Line data Source code
1 : //!
2 : //! Management HTTP API
3 : //!
4 : use std::collections::HashMap;
5 : use std::str::FromStr;
6 : use std::sync::Arc;
7 : use std::time::Duration;
8 :
9 : use anyhow::{anyhow, Context, Result};
10 : use enumset::EnumSet;
11 : use futures::TryFutureExt;
12 : use humantime::format_rfc3339;
13 : use hyper::header;
14 : use hyper::StatusCode;
15 : use hyper::{Body, Request, Response, Uri};
16 : use metrics::launch_timestamp::LaunchTimestamp;
17 : use pageserver_api::models::LocationConfig;
18 : use pageserver_api::models::LocationConfigListResponse;
19 : use pageserver_api::models::ShardParameters;
20 : use pageserver_api::models::TenantDetails;
21 : use pageserver_api::models::TenantLocationConfigResponse;
22 : use pageserver_api::models::TenantScanRemoteStorageResponse;
23 : use pageserver_api::models::TenantScanRemoteStorageShard;
24 : use pageserver_api::models::TenantShardLocation;
25 : use pageserver_api::models::TenantShardSplitRequest;
26 : use pageserver_api::models::TenantShardSplitResponse;
27 : use pageserver_api::models::TenantState;
28 : use pageserver_api::models::{
29 : DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
30 : TenantLoadRequest, TenantLocationConfigRequest,
31 : };
32 : use pageserver_api::shard::ShardCount;
33 : use pageserver_api::shard::TenantShardId;
34 : use remote_storage::DownloadError;
35 : use remote_storage::GenericRemoteStorage;
36 : use remote_storage::TimeTravelError;
37 : use tenant_size_model::{SizeResult, StorageModel};
38 : use tokio_util::sync::CancellationToken;
39 : use tracing::*;
40 : use utils::auth::JwtAuth;
41 : use utils::failpoint_support::failpoints_handler;
42 : use utils::http::endpoint::prometheus_metrics_handler;
43 : use utils::http::endpoint::request_span;
44 : use utils::http::json::json_request_or_empty_body;
45 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
46 :
47 : use crate::context::{DownloadBehavior, RequestContext};
48 : use crate::deletion_queue::DeletionQueueClient;
49 : use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
50 : use crate::pgdatadir_mapping::LsnForTimestamp;
51 : use crate::task_mgr::TaskKind;
52 : use crate::tenant::config::{LocationConf, TenantConfOpt};
53 : use crate::tenant::mgr::GetActiveTenantError;
54 : use crate::tenant::mgr::{
55 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
56 : TenantSlotUpsertError, TenantStateError,
57 : };
58 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
59 : use crate::tenant::remote_timeline_client;
60 : use crate::tenant::remote_timeline_client::download_index_part;
61 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
62 : use crate::tenant::remote_timeline_client::list_remote_timelines;
63 : use crate::tenant::secondary::SecondaryController;
64 : use crate::tenant::size::ModelInputs;
65 : use crate::tenant::storage_layer::LayerAccessStatsReset;
66 : use crate::tenant::storage_layer::LayerName;
67 : use crate::tenant::timeline::CompactFlags;
68 : use crate::tenant::timeline::Timeline;
69 : use crate::tenant::SpawnMode;
70 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
71 : use crate::{config::PageServerConf, tenant::mgr};
72 : use crate::{disk_usage_eviction_task, tenant};
73 : use pageserver_api::models::{
74 : StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
75 : TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
76 : };
77 : use utils::{
78 : auth::SwappableJwtAuth,
79 : generation::Generation,
80 : http::{
81 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
82 : error::{ApiError, HttpErrorBody},
83 : json::{json_request, json_response},
84 : request::parse_request_param,
85 : RequestExt, RouterBuilder,
86 : },
87 : id::{TenantId, TimelineId},
88 : lsn::Lsn,
89 : };
90 :
91 : // For APIs that require an Active tenant, how long should we block waiting for that state?
92 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
93 : // failed API calls while tenants are activating.
94 : #[cfg(not(feature = "testing"))]
95 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
96 :
97 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
98 : // finish attaching, if calls to remote storage are slow.
99 : #[cfg(feature = "testing")]
100 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
101 :
102 : pub struct State {
103 : conf: &'static PageServerConf,
104 : tenant_manager: Arc<TenantManager>,
105 : auth: Option<Arc<SwappableJwtAuth>>,
106 : allowlist_routes: Vec<Uri>,
107 : remote_storage: Option<GenericRemoteStorage>,
108 : broker_client: storage_broker::BrokerClientChannel,
109 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
110 : deletion_queue_client: DeletionQueueClient,
111 : secondary_controller: SecondaryController,
112 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
113 : }
114 :
115 : impl State {
116 : #[allow(clippy::too_many_arguments)]
117 0 : pub fn new(
118 0 : conf: &'static PageServerConf,
119 0 : tenant_manager: Arc<TenantManager>,
120 0 : auth: Option<Arc<SwappableJwtAuth>>,
121 0 : remote_storage: Option<GenericRemoteStorage>,
122 0 : broker_client: storage_broker::BrokerClientChannel,
123 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
124 0 : deletion_queue_client: DeletionQueueClient,
125 0 : secondary_controller: SecondaryController,
126 0 : ) -> anyhow::Result<Self> {
127 0 : let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
128 0 : .iter()
129 0 : .map(|v| v.parse().unwrap())
130 0 : .collect::<Vec<_>>();
131 0 : Ok(Self {
132 0 : conf,
133 0 : tenant_manager,
134 0 : auth,
135 0 : allowlist_routes,
136 0 : remote_storage,
137 0 : broker_client,
138 0 : disk_usage_eviction_state,
139 0 : deletion_queue_client,
140 0 : secondary_controller,
141 0 : latest_utilization: Default::default(),
142 0 : })
143 0 : }
144 : }
145 :
146 : #[inline(always)]
147 0 : fn get_state(request: &Request<Body>) -> &State {
148 0 : request
149 0 : .data::<Arc<State>>()
150 0 : .expect("unknown state type")
151 0 : .as_ref()
152 0 : }
153 :
154 : #[inline(always)]
155 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
156 0 : get_state(request).conf
157 0 : }
158 :
159 : /// Check that the requester is authorized to operate on given tenant
160 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
161 0 : check_permission_with(request, |claims| {
162 0 : crate::auth::check_permission(claims, tenant_id)
163 0 : })
164 0 : }
165 :
166 : impl From<PageReconstructError> for ApiError {
167 0 : fn from(pre: PageReconstructError) -> ApiError {
168 0 : match pre {
169 0 : PageReconstructError::Other(pre) => ApiError::InternalServerError(pre),
170 0 : PageReconstructError::MissingKey(e) => {
171 0 : ApiError::InternalServerError(anyhow::anyhow!("{e}"))
172 : }
173 : PageReconstructError::Cancelled => {
174 0 : ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
175 : }
176 : PageReconstructError::AncestorStopping(_) => {
177 0 : ApiError::ResourceUnavailable(format!("{pre}").into())
178 : }
179 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
180 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
181 : }
182 0 : }
183 : }
184 :
185 : impl From<TenantMapInsertError> for ApiError {
186 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
187 0 : match tmie {
188 0 : TenantMapInsertError::SlotError(e) => e.into(),
189 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
190 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
191 : }
192 0 : }
193 : }
194 :
195 : impl From<TenantSlotError> for ApiError {
196 0 : fn from(e: TenantSlotError) -> ApiError {
197 0 : use TenantSlotError::*;
198 0 : match e {
199 0 : NotFound(tenant_id) => {
200 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
201 : }
202 0 : e @ AlreadyExists(_, _) => ApiError::Conflict(format!("{e}")),
203 : InProgress => {
204 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
205 : }
206 0 : MapState(e) => e.into(),
207 : }
208 0 : }
209 : }
210 :
211 : impl From<TenantSlotUpsertError> for ApiError {
212 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
213 0 : use TenantSlotUpsertError::*;
214 0 : match e {
215 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
216 0 : MapState(e) => e.into(),
217 0 : ShuttingDown(_) => ApiError::ShuttingDown,
218 : }
219 0 : }
220 : }
221 :
222 : impl From<UpsertLocationError> for ApiError {
223 0 : fn from(e: UpsertLocationError) -> ApiError {
224 0 : use UpsertLocationError::*;
225 0 : match e {
226 0 : BadRequest(e) => ApiError::BadRequest(e),
227 0 : Unavailable(_) => ApiError::ShuttingDown,
228 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
229 0 : Flush(e) | Other(e) => ApiError::InternalServerError(e),
230 : }
231 0 : }
232 : }
233 :
234 : impl From<TenantMapError> for ApiError {
235 0 : fn from(e: TenantMapError) -> ApiError {
236 0 : use TenantMapError::*;
237 0 : match e {
238 : StillInitializing | ShuttingDown => {
239 0 : ApiError::ResourceUnavailable(format!("{e}").into())
240 0 : }
241 0 : }
242 0 : }
243 : }
244 :
245 : impl From<TenantStateError> for ApiError {
246 0 : fn from(tse: TenantStateError) -> ApiError {
247 0 : match tse {
248 : TenantStateError::IsStopping(_) => {
249 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
250 : }
251 0 : TenantStateError::SlotError(e) => e.into(),
252 0 : TenantStateError::SlotUpsertError(e) => e.into(),
253 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
254 : }
255 0 : }
256 : }
257 :
258 : impl From<GetTenantError> for ApiError {
259 0 : fn from(tse: GetTenantError) -> ApiError {
260 0 : match tse {
261 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
262 : GetTenantError::NotActive(_) => {
263 : // Why is this not `ApiError::NotFound`?
264 : // Because we must be careful to never return 404 for a tenant if it does
265 : // in fact exist locally. If we did, the caller could draw the conclusion
266 : // that it can attach the tenant to another PS and we'd be in split-brain.
