Line data Source code
1 : //!
2 : //! Management HTTP API
3 : //!
4 : use std::cmp::Reverse;
5 : use std::collections::BinaryHeap;
6 : use std::collections::HashMap;
7 : use std::str::FromStr;
8 : use std::sync::Arc;
9 : use std::time::Duration;
10 :
11 : use anyhow::{anyhow, Context, Result};
12 : use enumset::EnumSet;
13 : use futures::TryFutureExt;
14 : use humantime::format_rfc3339;
15 : use hyper::header;
16 : use hyper::StatusCode;
17 : use hyper::{Body, Request, Response, Uri};
18 : use metrics::launch_timestamp::LaunchTimestamp;
19 : use pageserver_api::models::AuxFilePolicy;
20 : use pageserver_api::models::IngestAuxFilesRequest;
21 : use pageserver_api::models::ListAuxFilesRequest;
22 : use pageserver_api::models::LocationConfig;
23 : use pageserver_api::models::LocationConfigListResponse;
24 : use pageserver_api::models::LsnLease;
25 : use pageserver_api::models::ShardParameters;
26 : use pageserver_api::models::TenantDetails;
27 : use pageserver_api::models::TenantLocationConfigResponse;
28 : use pageserver_api::models::TenantScanRemoteStorageResponse;
29 : use pageserver_api::models::TenantScanRemoteStorageShard;
30 : use pageserver_api::models::TenantShardLocation;
31 : use pageserver_api::models::TenantShardSplitRequest;
32 : use pageserver_api::models::TenantShardSplitResponse;
33 : use pageserver_api::models::TenantSorting;
34 : use pageserver_api::models::TenantState;
35 : use pageserver_api::models::TopTenantShardItem;
36 : use pageserver_api::models::TopTenantShardsRequest;
37 : use pageserver_api::models::TopTenantShardsResponse;
38 : use pageserver_api::models::{
39 : DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
40 : TenantLocationConfigRequest,
41 : };
42 : use pageserver_api::shard::ShardCount;
43 : use pageserver_api::shard::TenantShardId;
44 : use remote_storage::DownloadError;
45 : use remote_storage::GenericRemoteStorage;
46 : use remote_storage::TimeTravelError;
47 : use tenant_size_model::{SizeResult, StorageModel};
48 : use tokio_util::sync::CancellationToken;
49 : use tracing::*;
50 : use utils::auth::JwtAuth;
51 : use utils::failpoint_support::failpoints_handler;
52 : use utils::http::endpoint::prometheus_metrics_handler;
53 : use utils::http::endpoint::request_span;
54 : use utils::http::json::json_request_or_empty_body;
55 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
56 :
57 : use crate::context::{DownloadBehavior, RequestContext};
58 : use crate::deletion_queue::DeletionQueueClient;
59 : use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
60 : use crate::pgdatadir_mapping::LsnForTimestamp;
61 : use crate::task_mgr::TaskKind;
62 : use crate::tenant::config::{LocationConf, TenantConfOpt};
63 : use crate::tenant::mgr::GetActiveTenantError;
64 : use crate::tenant::mgr::{
65 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
66 : TenantSlotUpsertError, TenantStateError,
67 : };
68 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
69 : use crate::tenant::remote_timeline_client;
70 : use crate::tenant::remote_timeline_client::download_index_part;
71 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
72 : use crate::tenant::remote_timeline_client::list_remote_timelines;
73 : use crate::tenant::secondary::SecondaryController;
74 : use crate::tenant::size::ModelInputs;
75 : use crate::tenant::storage_layer::LayerAccessStatsReset;
76 : use crate::tenant::storage_layer::LayerName;
77 : use crate::tenant::timeline::CompactFlags;
78 : use crate::tenant::timeline::CompactionError;
79 : use crate::tenant::timeline::Timeline;
80 : use crate::tenant::GetTimelineError;
81 : use crate::tenant::SpawnMode;
82 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
83 : use crate::{config::PageServerConf, tenant::mgr};
84 : use crate::{disk_usage_eviction_task, tenant};
85 : use pageserver_api::models::{
86 : StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
87 : TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
88 : };
89 : use utils::{
90 : auth::SwappableJwtAuth,
91 : generation::Generation,
92 : http::{
93 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
94 : error::{ApiError, HttpErrorBody},
95 : json::{json_request, json_response},
96 : request::parse_request_param,
97 : RequestExt, RouterBuilder,
98 : },
99 : id::{TenantId, TimelineId},
100 : lsn::Lsn,
101 : };
102 :
103 : // For APIs that require an Active tenant, how long should we block waiting for that state?
104 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
105 : // failed API calls while tenants are activating.
106 : #[cfg(not(feature = "testing"))]
107 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
108 :
109 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
110 : // finish attaching, if calls to remote storage are slow.
111 : #[cfg(feature = "testing")]
112 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
113 :
114 : pub struct State {
115 : conf: &'static PageServerConf,
116 : tenant_manager: Arc<TenantManager>,
117 : auth: Option<Arc<SwappableJwtAuth>>,
118 : allowlist_routes: Vec<Uri>,
119 : remote_storage: GenericRemoteStorage,
120 : broker_client: storage_broker::BrokerClientChannel,
121 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
122 : deletion_queue_client: DeletionQueueClient,
123 : secondary_controller: SecondaryController,
124 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
125 : }
126 :
127 : impl State {
128 : #[allow(clippy::too_many_arguments)]
129 0 : pub fn new(
130 0 : conf: &'static PageServerConf,
131 0 : tenant_manager: Arc<TenantManager>,
132 0 : auth: Option<Arc<SwappableJwtAuth>>,
133 0 : remote_storage: GenericRemoteStorage,
134 0 : broker_client: storage_broker::BrokerClientChannel,
135 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
136 0 : deletion_queue_client: DeletionQueueClient,
137 0 : secondary_controller: SecondaryController,
138 0 : ) -> anyhow::Result<Self> {
139 0 : let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
140 0 : .iter()
141 0 : .map(|v| v.parse().unwrap())
142 0 : .collect::<Vec<_>>();
143 0 : Ok(Self {
144 0 : conf,
145 0 : tenant_manager,
146 0 : auth,
147 0 : allowlist_routes,
148 0 : remote_storage,
149 0 : broker_client,
150 0 : disk_usage_eviction_state,
151 0 : deletion_queue_client,
152 0 : secondary_controller,
153 0 : latest_utilization: Default::default(),
154 0 : })
155 0 : }
156 : }
157 :
158 : #[inline(always)]
159 0 : fn get_state(request: &Request<Body>) -> &State {
160 0 : request
161 0 : .data::<Arc<State>>()
162 0 : .expect("unknown state type")
163 0 : .as_ref()
164 0 : }
165 :
166 : #[inline(always)]
167 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
168 0 : get_state(request).conf
169 0 : }
170 :
171 : /// Check that the requester is authorized to operate on given tenant
172 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
173 0 : check_permission_with(request, |claims| {
174 0 : crate::auth::check_permission(claims, tenant_id)
175 0 : })
176 0 : }
177 :
178 : impl From<PageReconstructError> for ApiError {
179 0 : fn from(pre: PageReconstructError) -> ApiError {
180 0 : match pre {
181 0 : PageReconstructError::Other(pre) => ApiError::InternalServerError(pre),
182 0 : PageReconstructError::MissingKey(e) => {
183 0 : ApiError::InternalServerError(anyhow::anyhow!("{e}"))
184 : }
185 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
186 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
187 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
188 : }
189 0 : }
190 : }
191 :
192 : impl From<TenantMapInsertError> for ApiError {
193 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
194 0 : match tmie {
195 0 : TenantMapInsertError::SlotError(e) => e.into(),
196 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
197 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
198 : }
199 0 : }
200 : }
201 :
202 : impl From<TenantSlotError> for ApiError {
203 0 : fn from(e: TenantSlotError) -> ApiError {
204 0 : use TenantSlotError::*;
205 0 : match e {
206 0 : NotFound(tenant_id) => {
207 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
208 : }
209 : InProgress => {
210 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
211 : }
212 0 : MapState(e) => e.into(),
213 : }
214 0 : }
215 : }
216 :
217 : impl From<TenantSlotUpsertError> for ApiError {
218 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
219 0 : use TenantSlotUpsertError::*;
220 0 : match e {
221 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
222 0 : MapState(e) => e.into(),
223 0 : ShuttingDown(_) => ApiError::ShuttingDown,
224 : }
225 0 : }
226 : }
227 :
228 : impl From<UpsertLocationError> for ApiError {
229 0 : fn from(e: UpsertLocationError) -> ApiError {
230 0 : use UpsertLocationError::*;
231 0 : match e {
232 0 : BadRequest(e) => ApiError::BadRequest(e),
233 0 : Unavailable(_) => ApiError::ShuttingDown,
234 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
235 0 : Flush(e) | Other(e) => ApiError::InternalServerError(e),
236 : }
237 0 : }
238 : }
239 :
240 : impl From<TenantMapError> for ApiError {
241 0 : fn from(e: TenantMapError) -> ApiError {
242 0 : use TenantMapError::*;
243 0 : match e {
244 : StillInitializing | ShuttingDown => {
245 0 : ApiError::ResourceUnavailable(format!("{e}").into())
246 0 : }
247 0 : }
248 0 : }
249 : }
250 :
251 : impl From<TenantStateError> for ApiError {
252 0 : fn from(tse: TenantStateError) -> ApiError {
253 0 : match tse {
254 : TenantStateError::IsStopping(_) => {
255 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
256 : }
257 0 : TenantStateError::SlotError(e) => e.into(),
258 0 : TenantStateError::SlotUpsertError(e) => e.into(),
259 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
260 : }
261 0 : }
262 : }
263 :
264 : impl From<GetTenantError> for ApiError {
265 0 : fn from(tse: GetTenantError) -> ApiError {
266 0 : match tse {
267 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
268 : GetTenantError::NotActive(_) => {
269 : // Why is this not `ApiError::NotFound`?
270 : // Because we must be careful to never return 404 for a tenant if it does
271 : // in fact exist locally. If we did, the caller could draw the conclusion
272 : // that it can attach the tenant to another PS and we'd be in split-brain.
