Line data Source code
1 : use crate::metrics::{
2 : HttpRequestLatencyLabelGroup, HttpRequestStatusLabelGroup, PageserverRequestLabelGroup,
3 : METRICS_REGISTRY,
4 : };
5 : use crate::reconciler::ReconcileError;
6 : use crate::service::{Service, STARTUP_RECONCILE_TIMEOUT};
7 : use futures::Future;
8 : use hyper::header::CONTENT_TYPE;
9 : use hyper::{Body, Request, Response};
10 : use hyper::{StatusCode, Uri};
11 : use metrics::{BuildInfo, NeonMetrics};
12 : use pageserver_api::models::{
13 : TenantConfigRequest, TenantCreateRequest, TenantLocationConfigRequest, TenantShardSplitRequest,
14 : TenantTimeTravelRequest, TimelineCreateRequest,
15 : };
16 : use pageserver_api::shard::TenantShardId;
17 : use pageserver_client::mgmt_api;
18 : use std::sync::Arc;
19 : use std::time::{Duration, Instant};
20 : use tokio_util::sync::CancellationToken;
21 : use utils::auth::{Scope, SwappableJwtAuth};
22 : use utils::failpoint_support::failpoints_handler;
23 : use utils::http::endpoint::{auth_middleware, check_permission_with, request_span};
24 : use utils::http::request::{must_get_query_param, parse_query_param, parse_request_param};
25 : use utils::id::{TenantId, TimelineId};
26 :
27 : use utils::{
28 : http::{
29 : endpoint::{self},
30 : error::ApiError,
31 : json::{json_request, json_response},
32 : RequestExt, RouterBuilder,
33 : },
34 : id::NodeId,
35 : };
36 :
37 : use pageserver_api::controller_api::{
38 : NodeAvailability, NodeConfigureRequest, NodeRegisterRequest, TenantPolicyRequest,
39 : TenantShardMigrateRequest,
40 : };
41 : use pageserver_api::upcall_api::{ReAttachRequest, ValidateRequest};
42 :
43 : use control_plane::storage_controller::{AttachHookRequest, InspectRequest};
44 :
45 : use routerify::Middleware;
46 :
47 : /// State available to HTTP request handlers
48 : pub struct HttpState {
49 : service: Arc<crate::service::Service>,
50 : auth: Option<Arc<SwappableJwtAuth>>,
51 : neon_metrics: NeonMetrics,
52 : allowlist_routes: Vec<Uri>,
53 : }
54 :
55 : impl HttpState {
56 0 : pub fn new(
57 0 : service: Arc<crate::service::Service>,
58 0 : auth: Option<Arc<SwappableJwtAuth>>,
59 0 : build_info: BuildInfo,
60 0 : ) -> Self {
61 0 : let allowlist_routes = ["/status", "/ready", "/metrics"]
62 0 : .iter()
63 0 : .map(|v| v.parse().unwrap())
64 0 : .collect::<Vec<_>>();
65 0 : Self {
66 0 : service,
67 0 : auth,
68 0 : neon_metrics: NeonMetrics::new(build_info),
69 0 : allowlist_routes,
70 0 : }
71 0 : }
72 : }
73 :
74 : #[inline(always)]
75 0 : fn get_state(request: &Request<Body>) -> &HttpState {
76 0 : request
77 0 : .data::<Arc<HttpState>>()
78 0 : .expect("unknown state type")
79 0 : .as_ref()
80 0 : }
81 :
82 : /// Pageserver calls into this on startup, to learn which tenants it should attach
83 0 : async fn handle_re_attach(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
84 0 : check_permissions(&req, Scope::GenerationsApi)?;
85 :
86 0 : let reattach_req = json_request::<ReAttachRequest>(&mut req).await?;
87 0 : let state = get_state(&req);
88 0 : json_response(StatusCode::OK, state.service.re_attach(reattach_req).await?)
89 0 : }
90 :
91 : /// Pageserver calls into this before doing deletions, to confirm that it still
92 : /// holds the latest generation for the tenants with deletions enqueued
93 0 : async fn handle_validate(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
94 0 : check_permissions(&req, Scope::GenerationsApi)?;
95 :
96 0 : let validate_req = json_request::<ValidateRequest>(&mut req).await?;
97 0 : let state = get_state(&req);
98 0 : json_response(StatusCode::OK, state.service.validate(validate_req))
99 0 : }
100 :
101 : /// Call into this before attaching a tenant to a pageserver, to acquire a generation number
102 : /// (in the real control plane this is unnecessary, because the same program is managing
103 : /// generation numbers and doing attachments).
