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::ShardParameters;
25 : use pageserver_api::models::TenantDetails;
26 : use pageserver_api::models::TenantLocationConfigResponse;
27 : use pageserver_api::models::TenantScanRemoteStorageResponse;
28 : use pageserver_api::models::TenantScanRemoteStorageShard;
29 : use pageserver_api::models::TenantShardLocation;
30 : use pageserver_api::models::TenantShardSplitRequest;
31 : use pageserver_api::models::TenantShardSplitResponse;
32 : use pageserver_api::models::TenantSorting;
33 : use pageserver_api::models::TenantState;
34 : use pageserver_api::models::TopTenantShardItem;
35 : use pageserver_api::models::TopTenantShardsRequest;
36 : use pageserver_api::models::TopTenantShardsResponse;
37 : use pageserver_api::models::{
38 : DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
39 : TenantLoadRequest, TenantLocationConfigRequest,
40 : };
41 : use pageserver_api::shard::ShardCount;
42 : use pageserver_api::shard::TenantShardId;
43 : use remote_storage::DownloadError;
44 : use remote_storage::GenericRemoteStorage;
45 : use remote_storage::TimeTravelError;
46 : use tenant_size_model::{SizeResult, StorageModel};
47 : use tokio_util::sync::CancellationToken;
48 : use tracing::*;
49 : use utils::auth::JwtAuth;
50 : use utils::failpoint_support::failpoints_handler;
51 : use utils::http::endpoint::prometheus_metrics_handler;
52 : use utils::http::endpoint::request_span;
53 : use utils::http::json::json_request_or_empty_body;
54 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
55 :
56 : use crate::context::{DownloadBehavior, RequestContext};
57 : use crate::deletion_queue::DeletionQueueClient;
58 : use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
59 : use crate::pgdatadir_mapping::LsnForTimestamp;
60 : use crate::task_mgr::TaskKind;
61 : use crate::tenant::config::{LocationConf, TenantConfOpt};
62 : use crate::tenant::mgr::GetActiveTenantError;
63 : use crate::tenant::mgr::{
64 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
65 : TenantSlotUpsertError, TenantStateError,
66 : };
67 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
68 : use crate::tenant::remote_timeline_client;
69 : use crate::tenant::remote_timeline_client::download_index_part;
70 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
71 : use crate::tenant::remote_timeline_client::list_remote_timelines;
72 : use crate::tenant::secondary::SecondaryController;
73 : use crate::tenant::size::ModelInputs;
74 : use crate::tenant::storage_layer::LayerAccessStatsReset;
75 : use crate::tenant::storage_layer::LayerName;
76 : use crate::tenant::timeline::CompactFlags;
77 : use crate::tenant::timeline::CompactionError;
78 : use crate::tenant::timeline::Timeline;
79 : use crate::tenant::GetTimelineError;
80 : use crate::tenant::SpawnMode;
81 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
82 : use crate::{config::PageServerConf, tenant::mgr};
83 : use crate::{disk_usage_eviction_task, tenant};
84 : use pageserver_api::models::{
85 : StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
86 : TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
87 : };
88 : use utils::{
89 : auth::SwappableJwtAuth,
90 : generation::Generation,
91 : http::{
92 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
93 : error::{ApiError, HttpErrorBody},
94 : json::{json_request, json_response},
95 : request::parse_request_param,
96 : RequestExt, RouterBuilder,
97 : },
98 : id::{TenantId, TimelineId},
99 : lsn::Lsn,
100 : };
101 :
102 : // For APIs that require an Active tenant, how long should we block waiting for that state?
103 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
104 : // failed API calls while tenants are activating.
105 : #[cfg(not(feature = "testing"))]
106 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
107 :
108 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
109 : // finish attaching, if calls to remote storage are slow.
110 : #[cfg(feature = "testing")]
111 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
112 :
113 : pub struct State {
114 : conf: &'static PageServerConf,
115 : tenant_manager: Arc<TenantManager>,
116 : auth: Option<Arc<SwappableJwtAuth>>,
117 : allowlist_routes: Vec<Uri>,
118 : remote_storage: GenericRemoteStorage,
119 : broker_client: storage_broker::BrokerClientChannel,
120 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
121 : deletion_queue_client: DeletionQueueClient,
122 : secondary_controller: SecondaryController,
123 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
124 : }
125 :
126 : impl State {
127 : #[allow(clippy::too_many_arguments)]
128 0 : pub fn new(
129 0 : conf: &'static PageServerConf,
130 0 : tenant_manager: Arc<TenantManager>,
131 0 : auth: Option<Arc<SwappableJwtAuth>>,
132 0 : remote_storage: GenericRemoteStorage,
133 0 : broker_client: storage_broker::BrokerClientChannel,
134 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
135 0 : deletion_queue_client: DeletionQueueClient,
136 0 : secondary_controller: SecondaryController,
137 0 : ) -> anyhow::Result<Self> {
138 0 : let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
139 0 : .iter()
140 0 : .map(|v| v.parse().unwrap())
141 0 : .collect::<Vec<_>>();
142 0 : Ok(Self {
143 0 : conf,
144 0 : tenant_manager,
145 0 : auth,
146 0 : allowlist_routes,
147 0 : remote_storage,
148 0 : broker_client,
149 0 : disk_usage_eviction_state,
150 0 : deletion_queue_client,
151 0 : secondary_controller,
152 0 : latest_utilization: Default::default(),
153 0 : })
154 0 : }
155 : }
156 :
157 : #[inline(always)]
158 0 : fn get_state(request: &Request<Body>) -> &State {
159 0 : request
160 0 : .data::<Arc<State>>()
161 0 : .expect("unknown state type")
162 0 : .as_ref()
163 0 : }
164 :
165 : #[inline(always)]
166 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
167 0 : get_state(request).conf
168 0 : }
169 :
170 : /// Check that the requester is authorized to operate on given tenant
171 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
172 0 : check_permission_with(request, |claims| {
173 0 : crate::auth::check_permission(claims, tenant_id)
174 0 : })
175 0 : }
176 :
177 : impl From<PageReconstructError> for ApiError {
178 0 : fn from(pre: PageReconstructError) -> ApiError {
179 0 : match pre {