273 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
274 : }
275 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
276 : }
277 0 : }
278 : }
279 :
280 : impl From<GetTimelineError> for ApiError {
281 0 : fn from(gte: GetTimelineError) -> Self {
282 0 : // Rationale: tenant is activated only after eligble timelines activate
283 0 : ApiError::NotFound(gte.into())
284 0 : }
285 : }
286 :
287 : impl From<GetActiveTenantError> for ApiError {
288 0 : fn from(e: GetActiveTenantError) -> ApiError {
289 0 : match e {
290 0 : GetActiveTenantError::Broken(reason) => {
291 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
292 : }
293 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
294 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
295 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
296 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
297 0 : ApiError::ResourceUnavailable(format!("{}", e).into())
298 : }
299 : }
300 0 : }
301 : }
302 :
303 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
304 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
305 0 : use crate::tenant::DeleteTimelineError::*;
306 0 : match value {
307 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
308 0 : HasChildren(children) => ApiError::PreconditionFailed(
309 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
310 0 : .into_boxed_str(),
311 0 : ),
312 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
313 0 : Other(e) => ApiError::InternalServerError(e),
314 : }
315 0 : }
316 : }
317 :
318 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
319 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
320 : use crate::tenant::mgr::DeleteTimelineError::*;
321 0 : match value {
322 : // Report Precondition failed so client can distinguish between
323 : // "tenant is missing" case from "timeline is missing"
324 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
325 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
326 0 : ),
327 0 : Tenant(t) => ApiError::from(t),
328 0 : Timeline(t) => ApiError::from(t),
329 : }
330 0 : }
331 : }
332 :
333 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
334 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
335 0 : use crate::tenant::mgr::DeleteTenantError::*;
336 0 : match value {
337 0 : SlotError(e) => e.into(),
338 0 : Other(o) => ApiError::InternalServerError(o),
339 0 : Cancelled => ApiError::ShuttingDown,
340 : }
341 0 : }
342 : }
343 :
344 : // Helper function to construct a TimelineInfo struct for a timeline
345 0 : async fn build_timeline_info(
346 0 : timeline: &Arc<Timeline>,
347 0 : include_non_incremental_logical_size: bool,
348 0 : force_await_initial_logical_size: bool,
349 0 : ctx: &RequestContext,
350 0 : ) -> anyhow::Result<TimelineInfo> {
351 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
352 0 :
353 0 : if force_await_initial_logical_size {
354 0 : timeline.clone().await_initial_logical_size().await
355 0 : }
356 :
357 0 : let mut info = build_timeline_info_common(
358 0 : timeline,
359 0 : ctx,
360 0 : tenant::timeline::GetLogicalSizePriority::Background,
361 0 : )
362 0 : .await?;
363 0 : if include_non_incremental_logical_size {
364 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
365 : // Otherwise, if someone deletes the timeline / detaches the tenant while
366 : // we're executing this function, we will outlive the timeline on-disk state.
367 : info.current_logical_size_non_incremental = Some(
368 0 : timeline
369 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
370 0 : .await?,
371 : );
372 0 : }
373 0 : Ok(info)
374 0 : }
375 :
376 0 : async fn build_timeline_info_common(
377 0 : timeline: &Arc<Timeline>,
378 0 : ctx: &RequestContext,
379 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
380 0 : ) -> anyhow::Result<TimelineInfo> {
381 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
382 0 : let initdb_lsn = timeline.initdb_lsn;
383 0 : let last_record_lsn = timeline.get_last_record_lsn();
384 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
385 0 : let guard = timeline.last_received_wal.lock().unwrap();
386 0 : if let Some(info) = guard.as_ref() {
387 0 : (
388 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
389 0 : Some(info.last_received_msg_lsn),
390 0 : Some(info.last_received_msg_ts),
391 0 : )
392 : } else {
393 0 : (None, None, None)
394 : }
395 : };
396 :
397 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
398 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
399 0 : Lsn(0) => None,
400 0 : lsn @ Lsn(_) => Some(lsn),
401 : };
402 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
403 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
404 0 : let state = timeline.current_state();
405 0 : let remote_consistent_lsn_projected = timeline
406 0 : .get_remote_consistent_lsn_projected()
407 0 : .unwrap_or(Lsn(0));
408 0 : let remote_consistent_lsn_visible = timeline
409 0 : .get_remote_consistent_lsn_visible()
410 0 : .unwrap_or(Lsn(0));
411 0 :
412 0 : let walreceiver_status = timeline.walreceiver_status();
413 :
414 0 : let info = TimelineInfo {
415 0 : tenant_id: timeline.tenant_shard_id,
416 0 : timeline_id: timeline.timeline_id,
417 0 : ancestor_timeline_id,
418 0 : ancestor_lsn,
419 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
420 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
421 0 : remote_consistent_lsn_visible,
422 0 : initdb_lsn,
423 0 : last_record_lsn,
424 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
425 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
426 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
427 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
428 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
429 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
430 : },
431 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
432 0 : current_physical_size,
433 0 : current_logical_size_non_incremental: None,
434 0 : timeline_dir_layer_file_size_sum: None,
435 0 : wal_source_connstr,
436 0 : last_received_msg_lsn,
437 0 : last_received_msg_ts,
438 0 : pg_version: timeline.pg_version,
439 0 :
440 0 : state,
441 0 :
442 0 : walreceiver_status,
443 0 :
444 0 : last_aux_file_policy: timeline.last_aux_file_policy.load(),
445 0 : };
446 0 : Ok(info)
447 0 : }
448 :
449 : // healthcheck handler
450 0 : async fn status_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 : json_response(StatusCode::OK, StatusResponse { id: config.id })
457 0 : }
458 :
459 0 : async fn reload_auth_validation_keys_handler(
460 0 : request: Request<Body>,
461 0 : _cancel: CancellationToken,
462 0 : ) -> Result<Response<Body>, ApiError> {
463 0 : check_permission(&request, None)?;
464 0 : let config = get_config(&request);
465 0 : let state = get_state(&request);
466 0 : let Some(shared_auth) = &state.auth else {
467 0 : return json_response(StatusCode::BAD_REQUEST, ());
468 : };
469 : // unwrap is ok because check is performed when creating config, so path is set and exists
470 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
471 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
472 :
473 0 : match JwtAuth::from_key_path(key_path) {
474 0 : Ok(new_auth) => {
475 0 : shared_auth.swap(new_auth);
476 0 : json_response(StatusCode::OK, ())
477 : }
478 0 : Err(e) => {
479 0 : let err_msg = "Error reloading public keys";
480 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
481 0 : json_response(
482 0 : StatusCode::INTERNAL_SERVER_ERROR,
483 0 : HttpErrorBody::from_msg(err_msg.to_string()),
484 0 : )
485 : }
486 : }
487 0 : }
488 :
489 0 : async fn timeline_create_handler(
490 0 : mut request: Request<Body>,
491 0 : _cancel: CancellationToken,
492 0 : ) -> Result<Response<Body>, ApiError> {
493 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
494 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
495 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
496 :
497 0 : let new_timeline_id = request_data.new_timeline_id;
498 0 :
499 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
500 0 :
501 0 : let state = get_state(&request);
502 :
503 0 : async {
504 0 : let tenant = state
505 0 : .tenant_manager
506 0 : .get_attached_tenant_shard(tenant_shard_id)?;
507 :
508 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
509 :
510 0 : if let Some(ancestor_id) = request_data.ancestor_timeline_id.as_ref() {
511 0 : tracing::info!(%ancestor_id, "starting to branch");
512 : } else {
513 0 : tracing::info!("bootstrapping");
514 : }
515 :
516 0 : match tenant
517 0 : .create_timeline(
518 0 : new_timeline_id,
519 0 : request_data.ancestor_timeline_id,
520 0 : request_data.ancestor_start_lsn,
521 0 : request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
522 0 : request_data.existing_initdb_timeline_id,
523 0 : state.broker_client.clone(),
524 0 : &ctx,
525 0 : )
526 0 : .await
527 : {
528 0 : Ok(new_timeline) => {
529 : // Created. Construct a TimelineInfo for it.
530 0 : let timeline_info = build_timeline_info_common(
531 0 : &new_timeline,
532 0 : &ctx,
533 0 : tenant::timeline::GetLogicalSizePriority::User,
534 0 : )
535 0 : .await
536 0 : .map_err(ApiError::InternalServerError)?;
537 0 : json_response(StatusCode::CREATED, timeline_info)
538 : }
539 0 : Err(_) if tenant.cancel.is_cancelled() => {
540 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
541 0 : json_response(
542 0 : StatusCode::SERVICE_UNAVAILABLE,
543 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
544 0 : )
545 : }
546 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
547 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
548 : }
549 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
550 0 : StatusCode::TOO_MANY_REQUESTS,
551 0 : HttpErrorBody::from_msg(e.to_string()),
552 0 : ),
553 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
554 0 : StatusCode::NOT_ACCEPTABLE,
555 0 : HttpErrorBody::from_msg(format!("{err:#}")),
556 0 : ),
557 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
558 0 : StatusCode::SERVICE_UNAVAILABLE,
559 0 : HttpErrorBody::from_msg(e.to_string()),
560 0 : ),
561 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
562 0 : StatusCode::SERVICE_UNAVAILABLE,
563 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
564 0 : ),
565 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
566 : }
567 0 : }
568 0 : .instrument(info_span!("timeline_create",
569 : tenant_id = %tenant_shard_id.tenant_id,
570 0 : shard_id = %tenant_shard_id.shard_slug(),
571 : timeline_id = %new_timeline_id,
572 : lsn=?request_data.ancestor_start_lsn,
573 : pg_version=?request_data.pg_version
574 : ))
575 0 : .await
576 0 : }
577 :
578 0 : async fn timeline_list_handler(
579 0 : request: Request<Body>,
580 0 : _cancel: CancellationToken,
581 0 : ) -> Result<Response<Body>, ApiError> {
582 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
583 0 : let include_non_incremental_logical_size: Option<bool> =
584 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
585 0 : let force_await_initial_logical_size: Option<bool> =
586 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
587 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
588 :
589 0 : let state = get_state(&request);
590 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
591 :
592 0 : let response_data = async {
593 0 : let tenant = state
594 0 : .tenant_manager
595 0 : .get_attached_tenant_shard(tenant_shard_id)?;
596 :
597 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
598 :
599 0 : let timelines = tenant.list_timelines();
600 0 :
601 0 : let mut response_data = Vec::with_capacity(timelines.len());
602 0 : for timeline in timelines {
603 0 : let timeline_info = build_timeline_info(
604 0 : &timeline,
605 0 : include_non_incremental_logical_size.unwrap_or(false),
606 0 : force_await_initial_logical_size.unwrap_or(false),
607 0 : &ctx,
608 0 : )
609 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
610 0 : .await
611 0 : .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
612 0 : .map_err(ApiError::InternalServerError)?;
613 :
614 0 : response_data.push(timeline_info);
615 : }
616 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
617 0 : }
618 0 : .instrument(info_span!("timeline_list",
619 : tenant_id = %tenant_shard_id.tenant_id,
620 0 : shard_id = %tenant_shard_id.shard_slug()))
621 0 : .await?;
622 :
623 0 : json_response(StatusCode::OK, response_data)
624 0 : }
625 :
626 0 : async fn timeline_preserve_initdb_handler(
627 0 : request: Request<Body>,
628 0 : _cancel: CancellationToken,
629 0 : ) -> Result<Response<Body>, ApiError> {
630 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
631 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
632 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
633 0 : let state = get_state(&request);
634 :
635 : // Part of the process for disaster recovery from safekeeper-stored WAL:
636 : // If we don't recover into a new timeline but want to keep the timeline ID,
637 : // then the initdb archive is deleted. This endpoint copies it to a different
638 : // location where timeline recreation cand find it.
639 :
640 0 : async {
641 0 : let tenant = state
642 0 : .tenant_manager
643 0 : .get_attached_tenant_shard(tenant_shard_id)?;
644 :
645 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
646 :
647 0 : timeline
648 0 : .preserve_initdb_archive()
649 0 : .await
650 0 : .context("preserving initdb archive")
651 0 : .map_err(ApiError::InternalServerError)?;
652 :
653 0 : Ok::<_, ApiError>(())
654 0 : }
655 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
656 : tenant_id = %tenant_shard_id.tenant_id,
657 0 : shard_id = %tenant_shard_id.shard_slug(),
658 : %timeline_id))
659 0 : .await?;
660 :
661 0 : json_response(StatusCode::OK, ())
662 0 : }
663 :
664 0 : async fn timeline_detail_handler(
665 0 : request: Request<Body>,
666 0 : _cancel: CancellationToken,
667 0 : ) -> Result<Response<Body>, ApiError> {
668 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
669 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
670 0 : let include_non_incremental_logical_size: Option<bool> =
671 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
672 0 : let force_await_initial_logical_size: Option<bool> =
673 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
674 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
675 :
676 : // Logical size calculation needs downloading.