267 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
268 : }
269 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
270 : }
271 0 : }
272 : }
273 :
274 : impl From<GetActiveTenantError> for ApiError {
275 0 : fn from(e: GetActiveTenantError) -> ApiError {
276 0 : match e {
277 0 : GetActiveTenantError::Broken(reason) => {
278 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
279 : }
280 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
281 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
282 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
283 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
284 0 : ApiError::ResourceUnavailable(format!("{}", e).into())
285 : }
286 : }
287 0 : }
288 : }
289 :
290 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
291 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
292 0 : use crate::tenant::DeleteTimelineError::*;
293 0 : match value {
294 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
295 0 : HasChildren(children) => ApiError::PreconditionFailed(
296 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
297 0 : .into_boxed_str(),
298 0 : ),
299 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
300 0 : Other(e) => ApiError::InternalServerError(e),
301 : }
302 0 : }
303 : }
304 :
305 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
306 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
307 : use crate::tenant::mgr::DeleteTimelineError::*;
308 0 : match value {
309 : // Report Precondition failed so client can distinguish between
310 : // "tenant is missing" case from "timeline is missing"
311 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
312 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
313 0 : ),
314 0 : Tenant(t) => ApiError::from(t),
315 0 : Timeline(t) => ApiError::from(t),
316 : }
317 0 : }
318 : }
319 :
320 : impl From<crate::tenant::delete::DeleteTenantError> for ApiError {
321 0 : fn from(value: crate::tenant::delete::DeleteTenantError) -> Self {
322 0 : use crate::tenant::delete::DeleteTenantError::*;
323 0 : match value {
324 0 : Get(g) => ApiError::from(g),
325 0 : e @ AlreadyInProgress => ApiError::Conflict(e.to_string()),
326 0 : Timeline(t) => ApiError::from(t),
327 0 : NotAttached => ApiError::NotFound(anyhow::anyhow!("Tenant is not attached").into()),
328 0 : SlotError(e) => e.into(),
329 0 : SlotUpsertError(e) => e.into(),
330 0 : Other(o) => ApiError::InternalServerError(o),
331 0 : e @ InvalidState(_) => ApiError::PreconditionFailed(e.to_string().into_boxed_str()),
332 0 : Cancelled => ApiError::ShuttingDown,
333 : }
334 0 : }
335 : }
336 :
337 : // Helper function to construct a TimelineInfo struct for a timeline
338 0 : async fn build_timeline_info(
339 0 : timeline: &Arc<Timeline>,
340 0 : include_non_incremental_logical_size: bool,
341 0 : force_await_initial_logical_size: bool,
342 0 : ctx: &RequestContext,
343 0 : ) -> anyhow::Result<TimelineInfo> {
344 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
345 0 :
346 0 : if force_await_initial_logical_size {
347 0 : timeline.clone().await_initial_logical_size().await
348 0 : }
349 :
350 0 : let mut info = build_timeline_info_common(
351 0 : timeline,
352 0 : ctx,
353 0 : tenant::timeline::GetLogicalSizePriority::Background,
354 0 : )
355 0 : .await?;
356 0 : if include_non_incremental_logical_size {
357 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
358 : // Otherwise, if someone deletes the timeline / detaches the tenant while
359 : // we're executing this function, we will outlive the timeline on-disk state.
360 : info.current_logical_size_non_incremental = Some(
361 0 : timeline
362 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
363 0 : .await?,
364 : );
365 0 : }
366 0 : Ok(info)
367 0 : }
368 :
369 0 : async fn build_timeline_info_common(
370 0 : timeline: &Arc<Timeline>,
371 0 : ctx: &RequestContext,
372 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
373 0 : ) -> anyhow::Result<TimelineInfo> {
374 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
375 0 : let initdb_lsn = timeline.initdb_lsn;
376 0 : let last_record_lsn = timeline.get_last_record_lsn();
377 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
378 0 : let guard = timeline.last_received_wal.lock().unwrap();
379 0 : if let Some(info) = guard.as_ref() {
380 0 : (
381 0 : Some(format!("{:?}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
382 0 : Some(info.last_received_msg_lsn),
383 0 : Some(info.last_received_msg_ts),
384 0 : )
385 : } else {
386 0 : (None, None, None)
387 : }
388 : };
389 :
390 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
391 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
392 0 : Lsn(0) => None,
393 0 : lsn @ Lsn(_) => Some(lsn),
394 : };
395 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
396 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
397 0 : let state = timeline.current_state();
398 0 : let remote_consistent_lsn_projected = timeline
399 0 : .get_remote_consistent_lsn_projected()
400 0 : .unwrap_or(Lsn(0));
401 0 : let remote_consistent_lsn_visible = timeline
402 0 : .get_remote_consistent_lsn_visible()
403 0 : .unwrap_or(Lsn(0));
404 0 :
405 0 : let walreceiver_status = timeline.walreceiver_status();
406 :
407 0 : let info = TimelineInfo {
408 0 : tenant_id: timeline.tenant_shard_id,
409 0 : timeline_id: timeline.timeline_id,
410 0 : ancestor_timeline_id,
411 0 : ancestor_lsn,
412 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
413 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
414 0 : remote_consistent_lsn_visible,
415 0 : initdb_lsn,
416 0 : last_record_lsn,
417 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
418 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
419 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
420 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
421 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
422 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
423 : },
424 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
425 0 : current_physical_size,
426 0 : current_logical_size_non_incremental: None,
427 0 : timeline_dir_layer_file_size_sum: None,
428 0 : wal_source_connstr,
429 0 : last_received_msg_lsn,
430 0 : last_received_msg_ts,
431 0 : pg_version: timeline.pg_version,
432 0 :
433 0 : state,
434 0 :
435 0 : walreceiver_status,
436 0 : };
437 0 : Ok(info)
438 0 : }
439 :
440 : // healthcheck handler
441 0 : async fn status_handler(
442 0 : request: Request<Body>,
443 0 : _cancel: CancellationToken,
444 0 : ) -> Result<Response<Body>, ApiError> {
445 0 : check_permission(&request, None)?;
446 0 : let config = get_config(&request);
447 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
448 0 : }
449 :
450 0 : async fn reload_auth_validation_keys_handler(
451 0 : request: Request<Body>,
452 0 : _cancel: CancellationToken,
453 0 : ) -> Result<Response<Body>, ApiError> {
454 0 : check_permission(&request, None)?;
455 0 : let config = get_config(&request);
456 0 : let state = get_state(&request);
457 0 : let Some(shared_auth) = &state.auth else {
458 0 : return json_response(StatusCode::BAD_REQUEST, ());
459 : };
460 : // unwrap is ok because check is performed when creating config, so path is set and exists
461 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
462 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
463 :
464 0 : match JwtAuth::from_key_path(key_path) {
465 0 : Ok(new_auth) => {
466 0 : shared_auth.swap(new_auth);
467 0 : json_response(StatusCode::OK, ())
468 : }
469 0 : Err(e) => {
470 0 : let err_msg = "Error reloading public keys";
471 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
472 0 : json_response(
473 0 : StatusCode::INTERNAL_SERVER_ERROR,
474 0 : HttpErrorBody::from_msg(err_msg.to_string()),
475 0 : )
476 : }
477 : }
478 0 : }
479 :
480 0 : async fn timeline_create_handler(
481 0 : mut request: Request<Body>,
482 0 : _cancel: CancellationToken,
483 0 : ) -> Result<Response<Body>, ApiError> {
484 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
485 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
486 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
487 :
488 0 : let new_timeline_id = request_data.new_timeline_id;
489 0 :
490 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
491 0 :
492 0 : let state = get_state(&request);
493 :
494 0 : async {
495 0 : let tenant = state
496 0 : .tenant_manager
497 0 : .get_attached_tenant_shard(tenant_shard_id)?;
498 :
499 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
500 :
501 0 : if let Some(ancestor_id) = request_data.ancestor_timeline_id.as_ref() {
502 0 : tracing::info!(%ancestor_id, "starting to branch");
503 : } else {
504 0 : tracing::info!("bootstrapping");
505 : }
506 :
507 0 : match tenant
508 0 : .create_timeline(
509 0 : new_timeline_id,
510 0 : request_data.ancestor_timeline_id,
511 0 : request_data.ancestor_start_lsn,
512 0 : request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
513 0 : request_data.existing_initdb_timeline_id,
514 0 : state.broker_client.clone(),
515 0 : &ctx,
516 0 : )
517 0 : .await
518 : {
519 0 : Ok(new_timeline) => {
520 : // Created. Construct a TimelineInfo for it.
521 0 : let timeline_info = build_timeline_info_common(
522 0 : &new_timeline,
523 0 : &ctx,
524 0 : tenant::timeline::GetLogicalSizePriority::User,
525 0 : )
526 0 : .await
527 0 : .map_err(ApiError::InternalServerError)?;
528 0 : json_response(StatusCode::CREATED, timeline_info)
529 : }
530 0 : Err(_) if tenant.cancel.is_cancelled() => {
531 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
532 0 : json_response(
533 0 : StatusCode::SERVICE_UNAVAILABLE,
534 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
535 0 : )
536 : }
537 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
538 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
539 : }
540 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
541 0 : StatusCode::TOO_MANY_REQUESTS,
542 0 : HttpErrorBody::from_msg(e.to_string()),
543 0 : ),
544 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
545 0 : StatusCode::NOT_ACCEPTABLE,
546 0 : HttpErrorBody::from_msg(format!("{err:#}")),
547 0 : ),
548 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
549 0 : StatusCode::SERVICE_UNAVAILABLE,
550 0 : HttpErrorBody::from_msg(e.to_string()),
551 0 : ),
552 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
553 0 : StatusCode::SERVICE_UNAVAILABLE,
554 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
555 0 : ),
556 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
557 : }
558 0 : }
559 0 : .instrument(info_span!("timeline_create",
560 : tenant_id = %tenant_shard_id.tenant_id,
561 0 : shard_id = %tenant_shard_id.shard_slug(),
562 : timeline_id = %new_timeline_id,
563 : lsn=?request_data.ancestor_start_lsn,
564 : pg_version=?request_data.pg_version
565 : ))
566 0 : .await
567 0 : }
568 :
569 0 : async fn timeline_list_handler(
570 0 : request: Request<Body>,
571 0 : _cancel: CancellationToken,
572 0 : ) -> Result<Response<Body>, ApiError> {
573 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
574 0 : let include_non_incremental_logical_size: Option<bool> =
575 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
576 0 : let force_await_initial_logical_size: Option<bool> =
577 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
578 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
579 :
580 0 : let state = get_state(&request);
581 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
582 :
583 0 : let response_data = async {
584 0 : let tenant = state
585 0 : .tenant_manager
586 0 : .get_attached_tenant_shard(tenant_shard_id)?;
587 :
588 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
589 :
590 0 : let timelines = tenant.list_timelines();
591 0 :
592 0 : let mut response_data = Vec::with_capacity(timelines.len());
593 0 : for timeline in timelines {
594 0 : let timeline_info = build_timeline_info(
595 0 : &timeline,
596 0 : include_non_incremental_logical_size.unwrap_or(false),
597 0 : force_await_initial_logical_size.unwrap_or(false),
598 0 : &ctx,
599 0 : )
600 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
601 0 : .await
602 0 : .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
603 0 : .map_err(ApiError::InternalServerError)?;
604 :
605 0 : response_data.push(timeline_info);
606 : }
607 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
608 0 : }
609 0 : .instrument(info_span!("timeline_list",
610 : tenant_id = %tenant_shard_id.tenant_id,
611 0 : shard_id = %tenant_shard_id.shard_slug()))
612 0 : .await?;
613 :
614 0 : json_response(StatusCode::OK, response_data)
615 0 : }
616 :
617 0 : async fn timeline_preserve_initdb_handler(
618 0 : request: Request<Body>,
619 0 : _cancel: CancellationToken,
620 0 : ) -> Result<Response<Body>, ApiError> {
621 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
622 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
623 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
624 0 : let state = get_state(&request);
625 :
626 : // Part of the process for disaster recovery from safekeeper-stored WAL:
627 : // If we don't recover into a new timeline but want to keep the timeline ID,
628 : // then the initdb archive is deleted. This endpoint copies it to a different
629 : // location where timeline recreation cand find it.