180 0 : PageReconstructError::Other(pre) => ApiError::InternalServerError(pre),
181 0 : PageReconstructError::MissingKey(e) => {
182 0 : ApiError::InternalServerError(anyhow::anyhow!("{e}"))
183 : }
184 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
185 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
186 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
187 : }
188 0 : }
189 : }
190 :
191 : impl From<TenantMapInsertError> for ApiError {
192 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
193 0 : match tmie {
194 0 : TenantMapInsertError::SlotError(e) => e.into(),
195 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
196 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
197 : }
198 0 : }
199 : }
200 :
201 : impl From<TenantSlotError> for ApiError {
202 0 : fn from(e: TenantSlotError) -> ApiError {
203 0 : use TenantSlotError::*;
204 0 : match e {
205 0 : NotFound(tenant_id) => {
206 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
207 : }
208 0 : e @ AlreadyExists(_, _) => ApiError::Conflict(format!("{e}")),
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::delete::DeleteTenantError> for ApiError {
334 0 : fn from(value: crate::tenant::delete::DeleteTenantError) -> Self {
335 0 : use crate::tenant::delete::DeleteTenantError::*;
336 0 : match value {
337 0 : Get(g) => ApiError::from(g),
338 0 : e @ AlreadyInProgress => ApiError::Conflict(e.to_string()),
339 0 : Timeline(t) => ApiError::from(t),
340 0 : NotAttached => ApiError::NotFound(anyhow::anyhow!("Tenant is not attached").into()),
341 0 : SlotError(e) => e.into(),
342 0 : SlotUpsertError(e) => e.into(),
343 0 : Other(o) => ApiError::InternalServerError(o),
344 0 : e @ InvalidState(_) => ApiError::PreconditionFailed(e.to_string().into_boxed_str()),
345 0 : Cancelled => ApiError::ShuttingDown,
346 : }
347 0 : }
348 : }
349 :
350 : // Helper function to construct a TimelineInfo struct for a timeline
351 0 : async fn build_timeline_info(
352 0 : timeline: &Arc<Timeline>,
353 0 : include_non_incremental_logical_size: bool,
354 0 : force_await_initial_logical_size: bool,
355 0 : ctx: &RequestContext,
356 0 : ) -> anyhow::Result<TimelineInfo> {
357 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
358 0 :
359 0 : if force_await_initial_logical_size {
360 0 : timeline.clone().await_initial_logical_size().await
361 0 : }
362 :
363 0 : let mut info = build_timeline_info_common(
364 0 : timeline,
365 0 : ctx,
366 0 : tenant::timeline::GetLogicalSizePriority::Background,
367 0 : )
368 0 : .await?;
369 0 : if include_non_incremental_logical_size {
370 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
371 : // Otherwise, if someone deletes the timeline / detaches the tenant while
372 : // we're executing this function, we will outlive the timeline on-disk state.
373 : info.current_logical_size_non_incremental = Some(
374 0 : timeline
375 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
376 0 : .await?,
377 : );
378 0 : }
379 0 : Ok(info)
380 0 : }
381 :
382 0 : async fn build_timeline_info_common(
383 0 : timeline: &Arc<Timeline>,
384 0 : ctx: &RequestContext,
385 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
386 0 : ) -> anyhow::Result<TimelineInfo> {
387 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
388 0 : let initdb_lsn = timeline.initdb_lsn;
389 0 : let last_record_lsn = timeline.get_last_record_lsn();
390 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
391 0 : let guard = timeline.last_received_wal.lock().unwrap();
392 0 : if let Some(info) = guard.as_ref() {
393 0 : (
394 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
395 0 : Some(info.last_received_msg_lsn),
396 0 : Some(info.last_received_msg_ts),
397 0 : )
398 : } else {
399 0 : (None, None, None)
400 : }
401 : };
402 :
403 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
404 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
405 0 : Lsn(0) => None,
406 0 : lsn @ Lsn(_) => Some(lsn),
407 : };
408 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
409 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
410 0 : let state = timeline.current_state();
411 0 : let remote_consistent_lsn_projected = timeline
412 0 : .get_remote_consistent_lsn_projected()
413 0 : .unwrap_or(Lsn(0));
414 0 : let remote_consistent_lsn_visible = timeline
415 0 : .get_remote_consistent_lsn_visible()
416 0 : .unwrap_or(Lsn(0));
417 0 :
418 0 : let walreceiver_status = timeline.walreceiver_status();
419 :
420 0 : let info = TimelineInfo {
421 0 : tenant_id: timeline.tenant_shard_id,
422 0 : timeline_id: timeline.timeline_id,
423 0 : ancestor_timeline_id,
424 0 : ancestor_lsn,
425 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
426 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
427 0 : remote_consistent_lsn_visible,
428 0 : initdb_lsn,
429 0 : last_record_lsn,
430 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
431 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
432 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
433 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
434 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
435 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
436 : },
437 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
438 0 : current_physical_size,
439 0 : current_logical_size_non_incremental: None,
440 0 : timeline_dir_layer_file_size_sum: None,
441 0 : wal_source_connstr,
442 0 : last_received_msg_lsn,
443 0 : last_received_msg_ts,
444 0 : pg_version: timeline.pg_version,
445 0 :
446 0 : state,
447 0 :
448 0 : walreceiver_status,
449 0 :
450 0 : last_aux_file_policy: timeline.last_aux_file_policy.load(),
451 0 : };
452 0 : Ok(info)
453 0 : }
454 :
455 : // healthcheck handler
456 0 : async fn status_handler(
457 0 : request: Request<Body>,
458 0 : _cancel: CancellationToken,
459 0 : ) -> Result<Response<Body>, ApiError> {
460 0 : check_permission(&request, None)?;
461 0 : let config = get_config(&request);
462 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
463 0 : }
464 :
465 0 : async fn reload_auth_validation_keys_handler(
466 0 : request: Request<Body>,
467 0 : _cancel: CancellationToken,
468 0 : ) -> Result<Response<Body>, ApiError> {
469 0 : check_permission(&request, None)?;
470 0 : let config = get_config(&request);
471 0 : let state = get_state(&request);