677 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
678 0 : let state = get_state(&request);
679 :
680 0 : let timeline_info = async {
681 0 : let tenant = state
682 0 : .tenant_manager
683 0 : .get_attached_tenant_shard(tenant_shard_id)?;
684 :
685 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
686 :
687 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
688 :
689 0 : let timeline_info = build_timeline_info(
690 0 : &timeline,
691 0 : include_non_incremental_logical_size.unwrap_or(false),
692 0 : force_await_initial_logical_size.unwrap_or(false),
693 0 : &ctx,
694 0 : )
695 0 : .await
696 0 : .context("get local timeline info")
697 0 : .map_err(ApiError::InternalServerError)?;
698 :
699 0 : Ok::<_, ApiError>(timeline_info)
700 0 : }
701 0 : .instrument(info_span!("timeline_detail",
702 : tenant_id = %tenant_shard_id.tenant_id,
703 0 : shard_id = %tenant_shard_id.shard_slug(),
704 : %timeline_id))
705 0 : .await?;
706 :
707 0 : json_response(StatusCode::OK, timeline_info)
708 0 : }
709 :
710 0 : async fn get_lsn_by_timestamp_handler(
711 0 : request: Request<Body>,
712 0 : cancel: CancellationToken,
713 0 : ) -> Result<Response<Body>, ApiError> {
714 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
715 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
716 0 : let state = get_state(&request);
717 0 :
718 0 : if !tenant_shard_id.is_shard_zero() {
719 : // Requires SLRU contents, which are only stored on shard zero
720 0 : return Err(ApiError::BadRequest(anyhow!(
721 0 : "Size calculations are only available on shard zero"
722 0 : )));
723 0 : }
724 :
725 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
726 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
727 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
728 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
729 0 : .map_err(ApiError::BadRequest)?;
730 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
731 :
732 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
733 0 :
734 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
735 :
736 0 : let timeline =
737 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
738 0 : .await?;
739 0 : let result = timeline
740 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
741 0 : .await?;
742 :
743 : #[derive(serde::Serialize, Debug)]
744 : struct Result {
745 : lsn: Lsn,
746 : kind: &'static str,
747 : #[serde(default)]
748 : #[serde(skip_serializing_if = "Option::is_none")]
749 : #[serde(flatten)]
750 : lease: Option<LsnLease>,
751 : }
752 0 : let (lsn, kind) = match result {
753 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
754 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
755 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
756 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
757 : };
758 :
759 0 : let lease = if with_lease {
760 0 : timeline
761 0 : .make_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
762 0 : .inspect_err(|_| {
763 0 : warn!("fail to grant a lease to {}", lsn);
764 0 : })
765 0 : .ok()
766 : } else {
767 0 : None
768 : };
769 :
770 0 : let result = Result { lsn, kind, lease };
771 0 : let valid_until = result
772 0 : .lease
773 0 : .as_ref()
774 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
775 0 : tracing::info!(
776 : lsn=?result.lsn,
777 : kind=%result.kind,
778 : timestamp=%timestamp_raw,
779 : valid_until=?valid_until,
780 0 : "lsn_by_timestamp finished"
781 : );
782 0 : json_response(StatusCode::OK, result)
783 0 : }
784 :
785 0 : async fn get_timestamp_of_lsn_handler(
786 0 : request: Request<Body>,
787 0 : _cancel: CancellationToken,
788 0 : ) -> Result<Response<Body>, ApiError> {
789 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
790 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
791 0 : let state = get_state(&request);
792 0 :
793 0 : if !tenant_shard_id.is_shard_zero() {
794 : // Requires SLRU contents, which are only stored on shard zero
795 0 : return Err(ApiError::BadRequest(anyhow!(
796 0 : "Size calculations are only available on shard zero"
797 0 : )));
798 0 : }
799 :
800 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
801 :
802 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
803 0 : let lsn = Lsn::from_str(&lsn_str)
804 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
805 0 : .map_err(ApiError::BadRequest)?;
806 :
807 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
808 0 : let timeline =
809 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
810 0 : .await?;
811 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
812 :
813 0 : match result {
814 0 : Some(time) => {
815 0 : let time = format_rfc3339(postgres_ffi::from_pg_timestamp(time)).to_string();
816 0 : json_response(StatusCode::OK, time)
817 : }
818 0 : None => Err(ApiError::NotFound(
819 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
820 0 : )),
821 : }
822 0 : }
823 :
824 0 : async fn tenant_attach_handler(
825 0 : mut request: Request<Body>,
826 0 : _cancel: CancellationToken,
827 0 : ) -> Result<Response<Body>, ApiError> {
828 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
829 0 : check_permission(&request, Some(tenant_id))?;
830 :
831 0 : let maybe_body: Option<TenantAttachRequest> = json_request_or_empty_body(&mut request).await?;
832 0 : let tenant_conf = match &maybe_body {
833 0 : Some(request) => TenantConfOpt::try_from(&*request.config).map_err(ApiError::BadRequest)?,
834 0 : None => TenantConfOpt::default(),
835 : };
836 :
837 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
838 0 :
839 0 : info!("Handling tenant attach {tenant_id}");
840 :
841 0 : let state = get_state(&request);
842 :
843 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
844 :
845 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
846 0 : let shard_params = ShardParameters::default();
847 0 : let location_conf = LocationConf::attached_single(tenant_conf, generation, &shard_params);
848 :
849 0 : let tenant = state
850 0 : .tenant_manager
851 0 : .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
852 0 : .await?;
853 :
854 0 : let Some(tenant) = tenant else {
855 : // This should never happen: indicates a bug in upsert_location
856 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
857 0 : "Upsert succeeded but didn't return tenant!"
858 0 : )));
859 : };
860 :
861 : // We might have successfully constructed a Tenant, but it could still
862 : // end up in a broken state:
863 : if let TenantState::Broken {
864 0 : reason,
865 : backtrace: _,
866 0 : } = tenant.current_state()
867 : {
868 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
869 0 : "Tenant state is Broken: {reason}"
870 0 : )));
871 0 : }
872 0 :
873 0 : json_response(StatusCode::ACCEPTED, ())
874 0 : }
875 :
876 0 : async fn timeline_delete_handler(
877 0 : request: Request<Body>,
878 0 : _cancel: CancellationToken,
879 0 : ) -> Result<Response<Body>, ApiError> {
880 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
881 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
882 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
883 :
884 0 : let state = get_state(&request);
885 :
886 0 : let tenant = state
887 0 : .tenant_manager
888 0 : .get_attached_tenant_shard(tenant_shard_id)
889 0 : .map_err(|e| {
890 0 : match e {
891 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
892 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
893 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
894 0 : "Requested tenant is missing".to_string().into_boxed_str(),
895 0 : ),
896 0 : e => e.into(),
897 : }
898 0 : })?;
899 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
900 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))
901 0 : .await?;
902 :
903 0 : json_response(StatusCode::ACCEPTED, ())
904 0 : }
905 :
906 0 : async fn tenant_detach_handler(
907 0 : request: Request<Body>,
908 0 : _cancel: CancellationToken,
909 0 : ) -> Result<Response<Body>, ApiError> {
910 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
911 0 : check_permission(&request, Some(tenant_id))?;
912 : // This is a legacy API (`/location_conf` is the replacement). It only supports unsharded tenants
913 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
914 0 :
915 0 : let state = get_state(&request);
916 0 : let conf = state.conf;
917 0 : state
918 0 : .tenant_manager
919 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
920 0 : .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
921 0 : .await?;
922 :
923 0 : json_response(StatusCode::OK, ())
924 0 : }
925 :
926 0 : async fn tenant_reset_handler(
927 0 : request: Request<Body>,
928 0 : _cancel: CancellationToken,
929 0 : ) -> Result<Response<Body>, ApiError> {
930 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
931 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
932 :
933 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
934 :
935 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
936 0 : let state = get_state(&request);
937 0 : state
938 0 : .tenant_manager
939 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
940 0 : .await
941 0 : .map_err(ApiError::InternalServerError)?;
942 :
943 0 : json_response(StatusCode::OK, ())
944 0 : }
945 :
946 0 : async fn tenant_list_handler(
947 0 : request: Request<Body>,
948 0 : _cancel: CancellationToken,
949 0 : ) -> Result<Response<Body>, ApiError> {
950 0 : check_permission(&request, None)?;
951 0 : let state = get_state(&request);
952 :
953 0 : let response_data = state
954 0 : .tenant_manager
955 0 : .list_tenants()
956 0 : .map_err(|_| {
957 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
958 0 : })?
959 0 : .iter()
960 0 : .map(|(id, state, gen)| TenantInfo {
961 0 : id: *id,
962 0 : state: state.clone(),
963 0 : current_physical_size: None,
964 0 : attachment_status: state.attachment_status(),
965 0 : generation: (*gen).into(),
966 0 : })
967 0 : .collect::<Vec<TenantInfo>>();
968 0 :
969 0 : json_response(StatusCode::OK, response_data)
970 0 : }
971 :
972 0 : async fn tenant_status(
973 0 : request: Request<Body>,
974 0 : _cancel: CancellationToken,
975 0 : ) -> Result<Response<Body>, ApiError> {
976 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
977 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
978 0 : let state = get_state(&request);
979 0 :
980 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
981 0 : let activate = true;
982 : #[cfg(feature = "testing")]
983 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
984 :
985 0 : let tenant_info = async {
986 0 : let tenant = state
987 0 : .tenant_manager
988 0 : .get_attached_tenant_shard(tenant_shard_id)?;
989 :
990 0 : if activate {
991 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
992 : // called, but it is still valid to return 200 and describe the current state of the tenant
993 : // if it doesn't make it into an active state.
994 0 : tenant
995 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
996 0 : .await
997 0 : .ok();
998 0 : }
999 :
1000 : // Calculate total physical size of all timelines
1001 0 : let mut current_physical_size = 0;
1002 0 : for timeline in tenant.list_timelines().iter() {
1003 0 : current_physical_size += timeline.layer_size_sum().await;
1004 : }
1005 :
1006 0 : let state = tenant.current_state();
1007 0 : Result::<_, ApiError>::Ok(TenantDetails {
1008 0 : tenant_info: TenantInfo {
1009 0 : id: tenant_shard_id,
1010 0 : state: state.clone(),
1011 0 : current_physical_size: Some(current_physical_size),
1012 0 : attachment_status: state.attachment_status(),
1013 0 : generation: tenant.generation().into(),
1014 0 : },
1015 0 : walredo: tenant.wal_redo_manager_status(),
1016 0 : timelines: tenant.list_timeline_ids(),
1017 0 : })
1018 0 : }
1019 0 : .instrument(info_span!("tenant_status_handler",
1020 : tenant_id = %tenant_shard_id.tenant_id,
1021 0 : shard_id = %tenant_shard_id.shard_slug()))
1022 0 : .await?;
1023 :
1024 0 : json_response(StatusCode::OK, tenant_info)
1025 0 : }
1026 :
1027 0 : async fn tenant_delete_handler(
1028 0 : request: Request<Body>,
1029 0 : _cancel: CancellationToken,
1030 0 : ) -> Result<Response<Body>, ApiError> {
1031 : // TODO openapi spec
1032 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1033 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1034 :
1035 0 : let state = get_state(&request);
1036 0 :
1037 0 : state
1038 0 : .tenant_manager
1039 0 : .delete_tenant(tenant_shard_id)
1040 0 : .instrument(info_span!("tenant_delete_handler",
1041 : tenant_id = %tenant_shard_id.tenant_id,
1042 0 : shard_id = %tenant_shard_id.shard_slug()
1043 : ))
1044 0 : .await?;
1045 :
1046 0 : json_response(StatusCode::OK, ())
1047 0 : }
1048 :
1049 : /// HTTP endpoint to query the current tenant_size of a tenant.