630 :
631 0 : async {
632 0 : let tenant = state
633 0 : .tenant_manager
634 0 : .get_attached_tenant_shard(tenant_shard_id)?;
635 :
636 0 : let timeline = tenant
637 0 : .get_timeline(timeline_id, false)
638 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
639 :
640 0 : timeline
641 0 : .preserve_initdb_archive()
642 0 : .await
643 0 : .context("preserving initdb archive")
644 0 : .map_err(ApiError::InternalServerError)?;
645 :
646 0 : Ok::<_, ApiError>(())
647 0 : }
648 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
649 : tenant_id = %tenant_shard_id.tenant_id,
650 0 : shard_id = %tenant_shard_id.shard_slug(),
651 : %timeline_id))
652 0 : .await?;
653 :
654 0 : json_response(StatusCode::OK, ())
655 0 : }
656 :
657 0 : async fn timeline_detail_handler(
658 0 : request: Request<Body>,
659 0 : _cancel: CancellationToken,
660 0 : ) -> Result<Response<Body>, ApiError> {
661 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
662 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
663 0 : let include_non_incremental_logical_size: Option<bool> =
664 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
665 0 : let force_await_initial_logical_size: Option<bool> =
666 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
667 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
668 :
669 : // Logical size calculation needs downloading.
670 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
671 0 : let state = get_state(&request);
672 :
673 0 : let timeline_info = async {
674 0 : let tenant = state
675 0 : .tenant_manager
676 0 : .get_attached_tenant_shard(tenant_shard_id)?;
677 :
678 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
679 :
680 0 : let timeline = tenant
681 0 : .get_timeline(timeline_id, false)
682 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
683 :
684 0 : let timeline_info = build_timeline_info(
685 0 : &timeline,
686 0 : include_non_incremental_logical_size.unwrap_or(false),
687 0 : force_await_initial_logical_size.unwrap_or(false),
688 0 : &ctx,
689 0 : )
690 0 : .await
691 0 : .context("get local timeline info")
692 0 : .map_err(ApiError::InternalServerError)?;
693 :
694 0 : Ok::<_, ApiError>(timeline_info)
695 0 : }
696 0 : .instrument(info_span!("timeline_detail",
697 : tenant_id = %tenant_shard_id.tenant_id,
698 0 : shard_id = %tenant_shard_id.shard_slug(),
699 : %timeline_id))
700 0 : .await?;
701 :
702 0 : json_response(StatusCode::OK, timeline_info)
703 0 : }
704 :
705 0 : async fn get_lsn_by_timestamp_handler(
706 0 : request: Request<Body>,
707 0 : cancel: CancellationToken,
708 0 : ) -> Result<Response<Body>, ApiError> {
709 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
710 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
711 0 : let state = get_state(&request);
712 0 :
713 0 : if !tenant_shard_id.is_shard_zero() {
714 : // Requires SLRU contents, which are only stored on shard zero
715 0 : return Err(ApiError::BadRequest(anyhow!(
716 0 : "Size calculations are only available on shard zero"
717 0 : )));
718 0 : }
719 :
720 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
721 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
722 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
723 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
724 0 : .map_err(ApiError::BadRequest)?;
725 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
726 0 :
727 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
728 :
729 0 : let timeline =
730 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
731 0 : .await?;
732 0 : let result = timeline
733 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
734 0 : .await?;
735 : #[derive(serde::Serialize, Debug)]
736 : struct Result {
737 : lsn: Lsn,
738 : kind: &'static str,
739 : }
740 0 : let (lsn, kind) = match result {
741 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
742 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
743 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
744 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
745 : };
746 0 : let result = Result { lsn, kind };
747 0 : tracing::info!(
748 : lsn=?result.lsn,
749 : kind=%result.kind,
750 : timestamp=%timestamp_raw,
751 0 : "lsn_by_timestamp finished"
752 : );
753 0 : json_response(StatusCode::OK, result)
754 0 : }
755 :
756 0 : async fn get_timestamp_of_lsn_handler(
757 0 : request: Request<Body>,
758 0 : _cancel: CancellationToken,
759 0 : ) -> Result<Response<Body>, ApiError> {
760 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
761 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
762 0 : let state = get_state(&request);
763 0 :
764 0 : if !tenant_shard_id.is_shard_zero() {
765 : // Requires SLRU contents, which are only stored on shard zero
766 0 : return Err(ApiError::BadRequest(anyhow!(
767 0 : "Size calculations are only available on shard zero"
768 0 : )));
769 0 : }
770 :
771 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
772 :
773 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
774 0 : let lsn = Lsn::from_str(&lsn_str)
775 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
776 0 : .map_err(ApiError::BadRequest)?;
777 :
778 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
779 0 : let timeline =
780 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
781 0 : .await?;
782 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
783 :
784 0 : match result {
785 0 : Some(time) => {
786 0 : let time = format_rfc3339(postgres_ffi::from_pg_timestamp(time)).to_string();
787 0 : json_response(StatusCode::OK, time)
788 : }
789 0 : None => Err(ApiError::NotFound(
790 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
791 0 : )),
792 : }
793 0 : }
794 :
795 0 : async fn tenant_attach_handler(
796 0 : mut request: Request<Body>,
797 0 : _cancel: CancellationToken,
798 0 : ) -> Result<Response<Body>, ApiError> {
799 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
800 0 : check_permission(&request, Some(tenant_id))?;
801 :
802 0 : let maybe_body: Option<TenantAttachRequest> = json_request_or_empty_body(&mut request).await?;
803 0 : let tenant_conf = match &maybe_body {
804 0 : Some(request) => TenantConfOpt::try_from(&*request.config).map_err(ApiError::BadRequest)?,
805 0 : None => TenantConfOpt::default(),
806 : };
807 :
808 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
809 0 :
810 0 : info!("Handling tenant attach {tenant_id}");
811 :
812 0 : let state = get_state(&request);
813 :
814 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
815 :
816 0 : if state.remote_storage.is_none() {
817 0 : return Err(ApiError::BadRequest(anyhow!(
818 0 : "attach_tenant is not possible because pageserver was configured without remote storage"
819 0 : )));
820 0 : }
821 0 :
822 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
823 0 : let shard_params = ShardParameters::default();
824 0 : let location_conf = LocationConf::attached_single(tenant_conf, generation, &shard_params);
825 :
826 0 : let tenant = state
827 0 : .tenant_manager
828 0 : .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
829 0 : .await?;
830 :
831 0 : let Some(tenant) = tenant else {
832 : // This should never happen: indicates a bug in upsert_location
833 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
834 0 : "Upsert succeeded but didn't return tenant!"
835 0 : )));
836 : };
837 :
838 : // We might have successfully constructed a Tenant, but it could still
839 : // end up in a broken state:
840 : if let TenantState::Broken {
841 0 : reason,
842 : backtrace: _,
843 0 : } = tenant.current_state()
844 : {
845 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
846 0 : "Tenant state is Broken: {reason}"
847 0 : )));
848 0 : }
849 0 :
850 0 : json_response(StatusCode::ACCEPTED, ())
851 0 : }
852 :
853 0 : async fn timeline_delete_handler(
854 0 : request: Request<Body>,
855 0 : _cancel: CancellationToken,
856 0 : ) -> Result<Response<Body>, ApiError> {
857 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
858 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
859 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
860 :
861 0 : let state = get_state(&request);
862 :
863 0 : let tenant = state
864 0 : .tenant_manager
865 0 : .get_attached_tenant_shard(tenant_shard_id)
866 0 : .map_err(|e| {
867 0 : match e {
868 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
869 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
870 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
871 0 : "Requested tenant is missing".to_string().into_boxed_str(),
872 0 : ),
873 0 : e => e.into(),
874 : }
875 0 : })?;
876 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
877 0 : tenant.delete_timeline(timeline_id).instrument(info_span!("timeline_delete", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id))
878 0 : .await?;
879 :
880 0 : json_response(StatusCode::ACCEPTED, ())
881 0 : }
882 :
883 0 : async fn tenant_detach_handler(
884 0 : request: Request<Body>,
885 0 : _cancel: CancellationToken,
886 0 : ) -> Result<Response<Body>, ApiError> {
887 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
888 0 : check_permission(&request, Some(tenant_id))?;
889 0 : let detach_ignored: Option<bool> = parse_query_param(&request, "detach_ignored")?;
890 :
891 : // This is a legacy API (`/location_conf` is the replacement). It only supports unsharded tenants
892 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
893 0 :
894 0 : let state = get_state(&request);
895 0 : let conf = state.conf;
896 0 : state
897 0 : .tenant_manager
898 0 : .detach_tenant(
899 0 : conf,
900 0 : tenant_shard_id,
901 0 : detach_ignored.unwrap_or(false),
902 0 : &state.deletion_queue_client,
903 0 : )
904 0 : .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
905 0 : .await?;
906 :
907 0 : json_response(StatusCode::OK, ())
908 0 : }
909 :
910 0 : async fn tenant_reset_handler(
911 0 : request: Request<Body>,
912 0 : _cancel: CancellationToken,
913 0 : ) -> Result<Response<Body>, ApiError> {
914 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
915 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
916 :
917 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
918 :
919 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
920 0 : let state = get_state(&request);
921 0 : state
922 0 : .tenant_manager
923 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
924 0 : .await
925 0 : .map_err(ApiError::InternalServerError)?;
926 :
927 0 : json_response(StatusCode::OK, ())
928 0 : }
929 :
930 0 : async fn tenant_load_handler(
931 0 : mut request: Request<Body>,
932 0 : _cancel: CancellationToken,
933 0 : ) -> Result<Response<Body>, ApiError> {
934 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
935 0 : check_permission(&request, Some(tenant_id))?;
936 :
937 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
938 :
939 0 : let maybe_body: Option<TenantLoadRequest> = json_request_or_empty_body(&mut request).await?;
940 :
941 0 : let state = get_state(&request);
942 :
943 : // The /load request is only usable when control_plane_api is not set. Once it is set, callers
944 : // should always use /attach instead.
945 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
946 :
947 0 : mgr::load_tenant(
948 0 : state.conf,
949 0 : tenant_id,
950 0 : generation,
951 0 : state.broker_client.clone(),
952 0 : state.remote_storage.clone(),
953 0 : state.deletion_queue_client.clone(),
954 0 : &ctx,
955 0 : )
956 0 : .instrument(info_span!("load", %tenant_id))
957 0 : .await?;
958 :
959 0 : json_response(StatusCode::ACCEPTED, ())
960 0 : }
961 :
962 0 : async fn tenant_ignore_handler(
963 0 : request: Request<Body>,
964 0 : _cancel: CancellationToken,
965 0 : ) -> Result<Response<Body>, ApiError> {
966 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
967 0 : check_permission(&request, Some(tenant_id))?;
968 :
969 0 : let state = get_state(&request);
970 0 : let conf = state.conf;
971 0 : mgr::ignore_tenant(conf, tenant_id)
972 0 : .instrument(info_span!("ignore_tenant", %tenant_id))
973 0 : .await?;
974 :
975 0 : json_response(StatusCode::OK, ())
976 0 : }
977 :
978 0 : async fn tenant_list_handler(
979 0 : request: Request<Body>,
980 0 : _cancel: CancellationToken,
981 0 : ) -> Result<Response<Body>, ApiError> {
982 0 : check_permission(&request, None)?;
983 0 : let state = get_state(&request);
984 :
985 0 : let response_data = state
986 0 : .tenant_manager
987 0 : .list_tenants()
988 0 : .map_err(|_| {
989 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
990 0 : })?