104 0 : async fn handle_attach_hook(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
105 0 : check_permissions(&req, Scope::Admin)?;
106 :
107 0 : let attach_req = json_request::<AttachHookRequest>(&mut req).await?;
108 0 : let state = get_state(&req);
109 0 :
110 0 : json_response(
111 0 : StatusCode::OK,
112 0 : state
113 0 : .service
114 0 : .attach_hook(attach_req)
115 0 : .await
116 0 : .map_err(ApiError::InternalServerError)?,
117 : )
118 0 : }
119 :
120 0 : async fn handle_inspect(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
121 0 : check_permissions(&req, Scope::Admin)?;
122 :
123 0 : let inspect_req = json_request::<InspectRequest>(&mut req).await?;
124 :
125 0 : let state = get_state(&req);
126 0 :
127 0 : json_response(StatusCode::OK, state.service.inspect(inspect_req))
128 0 : }
129 :
130 0 : async fn handle_tenant_create(
131 0 : service: Arc<Service>,
132 0 : mut req: Request<Body>,
133 0 : ) -> Result<Response<Body>, ApiError> {
134 0 : check_permissions(&req, Scope::PageServerApi)?;
135 :
136 0 : let create_req = json_request::<TenantCreateRequest>(&mut req).await?;
137 :
138 : json_response(
139 : StatusCode::CREATED,
140 0 : service.tenant_create(create_req).await?,
141 : )
142 0 : }
143 :
144 : // For tenant and timeline deletions, which both implement an "initially return 202, then 404 once
145 : // we're done" semantic, we wrap with a retry loop to expose a simpler API upstream. This avoids
146 : // needing to track a "deleting" state for tenants.
147 0 : async fn deletion_wrapper<R, F>(service: Arc<Service>, f: F) -> Result<Response<Body>, ApiError>
148 0 : where
149 0 : R: std::future::Future<Output = Result<StatusCode, ApiError>> + Send + 'static,
150 0 : F: Fn(Arc<Service>) -> R + Send + Sync + 'static,
151 0 : {
152 0 : let started_at = Instant::now();
153 0 : // To keep deletion reasonably snappy for small tenants, initially check after 1 second if deletion
154 0 : // completed.
155 0 : let mut retry_period = Duration::from_secs(1);
156 0 : // On subsequent retries, wait longer.
157 0 : let max_retry_period = Duration::from_secs(5);
158 0 : // Enable callers with a 30 second request timeout to reliably get a response
159 0 : let max_wait = Duration::from_secs(25);
160 :
161 : loop {
162 0 : let status = f(service.clone()).await?;
163 0 : match status {
164 : StatusCode::ACCEPTED => {
165 0 : tracing::info!("Deletion accepted, waiting to try again...");
166 0 : tokio::time::sleep(retry_period).await;
167 0 : retry_period = max_retry_period;
168 : }
169 : StatusCode::NOT_FOUND => {
170 0 : tracing::info!("Deletion complete");
171 0 : return json_response(StatusCode::OK, ());
172 : }
173 : _ => {
174 0 : tracing::warn!("Unexpected status {status}");
175 0 : return json_response(status, ());
176 : }
177 : }
178 :
179 0 : let now = Instant::now();
180 0 : if now + retry_period > started_at + max_wait {
181 0 : tracing::info!("Deletion timed out waiting for 404");
182 : // REQUEST_TIMEOUT would be more appropriate, but CONFLICT is already part of
183 : // the pageserver's swagger definition for this endpoint, and has the same desired
184 : // effect of causing the control plane to retry later.
185 0 : return json_response(StatusCode::CONFLICT, ());
186 0 : }
187 : }
188 0 : }
189 :
190 0 : async fn handle_tenant_location_config(
191 0 : service: Arc<Service>,
192 0 : mut req: Request<Body>,
193 0 : ) -> Result<Response<Body>, ApiError> {
194 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
195 0 : check_permissions(&req, Scope::PageServerApi)?;
196 :
197 0 : let config_req = json_request::<TenantLocationConfigRequest>(&mut req).await?;
198 : json_response(
199 : StatusCode::OK,
200 0 : service
201 0 : .tenant_location_config(tenant_shard_id, config_req)
202 0 : .await?,
203 : )
204 0 : }
205 :
206 0 : async fn handle_tenant_config_set(
207 0 : service: Arc<Service>,
208 0 : mut req: Request<Body>,
209 0 : ) -> Result<Response<Body>, ApiError> {
210 0 : check_permissions(&req, Scope::PageServerApi)?;
211 :
212 0 : let config_req = json_request::<TenantConfigRequest>(&mut req).await?;
213 :
214 0 : json_response(StatusCode::OK, service.tenant_config_set(config_req).await?)
