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