472 0 : let Some(shared_auth) = &state.auth else {
473 0 : return json_response(StatusCode::BAD_REQUEST, ());
474 : };
475 : // unwrap is ok because check is performed when creating config, so path is set and exists
476 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
477 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
478 :
479 0 : match JwtAuth::from_key_path(key_path) {
480 0 : Ok(new_auth) => {
481 0 : shared_auth.swap(new_auth);
482 0 : json_response(StatusCode::OK, ())
483 : }
484 0 : Err(e) => {
485 0 : let err_msg = "Error reloading public keys";
486 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
487 0 : json_response(
488 0 : StatusCode::INTERNAL_SERVER_ERROR,
489 0 : HttpErrorBody::from_msg(err_msg.to_string()),
490 0 : )
491 : }
492 : }
493 0 : }
494 :
495 0 : async fn timeline_create_handler(
496 0 : mut request: Request<Body>,
497 0 : _cancel: CancellationToken,
498 0 : ) -> Result<Response<Body>, ApiError> {
499 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
500 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
501 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
502 :
503 0 : let new_timeline_id = request_data.new_timeline_id;
504 0 :
505 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
506 0 :
507 0 : let state = get_state(&request);
508 :
509 0 : async {
510 0 : let tenant = state
511 0 : .tenant_manager
512 0 : .get_attached_tenant_shard(tenant_shard_id)?;
513 :
514 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
515 :
516 0 : if let Some(ancestor_id) = request_data.ancestor_timeline_id.as_ref() {
517 0 : tracing::info!(%ancestor_id, "starting to branch");
518 : } else {
519 0 : tracing::info!("bootstrapping");
520 : }
521 :
522 0 : match tenant
523 0 : .create_timeline(
524 0 : new_timeline_id,
525 0 : request_data.ancestor_timeline_id,
526 0 : request_data.ancestor_start_lsn,
527 0 : request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
528 0 : request_data.existing_initdb_timeline_id,
529 0 : state.broker_client.clone(),
530 0 : &ctx,
531 0 : )
532 0 : .await
533 : {
534 0 : Ok(new_timeline) => {
535 : // Created. Construct a TimelineInfo for it.
536 0 : let timeline_info = build_timeline_info_common(
537 0 : &new_timeline,
538 0 : &ctx,
539 0 : tenant::timeline::GetLogicalSizePriority::User,
540 0 : )
541 0 : .await
542 0 : .map_err(ApiError::InternalServerError)?;
543 0 : json_response(StatusCode::CREATED, timeline_info)
544 : }
545 0 : Err(_) if tenant.cancel.is_cancelled() => {
546 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
547 0 : json_response(
548 0 : StatusCode::SERVICE_UNAVAILABLE,
549 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
550 0 : )
551 : }
552 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
553 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
554 : }
555 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
556 0 : StatusCode::TOO_MANY_REQUESTS,
557 0 : HttpErrorBody::from_msg(e.to_string()),
558 0 : ),
559 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
560 0 : StatusCode::NOT_ACCEPTABLE,
561 0 : HttpErrorBody::from_msg(format!("{err:#}")),
562 0 : ),
563 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
564 0 : StatusCode::SERVICE_UNAVAILABLE,
565 0 : HttpErrorBody::from_msg(e.to_string()),
566 0 : ),
567 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
568 0 : StatusCode::SERVICE_UNAVAILABLE,
569 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
570 0 : ),
571 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
572 : }
573 0 : }
574 0 : .instrument(info_span!("timeline_create",
575 : tenant_id = %tenant_shard_id.tenant_id,
576 0 : shard_id = %tenant_shard_id.shard_slug(),
577 : timeline_id = %new_timeline_id,
578 : lsn=?request_data.ancestor_start_lsn,
579 : pg_version=?request_data.pg_version
580 : ))
581 0 : .await
582 0 : }
583 :
584 0 : async fn timeline_list_handler(
585 0 : request: Request<Body>,
586 0 : _cancel: CancellationToken,
587 0 : ) -> Result<Response<Body>, ApiError> {
588 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
589 0 : let include_non_incremental_logical_size: Option<bool> =
590 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
591 0 : let force_await_initial_logical_size: Option<bool> =
592 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
593 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
594 :
595 0 : let state = get_state(&request);
596 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
597 :
598 0 : let response_data = async {
599 0 : let tenant = state
600 0 : .tenant_manager
601 0 : .get_attached_tenant_shard(tenant_shard_id)?;
602 :
603 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
604 :
605 0 : let timelines = tenant.list_timelines();
606 0 :
607 0 : let mut response_data = Vec::with_capacity(timelines.len());
608 0 : for timeline in timelines {
609 0 : let timeline_info = build_timeline_info(
610 0 : &timeline,
611 0 : include_non_incremental_logical_size.unwrap_or(false),
612 0 : force_await_initial_logical_size.unwrap_or(false),
613 0 : &ctx,
614 0 : )
615 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
616 0 : .await
617 0 : .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
618 0 : .map_err(ApiError::InternalServerError)?;
619 :
620 0 : response_data.push(timeline_info);
621 : }
622 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
623 0 : }
624 0 : .instrument(info_span!("timeline_list",
625 : tenant_id = %tenant_shard_id.tenant_id,
626 0 : shard_id = %tenant_shard_id.shard_slug()))
627 0 : .await?;
628 :
629 0 : json_response(StatusCode::OK, response_data)
630 0 : }
631 :
632 0 : async fn timeline_preserve_initdb_handler(
633 0 : request: Request<Body>,
634 0 : _cancel: CancellationToken,
635 0 : ) -> Result<Response<Body>, ApiError> {
636 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
637 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
638 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
639 0 : let state = get_state(&request);
640 :
641 : // Part of the process for disaster recovery from safekeeper-stored WAL:
642 : // If we don't recover into a new timeline but want to keep the timeline ID,
643 : // then the initdb archive is deleted. This endpoint copies it to a different
644 : // location where timeline recreation cand find it.