215 0 : }
216 :
217 0 : async fn handle_tenant_config_get(
218 0 : service: Arc<Service>,
219 0 : req: Request<Body>,
220 0 : ) -> Result<Response<Body>, ApiError> {
221 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
222 0 : check_permissions(&req, Scope::PageServerApi)?;
223 :
224 0 : json_response(StatusCode::OK, service.tenant_config_get(tenant_id)?)
225 0 : }
226 :
227 0 : async fn handle_tenant_time_travel_remote_storage(
228 0 : service: Arc<Service>,
229 0 : mut req: Request<Body>,
230 0 : ) -> Result<Response<Body>, ApiError> {
231 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
232 0 : check_permissions(&req, Scope::PageServerApi)?;
233 :
234 0 : let time_travel_req = json_request::<TenantTimeTravelRequest>(&mut req).await?;
235 :
236 0 : let timestamp_raw = must_get_query_param(&req, "travel_to")?;
237 0 : let _timestamp = humantime::parse_rfc3339(×tamp_raw).map_err(|_e| {
238 0 : ApiError::BadRequest(anyhow::anyhow!(
239 0 : "Invalid time for travel_to: {timestamp_raw:?}"
240 0 : ))
241 0 : })?;
242 :
243 0 : let done_if_after_raw = must_get_query_param(&req, "done_if_after")?;
244 0 : let _done_if_after = humantime::parse_rfc3339(&done_if_after_raw).map_err(|_e| {
245 0 : ApiError::BadRequest(anyhow::anyhow!(
246 0 : "Invalid time for done_if_after: {done_if_after_raw:?}"
247 0 : ))
248 0 : })?;
249 :
250 0 : service
251 0 : .tenant_time_travel_remote_storage(
252 0 : &time_travel_req,
253 0 : tenant_id,
254 0 : timestamp_raw,
255 0 : done_if_after_raw,
256 0 : )
257 0 : .await?;
258 0 : json_response(StatusCode::OK, ())
259 0 : }
260 :
261 0 : async fn handle_tenant_secondary_download(
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 : let wait = parse_query_param(&req, "wait_ms")?.map(Duration::from_millis);
267 :
268 0 : let (status, progress) = service.tenant_secondary_download(tenant_id, wait).await?;
269 0 : json_response(status, progress)
270 0 : }
271 :
272 0 : async fn handle_tenant_delete(
273 0 : service: Arc<Service>,
274 0 : req: Request<Body>,
275 0 : ) -> Result<Response<Body>, ApiError> {
276 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
277 0 : check_permissions(&req, Scope::PageServerApi)?;
278 :
279 0 : deletion_wrapper(service, move |service| async move {
280 0 : service.tenant_delete(tenant_id).await
281 0 : })
282 0 : .await
283 0 : }
284 :
285 0 : async fn handle_tenant_timeline_create(
286 0 : service: Arc<Service>,
287 0 : mut req: Request<Body>,
288 0 : ) -> Result<Response<Body>, ApiError> {
289 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
290 0 : check_permissions(&req, Scope::PageServerApi)?;
291 :
292 0 : let create_req = json_request::<TimelineCreateRequest>(&mut req).await?;
293 : json_response(
294 : StatusCode::CREATED,
295 0 : service
296 0 : .tenant_timeline_create(tenant_id, create_req)
297 0 : .await?,
298 : )
299 0 : }
300 :
301 0 : async fn handle_tenant_timeline_delete(
302 0 : service: Arc<Service>,
303 0 : req: Request<Body>,
304 0 : ) -> Result<Response<Body>, ApiError> {
305 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
306 0 : check_permissions(&req, Scope::PageServerApi)?;
307 :
308 0 : let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
309 :
310 0 : deletion_wrapper(service, move |service| async move {
311 0 : service.tenant_timeline_delete(tenant_id, timeline_id).await
312 0 : })
313 0 : .await
314 0 : }
315 :
316 0 : async fn handle_tenant_timeline_passthrough(
317 0 : service: Arc<Service>,
318 0 : req: Request<Body>,
319 0 : ) -> Result<Response<Body>, ApiError> {
320 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
321 0 : check_permissions(&req, Scope::PageServerApi)?;
322 :
323 0 : let Some(path) = req.uri().path_and_query() else {
324 : // This should never happen, our request router only calls us if there is a path
325 0 : return Err(ApiError::BadRequest(anyhow::anyhow!("Missing path")));
326 : };
327 :
328 0 : tracing::info!("Proxying request for tenant {} ({})", tenant_id, path);
329 :
330 : // Find the node that holds shard zero
331 0 : let (node, tenant_shard_id) = service.tenant_shard0_node(tenant_id)?;
332 :
333 : // Callers will always pass an unsharded tenant ID. Before proxying, we must
334 : // rewrite this to a shard-aware shard zero ID.