1050 : ///
1051 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1052 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1053 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1054 : /// values.
1055 : ///
1056 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1057 : /// (only if it is shorter than the real cutoff).
1058 : ///
1059 : /// Note: we don't update the cached size and prometheus metric here.
1060 : /// The retention period might be different, and it's nice to have a method to just calculate it
1061 : /// without modifying anything anyway.
1062 0 : async fn tenant_size_handler(
1063 0 : request: Request<Body>,
1064 0 : cancel: CancellationToken,
1065 0 : ) -> Result<Response<Body>, ApiError> {
1066 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1067 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1068 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1069 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1070 0 : let headers = request.headers();
1071 0 : let state = get_state(&request);
1072 0 :
1073 0 : if !tenant_shard_id.is_shard_zero() {
1074 0 : return Err(ApiError::BadRequest(anyhow!(
1075 0 : "Size calculations are only available on shard zero"
1076 0 : )));
1077 0 : }
1078 0 :
1079 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1080 0 : let tenant = state
1081 0 : .tenant_manager
1082 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1083 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1084 :
1085 : // this can be long operation
1086 0 : let inputs = tenant
1087 0 : .gather_size_inputs(
1088 0 : retention_period,
1089 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1090 0 : &cancel,
1091 0 : &ctx,
1092 0 : )
1093 0 : .await
1094 0 : .map_err(|e| match e {
1095 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1096 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1097 0 : })?;
1098 :
1099 0 : let mut sizes = None;
1100 0 : let accepts_html = headers
1101 0 : .get(header::ACCEPT)
1102 0 : .map(|v| v == "text/html")
1103 0 : .unwrap_or_default();
1104 0 : if !inputs_only.unwrap_or(false) {
1105 0 : let storage_model = inputs.calculate_model();
1106 0 : let size = storage_model.calculate();
1107 0 :
1108 0 : // If request header expects html, return html
1109 0 : if accepts_html {
1110 0 : return synthetic_size_html_response(inputs, storage_model, size);
1111 0 : }
1112 0 : sizes = Some(size);
1113 0 : } else if accepts_html {
1114 0 : return Err(ApiError::BadRequest(anyhow!(
1115 0 : "inputs_only parameter is incompatible with html output request"
1116 0 : )));
1117 0 : }
1118 :
1119 : /// The type resides in the pageserver not to expose `ModelInputs`.
1120 : #[derive(serde::Serialize)]
1121 : struct TenantHistorySize {
1122 : id: TenantId,
1123 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1124 : ///
1125 : /// Will be none if `?inputs_only=true` was given.
1126 : size: Option<u64>,
1127 : /// Size of each segment used in the model.
1128 : /// Will be null if `?inputs_only=true` was given.
1129 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1130 : inputs: crate::tenant::size::ModelInputs,
1131 : }
1132 :
1133 0 : json_response(
1134 0 : StatusCode::OK,
1135 0 : TenantHistorySize {
1136 0 : id: tenant_shard_id.tenant_id,
1137 0 : size: sizes.as_ref().map(|x| x.total_size),
1138 0 : segment_sizes: sizes.map(|x| x.segments),
1139 0 : inputs,
1140 0 : },
1141 0 : )
1142 0 : }
1143 :
1144 0 : async fn tenant_shard_split_handler(
1145 0 : mut request: Request<Body>,
1146 0 : _cancel: CancellationToken,
1147 0 : ) -> Result<Response<Body>, ApiError> {
1148 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1149 :
1150 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1151 0 : let state = get_state(&request);
1152 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1153 :
1154 0 : let tenant = state
1155 0 : .tenant_manager
1156 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1157 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1158 :
1159 0 : let new_shards = state
1160 0 : .tenant_manager
1161 0 : .shard_split(
1162 0 : tenant,
1163 0 : ShardCount::new(req.new_shard_count),
1164 0 : req.new_stripe_size,
1165 0 : &ctx,
1166 0 : )
1167 0 : .await
1168 0 : .map_err(ApiError::InternalServerError)?;
1169 :
1170 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1171 0 : }
1172 :
1173 0 : async fn layer_map_info_handler(
1174 0 : request: Request<Body>,
1175 0 : _cancel: CancellationToken,
1176 0 : ) -> Result<Response<Body>, ApiError> {
1177 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1178 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1179 0 : let reset: LayerAccessStatsReset =
1180 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1181 0 : let state = get_state(&request);
1182 0 :
1183 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1184 :
1185 0 : let timeline =
1186 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1187 0 : .await?;
1188 0 : let layer_map_info = timeline.layer_map_info(reset).await;
1189 :
1190 0 : json_response(StatusCode::OK, layer_map_info)
1191 0 : }
1192 :
1193 0 : async fn layer_download_handler(
1194 0 : request: Request<Body>,
1195 0 : _cancel: CancellationToken,
1196 0 : ) -> Result<Response<Body>, ApiError> {
1197 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1198 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1199 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1200 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1201 0 : let layer_name = LayerName::from_str(layer_file_name)
1202 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1203 0 : let state = get_state(&request);
1204 :
1205 0 : let timeline =
1206 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1207 0 : .await?;
1208 0 : let downloaded = timeline
1209 0 : .download_layer(&layer_name)
1210 0 : .await
1211 0 : .map_err(ApiError::InternalServerError)?;
1212 :
1213 0 : match downloaded {
1214 0 : Some(true) => json_response(StatusCode::OK, ()),
1215 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1216 0 : None => json_response(
1217 0 : StatusCode::BAD_REQUEST,
1218 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1219 0 : ),
1220 : }
1221 0 : }
1222 :
1223 0 : async fn evict_timeline_layer_handler(
1224 0 : request: Request<Body>,
1225 0 : _cancel: CancellationToken,
1226 0 : ) -> Result<Response<Body>, ApiError> {
1227 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1228 0 : check_permission(&request, Some(tenant_shard_id.tenant_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 : let state = get_state(&request);
1232 :
1233 0 : let layer_name = LayerName::from_str(layer_file_name)
1234 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
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 evicted = timeline
1240 0 : .evict_layer(&layer_name)
1241 0 : .await
1242 0 : .map_err(ApiError::InternalServerError)?;
1243 :
1244 0 : match evicted {
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 : /// Get tenant_size SVG graph along with the JSON data.
1255 0 : fn synthetic_size_html_response(
1256 0 : inputs: ModelInputs,
1257 0 : storage_model: StorageModel,
1258 0 : sizes: SizeResult,
1259 0 : ) -> Result<Response<Body>, ApiError> {
1260 0 : let mut timeline_ids: Vec<String> = Vec::new();
1261 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1262 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1263 0 : timeline_map.insert(ti.timeline_id, index);
1264 0 : timeline_ids.push(ti.timeline_id.to_string());
1265 0 : }
1266 0 : let seg_to_branch: Vec<usize> = inputs
1267 0 : .segments
1268 0 : .iter()
1269 0 : .map(|seg| *timeline_map.get(&seg.timeline_id).unwrap())
1270 0 : .collect();
1271 :
1272 0 : let svg =
1273 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1274 0 : .map_err(ApiError::InternalServerError)?;
1275 :
1276 0 : let mut response = String::new();
1277 0 :
1278 0 : use std::fmt::Write;
1279 0 : write!(response, "<html>\n<body>\n").unwrap();
1280 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1281 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1282 0 : writeln!(response, "<pre>").unwrap();
1283 0 : writeln!(
1284 0 : response,
1285 0 : "{}",
1286 0 : serde_json::to_string_pretty(&inputs).unwrap()
1287 0 : )
1288 0 : .unwrap();
1289 0 : writeln!(
1290 0 : response,
1291 0 : "{}",
1292 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1293 0 : )
1294 0 : .unwrap();
1295 0 : writeln!(response, "</pre>").unwrap();
1296 0 : write!(response, "</body>\n</html>\n").unwrap();
1297 0 :
1298 0 : html_response(StatusCode::OK, response)
1299 0 : }
1300 :
1301 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1302 0 : let response = Response::builder()
1303 0 : .status(status)
1304 0 : .header(header::CONTENT_TYPE, "text/html")
1305 0 : .body(Body::from(data.as_bytes().to_vec()))
1306 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1307 0 : Ok(response)
1308 0 : }
1309 :
1310 : /// Helper for requests that may take a generation, which is mandatory
1311 : /// when control_plane_api is set, but otherwise defaults to Generation::none()
1312 0 : fn get_request_generation(state: &State, req_gen: Option<u32>) -> Result<Generation, ApiError> {
1313 0 : if state.conf.control_plane_api.is_some() {
1314 0 : req_gen
1315 0 : .map(Generation::new)
1316 0 : .ok_or(ApiError::BadRequest(anyhow!(
1317 0 : "generation attribute missing"
1318 0 : )))
1319 : } else {
1320 : // Legacy mode: all tenants operate with no generation
1321 0 : Ok(Generation::none())
1322 : }
1323 0 : }
1324 :
1325 0 : async fn tenant_create_handler(
1326 0 : mut request: Request<Body>,
1327 0 : _cancel: CancellationToken,
1328 0 : ) -> Result<Response<Body>, ApiError> {
1329 0 : let request_data: TenantCreateRequest = json_request(&mut request).await?;
1330 0 : let target_tenant_id = request_data.new_tenant_id;
1331 0 : check_permission(&request, None)?;
1332 :
1333 0 : let _timer = STORAGE_TIME_GLOBAL
1334 0 : .get_metric_with_label_values(&[StorageTimeOperation::CreateTenant.into()])
1335 0 : .expect("bug")
1336 0 : .start_timer();
1337 :
1338 0 : let tenant_conf =
1339 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1340 :
1341 0 : let state = get_state(&request);
1342 :
1343 0 : let generation = get_request_generation(state, request_data.generation)?;
1344 :
1345 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1346 0 :
1347 0 : let location_conf =
1348 0 : LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
1349 :
1350 0 : let new_tenant = state
1351 0 : .tenant_manager
1352 0 : .upsert_location(
1353 0 : target_tenant_id,
1354 0 : location_conf,
1355 0 : None,
1356 0 : SpawnMode::Create,
1357 0 : &ctx,
1358 0 : )
1359 0 : .await?;
1360 :
1361 0 : let Some(new_tenant) = new_tenant else {
1362 : // This should never happen: indicates a bug in upsert_location
1363 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
1364 0 : "Upsert succeeded but didn't return tenant!"
1365 0 : )));
1366 : };
1367 : // We created the tenant. Existing API semantics are that the tenant
1368 : // is Active when this function returns.
