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