335 0 : let path = format!("{}", path);
336 0 : let tenant_str = tenant_id.to_string();
337 0 : let tenant_shard_str = format!("{}", tenant_shard_id);
338 0 : let path = path.replace(&tenant_str, &tenant_shard_str);
339 0 :
340 0 : let latency = &METRICS_REGISTRY
341 0 : .metrics_group
342 0 : .storage_controller_passthrough_request_latency;
343 0 :
344 0 : // This is a bit awkward. We remove the param from the request
345 0 : // and join the words by '_' to get a label for the request.
346 0 : let just_path = path.replace(&tenant_shard_str, "");
347 0 : let path_label = just_path
348 0 : .split('/')
349 0 : .filter(|token| !token.is_empty())
350 0 : .collect::<Vec<_>>()
351 0 : .join("_");
352 0 : let labels = PageserverRequestLabelGroup {
353 0 : pageserver_id: &node.get_id().to_string(),
354 0 : path: &path_label,
355 0 : method: crate::metrics::Method::Get,
356 0 : };
357 0 :
358 0 : let _timer = latency.start_timer(labels.clone());
359 0 :
360 0 : let client = mgmt_api::Client::new(node.base_url(), service.get_config().jwt_token.as_deref());
361 0 : let resp = client.get_raw(path).await.map_err(|_e|
362 : // FIXME: give APiError a proper Unavailable variant. We return 503 here because
363 : // if we can't successfully send a request to the pageserver, we aren't available.
364 0 : ApiError::ShuttingDown)?;
365 :
366 0 : if !resp.status().is_success() {
367 0 : let error_counter = &METRICS_REGISTRY
368 0 : .metrics_group
369 0 : .storage_controller_passthrough_request_error;
370 0 : error_counter.inc(labels);
371 0 : }
372 :
373 : // We have a reqest::Response, would like a http::Response
374 0 : let mut builder = hyper::Response::builder()
375 0 : .status(resp.status())
376 0 : .version(resp.version());
377 0 : for (k, v) in resp.headers() {
378 0 : builder = builder.header(k, v);
379 0 : }
380 :
381 0 : let response = builder
382 0 : .body(Body::wrap_stream(resp.bytes_stream()))
383 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
384 :
385 0 : Ok(response)
386 0 : }
387 :
388 0 : async fn handle_tenant_locate(
389 0 : service: Arc<Service>,
390 0 : req: Request<Body>,
391 0 : ) -> Result<Response<Body>, ApiError> {
392 0 : check_permissions(&req, Scope::Admin)?;
393 :
394 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
395 0 : json_response(StatusCode::OK, service.tenant_locate(tenant_id)?)
396 0 : }
397 :
398 0 : async fn handle_tenant_describe(
399 0 : service: Arc<Service>,
400 0 : req: Request<Body>,
401 0 : ) -> Result<Response<Body>, ApiError> {
402 0 : check_permissions(&req, Scope::Admin)?;
403 :
404 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
405 0 : json_response(StatusCode::OK, service.tenant_describe(tenant_id)?)
406 0 : }
407 :
408 0 : async fn handle_tenant_list(
409 0 : service: Arc<Service>,
410 0 : req: Request<Body>,
411 0 : ) -> Result<Response<Body>, ApiError> {
412 0 : check_permissions(&req, Scope::Admin)?;
413 :
414 0 : json_response(StatusCode::OK, service.tenant_list())
415 0 : }
416 :
417 0 : async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
418 0 : check_permissions(&req, Scope::Admin)?;
419 :
420 0 : let register_req = json_request::<NodeRegisterRequest>(&mut req).await?;
421 0 : let state = get_state(&req);
422 0 : state.service.node_register(register_req).await?;
423 0 : json_response(StatusCode::OK, ())
424 0 : }
425 :
426 0 : async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
427 0 : check_permissions(&req, Scope::Admin)?;
428 :
429 0 : let state = get_state(&req);
430 0 : let nodes = state.service.node_list().await?;
431 0 : let api_nodes = nodes.into_iter().map(|n| n.describe()).collect::<Vec<_>>();
432 0 :
433 0 : json_response(StatusCode::OK, api_nodes)
434 0 : }
435 :
436 0 : async fn handle_node_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
437 0 : check_permissions(&req, Scope::Admin)?;
438 :
439 0 : let state = get_state(&req);
440 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
441 0 : json_response(StatusCode::OK, state.service.node_drop(node_id).await?)