991 0 : .iter()
992 0 : .map(|(id, state, gen)| TenantInfo {
993 0 : id: *id,
994 0 : state: state.clone(),
995 0 : current_physical_size: None,
996 0 : attachment_status: state.attachment_status(),
997 0 : generation: (*gen).into(),
998 0 : })
999 0 : .collect::<Vec<TenantInfo>>();
1000 0 :
1001 0 : json_response(StatusCode::OK, response_data)
1002 0 : }
1003 :
1004 0 : async fn tenant_status(
1005 0 : request: Request<Body>,
1006 0 : _cancel: CancellationToken,
1007 0 : ) -> Result<Response<Body>, ApiError> {
1008 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1009 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1010 0 : let state = get_state(&request);
1011 0 :
1012 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1013 0 : let activate = true;
1014 : #[cfg(feature = "testing")]
1015 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1016 :
1017 0 : let tenant_info = async {
1018 0 : let tenant = state
1019 0 : .tenant_manager
1020 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1021 :
1022 0 : if activate {
1023 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1024 : // called, but it is still valid to return 200 and describe the current state of the tenant
1025 : // if it doesn't make it into an active state.
1026 0 : tenant
1027 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1028 0 : .await
1029 0 : .ok();
1030 0 : }
1031 :
1032 : // Calculate total physical size of all timelines
1033 0 : let mut current_physical_size = 0;
1034 0 : for timeline in tenant.list_timelines().iter() {
1035 0 : current_physical_size += timeline.layer_size_sum().await;
1036 : }
1037 :
1038 0 : let state = tenant.current_state();
1039 0 : Result::<_, ApiError>::Ok(TenantDetails {
1040 0 : tenant_info: TenantInfo {
1041 0 : id: tenant_shard_id,
1042 0 : state: state.clone(),
1043 0 : current_physical_size: Some(current_physical_size),
1044 0 : attachment_status: state.attachment_status(),
1045 0 : generation: tenant.generation().into(),
1046 0 : },
1047 0 : walredo: tenant.wal_redo_manager_status(),
1048 0 : timelines: tenant.list_timeline_ids(),
1049 0 : })
1050 0 : }
1051 0 : .instrument(info_span!("tenant_status_handler",
1052 : tenant_id = %tenant_shard_id.tenant_id,
1053 0 : shard_id = %tenant_shard_id.shard_slug()))
1054 0 : .await?;
1055 :
1056 0 : json_response(StatusCode::OK, tenant_info)
1057 0 : }
1058 :
1059 0 : async fn tenant_delete_handler(
1060 0 : request: Request<Body>,
1061 0 : _cancel: CancellationToken,
1062 0 : ) -> Result<Response<Body>, ApiError> {
1063 : // TODO openapi spec
1064 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1065 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1066 :
1067 0 : let state = get_state(&request);
1068 0 :
1069 0 : state
1070 0 : .tenant_manager
1071 0 : .delete_tenant(tenant_shard_id, ACTIVE_TENANT_TIMEOUT)
1072 0 : .instrument(info_span!("tenant_delete_handler",
1073 : tenant_id = %tenant_shard_id.tenant_id,
1074 0 : shard_id = %tenant_shard_id.shard_slug()
1075 : ))
1076 0 : .await?;
1077 :
1078 0 : json_response(StatusCode::ACCEPTED, ())
1079 0 : }
1080 :
1081 : /// HTTP endpoint to query the current tenant_size of a tenant.
1082 : ///
1083 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1084 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1085 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1086 : /// values.
1087 : ///
1088 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1089 : /// (only if it is shorter than the real cutoff).
1090 : ///
1091 : /// Note: we don't update the cached size and prometheus metric here.
1092 : /// The retention period might be different, and it's nice to have a method to just calculate it
1093 : /// without modifying anything anyway.
1094 0 : async fn tenant_size_handler(
1095 0 : request: Request<Body>,
1096 0 : cancel: CancellationToken,
1097 0 : ) -> Result<Response<Body>, ApiError> {
1098 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1099 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1100 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1101 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1102 0 : let headers = request.headers();
1103 0 : let state = get_state(&request);
1104 0 :
1105 0 : if !tenant_shard_id.is_shard_zero() {
1106 0 : return Err(ApiError::BadRequest(anyhow!(
1107 0 : "Size calculations are only available on shard zero"
1108 0 : )));
1109 0 : }
1110 0 :
1111 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1112 0 : let tenant = state
1113 0 : .tenant_manager
1114 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1115 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1116 :
1117 : // this can be long operation
1118 0 : let inputs = tenant
1119 0 : .gather_size_inputs(
1120 0 : retention_period,
1121 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1122 0 : &cancel,
1123 0 : &ctx,
1124 0 : )
1125 0 : .await
1126 0 : .map_err(ApiError::InternalServerError)?;
1127 :
1128 0 : let mut sizes = None;
1129 0 : let accepts_html = headers
1130 0 : .get(header::ACCEPT)
1131 0 : .map(|v| v == "text/html")
1132 0 : .unwrap_or_default();
1133 0 : if !inputs_only.unwrap_or(false) {
1134 0 : let storage_model = inputs
1135 0 : .calculate_model()
1136 0 : .map_err(ApiError::InternalServerError)?;
1137 0 : let size = storage_model.calculate();
1138 0 :
1139 0 : // If request header expects html, return html
1140 0 : if accepts_html {
1141 0 : return synthetic_size_html_response(inputs, storage_model, size);
1142 0 : }
1143 0 : sizes = Some(size);
1144 0 : } else if accepts_html {
1145 0 : return Err(ApiError::BadRequest(anyhow!(
1146 0 : "inputs_only parameter is incompatible with html output request"
1147 0 : )));
1148 0 : }
1149 :
1150 : /// The type resides in the pageserver not to expose `ModelInputs`.
1151 : #[derive(serde::Serialize)]
1152 : struct TenantHistorySize {
1153 : id: TenantId,
1154 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1155 : ///
1156 : /// Will be none if `?inputs_only=true` was given.
1157 : size: Option<u64>,
1158 : /// Size of each segment used in the model.
1159 : /// Will be null if `?inputs_only=true` was given.
1160 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1161 : inputs: crate::tenant::size::ModelInputs,
1162 : }
1163 :
1164 0 : json_response(
1165 0 : StatusCode::OK,
1166 0 : TenantHistorySize {
1167 0 : id: tenant_shard_id.tenant_id,
1168 0 : size: sizes.as_ref().map(|x| x.total_size),
1169 0 : segment_sizes: sizes.map(|x| x.segments),
1170 0 : inputs,
1171 0 : },
1172 0 : )
1173 0 : }
1174 :
1175 0 : async fn tenant_shard_split_handler(
1176 0 : mut request: Request<Body>,
1177 0 : _cancel: CancellationToken,
1178 0 : ) -> Result<Response<Body>, ApiError> {
1179 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1180 :
1181 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1182 0 : let state = get_state(&request);
1183 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1184 :
1185 0 : let tenant = state
1186 0 : .tenant_manager
1187 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1188 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1189 :
1190 0 : let new_shards = state
1191 0 : .tenant_manager
1192 0 : .shard_split(
1193 0 : tenant,
1194 0 : ShardCount::new(req.new_shard_count),
1195 0 : req.new_stripe_size,
1196 0 : &ctx,
1197 0 : )
1198 0 : .await
1199 0 : .map_err(ApiError::InternalServerError)?;
1200 :
1201 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1202 0 : }
1203 :
1204 0 : async fn layer_map_info_handler(
1205 0 : request: Request<Body>,
1206 0 : _cancel: CancellationToken,
1207 0 : ) -> Result<Response<Body>, ApiError> {
1208 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1209 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1210 0 : let reset: LayerAccessStatsReset =
1211 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1212 0 : let state = get_state(&request);
1213 0 :
1214 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1215 :
1216 0 : let timeline =
1217 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1218 0 : .await?;
1219 0 : let layer_map_info = timeline.layer_map_info(reset).await;
1220 :
1221 0 : json_response(StatusCode::OK, layer_map_info)
1222 0 : }
1223 :
1224 0 : async fn layer_download_handler(
1225 0 : request: Request<Body>,
1226 0 : _cancel: CancellationToken,
1227 0 : ) -> Result<Response<Body>, ApiError> {
1228 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1229 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1230 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1231 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1232 0 : let layer_name = LayerName::from_str(layer_file_name)
1233 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1234 0 : let state = get_state(&request);
1235 :
1236 0 : let timeline =
1237 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1238 0 : .await?;
1239 0 : let downloaded = timeline
1240 0 : .download_layer(&layer_name)
1241 0 : .await
1242 0 : .map_err(ApiError::InternalServerError)?;
1243 :
1244 0 : match downloaded {
1245 0 : Some(true) => json_response(StatusCode::OK, ()),
1246 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1247 0 : None => json_response(
1248 0 : StatusCode::BAD_REQUEST,
1249 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1250 0 : ),
1251 : }
1252 0 : }
1253 :
1254 0 : async fn evict_timeline_layer_handler(
1255 0 : request: Request<Body>,
1256 0 : _cancel: CancellationToken,
1257 0 : ) -> Result<Response<Body>, ApiError> {
1258 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1259 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1260 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1261 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1262 0 : let state = get_state(&request);
1263 :
1264 0 : let layer_name = LayerName::from_str(layer_file_name)
1265 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1266 :
1267 0 : let timeline =
1268 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1269 0 : .await?;
1270 0 : let evicted = timeline
1271 0 : .evict_layer(&layer_name)
1272 0 : .await
1273 0 : .map_err(ApiError::InternalServerError)?;
1274 :
1275 0 : match evicted {
1276 0 : Some(true) => json_response(StatusCode::OK, ()),
1277 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1278 0 : None => json_response(
1279 0 : StatusCode::BAD_REQUEST,
1280 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1281 0 : ),
1282 : }
1283 0 : }
1284 :
1285 : /// Get tenant_size SVG graph along with the JSON data.