645 :
646 0 : async {
647 0 : let tenant = state
648 0 : .tenant_manager
649 0 : .get_attached_tenant_shard(tenant_shard_id)?;
650 :
651 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
652 :
653 0 : timeline
654 0 : .preserve_initdb_archive()
655 0 : .await
656 0 : .context("preserving initdb archive")
657 0 : .map_err(ApiError::InternalServerError)?;
658 :
659 0 : Ok::<_, ApiError>(())
660 0 : }
661 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
662 : tenant_id = %tenant_shard_id.tenant_id,
663 0 : shard_id = %tenant_shard_id.shard_slug(),
664 : %timeline_id))
665 0 : .await?;
666 :
667 0 : json_response(StatusCode::OK, ())
668 0 : }
669 :
670 0 : async fn timeline_detail_handler(
671 0 : request: Request<Body>,
672 0 : _cancel: CancellationToken,
673 0 : ) -> Result<Response<Body>, ApiError> {
674 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
675 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
676 0 : let include_non_incremental_logical_size: Option<bool> =
677 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
678 0 : let force_await_initial_logical_size: Option<bool> =
679 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
680 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
681 :
682 : // Logical size calculation needs downloading.
683 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
684 0 : let state = get_state(&request);
685 :
686 0 : let timeline_info = async {
687 0 : let tenant = state
688 0 : .tenant_manager
689 0 : .get_attached_tenant_shard(tenant_shard_id)?;
690 :
691 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
692 :
693 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
694 :
695 0 : let timeline_info = build_timeline_info(
696 0 : &timeline,
697 0 : include_non_incremental_logical_size.unwrap_or(false),
698 0 : force_await_initial_logical_size.unwrap_or(false),
699 0 : &ctx,
700 0 : )
701 0 : .await
702 0 : .context("get local timeline info")
703 0 : .map_err(ApiError::InternalServerError)?;
704 :
705 0 : Ok::<_, ApiError>(timeline_info)
706 0 : }
707 0 : .instrument(info_span!("timeline_detail",
708 : tenant_id = %tenant_shard_id.tenant_id,
709 0 : shard_id = %tenant_shard_id.shard_slug(),
710 : %timeline_id))
711 0 : .await?;
712 :
713 0 : json_response(StatusCode::OK, timeline_info)
714 0 : }
715 :
716 0 : async fn get_lsn_by_timestamp_handler(
717 0 : request: Request<Body>,
718 0 : cancel: CancellationToken,
719 0 : ) -> Result<Response<Body>, ApiError> {
720 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
721 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
722 0 : let state = get_state(&request);
723 0 :
724 0 : if !tenant_shard_id.is_shard_zero() {
725 : // Requires SLRU contents, which are only stored on shard zero
726 0 : return Err(ApiError::BadRequest(anyhow!(
727 0 : "Size calculations are only available on shard zero"
728 0 : )));
729 0 : }
730 :
731 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
732 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
733 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
734 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
735 0 : .map_err(ApiError::BadRequest)?;
736 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
737 0 :
738 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
739 :
740 0 : let timeline =
741 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
742 0 : .await?;
743 0 : let result = timeline
744 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
745 0 : .await?;
746 : #[derive(serde::Serialize, Debug)]
747 : struct Result {
748 : lsn: Lsn,
749 : kind: &'static str,
750 : }
751 0 : let (lsn, kind) = match result {
752 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
753 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
754 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
755 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
756 : };
757 0 : let result = Result { lsn, kind };
758 0 : tracing::info!(
759 : lsn=?result.lsn,
760 : kind=%result.kind,
761 : timestamp=%timestamp_raw,
762 0 : "lsn_by_timestamp finished"
763 : );
764 0 : json_response(StatusCode::OK, result)
765 0 : }
766 :
767 0 : async fn get_timestamp_of_lsn_handler(
768 0 : request: Request<Body>,
769 0 : _cancel: CancellationToken,
770 0 : ) -> Result<Response<Body>, ApiError> {
771 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
772 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
773 0 : let state = get_state(&request);
774 0 :
775 0 : if !tenant_shard_id.is_shard_zero() {
776 : // Requires SLRU contents, which are only stored on shard zero
777 0 : return Err(ApiError::BadRequest(anyhow!(
778 0 : "Size calculations are only available on shard zero"
779 0 : )));
780 0 : }
781 :
782 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
783 :
784 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
785 0 : let lsn = Lsn::from_str(&lsn_str)
786 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
787 0 : .map_err(ApiError::BadRequest)?;
788 :
789 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
790 0 : let timeline =
791 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
792 0 : .await?;
793 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
794 :
795 0 : match result {
796 0 : Some(time) => {
797 0 : let time = format_rfc3339(postgres_ffi::from_pg_timestamp(time)).to_string();