442 0 : }
443 :
444 0 : async fn handle_node_configure(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
445 0 : check_permissions(&req, Scope::Admin)?;
446 :
447 0 : let node_id: NodeId = parse_request_param(&req, "node_id")?;
448 0 : let config_req = json_request::<NodeConfigureRequest>(&mut req).await?;
449 0 : if node_id != config_req.node_id {
450 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
451 0 : "Path and body node_id differ"
452 0 : )));
453 0 : }
454 0 : let state = get_state(&req);
455 0 :
456 0 : json_response(
457 0 : StatusCode::OK,
458 0 : state
459 0 : .service
460 0 : .node_configure(
461 0 : config_req.node_id,
462 0 : config_req.availability.map(NodeAvailability::from),
463 0 : config_req.scheduling,
464 0 : )
465 0 : .await?,
466 : )
467 0 : }
468 :
469 0 : async fn handle_tenant_shard_split(
470 0 : service: Arc<Service>,
471 0 : mut req: Request<Body>,
472 0 : ) -> Result<Response<Body>, ApiError> {
473 0 : check_permissions(&req, Scope::Admin)?;
474 :
475 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
476 0 : let split_req = json_request::<TenantShardSplitRequest>(&mut req).await?;
477 :
478 : json_response(
479 : StatusCode::OK,
480 0 : service.tenant_shard_split(tenant_id, split_req).await?,
481 : )
482 0 : }
483 :
484 0 : async fn handle_tenant_shard_migrate(
485 0 : service: Arc<Service>,
486 0 : mut req: Request<Body>,
487 0 : ) -> Result<Response<Body>, ApiError> {
488 0 : check_permissions(&req, Scope::Admin)?;
489 :
490 0 : let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
491 0 : let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
492 : json_response(
493 : StatusCode::OK,
494 0 : service
495 0 : .tenant_shard_migrate(tenant_shard_id, migrate_req)
496 0 : .await?,
497 : )
498 0 : }
499 :
500 0 : async fn handle_tenant_update_policy(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
501 0 : check_permissions(&req, Scope::Admin)?;
502 :
503 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
504 0 : let update_req = json_request::<TenantPolicyRequest>(&mut req).await?;
505 0 : let state = get_state(&req);
506 0 :
507 0 : json_response(
508 0 : StatusCode::OK,
509 0 : state
510 0 : .service
511 0 : .tenant_update_policy(tenant_id, update_req)
512 0 : .await?,
513 : )
514 0 : }
515 :
516 0 : async fn handle_tenant_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
517 0 : let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
518 0 : check_permissions(&req, Scope::PageServerApi)?;
519 :
520 0 : let state = get_state(&req);
521 0 :
522 0 : json_response(StatusCode::OK, state.service.tenant_drop(tenant_id).await?)
523 0 : }
524 :
525 0 : async fn handle_tenants_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
526 0 : check_permissions(&req, Scope::Admin)?;
527 :
528 0 : let state = get_state(&req);
529 0 : state.service.tenants_dump()
530 0 : }
531 :
532 0 : async fn handle_scheduler_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
533 0 : check_permissions(&req, Scope::Admin)?;
534 :
535 0 : let state = get_state(&req);
536 0 : state.service.scheduler_dump()
537 0 : }
538 :
539 0 : async fn handle_consistency_check(req: Request<Body>) -> Result<Response<Body>, ApiError> {
540 0 : check_permissions(&req, Scope::Admin)?;
541 :
542 0 : let state = get_state(&req);
543 0 :
544 0 : json_response(StatusCode::OK, state.service.consistency_check().await?)
545 0 : }
546 :
547 0 : async fn handle_reconcile_all(req: Request<Body>) -> Result<Response<Body>, ApiError> {
548 0 : check_permissions(&req, Scope::Admin)?;
549 :
550 0 : let state = get_state(&req);
551 0 :
552 0 : json_response(StatusCode::OK, state.service.reconcile_all_now().await?)
553 0 : }
554 :
555 : /// Status endpoint is just used for checking that our HTTP listener is up
556 0 : async fn handle_status(_req: Request<Body>) -> Result<Response<Body>, ApiError> {
557 0 : json_response(StatusCode::OK, ())
558 0 : }
559 :
560 : /// Readiness endpoint indicates when we're done doing startup I/O (e.g. reconciling
561 : /// with remote pageserver nodes). This is intended for use as a kubernetes readiness probe.
562 0 : async fn handle_ready(req: Request<Body>) -> Result<Response<Body>, ApiError> {
563 0 : let state = get_state(&req);
564 0 : if state.service.startup_complete.is_ready() {
565 0 : json_response(StatusCode::OK, ())
566 : } else {
567 0 : json_response(StatusCode::SERVICE_UNAVAILABLE, ())
568 : }
569 0 : }
570 :
571 : impl From<ReconcileError> for ApiError {
572 0 : fn from(value: ReconcileError) -> Self {
573 0 : ApiError::Conflict(format!("Reconciliation error: {}", value))
574 0 : }
575 : }
576 :
577 : /// Common wrapper for request handlers that call into Service and will operate on tenants: they must only
578 : /// be allowed to run if Service has finished its initial reconciliation.