1286 0 : fn synthetic_size_html_response(
1287 0 : inputs: ModelInputs,
1288 0 : storage_model: StorageModel,
1289 0 : sizes: SizeResult,
1290 0 : ) -> Result<Response<Body>, ApiError> {
1291 0 : let mut timeline_ids: Vec<String> = Vec::new();
1292 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1293 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1294 0 : timeline_map.insert(ti.timeline_id, index);
1295 0 : timeline_ids.push(ti.timeline_id.to_string());
1296 0 : }
1297 0 : let seg_to_branch: Vec<usize> = inputs
1298 0 : .segments
1299 0 : .iter()
1300 0 : .map(|seg| *timeline_map.get(&seg.timeline_id).unwrap())
1301 0 : .collect();
1302 :
1303 0 : let svg =
1304 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1305 0 : .map_err(ApiError::InternalServerError)?;
1306 :
1307 0 : let mut response = String::new();
1308 0 :
1309 0 : use std::fmt::Write;
1310 0 : write!(response, "<html>\n<body>\n").unwrap();
1311 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1312 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1313 0 : writeln!(response, "<pre>").unwrap();
1314 0 : writeln!(
1315 0 : response,
1316 0 : "{}",
1317 0 : serde_json::to_string_pretty(&inputs).unwrap()
1318 0 : )
1319 0 : .unwrap();
1320 0 : writeln!(
1321 0 : response,
1322 0 : "{}",
1323 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1324 0 : )
1325 0 : .unwrap();
1326 0 : writeln!(response, "</pre>").unwrap();
1327 0 : write!(response, "</body>\n</html>\n").unwrap();
1328 0 :
1329 0 : html_response(StatusCode::OK, response)
1330 0 : }
1331 :
1332 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1333 0 : let response = Response::builder()
1334 0 : .status(status)
1335 0 : .header(header::CONTENT_TYPE, "text/html")
1336 0 : .body(Body::from(data.as_bytes().to_vec()))
1337 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1338 0 : Ok(response)
1339 0 : }
1340 :
1341 : /// Helper for requests that may take a generation, which is mandatory
1342 : /// when control_plane_api is set, but otherwise defaults to Generation::none()
1343 0 : fn get_request_generation(state: &State, req_gen: Option<u32>) -> Result<Generation, ApiError> {
1344 0 : if state.conf.control_plane_api.is_some() {
1345 0 : req_gen
1346 0 : .map(Generation::new)
1347 0 : .ok_or(ApiError::BadRequest(anyhow!(
1348 0 : "generation attribute missing"
1349 0 : )))
1350 : } else {
1351 : // Legacy mode: all tenants operate with no generation
1352 0 : Ok(Generation::none())
1353 : }
1354 0 : }
1355 :
1356 0 : async fn tenant_create_handler(
1357 0 : mut request: Request<Body>,
1358 0 : _cancel: CancellationToken,
1359 0 : ) -> Result<Response<Body>, ApiError> {
1360 0 : let request_data: TenantCreateRequest = json_request(&mut request).await?;
1361 0 : let target_tenant_id = request_data.new_tenant_id;
1362 0 : check_permission(&request, None)?;
1363 :
1364 0 : let _timer = STORAGE_TIME_GLOBAL
1365 0 : .get_metric_with_label_values(&[StorageTimeOperation::CreateTenant.into()])
1366 0 : .expect("bug")
1367 0 : .start_timer();
1368 :
1369 0 : let tenant_conf =
1370 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1371 :
1372 0 : let state = get_state(&request);
1373 :
1374 0 : let generation = get_request_generation(state, request_data.generation)?;
1375 :
1376 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1377 0 :
1378 0 : let location_conf =
1379 0 : LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
1380 :
1381 0 : let new_tenant = state
1382 0 : .tenant_manager
1383 0 : .upsert_location(
1384 0 : target_tenant_id,
1385 0 : location_conf,
1386 0 : None,
1387 0 : SpawnMode::Create,
1388 0 : &ctx,
1389 0 : )
1390 0 : .await?;
1391 :
1392 0 : let Some(new_tenant) = new_tenant else {
1393 : // This should never happen: indicates a bug in upsert_location
1394 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
1395 0 : "Upsert succeeded but didn't return tenant!"
1396 0 : )));
1397 : };
1398 : // We created the tenant. Existing API semantics are that the tenant
1399 : // is Active when this function returns.
1400 0 : new_tenant
1401 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1402 0 : .await?;
1403 :
1404 0 : json_response(
1405 0 : StatusCode::CREATED,
1406 0 : TenantCreateResponse(new_tenant.tenant_shard_id().tenant_id),
1407 0 : )
1408 0 : }
1409 :
1410 0 : async fn get_tenant_config_handler(
1411 0 : request: Request<Body>,
1412 0 : _cancel: CancellationToken,
1413 0 : ) -> Result<Response<Body>, ApiError> {
1414 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1415 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1416 0 : let state = get_state(&request);
1417 :
1418 0 : let tenant = state
1419 0 : .tenant_manager
1420 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1421 :
1422 0 : let response = HashMap::from([
1423 : (
1424 : "tenant_specific_overrides",
1425 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1426 0 : .context("serializing tenant specific overrides")
1427 0 : .map_err(ApiError::InternalServerError)?,
1428 : ),
1429 : (
1430 0 : "effective_config",
1431 0 : serde_json::to_value(tenant.effective_config())
1432 0 : .context("serializing effective config")
1433 0 : .map_err(ApiError::InternalServerError)?,
1434 : ),
1435 : ]);
1436 :
1437 0 : json_response(StatusCode::OK, response)
1438 0 : }
1439 :
1440 0 : async fn update_tenant_config_handler(
1441 0 : mut request: Request<Body>,
1442 0 : _cancel: CancellationToken,
1443 0 : ) -> Result<Response<Body>, ApiError> {
1444 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1445 0 : let tenant_id = request_data.tenant_id;
1446 0 : check_permission(&request, Some(tenant_id))?;
1447 :
1448 0 : let new_tenant_conf =
1449 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1450 :
1451 0 : let state = get_state(&request);
1452 0 :
1453 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1454 :
1455 0 : let tenant = state
1456 0 : .tenant_manager
1457 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1458 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1459 :
1460 : // This is a legacy API that only operates on attached tenants: the preferred
1461 : // API to use is the location_config/ endpoint, which lets the caller provide
1462 : // the full LocationConf.
1463 0 : let location_conf = LocationConf::attached_single(
1464 0 : new_tenant_conf.clone(),
1465 0 : tenant.get_generation(),
1466 0 : &ShardParameters::default(),
1467 0 : );
1468 0 :
1469 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1470 0 : .await
1471 0 : .map_err(ApiError::InternalServerError)?;
1472 0 : tenant.set_new_tenant_config(new_tenant_conf);
1473 0 :
1474 0 : json_response(StatusCode::OK, ())
1475 0 : }
1476 :
1477 0 : async fn put_tenant_location_config_handler(
1478 0 : mut request: Request<Body>,
1479 0 : _cancel: CancellationToken,
1480 0 : ) -> Result<Response<Body>, ApiError> {
1481 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1482 :
1483 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1484 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1485 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1486 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1487 :
1488 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1489 0 : let state = get_state(&request);
1490 0 : let conf = state.conf;
1491 0 :
1492 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1493 0 : // its local disk content and drops it from memory.
1494 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1495 0 : if let Err(e) = state
1496 0 : .tenant_manager
1497 0 : .detach_tenant(conf, tenant_shard_id, true, &state.deletion_queue_client)
1498 0 : .instrument(info_span!("tenant_detach",
1499 : tenant_id = %tenant_shard_id.tenant_id,
1500 0 : shard_id = %tenant_shard_id.shard_slug()
1501 : ))
1502 0 : .await
1503 : {
1504 0 : match e {
1505 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1506 0 : // This API is idempotent: a NotFound on a detach is fine.
1507 0 : }
1508 0 : _ => return Err(e.into()),
1509 : }
1510 0 : }
1511 0 : return json_response(StatusCode::OK, ());
1512 0 : }
1513 :
1514 0 : let location_conf =
1515 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1516 :
1517 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1518 : // similar to at startup ordering.
1519 0 : let spawn_mode = if lazy {
1520 0 : tenant::SpawnMode::Lazy
1521 : } else {
1522 0 : tenant::SpawnMode::Eager
1523 : };
1524 :
1525 0 : let tenant = state
1526 0 : .tenant_manager
1527 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1528 0 : .await?;
1529 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1530 0 : let attached = tenant.is_some();
1531 :
1532 0 : if let Some(_flush_ms) = flush {
1533 0 : match state
1534 0 : .secondary_controller
1535 0 : .upload_tenant(tenant_shard_id)
1536 0 : .await
1537 : {
1538 : Ok(()) => {
1539 0 : tracing::info!("Uploaded heatmap during flush");
1540 : }
1541 0 : Err(e) => {
1542 0 : tracing::warn!("Failed to flush heatmap: {e}");
1543 : }
1544 : }
1545 : } else {
1546 0 : tracing::info!("No flush requested when configuring");
1547 : }
1548 :
1549 : // This API returns a vector of pageservers where the tenant is attached: this is
1550 : // primarily for use in the sharding service. For compatibilty, we also return this
1551 : // when called directly on a pageserver, but the payload is always zero or one shards.
1552 0 : let mut response = TenantLocationConfigResponse {
1553 0 : shards: Vec::new(),
1554 0 : stripe_size: None,
1555 0 : };
1556 0 : if attached {
1557 0 : response.shards.push(TenantShardLocation {
1558 0 : shard_id: tenant_shard_id,
1559 0 : node_id: state.conf.id,
1560 0 : });
1561 0 : if tenant_shard_id.shard_count.count() > 1 {
1562 : // Stripe size should be set if we are attached
1563 0 : debug_assert!(stripe_size.is_some());
1564 0 : response.stripe_size = stripe_size;
1565 0 : }
1566 0 : }
1567 :
1568 0 : json_response(StatusCode::OK, response)
1569 0 : }
1570 :
1571 0 : async fn list_location_config_handler(
1572 0 : request: Request<Body>,
1573 0 : _cancel: CancellationToken,
1574 0 : ) -> Result<Response<Body>, ApiError> {
1575 0 : let state = get_state(&request);
1576 0 : let slots = state.tenant_manager.list();
1577 0 : let result = LocationConfigListResponse {
1578 0 : tenant_shards: slots
1579 0 : .into_iter()
1580 0 : .map(|(tenant_shard_id, slot)| {
1581 0 : let v = match slot {
1582 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1583 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1584 0 : TenantSlot::InProgress(_) => None,
1585 : };
1586 0 : (tenant_shard_id, v)
1587 0 : })
1588 0 : .collect(),
1589 0 : };
1590 0 : json_response(StatusCode::OK, result)
1591 0 : }
1592 :
1593 0 : async fn get_location_config_handler(
1594 0 : request: Request<Body>,
1595 0 : _cancel: CancellationToken,
1596 0 : ) -> Result<Response<Body>, ApiError> {
1597 0 : let state = get_state(&request);
1598 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1599 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1600 :
1601 0 : let Some(slot) = slot else {
1602 0 : return Err(ApiError::NotFound(
1603 0 : anyhow::anyhow!("Tenant shard not found").into(),
1604 0 : ));
1605 : };
1606 :
1607 0 : let result: Option<LocationConfig> = match slot {
1608 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1609 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1610 0 : TenantSlot::InProgress(_) => None,
1611 : };
1612 :
1613 0 : json_response(StatusCode::OK, result)
1614 0 : }
1615 :
1616 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1617 : // (from all pageservers) as it invalidates consistency assumptions.