798 0 : json_response(StatusCode::OK, time)
799 : }
800 0 : None => Err(ApiError::NotFound(
801 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
802 0 : )),
803 : }
804 0 : }
805 :
806 0 : async fn tenant_attach_handler(
807 0 : mut request: Request<Body>,
808 0 : _cancel: CancellationToken,
809 0 : ) -> Result<Response<Body>, ApiError> {
810 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
811 0 : check_permission(&request, Some(tenant_id))?;
812 :
813 0 : let maybe_body: Option<TenantAttachRequest> = json_request_or_empty_body(&mut request).await?;
814 0 : let tenant_conf = match &maybe_body {
815 0 : Some(request) => TenantConfOpt::try_from(&*request.config).map_err(ApiError::BadRequest)?,
816 0 : None => TenantConfOpt::default(),
817 : };
818 :
819 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
820 0 :
821 0 : info!("Handling tenant attach {tenant_id}");
822 :
823 0 : let state = get_state(&request);
824 :
825 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
826 :
827 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
828 0 : let shard_params = ShardParameters::default();
829 0 : let location_conf = LocationConf::attached_single(tenant_conf, generation, &shard_params);
830 :
831 0 : let tenant = state
832 0 : .tenant_manager
833 0 : .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
834 0 : .await?;
835 :
836 0 : let Some(tenant) = tenant else {
837 : // This should never happen: indicates a bug in upsert_location
838 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
839 0 : "Upsert succeeded but didn't return tenant!"
840 0 : )));
841 : };
842 :
843 : // We might have successfully constructed a Tenant, but it could still
844 : // end up in a broken state:
845 : if let TenantState::Broken {
846 0 : reason,
847 : backtrace: _,
848 0 : } = tenant.current_state()
849 : {
850 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
851 0 : "Tenant state is Broken: {reason}"
852 0 : )));
853 0 : }
854 0 :
855 0 : json_response(StatusCode::ACCEPTED, ())
856 0 : }
857 :
858 0 : async fn timeline_delete_handler(
859 0 : request: Request<Body>,
860 0 : _cancel: CancellationToken,
861 0 : ) -> Result<Response<Body>, ApiError> {
862 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
863 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
864 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
865 :
866 0 : let state = get_state(&request);
867 :
868 0 : let tenant = state
869 0 : .tenant_manager
870 0 : .get_attached_tenant_shard(tenant_shard_id)
871 0 : .map_err(|e| {
872 0 : match e {
873 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
874 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
875 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
876 0 : "Requested tenant is missing".to_string().into_boxed_str(),
877 0 : ),
878 0 : e => e.into(),
879 : }
880 0 : })?;
881 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
882 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))
883 0 : .await?;
884 :
885 0 : json_response(StatusCode::ACCEPTED, ())
886 0 : }
887 :
888 0 : async fn tenant_detach_handler(
889 0 : request: Request<Body>,
890 0 : _cancel: CancellationToken,
891 0 : ) -> Result<Response<Body>, ApiError> {
892 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
893 0 : check_permission(&request, Some(tenant_id))?;
894 0 : let detach_ignored: Option<bool> = parse_query_param(&request, "detach_ignored")?;
895 :
896 : // This is a legacy API (`/location_conf` is the replacement). It only supports unsharded tenants
897 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
898 0 :
899 0 : let state = get_state(&request);
900 0 : let conf = state.conf;
901 0 : state
902 0 : .tenant_manager
903 0 : .detach_tenant(
904 0 : conf,
905 0 : tenant_shard_id,
906 0 : detach_ignored.unwrap_or(false),
907 0 : &state.deletion_queue_client,
908 0 : )
909 0 : .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
910 0 : .await?;
911 :
912 0 : json_response(StatusCode::OK, ())
913 0 : }
914 :
915 0 : async fn tenant_reset_handler(
916 0 : request: Request<Body>,
917 0 : _cancel: CancellationToken,
918 0 : ) -> Result<Response<Body>, ApiError> {
919 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
920 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
921 :
922 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
923 :
924 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
925 0 : let state = get_state(&request);
926 0 : state
927 0 : .tenant_manager
928 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
929 0 : .await
930 0 : .map_err(ApiError::InternalServerError)?;
931 :
932 0 : json_response(StatusCode::OK, ())
933 0 : }
934 :
935 0 : async fn tenant_load_handler(
936 0 : mut request: Request<Body>,
937 0 : _cancel: CancellationToken,
938 0 : ) -> Result<Response<Body>, ApiError> {
939 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
940 0 : check_permission(&request, Some(tenant_id))?;
941 :
942 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
943 :
944 0 : let maybe_body: Option<TenantLoadRequest> = json_request_or_empty_body(&mut request).await?;
945 :
946 0 : let state = get_state(&request);
947 :
948 : // The /load request is only usable when control_plane_api is not set. Once it is set, callers
949 : // should always use /attach instead.