579 0 : async fn tenant_service_handler<R, H>(
580 0 : request: Request<Body>,
581 0 : handler: H,
582 0 : request_name: RequestName,
583 0 : ) -> R::Output
584 0 : where
585 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
586 0 : H: FnOnce(Arc<Service>, Request<Body>) -> R + Send + Sync + 'static,
587 0 : {
588 0 : let state = get_state(&request);
589 0 : let service = state.service.clone();
590 0 :
591 0 : let startup_complete = service.startup_complete.clone();
592 0 : if tokio::time::timeout(STARTUP_RECONCILE_TIMEOUT, startup_complete.wait())
593 0 : .await
594 0 : .is_err()
595 : {
596 : // This shouldn't happen: it is the responsibilty of [`Service::startup_reconcile`] to use appropriate
597 : // timeouts around its remote calls, to bound its runtime.
598 0 : return Err(ApiError::Timeout(
599 0 : "Timed out waiting for service readiness".into(),
600 0 : ));
601 0 : }
602 0 :
603 0 : named_request_span(
604 0 : request,
605 0 : |request| async move { handler(service, request).await },
606 0 : request_name,
607 0 : )
608 0 : .await
609 0 : }
610 :
611 : /// Check if the required scope is held in the request's token, or if the request has
612 : /// a token with 'admin' scope then always permit it.
613 0 : fn check_permissions(request: &Request<Body>, required_scope: Scope) -> Result<(), ApiError> {
614 0 : check_permission_with(request, |claims| {
615 0 : match crate::auth::check_permission(claims, required_scope) {
616 0 : Err(e) => match crate::auth::check_permission(claims, Scope::Admin) {
617 0 : Ok(()) => Ok(()),
618 0 : Err(_) => Err(e),
619 : },
620 0 : Ok(()) => Ok(()),
621 : }
622 0 : })
623 0 : }
624 :
625 : #[derive(Clone, Debug)]
626 : struct RequestMeta {
627 : method: hyper::http::Method,
628 : at: Instant,
629 : }
630 :
631 0 : fn prologue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>(
632 0 : ) -> Middleware<B, ApiError> {
633 0 : Middleware::pre(move |req| async move {
634 0 : let meta = RequestMeta {
635 0 : method: req.method().clone(),
636 0 : at: Instant::now(),
637 0 : };
638 0 :
639 0 : req.set_context(meta);
640 0 :
641 0 : Ok(req)
642 0 : })
643 0 : }
644 :
645 0 : fn epilogue_metrics_middleware<B: hyper::body::HttpBody + Send + Sync + 'static>(
646 0 : ) -> Middleware<B, ApiError> {
647 0 : Middleware::post_with_info(move |resp, req_info| async move {
648 0 : let request_name = match req_info.context::<RequestName>() {
649 0 : Some(name) => name,
650 : None => {
651 0 : return Ok(resp);
652 : }
653 : };
654 :
655 0 : if let Some(meta) = req_info.context::<RequestMeta>() {
656 0 : let status = &crate::metrics::METRICS_REGISTRY
657 0 : .metrics_group
658 0 : .storage_controller_http_request_status;
659 0 : let latency = &crate::metrics::METRICS_REGISTRY
660 0 : .metrics_group
661 0 : .storage_controller_http_request_latency;
662 0 :
663 0 : status.inc(HttpRequestStatusLabelGroup {
664 0 : path: request_name.0,
665 0 : method: meta.method.clone().into(),
666 0 : status: crate::metrics::StatusCode(resp.status()),
667 0 : });
668 0 :
669 0 : latency.observe(
670 0 : HttpRequestLatencyLabelGroup {
671 0 : path: request_name.0,
672 0 : method: meta.method.into(),
673 0 : },
674 0 : meta.at.elapsed().as_secs_f64(),
675 0 : );
676 0 : }
677 0 : Ok(resp)
678 0 : })
679 0 : }
680 :
681 0 : pub async fn measured_metrics_handler(req: Request<Body>) -> Result<Response<Body>, ApiError> {
682 0 : pub const TEXT_FORMAT: &str = "text/plain; version=0.0.4";
683 0 :
684 0 : let state = get_state(&req);
685 0 : let payload = crate::metrics::METRICS_REGISTRY.encode(&state.neon_metrics);
686 0 : let response = Response::builder()
687 0 : .status(200)
688 0 : .header(CONTENT_TYPE, TEXT_FORMAT)
689 0 : .body(payload.into())
690 0 : .unwrap();
691 0 :
692 0 : Ok(response)
693 0 : }
694 :