1618 0 : async fn tenant_time_travel_remote_storage_handler(
1619 0 : request: Request<Body>,
1620 0 : cancel: CancellationToken,
1621 0 : ) -> Result<Response<Body>, ApiError> {
1622 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1623 :
1624 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1625 :
1626 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1627 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1628 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1629 0 : .map_err(ApiError::BadRequest)?;
1630 :
1631 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1632 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1633 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1634 0 : .map_err(ApiError::BadRequest)?;
1635 :
1636 : // This is just a sanity check to fend off naive wrong usages of the API:
1637 : // the tenant needs to be detached *everywhere*
1638 0 : let state = get_state(&request);
1639 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1640 0 : if we_manage_tenant {
1641 0 : return Err(ApiError::BadRequest(anyhow!(
1642 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1643 0 : )));
1644 0 : }
1645 :
1646 0 : let Some(storage) = state.remote_storage.as_ref() else {
1647 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
1648 0 : "remote storage not configured, cannot run time travel"
1649 0 : )));
1650 : };
1651 :
1652 0 : if timestamp > done_if_after {
1653 0 : return Err(ApiError::BadRequest(anyhow!(
1654 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1655 0 : )));
1656 0 : }
1657 0 :
1658 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1659 :
1660 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1661 0 : storage,
1662 0 : &tenant_shard_id,
1663 0 : timestamp,
1664 0 : done_if_after,
1665 0 : &cancel,
1666 0 : )
1667 0 : .await
1668 0 : .map_err(|e| match e {
1669 0 : TimeTravelError::BadInput(e) => {
1670 0 : warn!("bad input error: {e}");
1671 0 : ApiError::BadRequest(anyhow!("bad input error"))
1672 : }
1673 : TimeTravelError::Unimplemented => {
1674 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1675 : }
1676 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1677 : TimeTravelError::TooManyVersions => {
1678 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1679 : }
1680 0 : TimeTravelError::Other(e) => {
1681 0 : warn!("internal error: {e}");
1682 0 : ApiError::InternalServerError(anyhow!("internal error"))
1683 : }
1684 0 : })?;
1685 :
1686 0 : json_response(StatusCode::OK, ())
1687 0 : }
1688 :
1689 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1690 0 : async fn handle_tenant_break(
1691 0 : r: Request<Body>,
1692 0 : _cancel: CancellationToken,
1693 0 : ) -> Result<Response<Body>, ApiError> {
1694 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1695 :
1696 0 : let state = get_state(&r);
1697 0 : state
1698 0 : .tenant_manager
1699 0 : .get_attached_tenant_shard(tenant_shard_id)?
1700 0 : .set_broken("broken from test".to_owned())
1701 0 : .await;
1702 :
1703 0 : json_response(StatusCode::OK, ())
1704 0 : }
1705 :
1706 : // Run GC immediately on given timeline.
1707 0 : async fn timeline_gc_handler(
1708 0 : mut request: Request<Body>,
1709 0 : cancel: CancellationToken,
1710 0 : ) -> Result<Response<Body>, ApiError> {
1711 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1712 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1713 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1714 :
1715 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1716 :
1717 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1718 0 : let wait_task_done = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)?;
1719 0 : let gc_result = wait_task_done
1720 0 : .await
1721 0 : .context("wait for gc task")
1722 0 : .map_err(ApiError::InternalServerError)?
1723 0 : .map_err(ApiError::InternalServerError)?;
1724 :
1725 0 : json_response(StatusCode::OK, gc_result)
1726 0 : }
1727 :
1728 : // Run compaction immediately on given timeline.
1729 0 : async fn timeline_compact_handler(
1730 0 : request: Request<Body>,
1731 0 : cancel: CancellationToken,
1732 0 : ) -> Result<Response<Body>, ApiError> {
1733 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1734 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1735 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1736 :
1737 0 : let state = get_state(&request);
1738 0 :
1739 0 : let mut flags = EnumSet::empty();
1740 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1741 0 : flags |= CompactFlags::ForceRepartition;
1742 0 : }
1743 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1744 0 : flags |= CompactFlags::ForceImageLayerCreation;
1745 0 : }
1746 :
1747 0 : async {
1748 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1749 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1750 0 : timeline
1751 0 : .compact(&cancel, flags, &ctx)
1752 0 : .await
1753 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1754 0 : json_response(StatusCode::OK, ())
1755 0 : }
1756 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1757 0 : .await
1758 0 : }
1759 :
1760 : // Run checkpoint immediately on given timeline.
1761 0 : async fn timeline_checkpoint_handler(
1762 0 : request: Request<Body>,
1763 0 : cancel: CancellationToken,
1764 0 : ) -> Result<Response<Body>, ApiError> {
1765 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1766 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1767 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1768 :
1769 0 : let state = get_state(&request);
1770 0 :
1771 0 : let mut flags = EnumSet::empty();
1772 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1773 0 : flags |= CompactFlags::ForceRepartition;
1774 0 : }
1775 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1776 0 : flags |= CompactFlags::ForceImageLayerCreation;
1777 0 : }
1778 :
1779 0 : async {
1780 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1781 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1782 0 : timeline
1783 0 : .freeze_and_flush()
1784 0 : .await
1785 0 : .map_err(ApiError::InternalServerError)?;
1786 0 : timeline
1787 0 : .compact(&cancel, flags, &ctx)
1788 0 : .await
1789 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1790 :
1791 0 : json_response(StatusCode::OK, ())
1792 0 : }
1793 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1794 0 : .await
1795 0 : }
1796 :
1797 0 : async fn timeline_download_remote_layers_handler_post(
1798 0 : mut request: Request<Body>,
1799 0 : _cancel: CancellationToken,
1800 0 : ) -> Result<Response<Body>, ApiError> {
1801 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1802 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1803 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
1804 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1805 :
1806 0 : let state = get_state(&request);
1807 :
1808 0 : let timeline =
1809 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1810 0 : .await?;
1811 0 : match timeline.spawn_download_all_remote_layers(body).await {
1812 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
1813 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
1814 : }
1815 0 : }
1816 :
1817 0 : async fn timeline_download_remote_layers_handler_get(
1818 0 : request: Request<Body>,
1819 0 : _cancel: CancellationToken,
1820 0 : ) -> Result<Response<Body>, ApiError> {
1821 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1822 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1823 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1824 0 : let state = get_state(&request);
1825 :
1826 0 : let timeline =
1827 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1828 0 : .await?;
1829 0 : let info = timeline
1830 0 : .get_download_all_remote_layers_task_info()
1831 0 : .context("task never started since last pageserver process start")
1832 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1833 0 : json_response(StatusCode::OK, info)
1834 0 : }
1835 :
1836 0 : async fn timeline_detach_ancestor_handler(
1837 0 : request: Request<Body>,
1838 0 : _cancel: CancellationToken,
1839 0 : ) -> Result<Response<Body>, ApiError> {
1840 : use crate::tenant::timeline::detach_ancestor::Options;
1841 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1842 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1843 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1844 :
1845 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
1846 :
1847 0 : async move {
1848 0 : let mut options = Options::default();
1849 :
1850 0 : let rewrite_concurrency =
1851 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
1852 0 : let copy_concurrency =
1853 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
1854 :
1855 0 : [
1856 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
1857 0 : (&mut options.copy_concurrency, copy_concurrency),
1858 0 : ]
1859 0 : .into_iter()
1860 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
1861 0 : .for_each(|(target, val)| *target = val);
1862 0 :
1863 0 : let state = get_state(&request);
1864 :
1865 0 : let tenant = state
1866 0 : .tenant_manager
1867 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1868 :
1869 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1870 :
1871 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
1872 0 : let ctx = &ctx;
1873 :
1874 0 : let timeline = tenant
1875 0 : .get_timeline(timeline_id, true)
1876 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1877 :
1878 0 : let (_guard, prepared) = timeline
1879 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
1880 0 : .await
1881 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1882 :
1883 0 : let res = state
1884 0 : .tenant_manager
1885 0 : .complete_detaching_timeline_ancestor(tenant_shard_id, timeline_id, prepared, ctx)
1886 0 : .await;
1887 :
1888 0 : match res {
1889 0 : Ok(reparented_timelines) => {
1890 0 : let resp = pageserver_api::models::detach_ancestor::AncestorDetached {
1891 0 : reparented_timelines,
1892 0 : };
1893 0 :
1894 0 : json_response(StatusCode::OK, resp)
1895 : }
1896 0 : Err(e) => Err(ApiError::InternalServerError(
1897 0 : e.context("timeline detach completion"),
1898 0 : )),
1899 : }
1900 0 : }
1901 0 : .instrument(span)
1902 0 : .await
1903 0 : }
1904 :
1905 0 : async fn deletion_queue_flush(
1906 0 : r: Request<Body>,
1907 0 : cancel: CancellationToken,
1908 0 : ) -> Result<Response<Body>, ApiError> {
1909 0 : let state = get_state(&r);
1910 0 :
1911 0 : if state.remote_storage.is_none() {
1912 : // Nothing to do if remote storage is disabled.
1913 0 : return json_response(StatusCode::OK, ());
1914 0 : }
1915 :
1916 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
1917 0 :
1918 0 : let flush = async {
1919 0 : if execute {
1920 0 : state.deletion_queue_client.flush_execute().await
1921 : } else {
1922 0 : state.deletion_queue_client.flush().await
1923 : }
1924 0 : }
1925 : // DeletionQueueError's only case is shutting down.
1926 0 : .map_err(|_| ApiError::ShuttingDown);
1927 :
1928 : tokio::select! {
1929 : res = flush => {
1930 0 : res.map(|()| json_response(StatusCode::OK, ()))?
1931 : }
1932 : _ = cancel.cancelled() => {
1933 : Err(ApiError::ShuttingDown)
1934 : }
1935 : }
1936 0 : }
1937 :
1938 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
1939 0 : async fn getpage_at_lsn_handler(
1940 0 : request: Request<Body>,
1941 0 : _cancel: CancellationToken,
1942 0 : ) -> Result<Response<Body>, ApiError> {
1943 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1944 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1945 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1946 0 : let state = get_state(&request);
1947 :
1948 : struct Key(crate::repository::Key);
1949 :
1950 : impl std::str::FromStr for Key {
1951 : type Err = anyhow::Error;
1952 :
1953 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
1954 0 : crate::repository::Key::from_hex(s).map(Key)
1955 0 : }
1956 : }
1957 :
1958 0 : let key: Key = parse_query_param(&request, "key")?
1959 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
1960 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1961 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1962 :
1963 0 : async {
1964 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1965 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1966 :
1967 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
1968 :
1969 0 : Result::<_, ApiError>::Ok(
1970 0 : Response::builder()
1971 0 : .status(StatusCode::OK)
1972 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
1973 0 : .body(hyper::Body::from(page))
1974 0 : .unwrap(),
1975 0 : )
1976 0 : }
1977 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1978 0 : .await
1979 0 : }
1980 :
1981 0 : async fn timeline_collect_keyspace(
1982 0 : request: Request<Body>,
1983 0 : _cancel: CancellationToken,
1984 0 : ) -> Result<Response<Body>, ApiError> {
1985 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1986 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1987 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1988 0 : let state = get_state(&request);
1989 :
1990 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
1991 :
1992 0 : async {
1993 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1994 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1995 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
1996 0 : let (dense_ks, sparse_ks) = timeline
1997 0 : .collect_keyspace(at_lsn, &ctx)
1998 0 : .await
1999 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2000 :
2001 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2002 : // Therefore, we split dense/sparse keys in this API.
