 0 : )
2230 0 : })
2231 0 : .put("/control/v1/step_down", |r| {
2232 0 : named_request_span(r, handle_step_down, RequestName("control_v1_step_down"))
2233 0 : })
2234 0 : // Tenant operations
2235 0 : // The ^/v1/ endpoints act as a "Virtual Pageserver", enabling shard-naive clients to call into
2236 0 : // this service to manage tenants that actually consist of many tenant shards, as if they are a single entity.
2237 0 : .post("/v1/tenant", |r| {
2238 0 : tenant_service_handler(r, handle_tenant_create, RequestName("v1_tenant"))
2239 0 : })
2240 0 : .delete("/v1/tenant/:tenant_id", |r| {
2241 0 : tenant_service_handler(r, handle_tenant_delete, RequestName("v1_tenant"))
2242 0 : })
2243 0 : .patch("/v1/tenant/config", |r| {
2244 0 : tenant_service_handler(
2245 0 : r,
2246 0 : handle_tenant_config_patch,
2247 0 : RequestName("v1_tenant_config"),
2248 0 : )
2249 0 : })
2250 0 : .put("/v1/tenant/config", |r| {
2251 0 : tenant_service_handler(r, handle_tenant_config_set, RequestName("v1_tenant_config"))
2252 0 : })
2253 0 : .get("/v1/tenant/:tenant_id/config", |r| {
2254 0 : tenant_service_handler(r, handle_tenant_config_get, RequestName("v1_tenant_config"))
2255 0 : })
2256 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2257 0 : tenant_service_handler(
2258 0 : r,
2259 0 : handle_tenant_location_config,
2260 0 : RequestName("v1_tenant_location_config"),
2261 0 : )
2262 0 : })
2263 0 : .put("/v1/tenant/:tenant_id/time_travel_remote_storage", |r| {
2264 0 : tenant_service_handler(
2265 0 : r,
2266 0 : handle_tenant_time_travel_remote_storage,
2267 0 : RequestName("v1_tenant_time_travel_remote_storage"),
2268 0 : )
2269 0 : })
2270 0 : .post("/v1/tenant/:tenant_id/secondary/download", |r| {
2271 0 : tenant_service_handler(
2272 0 : r,
2273 0 : handle_tenant_secondary_download,
2274 0 : RequestName("v1_tenant_secondary_download"),
2275 0 : )
2276 0 : })
2277 0 : // Timeline operations
2278 0 : .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
2279 0 : tenant_service_handler(
2280 0 : r,
2281 0 : handle_tenant_timeline_delete,
2282 0 : RequestName("v1_tenant_timeline"),
2283 0 : )
2284 0 : })
2285 0 : .post("/v1/tenant/:tenant_id/timeline", |r| {
2286 0 : tenant_service_handler(
2287 0 : r,
2288 0 : handle_tenant_timeline_create,
2289 0 : RequestName("v1_tenant_timeline"),
2290 0 : )
2291 0 : })
2292 0 : .put(
2293 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/archival_config",
2294 0 : |r| {
2295 0 : tenant_service_handler(
2296 0 : r,
2297 0 : handle_tenant_timeline_archival_config,
2298 0 : RequestName("v1_tenant_timeline_archival_config"),
2299 0 : )
2300 0 : },
2301 0 : )
2302 0 : .put(
2303 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/detach_ancestor",
2304 0 : |r| {
2305 0 : tenant_service_handler(
2306 0 : r,
2307 0 : handle_tenant_timeline_detach_ancestor,
2308 0 : RequestName("v1_tenant_timeline_detach_ancestor"),
2309 0 : )
2310 0 : },
2311 0 : )
2312 0 : .post(
2313 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/block_gc",
2314 0 : |r| {
2315 0 : tenant_service_handler(
2316 0 : r,
2317 0 : |s, r| handle_tenant_timeline_block_unblock_gc(s, r, BlockUnblock::Block),
2318 0 : RequestName("v1_tenant_timeline_block_unblock_gc"),
2319 0 : )
2320 0 : },
2321 0 : )
2322 0 : .post(
2323 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/unblock_gc",
2324 0 : |r| {
2325 0 : tenant_service_handler(
2326 0 : r,
2327 0 : |s, r| handle_tenant_timeline_block_unblock_gc(s, r, BlockUnblock::Unblock),
2328 0 : RequestName("v1_tenant_timeline_block_unblock_gc"),
2329 0 : )
2330 0 : },
2331 0 : )
2332 0 : .post(
2333 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
2334 0 : |r| {
2335 0 : tenant_service_handler(
2336 0 : r,
2337 0 : handle_tenant_timeline_download_heatmap_layers,
2338 0 : RequestName("v1_tenant_timeline_download_heatmap_layers"),
2339 0 : )
2340 0 : },
2341 0 : )
2342 0 : // LSN lease passthrough to all shards
2343 0 : .post(
2344 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/lsn_lease",
2345 0 : |r| {
2346 0 : tenant_service_handler(
2347 0 : r,
2348 0 : handle_tenant_timeline_lsn_lease,
2349 0 : RequestName("v1_tenant_timeline_lsn_lease"),
2350 0 : )
2351 0 : },
2352 0 : )
2353 0 : // Tenant detail GET passthrough to shard zero:
2354 0 : .get("/v1/tenant/:tenant_id", |r| {
2355 0 : tenant_service_handler(
2356 0 : r,
2357 0 : handle_tenant_timeline_passthrough,
2358 0 : RequestName("v1_tenant_passthrough"),
2359 0 : )
2360 0 : })
2361 0 : // The `*` in the URL is a wildcard: any tenant/timeline GET APIs on the pageserver
2362 0 : // are implicitly exposed here. This must be last in the list to avoid
2363 0 : // taking precedence over other GET methods we might implement by hand.
2364 0 : .get("/v1/tenant/:tenant_id/*", |r| {
2365 0 : tenant_service_handler(
2366 0 : r,
2367 0 : handle_tenant_timeline_passthrough,
2368 0 : RequestName("v1_tenant_passthrough"),
2369 0 : )
2370 0 : })
2371 0 : // Tenant timeline mark_invisible passthrough to shard zero
2372 0 : .put(
2373 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/mark_invisible",
2374 0 : |r| {
2375 0 : tenant_service_handler(
2376 0 : r,
2377 0 : handle_tenant_timeline_passthrough,
2378 0 : RequestName("v1_tenant_timeline_mark_invisible_passthrough"),
2379 0 : )
2380 0 : },
2381 0 : )
2382 0 : }
2383 :
2384 : #[cfg(test)]
2385 : mod test {
2386 :
2387 : use super::path_without_ids;
2388 :
2389 : #[test]
2390 1 : fn test_path_without_ids() {
2391 1 : assert_eq!(
2392 1 : path_without_ids(
2393 1 : "/v1/tenant/1a2b3344556677881122334455667788/timeline/AA223344556677881122334455667788"
2394 1 : ),
2395 1 : "/v1/tenant//timeline/"
2396 1 : );
2397 1 : assert_eq!(
2398 1 : path_without_ids(
2399 1 : "/v1/tenant/1a2b3344556677881122334455667788-0108/timeline/AA223344556677881122334455667788"
2400 1 : ),
2401 1 : "/v1/tenant//timeline/"
2402 1 : );
2403 1 : assert_eq!(
2404 1 : path_without_ids(
2405 1 : "/v1/tenant/1a2b3344556677881122334455667788-0108/timeline/AA223344556677881122334455667788?parameter=foo"
2406 1 : ),
2407 1 : "/v1/tenant//timeline/"
2408 1 : );
2409 1 : }
2410 : }
|