1369 0 : new_tenant
1370 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1371 0 : .await?;
1372 :
1373 0 : json_response(
1374 0 : StatusCode::CREATED,
1375 0 : TenantCreateResponse(new_tenant.tenant_shard_id().tenant_id),
1376 0 : )
1377 0 : }
1378 :
1379 0 : async fn get_tenant_config_handler(
1380 0 : request: Request<Body>,
1381 0 : _cancel: CancellationToken,
1382 0 : ) -> Result<Response<Body>, ApiError> {
1383 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1384 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1385 0 : let state = get_state(&request);
1386 :
1387 0 : let tenant = state
1388 0 : .tenant_manager
1389 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1390 :
1391 0 : let response = HashMap::from([
1392 : (
1393 : "tenant_specific_overrides",
1394 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1395 0 : .context("serializing tenant specific overrides")
1396 0 : .map_err(ApiError::InternalServerError)?,
1397 : ),
1398 : (
1399 0 : "effective_config",
1400 0 : serde_json::to_value(tenant.effective_config())
1401 0 : .context("serializing effective config")
1402 0 : .map_err(ApiError::InternalServerError)?,
1403 : ),
1404 : ]);
1405 :
1406 0 : json_response(StatusCode::OK, response)
1407 0 : }
1408 :
1409 0 : async fn update_tenant_config_handler(
1410 0 : mut request: Request<Body>,
1411 0 : _cancel: CancellationToken,
1412 0 : ) -> Result<Response<Body>, ApiError> {
1413 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1414 0 : let tenant_id = request_data.tenant_id;
1415 0 : check_permission(&request, Some(tenant_id))?;
1416 :
1417 0 : let new_tenant_conf =
1418 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1419 :
1420 0 : let state = get_state(&request);
1421 0 :
1422 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1423 :
1424 0 : let tenant = state
1425 0 : .tenant_manager
1426 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1427 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1428 :
1429 : // This is a legacy API that only operates on attached tenants: the preferred
1430 : // API to use is the location_config/ endpoint, which lets the caller provide
1431 : // the full LocationConf.
1432 0 : let location_conf = LocationConf::attached_single(
1433 0 : new_tenant_conf.clone(),
1434 0 : tenant.get_generation(),
1435 0 : &ShardParameters::default(),
1436 0 : );
1437 0 :
1438 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1439 0 : .await
1440 0 : .map_err(ApiError::InternalServerError)?;
1441 0 : tenant.set_new_tenant_config(new_tenant_conf);
1442 0 :
1443 0 : json_response(StatusCode::OK, ())
1444 0 : }
1445 :
1446 0 : async fn put_tenant_location_config_handler(
1447 0 : mut request: Request<Body>,
1448 0 : _cancel: CancellationToken,
1449 0 : ) -> Result<Response<Body>, ApiError> {
1450 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1451 :
1452 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1453 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1454 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1455 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1456 :
1457 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1458 0 : let state = get_state(&request);
1459 0 : let conf = state.conf;
1460 0 :
1461 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1462 0 : // its local disk content and drops it from memory.
1463 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1464 0 : if let Err(e) = state
1465 0 : .tenant_manager
1466 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
1467 0 : .instrument(info_span!("tenant_detach",
1468 : tenant_id = %tenant_shard_id.tenant_id,
1469 0 : shard_id = %tenant_shard_id.shard_slug()
1470 : ))
1471 0 : .await
1472 : {
1473 0 : match e {
1474 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1475 0 : // This API is idempotent: a NotFound on a detach is fine.
1476 0 : }
1477 0 : _ => return Err(e.into()),
1478 : }
1479 0 : }
1480 0 : return json_response(StatusCode::OK, ());
1481 0 : }
1482 :
1483 0 : let location_conf =
1484 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1485 :
1486 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1487 : // similar to at startup ordering.
1488 0 : let spawn_mode = if lazy {
1489 0 : tenant::SpawnMode::Lazy
1490 : } else {
1491 0 : tenant::SpawnMode::Eager
1492 : };
1493 :
1494 0 : let tenant = state
1495 0 : .tenant_manager
1496 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1497 0 : .await?;
1498 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1499 0 : let attached = tenant.is_some();
1500 :
1501 0 : if let Some(_flush_ms) = flush {
1502 0 : match state
1503 0 : .secondary_controller
1504 0 : .upload_tenant(tenant_shard_id)
1505 0 : .await
1506 : {
1507 : Ok(()) => {
1508 0 : tracing::info!("Uploaded heatmap during flush");
1509 : }
1510 0 : Err(e) => {
1511 0 : tracing::warn!("Failed to flush heatmap: {e}");
1512 : }
1513 : }
1514 : } else {
1515 0 : tracing::info!("No flush requested when configuring");
1516 : }
1517 :
1518 : // This API returns a vector of pageservers where the tenant is attached: this is
1519 : // primarily for use in the sharding service. For compatibilty, we also return this
1520 : // when called directly on a pageserver, but the payload is always zero or one shards.
1521 0 : let mut response = TenantLocationConfigResponse {
1522 0 : shards: Vec::new(),
1523 0 : stripe_size: None,
1524 0 : };
1525 0 : if attached {
1526 0 : response.shards.push(TenantShardLocation {
1527 0 : shard_id: tenant_shard_id,
1528 0 : node_id: state.conf.id,
1529 0 : });
1530 0 : if tenant_shard_id.shard_count.count() > 1 {
1531 : // Stripe size should be set if we are attached
1532 0 : debug_assert!(stripe_size.is_some());
1533 0 : response.stripe_size = stripe_size;
1534 0 : }
1535 0 : }
1536 :
1537 0 : json_response(StatusCode::OK, response)
1538 0 : }
1539 :
1540 0 : async fn list_location_config_handler(
1541 0 : request: Request<Body>,
1542 0 : _cancel: CancellationToken,
1543 0 : ) -> Result<Response<Body>, ApiError> {
1544 0 : let state = get_state(&request);
1545 0 : let slots = state.tenant_manager.list();
1546 0 : let result = LocationConfigListResponse {
1547 0 : tenant_shards: slots
1548 0 : .into_iter()
1549 0 : .map(|(tenant_shard_id, slot)| {
1550 0 : let v = match slot {
1551 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1552 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1553 0 : TenantSlot::InProgress(_) => None,
1554 : };
1555 0 : (tenant_shard_id, v)
1556 0 : })
1557 0 : .collect(),
1558 0 : };
1559 0 : json_response(StatusCode::OK, result)
1560 0 : }
1561 :
1562 0 : async fn get_location_config_handler(
1563 0 : request: Request<Body>,
1564 0 : _cancel: CancellationToken,
1565 0 : ) -> Result<Response<Body>, ApiError> {
1566 0 : let state = get_state(&request);
1567 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1568 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1569 :
1570 0 : let Some(slot) = slot else {
1571 0 : return Err(ApiError::NotFound(
1572 0 : anyhow::anyhow!("Tenant shard not found").into(),
1573 0 : ));
1574 : };
1575 :
1576 0 : let result: Option<LocationConfig> = match slot {
1577 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1578 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1579 0 : TenantSlot::InProgress(_) => None,
1580 : };
1581 :
1582 0 : json_response(StatusCode::OK, result)
1583 0 : }
1584 :
1585 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1586 : // (from all pageservers) as it invalidates consistency assumptions.
1587 0 : async fn tenant_time_travel_remote_storage_handler(
1588 0 : request: Request<Body>,
1589 0 : cancel: CancellationToken,
1590 0 : ) -> Result<Response<Body>, ApiError> {
1591 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1592 :
1593 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1594 :
1595 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1596 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1597 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1598 0 : .map_err(ApiError::BadRequest)?;
1599 :
1600 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1601 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1602 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1603 0 : .map_err(ApiError::BadRequest)?;
1604 :
1605 : // This is just a sanity check to fend off naive wrong usages of the API:
1606 : // the tenant needs to be detached *everywhere*
1607 0 : let state = get_state(&request);
1608 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1609 0 : if we_manage_tenant {
1610 0 : return Err(ApiError::BadRequest(anyhow!(
1611 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1612 0 : )));
1613 0 : }
1614 0 :
1615 0 : if timestamp > done_if_after {
1616 0 : return Err(ApiError::BadRequest(anyhow!(
1617 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1618 0 : )));
1619 0 : }
1620 0 :
1621 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1622 :
1623 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1624 0 : &state.remote_storage,
1625 0 : &tenant_shard_id,
1626 0 : timestamp,
1627 0 : done_if_after,
1628 0 : &cancel,
1629 0 : )
1630 0 : .await
1631 0 : .map_err(|e| match e {
1632 0 : TimeTravelError::BadInput(e) => {
1633 0 : warn!("bad input error: {e}");
1634 0 : ApiError::BadRequest(anyhow!("bad input error"))
1635 : }
1636 : TimeTravelError::Unimplemented => {
1637 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1638 : }
1639 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1640 : TimeTravelError::TooManyVersions => {
1641 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1642 : }
1643 0 : TimeTravelError::Other(e) => {
1644 0 : warn!("internal error: {e}");
1645 0 : ApiError::InternalServerError(anyhow!("internal error"))
1646 : }
1647 0 : })?;
1648 :
1649 0 : json_response(StatusCode::OK, ())
1650 0 : }
1651 :
1652 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1653 0 : async fn handle_tenant_break(
1654 0 : r: Request<Body>,
1655 0 : _cancel: CancellationToken,
1656 0 : ) -> Result<Response<Body>, ApiError> {
1657 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1658 :
1659 0 : let state = get_state(&r);
1660 0 : state
1661 0 : .tenant_manager
1662 0 : .get_attached_tenant_shard(tenant_shard_id)?
1663 0 : .set_broken("broken from test".to_owned())
1664 0 : .await;
1665 :
1666 0 : json_response(StatusCode::OK, ())
1667 0 : }
1668 :
1669 : // Obtains an lsn lease on the given timeline.
1670 0 : async fn lsn_lease_handler(
1671 0 : request: Request<Body>,
1672 0 : _cancel: CancellationToken,
1673 0 : ) -> Result<Response<Body>, ApiError> {
1674 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1675 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1676 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1677 :
1678 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1679 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1680 :
1681 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1682 0 :
1683 0 : let state = get_state(&request);
1684 :
1685 0 : let timeline =
1686 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1687 0 : .await?;
1688 0 : let result = timeline
1689 0 : .make_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
1690 0 : .map_err(|e| ApiError::InternalServerError(e.context("lsn lease http handler")))?;
1691 :
1692 0 : json_response(StatusCode::OK, result)
1693 0 : }
1694 :
1695 : // Run GC immediately on given timeline.
1696 0 : async fn timeline_gc_handler(
1697 0 : mut request: Request<Body>,
1698 0 : cancel: CancellationToken,
1699 0 : ) -> Result<Response<Body>, ApiError> {
1700 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1701 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1702 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1703 :
1704 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1705 :
1706 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1707 0 : let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
1708 :
1709 0 : json_response(StatusCode::OK, gc_result)
1710 0 : }
1711 :
1712 : // Run compaction immediately on given timeline.
1713 0 : async fn timeline_compact_handler(
1714 0 : request: Request<Body>,
1715 0 : cancel: CancellationToken,
1716 0 : ) -> Result<Response<Body>, ApiError> {
1717 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1718 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1719 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1720 :
1721 0 : let state = get_state(&request);
1722 0 :
1723 0 : let mut flags = EnumSet::empty();
1724 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1725 0 : flags |= CompactFlags::ForceRepartition;
1726 0 : }
1727 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1728 0 : flags |= CompactFlags::ForceImageLayerCreation;
1729 0 : }
1730 0 : let wait_until_uploaded =
1731 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1732 :
1733 0 : async {
1734 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1735 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1736 0 : timeline
1737 0 : .compact(&cancel, flags, &ctx)
1738 0 : .await
1739 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1740 0 : if wait_until_uploaded {
1741 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1742 0 : }
1743 0 : json_response(StatusCode::OK, ())
1744 0 : }
1745 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1746 0 : .await
1747 0 : }
1748 :
1749 : // Run checkpoint immediately on given timeline.