2003 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2004 0 :
2005 0 : json_response(StatusCode::OK, res)
2006 0 : }
2007 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2008 0 : .await
2009 0 : }
2010 :
2011 0 : async fn active_timeline_of_active_tenant(
2012 0 : tenant_manager: &TenantManager,
2013 0 : tenant_shard_id: TenantShardId,
2014 0 : timeline_id: TimelineId,
2015 0 : ) -> Result<Arc<Timeline>, ApiError> {
2016 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2017 :
2018 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2019 :
2020 0 : tenant
2021 0 : .get_timeline(timeline_id, true)
2022 0 : .map_err(|e| ApiError::NotFound(e.into()))
2023 0 : }
2024 :
2025 0 : async fn always_panic_handler(
2026 0 : req: Request<Body>,
2027 0 : _cancel: CancellationToken,
2028 0 : ) -> Result<Response<Body>, ApiError> {
2029 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2030 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2031 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2032 0 : let query = req.uri().query();
2033 0 : let _ = std::panic::catch_unwind(|| {
2034 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2035 0 : });
2036 0 : json_response(StatusCode::NO_CONTENT, ())
2037 0 : }
2038 :
2039 0 : async fn disk_usage_eviction_run(
2040 0 : mut r: Request<Body>,
2041 0 : cancel: CancellationToken,
2042 0 : ) -> Result<Response<Body>, ApiError> {
2043 0 : check_permission(&r, None)?;
2044 :
2045 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2046 : struct Config {
2047 : /// How many bytes to evict before reporting that pressure is relieved.
2048 : evict_bytes: u64,
2049 :
2050 : #[serde(default)]
2051 : eviction_order: crate::disk_usage_eviction_task::EvictionOrder,
2052 : }
2053 :
2054 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2055 : struct Usage {
2056 : // remains unchanged after instantiation of the struct
2057 : evict_bytes: u64,
2058 : // updated by `add_available_bytes`
2059 : freed_bytes: u64,
2060 : }
2061 :
2062 : impl crate::disk_usage_eviction_task::Usage for Usage {
2063 0 : fn has_pressure(&self) -> bool {
2064 0 : self.evict_bytes > self.freed_bytes
2065 0 : }
2066 :
2067 0 : fn add_available_bytes(&mut self, bytes: u64) {
2068 0 : self.freed_bytes += bytes;
2069 0 : }
2070 : }
2071 :
2072 0 : let config = json_request::<Config>(&mut r).await?;
2073 :
2074 0 : let usage = Usage {
2075 0 : evict_bytes: config.evict_bytes,
2076 0 : freed_bytes: 0,
2077 0 : };
2078 0 :
2079 0 : let state = get_state(&r);
2080 :
2081 0 : let Some(storage) = state.remote_storage.as_ref() else {
2082 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
2083 0 : "remote storage not configured, cannot run eviction iteration"
2084 0 : )));
2085 : };
2086 :
2087 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2088 :
2089 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2090 0 : &eviction_state,
2091 0 : storage,
2092 0 : usage,
2093 0 : &state.tenant_manager,
2094 0 : config.eviction_order,
2095 0 : &cancel,
2096 0 : )
2097 0 : .await;
2098 :
2099 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2100 :
2101 0 : let res = res.map_err(ApiError::InternalServerError)?;
2102 :
2103 0 : json_response(StatusCode::OK, res)
2104 0 : }
2105 :
2106 0 : async fn secondary_upload_handler(
2107 0 : request: Request<Body>,
2108 0 : _cancel: CancellationToken,
2109 0 : ) -> Result<Response<Body>, ApiError> {
2110 0 : let state = get_state(&request);
2111 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2112 0 : state
2113 0 : .secondary_controller
2114 0 : .upload_tenant(tenant_shard_id)
2115 0 : .await
2116 0 : .map_err(ApiError::InternalServerError)?;
2117 :
2118 0 : json_response(StatusCode::OK, ())
2119 0 : }
2120 :
2121 0 : async fn tenant_scan_remote_handler(
2122 0 : request: Request<Body>,
2123 0 : cancel: CancellationToken,
2124 0 : ) -> Result<Response<Body>, ApiError> {
2125 0 : let state = get_state(&request);
2126 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2127 :
2128 0 : let Some(remote_storage) = state.remote_storage.as_ref() else {
2129 0 : return Err(ApiError::BadRequest(anyhow::anyhow!(
2130 0 : "Remote storage not configured"
2131 0 : )));
2132 : };
2133 :
2134 0 : let mut response = TenantScanRemoteStorageResponse::default();
2135 :
2136 0 : let (shards, _other_keys) =
2137 0 : list_remote_tenant_shards(remote_storage, tenant_id, cancel.clone())
2138 0 : .await
2139 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2140 :
2141 0 : for tenant_shard_id in shards {
2142 0 : let (timeline_ids, _other_keys) =
2143 0 : list_remote_timelines(remote_storage, tenant_shard_id, cancel.clone())
2144 0 : .await
2145 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2146 :
2147 0 : let mut generation = Generation::none();
2148 0 : for timeline_id in timeline_ids {
2149 0 : match download_index_part(
2150 0 : remote_storage,
2151 0 : &tenant_shard_id,
2152 0 : &timeline_id,
2153 0 : Generation::MAX,
2154 0 : &cancel,
2155 0 : )
2156 0 : .instrument(info_span!("download_index_part",
2157 : tenant_id=%tenant_shard_id.tenant_id,
2158 0 : shard_id=%tenant_shard_id.shard_slug(),
2159 : %timeline_id))
2160 0 : .await
2161 : {
2162 0 : Ok((index_part, index_generation)) => {
2163 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2164 0 : index_part.layer_metadata.len(), index_part.get_disk_consistent_lsn());
2165 0 : generation = std::cmp::max(generation, index_generation);
2166 : }
2167 : Err(DownloadError::NotFound) => {
2168 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2169 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2170 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2171 0 : continue;
2172 : }
2173 0 : Err(e) => {
2174 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2175 : }
2176 : };
2177 : }
2178 :
2179 0 : response.shards.push(TenantScanRemoteStorageShard {
2180 0 : tenant_shard_id,
2181 0 : generation: generation.into(),
2182 0 : });
2183 : }
2184 :
2185 0 : if response.shards.is_empty() {
2186 0 : return Err(ApiError::NotFound(
2187 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2188 0 : ));
2189 0 : }
2190 0 :
2191 0 : json_response(StatusCode::OK, response)
2192 0 : }
2193 :
2194 0 : async fn secondary_download_handler(
2195 0 : request: Request<Body>,
2196 0 : _cancel: CancellationToken,
2197 0 : ) -> Result<Response<Body>, ApiError> {
2198 0 : let state = get_state(&request);
2199 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2200 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2201 :
2202 : // We don't need this to issue the download request, but:
2203 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2204 : // - we will use this to provide status feedback in the response
2205 0 : let Some(secondary_tenant) = state
2206 0 : .tenant_manager
2207 0 : .get_secondary_tenant_shard(tenant_shard_id)
2208 : else {
2209 0 : return Err(ApiError::NotFound(
2210 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2211 0 : ));
2212 : };
2213 :
2214 0 : let timeout = wait.unwrap_or(Duration::MAX);
2215 :
2216 0 : let status = match tokio::time::timeout(
2217 0 : timeout,
2218 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2219 0 : )
2220 0 : .await
2221 : {
2222 : // Download job ran to completion.
2223 0 : Ok(Ok(())) => StatusCode::OK,
2224 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2225 : // okay. We could get an error here in the unlikely edge case that the tenant
2226 : // was detached between our check above and executing the download job.
2227 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2228 : // A timeout is not an error: we have started the download, we're just not done
2229 : // yet. The caller will get a response body indicating status.
2230 0 : Err(_) => StatusCode::ACCEPTED,
2231 : };
2232 :
2233 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2234 0 :
2235 0 : json_response(status, progress)
2236 0 : }
2237 :
2238 0 : async fn secondary_status_handler(
2239 0 : request: Request<Body>,
2240 0 : _cancel: CancellationToken,
2241 0 : ) -> Result<Response<Body>, ApiError> {
2242 0 : let state = get_state(&request);
2243 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2244 :
2245 0 : let Some(secondary_tenant) = state
2246 0 : .tenant_manager
2247 0 : .get_secondary_tenant_shard(tenant_shard_id)
2248 : else {
2249 0 : return Err(ApiError::NotFound(
2250 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2251 0 : ));
2252 : };
2253 :
2254 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2255 0 :
2256 0 : json_response(StatusCode::OK, progress)
2257 0 : }
2258 :
2259 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2260 0 : json_response(
2261 0 : StatusCode::NOT_FOUND,
2262 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2263 0 : )
2264 0 : }
2265 :
2266 0 : async fn post_tracing_event_handler(
2267 0 : mut r: Request<Body>,
2268 0 : _cancel: CancellationToken,
2269 0 : ) -> Result<Response<Body>, ApiError> {
2270 0 : #[derive(Debug, serde::Deserialize)]
2271 : #[serde(rename_all = "lowercase")]
2272 : enum Level {
2273 : Error,
2274 : Warn,
2275 : Info,
2276 : Debug,
2277 : Trace,
2278 : }
2279 0 : #[derive(Debug, serde::Deserialize)]
2280 : struct Request {
2281 : level: Level,
2282 : message: String,
2283 : }
2284 0 : let body: Request = json_request(&mut r)
2285 0 : .await
2286 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2287 :
2288 0 : match body.level {
2289 0 : Level::Error => tracing::error!(?body.message),
2290 0 : Level::Warn => tracing::warn!(?body.message),
2291 0 : Level::Info => tracing::info!(?body.message),
2292 0 : Level::Debug => tracing::debug!(?body.message),
2293 0 : Level::Trace => tracing::trace!(?body.message),
2294 : }
2295 :
2296 0 : json_response(StatusCode::OK, ())
2297 0 : }
2298 :
2299 0 : async fn put_io_engine_handler(
2300 0 : mut r: Request<Body>,
2301 0 : _cancel: CancellationToken,
2302 0 : ) -> Result<Response<Body>, ApiError> {
2303 0 : check_permission(&r, None)?;
2304 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2305 0 : crate::virtual_file::io_engine::set(kind);
2306 0 : json_response(StatusCode::OK, ())
2307 0 : }
2308 :
2309 : /// Polled by control plane.
2310 : ///
2311 : /// See [`crate::utilization`].
2312 0 : async fn get_utilization(
2313 0 : r: Request<Body>,
2314 0 : _cancel: CancellationToken,
2315 0 : ) -> Result<Response<Body>, ApiError> {
2316 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2317 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2318 0 : });
2319 :
2320 : // this probably could be completely public, but lets make that change later.