950 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
951 :
952 0 : mgr::load_tenant(
953 0 : state.conf,
954 0 : tenant_id,
955 0 : generation,
956 0 : state.broker_client.clone(),
957 0 : state.remote_storage.clone(),
958 0 : state.deletion_queue_client.clone(),
959 0 : &ctx,
960 0 : )
961 0 : .instrument(info_span!("load", %tenant_id))
962 0 : .await?;
963 :
964 0 : json_response(StatusCode::ACCEPTED, ())
965 0 : }
966 :
967 0 : async fn tenant_ignore_handler(
968 0 : request: Request<Body>,
969 0 : _cancel: CancellationToken,
970 0 : ) -> Result<Response<Body>, ApiError> {
971 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
972 0 : check_permission(&request, Some(tenant_id))?;
973 :
974 0 : let state = get_state(&request);
975 0 : let conf = state.conf;
976 0 : mgr::ignore_tenant(conf, tenant_id)
977 0 : .instrument(info_span!("ignore_tenant", %tenant_id))
978 0 : .await?;
979 :
980 0 : json_response(StatusCode::OK, ())
981 0 : }
982 :
983 0 : async fn tenant_list_handler(
984 0 : request: Request<Body>,
985 0 : _cancel: CancellationToken,
986 0 : ) -> Result<Response<Body>, ApiError> {
987 0 : check_permission(&request, None)?;
988 0 : let state = get_state(&request);
989 :
990 0 : let response_data = state
991 0 : .tenant_manager
992 0 : .list_tenants()
993 0 : .map_err(|_| {
994 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
995 0 : })?
996 0 : .iter()
997 0 : .map(|(id, state, gen)| TenantInfo {
998 0 : id: *id,
999 0 : state: state.clone(),
1000 0 : current_physical_size: None,
1001 0 : attachment_status: state.attachment_status(),
1002 0 : generation: (*gen).into(),
1003 0 : })
1004 0 : .collect::<Vec<TenantInfo>>();
1005 0 :
1006 0 : json_response(StatusCode::OK, response_data)
1007 0 : }
1008 :
1009 0 : async fn tenant_status(
1010 0 : request: Request<Body>,
1011 0 : _cancel: CancellationToken,
1012 0 : ) -> Result<Response<Body>, ApiError> {
1013 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1014 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1015 0 : let state = get_state(&request);
1016 0 :
1017 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1018 0 : let activate = true;
1019 : #[cfg(feature = "testing")]
1020 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1021 :
1022 0 : let tenant_info = async {
1023 0 : let tenant = state
1024 0 : .tenant_manager
1025 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1026 :
1027 0 : if activate {
1028 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1029 : // called, but it is still valid to return 200 and describe the current state of the tenant
1030 : // if it doesn't make it into an active state.
1031 0 : tenant
1032 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1033 0 : .await
1034 0 : .ok();
1035 0 : }
1036 :
1037 : // Calculate total physical size of all timelines
1038 0 : let mut current_physical_size = 0;
1039 0 : for timeline in tenant.list_timelines().iter() {
1040 0 : current_physical_size += timeline.layer_size_sum().await;
1041 : }
1042 :
1043 0 : let state = tenant.current_state();
1044 0 : Result::<_, ApiError>::Ok(TenantDetails {
1045 0 : tenant_info: TenantInfo {
1046 0 : id: tenant_shard_id,
1047 0 : state: state.clone(),
1048 0 : current_physical_size: Some(current_physical_size),
1049 0 : attachment_status: state.attachment_status(),
1050 0 : generation: tenant.generation().into(),
1051 0 : },
1052 0 : walredo: tenant.wal_redo_manager_status(),
1053 0 : timelines: tenant.list_timeline_ids(),
1054 0 : })
1055 0 : }
1056 0 : .instrument(info_span!("tenant_status_handler",
1057 : tenant_id = %tenant_shard_id.tenant_id,
1058 0 : shard_id = %tenant_shard_id.shard_slug()))
1059 0 : .await?;
1060 :
1061 0 : json_response(StatusCode::OK, tenant_info)
1062 0 : }
1063 :
1064 0 : async fn tenant_delete_handler(
1065 0 : request: Request<Body>,
1066 0 : _cancel: CancellationToken,
1067 0 : ) -> Result<Response<Body>, ApiError> {
1068 : // TODO openapi spec
1069 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1070 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1071 :
1072 0 : let state = get_state(&request);
1073 :
1074 0 : let status = state
1075 0 : .tenant_manager
1076 0 : .delete_tenant(tenant_shard_id, ACTIVE_TENANT_TIMEOUT)
1077 0 : .instrument(info_span!("tenant_delete_handler",
1078 : tenant_id = %tenant_shard_id.tenant_id,
1079 0 : shard_id = %tenant_shard_id.shard_slug()
1080 : ))
1081 0 : .await?;
1082 :
1083 : // Callers use 404 as success for deletions, for historical reasons.
1084 0 : if status == StatusCode::NOT_FOUND {
1085 0 : return Err(ApiError::NotFound(
1086 0 : anyhow::anyhow!("Deletion complete").into(),
1087 0 : ));
1088 0 : }
1089 0 :
1090 0 : json_response(status, ())
1091 0 : }
1092 :
1093 : /// HTTP endpoint to query the current tenant_size of a tenant.
1094 : ///
1095 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1096 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1097 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1098 : /// values.
1099 : ///
1100 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1101 : /// (only if it is shorter than the real cutoff).
1102 : ///
1103 : /// Note: we don't update the cached size and prometheus metric here.
1104 : /// The retention period might be different, and it's nice to have a method to just calculate it
1105 : /// without modifying anything anyway.