695 : #[derive(Clone)]
696 : struct RequestName(&'static str);
697 :
698 0 : async fn named_request_span<R, H>(
699 0 : request: Request<Body>,
700 0 : handler: H,
701 0 : name: RequestName,
702 0 : ) -> R::Output
703 0 : where
704 0 : R: Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
705 0 : H: FnOnce(Request<Body>) -> R + Send + Sync + 'static,
706 0 : {
707 0 : request.set_context(name);
708 0 : request_span(request, handler).await
709 0 : }
710 :
711 0 : pub fn make_router(
712 0 : service: Arc<Service>,
713 0 : auth: Option<Arc<SwappableJwtAuth>>,
714 0 : build_info: BuildInfo,
715 0 : ) -> RouterBuilder<hyper::Body, ApiError> {
716 0 : let mut router = endpoint::make_router()
717 0 : .middleware(prologue_metrics_middleware())
718 0 : .middleware(epilogue_metrics_middleware());
719 0 : if auth.is_some() {
720 0 : router = router.middleware(auth_middleware(|request| {
721 0 : let state = get_state(request);
722 0 : if state.allowlist_routes.contains(request.uri()) {
723 0 : None
724 : } else {
725 0 : state.auth.as_deref()
726 : }
727 0 : }));
728 0 : }
729 :
730 0 : router
731 0 : .data(Arc::new(HttpState::new(service, auth, build_info)))
732 0 : .get("/metrics", |r| {
733 0 : named_request_span(r, measured_metrics_handler, RequestName("metrics"))
734 0 : })
735 0 : // Non-prefixed generic endpoints (status, metrics)
736 0 : .get("/status", |r| {
737 0 : named_request_span(r, handle_status, RequestName("status"))
738 0 : })
739 0 : .get("/ready", |r| {
740 0 : named_request_span(r, handle_ready, RequestName("ready"))
741 0 : })
742 0 : // Upcalls for the pageserver: point the pageserver's `control_plane_api` config to this prefix
743 0 : .post("/upcall/v1/re-attach", |r| {
744 0 : named_request_span(r, handle_re_attach, RequestName("upcall_v1_reattach"))
745 0 : })
746 0 : .post("/upcall/v1/validate", |r| {
747 0 : named_request_span(r, handle_validate, RequestName("upcall_v1_validate"))
748 0 : })
749 0 : // Test/dev/debug endpoints
750 0 : .post("/debug/v1/attach-hook", |r| {
751 0 : named_request_span(r, handle_attach_hook, RequestName("debug_v1_attach_hook"))
752 0 : })
753 0 : .post("/debug/v1/inspect", |r| {
754 0 : named_request_span(r, handle_inspect, RequestName("debug_v1_inspect"))
755 0 : })
756 0 : .post("/debug/v1/tenant/:tenant_id/drop", |r| {
757 0 : named_request_span(r, handle_tenant_drop, RequestName("debug_v1_tenant_drop"))
758 0 : })
759 0 : .post("/debug/v1/node/:node_id/drop", |r| {
760 0 : named_request_span(r, handle_node_drop, RequestName("debug_v1_node_drop"))
761 0 : })
762 0 : .get("/debug/v1/tenant", |r| {
763 0 : named_request_span(r, handle_tenants_dump, RequestName("debug_v1_tenant"))
764 0 : })
765 0 : .get("/debug/v1/tenant/:tenant_id/locate", |r| {
766 0 : tenant_service_handler(
767 0 : r,
768 0 : handle_tenant_locate,
769 0 : RequestName("debug_v1_tenant_locate"),
770 0 : )
771 0 : })
772 0 : .get("/debug/v1/scheduler", |r| {
773 0 : named_request_span(r, handle_scheduler_dump, RequestName("debug_v1_scheduler"))
774 0 : })
775 0 : .post("/debug/v1/consistency_check", |r| {
776 0 : named_request_span(
777 0 : r,
778 0 : handle_consistency_check,
779 0 : RequestName("debug_v1_consistency_check"),
780 0 : )
781 0 : })
782 0 : .post("/debug/v1/reconcile_all", |r| {
783 0 : request_span(r, handle_reconcile_all)
784 0 : })
785 0 : .put("/debug/v1/failpoints", |r| {
786 0 : request_span(r, |r| failpoints_handler(r, CancellationToken::new()))
787 0 : })
788 0 : // Node operations
789 0 : .post("/control/v1/node", |r| {
790 0 : named_request_span(r, handle_node_register, RequestName("control_v1_node"))
791 0 : })
792 0 : .get("/control/v1/node", |r| {
793 0 : named_request_span(r, handle_node_list, RequestName("control_v1_node"))
794 0 : })