1750 0 : async fn timeline_checkpoint_handler(
1751 0 : request: Request<Body>,
1752 0 : cancel: CancellationToken,
1753 0 : ) -> Result<Response<Body>, ApiError> {
1754 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1755 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1756 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1757 :
1758 0 : let state = get_state(&request);
1759 0 :
1760 0 : let mut flags = EnumSet::empty();
1761 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1762 0 : flags |= CompactFlags::ForceRepartition;
1763 0 : }
1764 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1765 0 : flags |= CompactFlags::ForceImageLayerCreation;
1766 0 : }
1767 0 : let wait_until_uploaded =
1768 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1769 :
1770 0 : async {
1771 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1772 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1773 0 : timeline
1774 0 : .freeze_and_flush()
1775 0 : .await
1776 0 : .map_err(|e| {
1777 0 : match e {
1778 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
1779 0 : other => ApiError::InternalServerError(other.into()),
1780 :
1781 : }
1782 0 : })?;
1783 0 : timeline
1784 0 : .compact(&cancel, flags, &ctx)
1785 0 : .await
1786 0 : .map_err(|e|
1787 0 : match e {
1788 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
1789 0 : CompactionError::Other(e) => ApiError::InternalServerError(e)
1790 0 : }
1791 0 : )?;
1792 :
1793 0 : if wait_until_uploaded {
1794 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1795 0 : }
1796 :
1797 0 : json_response(StatusCode::OK, ())
1798 0 : }
1799 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1800 0 : .await
1801 0 : }
1802 :
1803 0 : async fn timeline_download_remote_layers_handler_post(
1804 0 : mut request: Request<Body>,
1805 0 : _cancel: CancellationToken,
1806 0 : ) -> Result<Response<Body>, ApiError> {
1807 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1808 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1809 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
1810 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1811 :
1812 0 : let state = get_state(&request);
1813 :
1814 0 : let timeline =
1815 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1816 0 : .await?;
1817 0 : match timeline.spawn_download_all_remote_layers(body).await {
1818 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
1819 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
1820 : }
1821 0 : }
1822 :
1823 0 : async fn timeline_download_remote_layers_handler_get(
1824 0 : request: Request<Body>,
1825 0 : _cancel: CancellationToken,
1826 0 : ) -> Result<Response<Body>, ApiError> {
1827 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1828 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1829 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1830 0 : let state = get_state(&request);
1831 :
1832 0 : let timeline =
1833 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1834 0 : .await?;
1835 0 : let info = timeline
1836 0 : .get_download_all_remote_layers_task_info()
1837 0 : .context("task never started since last pageserver process start")
1838 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1839 0 : json_response(StatusCode::OK, info)
1840 0 : }
1841 :
1842 0 : async fn timeline_detach_ancestor_handler(
1843 0 : request: Request<Body>,
1844 0 : _cancel: CancellationToken,
1845 0 : ) -> Result<Response<Body>, ApiError> {
1846 : use crate::tenant::timeline::detach_ancestor::Options;
1847 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1848 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1849 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1850 :
1851 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
1852 :
1853 0 : async move {
1854 0 : let mut options = Options::default();
1855 :
1856 0 : let rewrite_concurrency =
1857 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
1858 0 : let copy_concurrency =
1859 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
1860 :
1861 0 : [
1862 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
1863 0 : (&mut options.copy_concurrency, copy_concurrency),
1864 0 : ]
1865 0 : .into_iter()
1866 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
1867 0 : .for_each(|(target, val)| *target = val);
1868 0 :
1869 0 : let state = get_state(&request);
1870 :
1871 0 : let tenant = state
1872 0 : .tenant_manager
1873 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1874 :
1875 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1876 :
1877 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
1878 0 : let ctx = &ctx;
1879 :
1880 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1881 :
1882 0 : let (_guard, prepared) = timeline
1883 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
1884 0 : .await?;
1885 :
1886 0 : let res = state
1887 0 : .tenant_manager
1888 0 : .complete_detaching_timeline_ancestor(tenant_shard_id, timeline_id, prepared, ctx)
1889 0 : .await;
1890 :
1891 0 : match res {
1892 0 : Ok(reparented_timelines) => {
1893 0 : let resp = pageserver_api::models::detach_ancestor::AncestorDetached {
1894 0 : reparented_timelines,
1895 0 : };
1896 0 :
1897 0 : json_response(StatusCode::OK, resp)
1898 : }
1899 0 : Err(e) => Err(ApiError::InternalServerError(
1900 0 : e.context("timeline detach completion"),
1901 0 : )),
1902 : }
1903 0 : }
1904 0 : .instrument(span)
1905 0 : .await
1906 0 : }
1907 :
1908 0 : async fn deletion_queue_flush(
1909 0 : r: Request<Body>,
1910 0 : cancel: CancellationToken,
1911 0 : ) -> Result<Response<Body>, ApiError> {
1912 0 : let state = get_state(&r);
1913 :
1914 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
1915 0 :
1916 0 : let flush = async {
1917 0 : if execute {
1918 0 : state.deletion_queue_client.flush_execute().await
1919 : } else {
1920 0 : state.deletion_queue_client.flush().await
1921 : }
1922 0 : }
1923 : // DeletionQueueError's only case is shutting down.
1924 0 : .map_err(|_| ApiError::ShuttingDown);
1925 :
1926 : tokio::select! {
1927 : res = flush => {
1928 0 : res.map(|()| json_response(StatusCode::OK, ()))?
1929 : }
1930 : _ = cancel.cancelled() => {
1931 : Err(ApiError::ShuttingDown)
1932 : }
1933 : }
1934 0 : }
1935 :
1936 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
1937 0 : async fn getpage_at_lsn_handler(
1938 0 : request: Request<Body>,
1939 0 : _cancel: CancellationToken,
1940 0 : ) -> Result<Response<Body>, ApiError> {
1941 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1942 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1943 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1944 0 : let state = get_state(&request);
1945 :
1946 : struct Key(crate::repository::Key);
1947 :
1948 : impl std::str::FromStr for Key {
1949 : type Err = anyhow::Error;
1950 :
1951 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
1952 0 : crate::repository::Key::from_hex(s).map(Key)
1953 0 : }
1954 : }
1955 :
1956 0 : let key: Key = parse_query_param(&request, "key")?
1957 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
1958 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1959 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1960 :
1961 0 : async {
1962 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1963 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1964 :
1965 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
1966 :
1967 0 : Result::<_, ApiError>::Ok(
1968 0 : Response::builder()
1969 0 : .status(StatusCode::OK)
1970 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
1971 0 : .body(hyper::Body::from(page))
1972 0 : .unwrap(),
1973 0 : )
1974 0 : }
1975 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1976 0 : .await
1977 0 : }
1978 :
1979 0 : async fn timeline_collect_keyspace(
1980 0 : request: Request<Body>,
1981 0 : _cancel: CancellationToken,
1982 0 : ) -> Result<Response<Body>, ApiError> {
1983 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1984 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1985 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1986 0 : let state = get_state(&request);
1987 :
1988 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
1989 :
1990 0 : async {
1991 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1992 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1993 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
1994 0 : let (dense_ks, sparse_ks) = timeline
1995 0 : .collect_keyspace(at_lsn, &ctx)
1996 0 : .await
1997 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1998 :
1999 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2000 : // Therefore, we split dense/sparse keys in this API.
2001 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2002 0 :
2003 0 : json_response(StatusCode::OK, res)
2004 0 : }
2005 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2006 0 : .await
2007 0 : }
2008 :
2009 0 : async fn active_timeline_of_active_tenant(
2010 0 : tenant_manager: &TenantManager,
2011 0 : tenant_shard_id: TenantShardId,
2012 0 : timeline_id: TimelineId,
2013 0 : ) -> Result<Arc<Timeline>, ApiError> {
2014 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2015 :
2016 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2017 :
2018 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2019 0 : }
2020 :
2021 0 : async fn always_panic_handler(
2022 0 : req: Request<Body>,
2023 0 : _cancel: CancellationToken,
2024 0 : ) -> Result<Response<Body>, ApiError> {
2025 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2026 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2027 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2028 0 : let query = req.uri().query();
2029 0 : let _ = std::panic::catch_unwind(|| {
2030 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2031 0 : });
2032 0 : json_response(StatusCode::NO_CONTENT, ())
2033 0 : }
2034 :
2035 0 : async fn disk_usage_eviction_run(
2036 0 : mut r: Request<Body>,
2037 0 : cancel: CancellationToken,
2038 0 : ) -> Result<Response<Body>, ApiError> {
2039 0 : check_permission(&r, None)?;
2040 :
2041 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2042 : struct Config {
2043 : /// How many bytes to evict before reporting that pressure is relieved.