1106 0 : async fn tenant_size_handler(
1107 0 : request: Request<Body>,
1108 0 : cancel: CancellationToken,
1109 0 : ) -> Result<Response<Body>, ApiError> {
1110 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1111 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1112 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1113 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1114 0 : let headers = request.headers();
1115 0 : let state = get_state(&request);
1116 0 :
1117 0 : if !tenant_shard_id.is_shard_zero() {
1118 0 : return Err(ApiError::BadRequest(anyhow!(
1119 0 : "Size calculations are only available on shard zero"
1120 0 : )));
1121 0 : }
1122 0 :
1123 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1124 0 : let tenant = state
1125 0 : .tenant_manager
1126 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1127 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1128 :
1129 : // this can be long operation
1130 0 : let inputs = tenant
1131 0 : .gather_size_inputs(
1132 0 : retention_period,
1133 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1134 0 : &cancel,
1135 0 : &ctx,
1136 0 : )
1137 0 : .await
1138 0 : .map_err(|e| match e {
1139 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1140 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1141 0 : })?;
1142 :
1143 0 : let mut sizes = None;
1144 0 : let accepts_html = headers
1145 0 : .get(header::ACCEPT)
1146 0 : .map(|v| v == "text/html")
1147 0 : .unwrap_or_default();
1148 0 : if !inputs_only.unwrap_or(false) {
1149 0 : let storage_model = inputs.calculate_model();
1150 0 : let size = storage_model.calculate();
1151 0 :
1152 0 : // If request header expects html, return html
1153 0 : if accepts_html {
1154 0 : return synthetic_size_html_response(inputs, storage_model, size);
1155 0 : }
1156 0 : sizes = Some(size);
1157 0 : } else if accepts_html {
1158 0 : return Err(ApiError::BadRequest(anyhow!(
1159 0 : "inputs_only parameter is incompatible with html output request"
1160 0 : )));
1161 0 : }
1162 :
1163 : /// The type resides in the pageserver not to expose `ModelInputs`.
1164 : #[derive(serde::Serialize)]
1165 : struct TenantHistorySize {
1166 : id: TenantId,
1167 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1168 : ///
1169 : /// Will be none if `?inputs_only=true` was given.
1170 : size: Option<u64>,
1171 : /// Size of each segment used in the model.
1172 : /// Will be null if `?inputs_only=true` was given.
1173 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1174 : inputs: crate::tenant::size::ModelInputs,
1175 : }
1176 :
1177 0 : json_response(
1178 0 : StatusCode::OK,
1179 0 : TenantHistorySize {
1180 0 : id: tenant_shard_id.tenant_id,
1181 0 : size: sizes.as_ref().map(|x| x.total_size),
1182 0 : segment_sizes: sizes.map(|x| x.segments),
1183 0 : inputs,
1184 0 : },
1185 0 : )
1186 0 : }
1187 :
1188 0 : async fn tenant_shard_split_handler(
1189 0 : mut request: Request<Body>,
1190 0 : _cancel: CancellationToken,
1191 0 : ) -> Result<Response<Body>, ApiError> {
1192 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1193 :
1194 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1195 0 : let state = get_state(&request);
1196 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1197 :
1198 0 : let tenant = state
1199 0 : .tenant_manager
1200 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1201 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1202 :
1203 0 : let new_shards = state
1204 0 : .tenant_manager
1205 0 : .shard_split(
1206 0 : tenant,
1207 0 : ShardCount::new(req.new_shard_count),
1208 0 : req.new_stripe_size,
1209 0 : &ctx,
1210 0 : )
1211 0 : .await
1212 0 : .map_err(ApiError::InternalServerError)?;
1213 :
1214 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1215 0 : }
1216 :
1217 0 : async fn layer_map_info_handler(
1218 0 : request: Request<Body>,
1219 0 : _cancel: CancellationToken,
1220 0 : ) -> Result<Response<Body>, ApiError> {
1221 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1222 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1223 0 : let reset: LayerAccessStatsReset =
1224 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1225 0 : let state = get_state(&request);
1226 0 :
1227 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1228 :
1229 0 : let timeline =
1230 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1231 0 : .await?;
1232 0 : let layer_map_info = timeline.layer_map_info(reset).await;
1233 :
1234 0 : json_response(StatusCode::OK, layer_map_info)
1235 0 : }
1236 :
1237 0 : async fn layer_download_handler(
1238 0 : request: Request<Body>,
1239 0 : _cancel: CancellationToken,
1240 0 : ) -> Result<Response<Body>, ApiError> {
1241 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1242 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1243 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1244 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1245 0 : let layer_name = LayerName::from_str(layer_file_name)
1246 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1247 0 : let state = get_state(&request);
1248 :
1249 0 : let timeline =
1250 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1251 0 : .await?;
1252 0 : let downloaded = timeline
1253 0 : .download_layer(&layer_name)
1254 0 : .await
1255 0 : .map_err(ApiError::InternalServerError)?;
1256 :
1257 0 : match downloaded {
1258 0 : Some(true) => json_response(StatusCode::OK, ()),
1259 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1260 0 : None => json_response(
1261 0 : StatusCode::BAD_REQUEST,
1262 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1263 0 : ),
1264 : }
1265 0 : }
1266 :
1267 0 : async fn evict_timeline_layer_handler(
1268 0 : request: Request<Body>,
1269 0 : _cancel: CancellationToken,
1270 0 : ) -> Result<Response<Body>, ApiError> {
1271 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1272 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1273 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1274 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1275 0 : let state = get_state(&request);
1276 :
1277 0 : let layer_name = LayerName::from_str(layer_file_name)
1278 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1279 :
1280 0 : let timeline =
1281 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1282 0 : .await?;
1283 0 : let evicted = timeline
1284 0 : .evict_layer(&layer_name)
1285 0 : .await
1286 0 : .map_err(ApiError::InternalServerError)?;
1287 :
1288 0 : match evicted {
1289 0 : Some(true) => json_response(StatusCode::OK, ()),
1290 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1291 0 : None => json_response(
1292 0 : StatusCode::BAD_REQUEST,
1293 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1294 0 : ),
1295 : }
1296 0 : }
1297 :
1298 : /// Get tenant_size SVG graph along with the JSON data.