795 0 : .put("/control/v1/node/:node_id/config", |r| {
796 0 : named_request_span(
797 0 : r,
798 0 : handle_node_configure,
799 0 : RequestName("control_v1_node_config"),
800 0 : )
801 0 : })
802 0 : // Tenant Shard operations
803 0 : .put("/control/v1/tenant/:tenant_shard_id/migrate", |r| {
804 0 : tenant_service_handler(
805 0 : r,
806 0 : handle_tenant_shard_migrate,
807 0 : RequestName("control_v1_tenant_migrate"),
808 0 : )
809 0 : })
810 0 : .put("/control/v1/tenant/:tenant_id/shard_split", |r| {
811 0 : tenant_service_handler(
812 0 : r,
813 0 : handle_tenant_shard_split,
814 0 : RequestName("control_v1_tenant_shard_split"),
815 0 : )
816 0 : })
817 0 : .get("/control/v1/tenant/:tenant_id", |r| {
818 0 : tenant_service_handler(
819 0 : r,
820 0 : handle_tenant_describe,
821 0 : RequestName("control_v1_tenant_describe"),
822 0 : )
823 0 : })
824 0 : .get("/control/v1/tenant", |r| {
825 0 : tenant_service_handler(r, handle_tenant_list, RequestName("control_v1_tenant_list"))
826 0 : })
827 0 : .put("/control/v1/tenant/:tenant_id/policy", |r| {
828 0 : named_request_span(
829 0 : r,
830 0 : handle_tenant_update_policy,
831 0 : RequestName("control_v1_tenant_policy"),
832 0 : )
833 0 : })
834 0 : // Tenant operations
835 0 : // The ^/v1/ endpoints act as a "Virtual Pageserver", enabling shard-naive clients to call into
836 0 : // this service to manage tenants that actually consist of many tenant shards, as if they are a single entity.
837 0 : .post("/v1/tenant", |r| {
838 0 : tenant_service_handler(r, handle_tenant_create, RequestName("v1_tenant"))
839 0 : })
840 0 : .delete("/v1/tenant/:tenant_id", |r| {
841 0 : tenant_service_handler(r, handle_tenant_delete, RequestName("v1_tenant"))
842 0 : })
843 0 : .put("/v1/tenant/config", |r| {
844 0 : tenant_service_handler(r, handle_tenant_config_set, RequestName("v1_tenant_config"))
845 0 : })
846 0 : .get("/v1/tenant/:tenant_id/config", |r| {
847 0 : tenant_service_handler(r, handle_tenant_config_get, RequestName("v1_tenant_config"))
848 0 : })
849 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
850 0 : tenant_service_handler(
851 0 : r,
852 0 : handle_tenant_location_config,
853 0 : RequestName("v1_tenant_location_config"),
854 0 : )
855 0 : })
856 0 : .put("/v1/tenant/:tenant_id/time_travel_remote_storage", |r| {
857 0 : tenant_service_handler(
858 0 : r,
859 0 : handle_tenant_time_travel_remote_storage,
860 0 : RequestName("v1_tenant_time_travel_remote_storage"),
861 0 : )
862 0 : })
863 0 : .post("/v1/tenant/:tenant_id/secondary/download", |r| {
864 0 : tenant_service_handler(
865 0 : r,
866 0 : handle_tenant_secondary_download,
867 0 : RequestName("v1_tenant_secondary_download"),
868 0 : )
869 0 : })
870 0 : // Timeline operations
871 0 : .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
872 0 : tenant_service_handler(
873 0 : r,
874 0 : handle_tenant_timeline_delete,
875 0 : RequestName("v1_tenant_timeline"),
876 0 : )
877 0 : })
878 0 : .post("/v1/tenant/:tenant_id/timeline", |r| {
879 0 : tenant_service_handler(
880 0 : r,
881 0 : handle_tenant_timeline_create,
882 0 : RequestName("v1_tenant_timeline"),
883 0 : )
884 0 : })
885 0 : // Tenant detail GET passthrough to shard zero
886 0 : .get("/v1/tenant/:tenant_id", |r| {
887 0 : tenant_service_handler(
888 0 : r,
889 0 : handle_tenant_timeline_passthrough,
890 0 : RequestName("v1_tenant_passthrough"),
891 0 : )
892 0 : })
893 0 : // Timeline GET passthrough to shard zero. Note that the `*` in the URL is a wildcard: any future
894 0 : // timeline GET APIs will be implicitly included.
895 0 : .get("/v1/tenant/:tenant_id/timeline*", |r| {
896 0 : tenant_service_handler(
897 0 : r,
898 0 : handle_tenant_timeline_passthrough,
899 0 : RequestName("v1_tenant_timeline_passthrough"),
900 0 : )
901 0 : })
902 0 : }
|