2044 : evict_bytes: u64,
2045 :
2046 : #[serde(default)]
2047 : eviction_order: crate::disk_usage_eviction_task::EvictionOrder,
2048 : }
2049 :
2050 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2051 : struct Usage {
2052 : // remains unchanged after instantiation of the struct
2053 : evict_bytes: u64,
2054 : // updated by `add_available_bytes`
2055 : freed_bytes: u64,
2056 : }
2057 :
2058 : impl crate::disk_usage_eviction_task::Usage for Usage {
2059 0 : fn has_pressure(&self) -> bool {
2060 0 : self.evict_bytes > self.freed_bytes
2061 0 : }
2062 :
2063 0 : fn add_available_bytes(&mut self, bytes: u64) {
2064 0 : self.freed_bytes += bytes;
2065 0 : }
2066 : }
2067 :
2068 0 : let config = json_request::<Config>(&mut r).await?;
2069 :
2070 0 : let usage = Usage {
2071 0 : evict_bytes: config.evict_bytes,
2072 0 : freed_bytes: 0,
2073 0 : };
2074 0 :
2075 0 : let state = get_state(&r);
2076 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2077 :
2078 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2079 0 : &eviction_state,
2080 0 : &state.remote_storage,
2081 0 : usage,
2082 0 : &state.tenant_manager,
2083 0 : config.eviction_order,
2084 0 : &cancel,
2085 0 : )
2086 0 : .await;
2087 :
2088 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2089 :
2090 0 : let res = res.map_err(ApiError::InternalServerError)?;
2091 :
2092 0 : json_response(StatusCode::OK, res)
2093 0 : }
2094 :
2095 0 : async fn secondary_upload_handler(
2096 0 : request: Request<Body>,
2097 0 : _cancel: CancellationToken,
2098 0 : ) -> Result<Response<Body>, ApiError> {
2099 0 : let state = get_state(&request);
2100 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2101 0 : state
2102 0 : .secondary_controller
2103 0 : .upload_tenant(tenant_shard_id)
2104 0 : .await
2105 0 : .map_err(ApiError::InternalServerError)?;
2106 :
2107 0 : json_response(StatusCode::OK, ())
2108 0 : }
2109 :
2110 0 : async fn tenant_scan_remote_handler(
2111 0 : request: Request<Body>,
2112 0 : cancel: CancellationToken,
2113 0 : ) -> Result<Response<Body>, ApiError> {
2114 0 : let state = get_state(&request);
2115 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2116 :
2117 0 : let mut response = TenantScanRemoteStorageResponse::default();
2118 :
2119 0 : let (shards, _other_keys) =
2120 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2121 0 : .await
2122 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2123 :
2124 0 : for tenant_shard_id in shards {
2125 0 : let (timeline_ids, _other_keys) =
2126 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2127 0 : .await
2128 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2129 :
2130 0 : let mut generation = Generation::none();
2131 0 : for timeline_id in timeline_ids {
2132 0 : match download_index_part(
2133 0 : &state.remote_storage,
2134 0 : &tenant_shard_id,
2135 0 : &timeline_id,
2136 0 : Generation::MAX,
2137 0 : &cancel,
2138 0 : )
2139 0 : .instrument(info_span!("download_index_part",
2140 : tenant_id=%tenant_shard_id.tenant_id,
2141 0 : shard_id=%tenant_shard_id.shard_slug(),
2142 : %timeline_id))
2143 0 : .await
2144 : {
2145 0 : Ok((index_part, index_generation)) => {
2146 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2147 0 : index_part.layer_metadata.len(), index_part.metadata.disk_consistent_lsn());
2148 0 : generation = std::cmp::max(generation, index_generation);
2149 : }
2150 : Err(DownloadError::NotFound) => {
2151 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2152 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2153 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2154 0 : continue;
2155 : }
2156 0 : Err(e) => {
2157 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2158 : }
2159 : };
2160 : }
2161 :
2162 0 : response.shards.push(TenantScanRemoteStorageShard {
2163 0 : tenant_shard_id,
2164 0 : generation: generation.into(),
2165 0 : });
2166 : }
2167 :
2168 0 : if response.shards.is_empty() {
2169 0 : return Err(ApiError::NotFound(
2170 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2171 0 : ));
2172 0 : }
2173 0 :
2174 0 : json_response(StatusCode::OK, response)
2175 0 : }
2176 :
2177 0 : async fn secondary_download_handler(
2178 0 : request: Request<Body>,
2179 0 : _cancel: CancellationToken,
2180 0 : ) -> Result<Response<Body>, ApiError> {
2181 0 : let state = get_state(&request);
2182 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2183 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2184 :
2185 : // We don't need this to issue the download request, but:
2186 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2187 : // - we will use this to provide status feedback in the response
2188 0 : let Some(secondary_tenant) = state
2189 0 : .tenant_manager
2190 0 : .get_secondary_tenant_shard(tenant_shard_id)
2191 : else {
2192 0 : return Err(ApiError::NotFound(
2193 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2194 0 : ));
2195 : };
2196 :
2197 0 : let timeout = wait.unwrap_or(Duration::MAX);
2198 :
2199 0 : let status = match tokio::time::timeout(
2200 0 : timeout,
2201 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2202 0 : )
2203 0 : .await
2204 : {
2205 : // Download job ran to completion.
2206 0 : Ok(Ok(())) => StatusCode::OK,
2207 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2208 : // okay. We could get an error here in the unlikely edge case that the tenant
2209 : // was detached between our check above and executing the download job.
2210 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2211 : // A timeout is not an error: we have started the download, we're just not done
2212 : // yet. The caller will get a response body indicating status.
2213 0 : Err(_) => StatusCode::ACCEPTED,
2214 : };
2215 :
2216 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2217 0 :
2218 0 : json_response(status, progress)
2219 0 : }
2220 :
2221 0 : async fn secondary_status_handler(
2222 0 : request: Request<Body>,
2223 0 : _cancel: CancellationToken,
2224 0 : ) -> Result<Response<Body>, ApiError> {
2225 0 : let state = get_state(&request);
2226 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2227 :
2228 0 : let Some(secondary_tenant) = state
2229 0 : .tenant_manager
2230 0 : .get_secondary_tenant_shard(tenant_shard_id)
2231 : else {
2232 0 : return Err(ApiError::NotFound(
2233 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2234 0 : ));
2235 : };
2236 :
2237 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2238 0 :
2239 0 : json_response(StatusCode::OK, progress)
2240 0 : }
2241 :
2242 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2243 0 : json_response(
2244 0 : StatusCode::NOT_FOUND,
2245 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2246 0 : )
2247 0 : }
2248 :
2249 0 : async fn post_tracing_event_handler(
2250 0 : mut r: Request<Body>,
2251 0 : _cancel: CancellationToken,
2252 0 : ) -> Result<Response<Body>, ApiError> {
2253 0 : #[derive(Debug, serde::Deserialize)]
2254 : #[serde(rename_all = "lowercase")]
2255 : enum Level {
2256 : Error,
2257 : Warn,
2258 : Info,
2259 : Debug,
2260 : Trace,
2261 : }
2262 0 : #[derive(Debug, serde::Deserialize)]
2263 : struct Request {
2264 : level: Level,
2265 : message: String,
2266 : }
2267 0 : let body: Request = json_request(&mut r)
2268 0 : .await
2269 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2270 :
2271 0 : match body.level {
2272 0 : Level::Error => tracing::error!(?body.message),
2273 0 : Level::Warn => tracing::warn!(?body.message),
2274 0 : Level::Info => tracing::info!(?body.message),
2275 0 : Level::Debug => tracing::debug!(?body.message),
2276 0 : Level::Trace => tracing::trace!(?body.message),
2277 : }
2278 :
2279 0 : json_response(StatusCode::OK, ())
2280 0 : }
2281 :
2282 0 : async fn force_aux_policy_switch_handler(
2283 0 : mut r: Request<Body>,
2284 0 : _cancel: CancellationToken,
2285 0 : ) -> Result<Response<Body>, ApiError> {
2286 0 : check_permission(&r, None)?;
2287 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
2288 0 : let timeline_id: TimelineId = parse_request_param(&r, "timeline_id")?;
2289 0 : let policy: AuxFilePolicy = json_request(&mut r).await?;
2290 :
2291 0 : let state = get_state(&r);
2292 :
2293 0 : let tenant = state
2294 0 : .tenant_manager
2295 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2296 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2297 0 : let timeline =
2298 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2299 0 : .await?;
2300 0 : timeline
2301 0 : .do_switch_aux_policy(policy)
2302 0 : .map_err(ApiError::InternalServerError)?;
2303 :
2304 0 : json_response(StatusCode::OK, ())
2305 0 : }
2306 :
2307 0 : async fn put_io_engine_handler(
2308 0 : mut r: Request<Body>,
2309 0 : _cancel: CancellationToken,
2310 0 : ) -> Result<Response<Body>, ApiError> {
2311 0 : check_permission(&r, None)?;
2312 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2313 0 : crate::virtual_file::io_engine::set(kind);
2314 0 : json_response(StatusCode::OK, ())
2315 0 : }
2316 :
2317 : /// Polled by control plane.
2318 : ///
2319 : /// See [`crate::utilization`].
2320 0 : async fn get_utilization(
2321 0 : r: Request<Body>,
2322 0 : _cancel: CancellationToken,
2323 0 : ) -> Result<Response<Body>, ApiError> {
2324 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2325 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2326 0 : });
2327 :
2328 : // this probably could be completely public, but lets make that change later.
2329 0 : check_permission(&r, None)?;
2330 :
2331 0 : let state = get_state(&r);
2332 0 : let mut g = state.latest_utilization.lock().await;
2333 :
2334 0 : let regenerate_every = Duration::from_secs(1);
2335 0 : let still_valid = g
2336 0 : .as_ref()
2337 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2338 0 :
2339 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2340 0 : // regenerate at most 1Hz to allow polling at any rate.
2341 0 : if !still_valid {
2342 0 : let path = state.conf.tenants_path();
2343 0 : let doc = crate::utilization::regenerate(path.as_std_path())
2344 0 : .map_err(ApiError::InternalServerError)?;
2345 :
2346 0 : let mut buf = Vec::new();
2347 0 : serde_json::to_writer(&mut buf, &doc)
2348 0 : .context("serialize")
2349 0 : .map_err(ApiError::InternalServerError)?;
2350 :
2351 0 : let body = bytes::Bytes::from(buf);
2352 0 :
2353 0 : *g = Some((std::time::Instant::now(), body));
2354 0 : }
2355 :
2356 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2357 0 : let cached = g.as_ref().expect("just set").1.clone();
2358 0 :
2359 0 : Response::builder()
2360 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2361 0 : // thought of using http date header, but that is second precision which does not give any
2362 0 : // debugging aid
2363 0 : .status(StatusCode::OK)
2364 0 : .body(hyper::Body::from(cached))
2365 0 : .context("build response")
2366 0 : .map_err(ApiError::InternalServerError)
2367 0 : }
2368 :
2369 0 : async fn list_aux_files(
2370 0 : mut request: Request<Body>,
2371 0 : _cancel: CancellationToken,
2372 0 : ) -> Result<Response<Body>, ApiError> {
2373 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2374 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2375 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2376 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2377 :
2378 0 : let state = get_state(&request);
2379 :
2380 0 : let timeline =
2381 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2382 0 : .await?;
2383 :
2384 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2385 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2386 0 : json_response(StatusCode::OK, files)
2387 0 : }
2388 :
2389 0 : async fn perf_info(
2390 0 : request: Request<Body>,
2391 0 : _cancel: CancellationToken,
2392 0 : ) -> Result<Response<Body>, ApiError> {
2393 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2394 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2395 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2396 :
2397 0 : let state = get_state(&request);
2398 :
2399 0 : let timeline =
2400 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2401 0 : .await?;
2402 :
2403 0 : let result = timeline.perf_info().await;
2404 :
2405 0 : json_response(StatusCode::OK, result)
2406 0 : }
2407 :
2408 0 : async fn ingest_aux_files(
2409 0 : mut request: Request<Body>,
2410 0 : _cancel: CancellationToken,
2411 0 : ) -> Result<Response<Body>, ApiError> {
2412 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2413 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2414 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2415 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2416 :
2417 0 : let state = get_state(&request);
2418 :
2419 0 : let timeline =
2420 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2421 0 : .await?;
2422 :
2423 0 : let mut modification = timeline.begin_modification(
2424 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
2425 0 : );
2426 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2427 0 : for (fname, content) in body.aux_files {
2428 0 : modification
2429 0 : .put_file(&fname, content.as_bytes(), &ctx)
2430 0 : .await
2431 0 : .map_err(ApiError::InternalServerError)?;
2432 : }
2433 0 : modification
2434 0 : .commit(&ctx)
2435 0 : .await
2436 0 : .map_err(ApiError::InternalServerError)?;
2437 :
2438 0 : json_response(StatusCode::OK, ())
2439 0 : }
2440 :
2441 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2442 : /// candidates for splitting
2443 0 : async fn post_top_tenants(
2444 0 : mut r: Request<Body>,
2445 0 : _cancel: CancellationToken,
2446 0 : ) -> Result<Response<Body>, ApiError> {
2447 0 : check_permission(&r, None)?;
2448 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2449 0 : let state = get_state(&r);
2450 0 :
2451 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2452 0 : match order_by {
2453 0 : TenantSorting::ResidentSize => sizes.resident_size,
2454 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2455 0 : }
2456 0 : }
2457 0 :
2458 0 : #[derive(Eq, PartialEq)]
2459 0 : struct HeapItem {
2460 0 : metric: u64,
2461 0 : sizes: TopTenantShardItem,
2462 0 : }
2463 0 :
2464 0 : impl PartialOrd for HeapItem {
2465 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2466 0 : Some(self.cmp(other))
2467 0 : }
2468 0 : }
2469 0 :
2470 0 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2471 0 : /// supports popping the greatest item but not the smallest.