2321 0 : check_permission(&r, None)?;
2322 :
2323 0 : let state = get_state(&r);
2324 0 : let mut g = state.latest_utilization.lock().await;
2325 :
2326 0 : let regenerate_every = Duration::from_secs(1);
2327 0 : let still_valid = g
2328 0 : .as_ref()
2329 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2330 0 :
2331 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2332 0 : // regenerate at most 1Hz to allow polling at any rate.
2333 0 : if !still_valid {
2334 0 : let path = state.conf.tenants_path();
2335 0 : let doc = crate::utilization::regenerate(path.as_std_path())
2336 0 : .map_err(ApiError::InternalServerError)?;
2337 :
2338 0 : let mut buf = Vec::new();
2339 0 : serde_json::to_writer(&mut buf, &doc)
2340 0 : .context("serialize")
2341 0 : .map_err(ApiError::InternalServerError)?;
2342 :
2343 0 : let body = bytes::Bytes::from(buf);
2344 0 :
2345 0 : *g = Some((std::time::Instant::now(), body));
2346 0 : }
2347 :
2348 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2349 0 : let cached = g.as_ref().expect("just set").1.clone();
2350 0 :
2351 0 : Response::builder()
2352 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2353 0 : // thought of using http date header, but that is second precision which does not give any
2354 0 : // debugging aid
2355 0 : .status(StatusCode::OK)
2356 0 : .body(hyper::Body::from(cached))
2357 0 : .context("build response")
2358 0 : .map_err(ApiError::InternalServerError)
2359 0 : }
2360 :
2361 : /// Common functionality of all the HTTP API handlers.
2362 : ///
2363 : /// - Adds a tracing span to each request (by `request_span`)
2364 : /// - Logs the request depending on the request method (by `request_span`)
2365 : /// - Logs the response if it was not successful (by `request_span`
2366 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
2367 : /// Future if the connection to the client is lost, but most of the pageserver code is
2368 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
2369 : /// request with a CancellationToken.
2370 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
2371 0 : where
2372 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2373 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2374 0 : {
2375 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
2376 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
2377 0 : "failpoint".into()
2378 0 : )));
2379 :
2380 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
2381 0 : anyhow::anyhow!("failpoint")
2382 0 : )));
2383 0 : }
2384 :
2385 : // Spawn a new task to handle the request, to protect the handler from unexpected
2386 : // async cancellations. Most pageserver functions are not async cancellation safe.
2387 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
2388 : // with the cancellation token.
2389 0 : let token = CancellationToken::new();
2390 0 : let cancel_guard = token.clone().drop_guard();
2391 0 : let result = request_span(request, move |r| async {
2392 0 : let handle = tokio::spawn(
2393 0 : async {
2394 0 : let token_cloned = token.clone();
2395 0 : let result = handler(r, token).await;
2396 0 : if token_cloned.is_cancelled() {
2397 0 : // dropguard has executed: we will never turn this result into response.
2398 0 : //
2399 0 : // at least temporarily do {:?} logging; these failures are rare enough but
2400 0 : // could hide difficult errors.
2401 0 : match &result {
2402 0 : Ok(response) => {
2403 0 : let status = response.status();
2404 0 : info!(%status, "Cancelled request finished successfully")
2405 0 : }
2406 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
2407 0 : }
2408 0 : }
2409 0 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
2410 0 : // which should suffice.
2411 0 : //
2412 0 : // there is still a chance to lose the result due to race between
2413 0 : // returning from here and the actual connection closing happening
2414 0 : // before outer task gets to execute. leaving that up for #5815.
2415 0 : result
2416 0 : }
2417 0 : .in_current_span(),
2418 0 : );
2419 0 :
2420 0 : match handle.await {
2421 0 : // TODO: never actually return Err from here, always Ok(...) so that we can log
2422 0 : // spanned errors. Call api_error_handler instead and return appropriate Body.
2423 0 : Ok(result) => result,
2424 0 : Err(e) => {
2425 0 : // The handler task panicked. We have a global panic handler that logs the
2426 0 : // panic with its backtrace, so no need to log that here. Only log a brief
2427 0 : // message to make it clear that we returned the error to the client.
2428 0 : error!("HTTP request handler task panicked: {e:#}");
2429 0 :
2430 0 : // Don't return an Error here, because then fallback error handler that was
2431 0 : // installed in make_router() will print the error. Instead, construct the
2432 0 : // HTTP error response and return that.
2433 0 : Ok(
2434 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
2435 0 : .into_response(),
2436 0 : )
2437 0 : }
2438 0 : }
2439 0 : })
2440 0 : .await;
2441 :
2442 0 : cancel_guard.disarm();
2443 0 :
2444 0 : result
2445 0 : }
2446 :
2447 : /// Like api_handler, but returns an error response if the server is built without
2448 : /// the 'testing' feature.
2449 0 : async fn testing_api_handler<R, H>(
2450 0 : desc: &str,
2451 0 : request: Request<Body>,
2452 0 : handler: H,
2453 0 : ) -> Result<Response<Body>, ApiError>
2454 0 : where
2455 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2456 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2457 0 : {
2458 0 : if cfg!(feature = "testing") {
2459 0 : api_handler(request, handler).await
2460 : } else {
2461 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
2462 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
2463 0 : ))))
2464 0 : .await
2465 : }
2466 0 : }
2467 :
2468 0 : pub fn make_router(
2469 0 : state: Arc<State>,
2470 0 : launch_ts: &'static LaunchTimestamp,
2471 0 : auth: Option<Arc<SwappableJwtAuth>>,
2472 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
2473 0 : let spec = include_bytes!("openapi_spec.yml");
2474 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
2475 0 : if auth.is_some() {
2476 0 : router = router.middleware(auth_middleware(|request| {
2477 0 : let state = get_state(request);
2478 0 : if state.allowlist_routes.contains(request.uri()) {
2479 0 : None
2480 : } else {
2481 0 : state.auth.as_deref()
2482 : }
2483 0 : }))
2484 0 : }
2485 :
2486 0 : router = router.middleware(
2487 0 : endpoint::add_response_header_middleware(
2488 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
2489 0 : &launch_ts.to_string(),
2490 0 : )
2491 0 : .expect("construct launch timestamp header middleware"),
2492 0 : );
2493 0 :
2494 0 : Ok(router
2495 0 : .data(state)
2496 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
2497 0 : .get("/v1/status", |r| api_handler(r, status_handler))
2498 0 : .put("/v1/failpoints", |r| {
2499 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
2500 0 : })
2501 0 : .post("/v1/reload_auth_validation_keys", |r| {
2502 0 : api_handler(r, reload_auth_validation_keys_handler)
2503 0 : })
2504 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
2505 0 : .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
2506 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
2507 0 : api_handler(r, tenant_status)
2508 0 : })
2509 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
2510 0 : api_handler(r, tenant_delete_handler)
2511 0 : })
2512 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
2513 0 : api_handler(r, tenant_size_handler)
2514 0 : })
2515 0 : .put("/v1/tenant/config", |r| {
2516 0 : api_handler(r, update_tenant_config_handler)
2517 0 : })
2518 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
2519 0 : api_handler(r, tenant_shard_split_handler)
2520 0 : })
2521 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
2522 0 : api_handler(r, get_tenant_config_handler)
2523 0 : })
2524 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2525 0 : api_handler(r, put_tenant_location_config_handler)
2526 0 : })
2527 0 : .get("/v1/location_config", |r| {
2528 0 : api_handler(r, list_location_config_handler)
2529 0 : })
2530 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
2531 0 : api_handler(r, get_location_config_handler)
2532 0 : })
2533 0 : .put(
2534 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
2535 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
2536 0 : )
2537 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
2538 0 : api_handler(r, timeline_list_handler)
2539 0 : })
2540 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
2541 0 : api_handler(r, timeline_create_handler)
2542 0 : })
2543 0 : .post("/v1/tenant/:tenant_id/attach", |r| {
2544 0 : api_handler(r, tenant_attach_handler)
2545 0 : })
2546 0 : .post("/v1/tenant/:tenant_id/detach", |r| {
2547 0 : api_handler(r, tenant_detach_handler)
2548 0 : })
2549 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
2550 0 : api_handler(r, tenant_reset_handler)
2551 0 : })
2552 0 : .post("/v1/tenant/:tenant_id/load", |r| {
2553 0 : api_handler(r, tenant_load_handler)
2554 0 : })
2555 0 : .post("/v1/tenant/:tenant_id/ignore", |r| {
2556 0 : api_handler(r, tenant_ignore_handler)
2557 0 : })
2558 0 : .post(
2559 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
2560 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
2561 0 : )
2562 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2563 0 : api_handler(r, timeline_detail_handler)
2564 0 : })
2565 0 : .get(
2566 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
2567 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
2568 0 : )
2569 0 : .get(
2570 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
2571 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
2572 0 : )
2573 0 : .put(
2574 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
2575 0 : |r| api_handler(r, timeline_gc_handler),
2576 0 : )
2577 0 : .put(
2578 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
2579 0 : |r| testing_api_handler("run timeline compaction", r, timeline_compact_handler),
2580 0 : )
2581 0 : .put(
2582 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
2583 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
2584 0 : )
2585 0 : .post(
2586 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2587 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
2588 0 : )
2589 0 : .get(
2590 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2591 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
2592 0 : )
2593 0 : .put(
2594 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
2595 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
2596 0 : )
2597 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2598 0 : api_handler(r, timeline_delete_handler)
2599 0 : })
2600 0 : .get(
2601 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
2602 0 : |r| api_handler(r, layer_map_info_handler),
2603 0 : )
2604 0 : .get(
2605 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2606 0 : |r| api_handler(r, layer_download_handler),
2607 0 : )
2608 0 : .delete(
2609 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2610 0 : |r| api_handler(r, evict_timeline_layer_handler),
2611 0 : )
2612 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
2613 0 : api_handler(r, secondary_upload_handler)
2614 0 : })
2615 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
2616 0 : api_handler(r, tenant_scan_remote_handler)
2617 0 : })
2618 0 : .put("/v1/disk_usage_eviction/run", |r| {
2619 0 : api_handler(r, disk_usage_eviction_run)
2620 0 : })
2621 0 : .put("/v1/deletion_queue/flush", |r| {
2622 0 : api_handler(r, deletion_queue_flush)
2623 0 : })
2624 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
2625 0 : api_handler(r, secondary_status_handler)
2626 0 : })
2627 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
2628 0 : api_handler(r, secondary_download_handler)
2629 0 : })
2630 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
2631 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
2632 0 : })
2633 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
2634 0 : .post("/v1/tracing/event", |r| {
2635 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
2636 0 : })
2637 0 : .get(
2638 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
2639 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
2640 0 : )
2641 0 : .get(
2642 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
2643 0 : |r| api_handler(r, timeline_collect_keyspace),
2644 0 : )
2645 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
2646 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
2647 0 : .any(handler_404))
2648 0 : }
|