2472 0 : impl Ord for HeapItem {
2473 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2474 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2475 0 : }
2476 0 : }
2477 0 :
2478 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2479 :
2480 : // FIXME: this is a lot of clones to take this tenant list
2481 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2482 0 : if let Some(shards_lt) = request.where_shards_lt {
2483 : // Ignore tenants which already have >= this many shards
2484 0 : if tenant_shard_id.shard_count >= shards_lt {
2485 0 : continue;
2486 0 : }
2487 0 : }
2488 :
2489 0 : let sizes = match tenant_slot {
2490 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2491 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2492 0 : continue;
2493 : }
2494 : };
2495 0 : let metric = get_size_metric(&sizes, &request.order_by);
2496 :
2497 0 : if let Some(gt) = request.where_gt {
2498 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2499 0 : if metric <= gt {
2500 0 : continue;
2501 0 : }
2502 0 : };
2503 :
2504 0 : match top_n.peek() {
2505 0 : None => {
2506 0 : // Top N list is empty: candidate becomes first member
2507 0 : top_n.push(HeapItem { metric, sizes });
2508 0 : }
2509 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
2510 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
2511 0 : top_n.push(HeapItem { metric, sizes });
2512 0 : }
2513 0 : Some(i) if i.metric > metric => {
2514 0 : // List is at limit and lowest value is greater than our candidate, drop it.
2515 0 : }
2516 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
2517 : }
2518 :
2519 0 : while top_n.len() > request.limit {
2520 0 : top_n.pop();
2521 0 : }
2522 : }
2523 :
2524 0 : json_response(
2525 0 : StatusCode::OK,
2526 0 : TopTenantShardsResponse {
2527 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
2528 0 : },
2529 0 : )
2530 0 : }
2531 :
2532 : /// Common functionality of all the HTTP API handlers.
2533 : ///
2534 : /// - Adds a tracing span to each request (by `request_span`)
2535 : /// - Logs the request depending on the request method (by `request_span`)
2536 : /// - Logs the response if it was not successful (by `request_span`
2537 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
2538 : /// Future if the connection to the client is lost, but most of the pageserver code is
2539 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
2540 : /// request with a CancellationToken.
2541 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
2542 0 : where
2543 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2544 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2545 0 : {
2546 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
2547 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
2548 0 : "failpoint".into()
2549 0 : )));
2550 :
2551 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
2552 0 : anyhow::anyhow!("failpoint")
2553 0 : )));
2554 0 : }
2555 :
2556 : // Spawn a new task to handle the request, to protect the handler from unexpected
2557 : // async cancellations. Most pageserver functions are not async cancellation safe.
2558 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
2559 : // with the cancellation token.
2560 0 : let token = CancellationToken::new();
2561 0 : let cancel_guard = token.clone().drop_guard();
2562 0 : let result = request_span(request, move |r| async {
2563 0 : let handle = tokio::spawn(
2564 0 : async {
2565 0 : let token_cloned = token.clone();
2566 0 : let result = handler(r, token).await;
2567 0 : if token_cloned.is_cancelled() {
2568 0 : // dropguard has executed: we will never turn this result into response.
2569 0 : //
2570 0 : // at least temporarily do {:?} logging; these failures are rare enough but
2571 0 : // could hide difficult errors.
2572 0 : match &result {
2573 0 : Ok(response) => {
2574 0 : let status = response.status();
2575 0 : info!(%status, "Cancelled request finished successfully")
2576 0 : }
2577 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
2578 0 : }
2579 0 : }
2580 0 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
2581 0 : // which should suffice.
2582 0 : //
2583 0 : // there is still a chance to lose the result due to race between
2584 0 : // returning from here and the actual connection closing happening
2585 0 : // before outer task gets to execute. leaving that up for #5815.
2586 0 : result
2587 0 : }
2588 0 : .in_current_span(),
2589 0 : );
2590 0 :
2591 0 : match handle.await {
2592 0 : // TODO: never actually return Err from here, always Ok(...) so that we can log
2593 0 : // spanned errors. Call api_error_handler instead and return appropriate Body.
2594 0 : Ok(result) => result,
2595 0 : Err(e) => {
2596 0 : // The handler task panicked. We have a global panic handler that logs the
2597 0 : // panic with its backtrace, so no need to log that here. Only log a brief
2598 0 : // message to make it clear that we returned the error to the client.
2599 0 : error!("HTTP request handler task panicked: {e:#}");
2600 0 :
2601 0 : // Don't return an Error here, because then fallback error handler that was
2602 0 : // installed in make_router() will print the error. Instead, construct the
2603 0 : // HTTP error response and return that.
2604 0 : Ok(
2605 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
2606 0 : .into_response(),
2607 0 : )
2608 0 : }
2609 0 : }
2610 0 : })
2611 0 : .await;
2612 :
2613 0 : cancel_guard.disarm();
2614 0 :
2615 0 : result
2616 0 : }
2617 :
2618 : /// Like api_handler, but returns an error response if the server is built without
2619 : /// the 'testing' feature.
2620 0 : async fn testing_api_handler<R, H>(
2621 0 : desc: &str,
2622 0 : request: Request<Body>,
2623 0 : handler: H,
2624 0 : ) -> Result<Response<Body>, ApiError>
2625 0 : where
2626 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2627 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2628 0 : {
2629 0 : if cfg!(feature = "testing") {
2630 0 : api_handler(request, handler).await
2631 : } else {
2632 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
2633 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
2634 0 : ))))
2635 0 : .await
2636 : }
2637 0 : }
2638 :
2639 0 : pub fn make_router(
2640 0 : state: Arc<State>,
2641 0 : launch_ts: &'static LaunchTimestamp,
2642 0 : auth: Option<Arc<SwappableJwtAuth>>,
2643 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
2644 0 : let spec = include_bytes!("openapi_spec.yml");
2645 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
2646 0 : if auth.is_some() {
2647 0 : router = router.middleware(auth_middleware(|request| {
2648 0 : let state = get_state(request);
2649 0 : if state.allowlist_routes.contains(request.uri()) {
2650 0 : None
2651 : } else {
2652 0 : state.auth.as_deref()
2653 : }
2654 0 : }))
2655 0 : }
2656 :
2657 0 : router = router.middleware(
2658 0 : endpoint::add_response_header_middleware(
2659 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
2660 0 : &launch_ts.to_string(),
2661 0 : )
2662 0 : .expect("construct launch timestamp header middleware"),
2663 0 : );
2664 0 :
2665 0 : Ok(router
2666 0 : .data(state)
2667 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
2668 0 : .get("/v1/status", |r| api_handler(r, status_handler))
2669 0 : .put("/v1/failpoints", |r| {
2670 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
2671 0 : })
2672 0 : .post("/v1/reload_auth_validation_keys", |r| {
2673 0 : api_handler(r, reload_auth_validation_keys_handler)
2674 0 : })
2675 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
2676 0 : .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
2677 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
2678 0 : api_handler(r, tenant_status)
2679 0 : })
2680 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
2681 0 : api_handler(r, tenant_delete_handler)
2682 0 : })
2683 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
2684 0 : api_handler(r, tenant_size_handler)
2685 0 : })
2686 0 : .put("/v1/tenant/config", |r| {
2687 0 : api_handler(r, update_tenant_config_handler)
2688 0 : })
2689 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
2690 0 : api_handler(r, tenant_shard_split_handler)
2691 0 : })
2692 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
2693 0 : api_handler(r, get_tenant_config_handler)
2694 0 : })
2695 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2696 0 : api_handler(r, put_tenant_location_config_handler)
2697 0 : })
2698 0 : .get("/v1/location_config", |r| {
2699 0 : api_handler(r, list_location_config_handler)
2700 0 : })
2701 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
2702 0 : api_handler(r, get_location_config_handler)
2703 0 : })
2704 0 : .put(
2705 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
2706 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
2707 0 : )
2708 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
2709 0 : api_handler(r, timeline_list_handler)
2710 0 : })
2711 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
2712 0 : api_handler(r, timeline_create_handler)
2713 0 : })
2714 0 : .post("/v1/tenant/:tenant_id/attach", |r| {
2715 0 : api_handler(r, tenant_attach_handler)
2716 0 : })
2717 0 : .post("/v1/tenant/:tenant_id/detach", |r| {
2718 0 : api_handler(r, tenant_detach_handler)
2719 0 : })
2720 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
2721 0 : api_handler(r, tenant_reset_handler)
2722 0 : })
2723 0 : .post(
2724 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
2725 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
2726 0 : )
2727 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2728 0 : api_handler(r, timeline_detail_handler)
2729 0 : })
2730 0 : .get(
2731 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
2732 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
2733 0 : )
2734 0 : .get(
2735 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
2736 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
2737 0 : )
2738 0 : .post(
2739 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
2740 0 : |r| api_handler(r, lsn_lease_handler),
2741 0 : )
2742 0 : .put(
2743 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
2744 0 : |r| api_handler(r, timeline_gc_handler),
2745 0 : )
2746 0 : .put(
2747 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
2748 0 : |r| testing_api_handler("run timeline compaction", r, timeline_compact_handler),
2749 0 : )
2750 0 : .put(
2751 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
2752 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
2753 0 : )
2754 0 : .post(
2755 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2756 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
2757 0 : )
2758 0 : .get(
2759 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2760 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
2761 0 : )
2762 0 : .put(
2763 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
2764 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
2765 0 : )
2766 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2767 0 : api_handler(r, timeline_delete_handler)
2768 0 : })
2769 0 : .get(
2770 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
2771 0 : |r| api_handler(r, layer_map_info_handler),
2772 0 : )
2773 0 : .get(
2774 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2775 0 : |r| api_handler(r, layer_download_handler),
2776 0 : )
2777 0 : .delete(
2778 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2779 0 : |r| api_handler(r, evict_timeline_layer_handler),
2780 0 : )
2781 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
2782 0 : api_handler(r, secondary_upload_handler)
2783 0 : })
2784 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
2785 0 : api_handler(r, tenant_scan_remote_handler)
2786 0 : })
2787 0 : .put("/v1/disk_usage_eviction/run", |r| {
2788 0 : api_handler(r, disk_usage_eviction_run)
2789 0 : })
2790 0 : .put("/v1/deletion_queue/flush", |r| {
2791 0 : api_handler(r, deletion_queue_flush)
2792 0 : })
2793 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
2794 0 : api_handler(r, secondary_status_handler)
2795 0 : })
2796 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
2797 0 : api_handler(r, secondary_download_handler)
2798 0 : })
2799 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
2800 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
2801 0 : })
2802 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
2803 0 : .post("/v1/tracing/event", |r| {
2804 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
2805 0 : })
2806 0 : .get(
2807 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
2808 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
2809 0 : )
2810 0 : .get(
2811 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
2812 0 : |r| api_handler(r, timeline_collect_keyspace),
2813 0 : )
2814 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
2815 0 : .put(
2816 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/force_aux_policy_switch",
2817 0 : |r| api_handler(r, force_aux_policy_switch_handler),
2818 0 : )
2819 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
2820 0 : .post(
2821 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
2822 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
2823 0 : )
2824 0 : .post(
2825 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
2826 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
2827 0 : )
2828 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
2829 0 : .post(
2830 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
2831 0 : |r| testing_api_handler("perf_info", r, perf_info),
2832 0 : )
2833 0 : .any(handler_404))
2834 0 : }
|