1299 0 : fn synthetic_size_html_response(
1300 0 : inputs: ModelInputs,
1301 0 : storage_model: StorageModel,
1302 0 : sizes: SizeResult,
1303 0 : ) -> Result<Response<Body>, ApiError> {
1304 0 : let mut timeline_ids: Vec<String> = Vec::new();
1305 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1306 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1307 0 : timeline_map.insert(ti.timeline_id, index);
1308 0 : timeline_ids.push(ti.timeline_id.to_string());
1309 0 : }
1310 0 : let seg_to_branch: Vec<usize> = inputs
1311 0 : .segments
1312 0 : .iter()
1313 0 : .map(|seg| *timeline_map.get(&seg.timeline_id).unwrap())
1314 0 : .collect();
1315 :
1316 0 : let svg =
1317 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1318 0 : .map_err(ApiError::InternalServerError)?;
1319 :
1320 0 : let mut response = String::new();
1321 0 :
1322 0 : use std::fmt::Write;
1323 0 : write!(response, "<html>\n<body>\n").unwrap();
1324 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1325 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1326 0 : writeln!(response, "<pre>").unwrap();
1327 0 : writeln!(
1328 0 : response,
1329 0 : "{}",
1330 0 : serde_json::to_string_pretty(&inputs).unwrap()
1331 0 : )
1332 0 : .unwrap();
1333 0 : writeln!(
1334 0 : response,
1335 0 : "{}",
1336 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1337 0 : )
1338 0 : .unwrap();
1339 0 : writeln!(response, "</pre>").unwrap();
1340 0 : write!(response, "</body>\n</html>\n").unwrap();
1341 0 :
1342 0 : html_response(StatusCode::OK, response)
1343 0 : }
1344 :
1345 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1346 0 : let response = Response::builder()
1347 0 : .status(status)
1348 0 : .header(header::CONTENT_TYPE, "text/html")
1349 0 : .body(Body::from(data.as_bytes().to_vec()))
1350 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1351 0 : Ok(response)
1352 0 : }
1353 :
1354 : /// Helper for requests that may take a generation, which is mandatory
1355 : /// when control_plane_api is set, but otherwise defaults to Generation::none()
1356 0 : fn get_request_generation(state: &State, req_gen: Option<u32>) -> Result<Generation, ApiError> {
1357 0 : if state.conf.control_plane_api.is_some() {
1358 0 : req_gen
1359 0 : .map(Generation::new)
1360 0 : .ok_or(ApiError::BadRequest(anyhow!(
1361 0 : "generation attribute missing"
1362 0 : )))
1363 : } else {
1364 : // Legacy mode: all tenants operate with no generation
1365 0 : Ok(Generation::none())
1366 : }
1367 0 : }
1368 :
1369 0 : async fn tenant_create_handler(
1370 0 : mut request: Request<Body>,
1371 0 : _cancel: CancellationToken,
1372 0 : ) -> Result<Response<Body>, ApiError> {
1373 0 : let request_data: TenantCreateRequest = json_request(&mut request).await?;
1374 0 : let target_tenant_id = request_data.new_tenant_id;
1375 0 : check_permission(&request, None)?;
1376 :
1377 0 : let _timer = STORAGE_TIME_GLOBAL
1378 0 : .get_metric_with_label_values(&[StorageTimeOperation::CreateTenant.into()])
1379 0 : .expect("bug")
1380 0 : .start_timer();
1381 :
1382 0 : let tenant_conf =
1383 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1384 :
1385 0 : let state = get_state(&request);
1386 :
1387 0 : let generation = get_request_generation(state, request_data.generation)?;
1388 :
1389 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1390 0 :
1391 0 : let location_conf =
1392 0 : LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
1393 :
1394 0 : let new_tenant = state
1395 0 : .tenant_manager
1396 0 : .upsert_location(
1397 0 : target_tenant_id,
1398 0 : location_conf,
1399 0 : None,
1400 0 : SpawnMode::Create,
1401 0 : &ctx,
1402 0 : )
1403 0 : .await?;
1404 :
1405 0 : let Some(new_tenant) = new_tenant else {
1406 : // This should never happen: indicates a bug in upsert_location
1407 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
1408 0 : "Upsert succeeded but didn't return tenant!"
1409 0 : )));
1410 : };
1411 : // We created the tenant. Existing API semantics are that the tenant
1412 : // is Active when this function returns.
1413 0 : new_tenant
1414 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1415 0 : .await?;
1416 :
1417 0 : json_response(
1418 0 : StatusCode::CREATED,
1419 0 : TenantCreateResponse(new_tenant.tenant_shard_id().tenant_id),
1420 0 : )
1421 0 : }
1422 :
1423 0 : async fn get_tenant_config_handler(
1424 0 : request: Request<Body>,
1425 0 : _cancel: CancellationToken,
1426 0 : ) -> Result<Response<Body>, ApiError> {
1427 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1428 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1429 0 : let state = get_state(&request);
1430 :
1431 0 : let tenant = state
1432 0 : .tenant_manager
1433 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1434 :
1435 0 : let response = HashMap::from([
1436 : (
1437 : "tenant_specific_overrides",
1438 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1439 0 : .context("serializing tenant specific overrides")
1440 0 : .map_err(ApiError::InternalServerError)?,
1441 : ),
1442 : (
1443 0 : "effective_config",
1444 0 : serde_json::to_value(tenant.effective_config())
1445 0 : .context("serializing effective config")
1446 0 : .map_err(ApiError::InternalServerError)?,
1447 : ),
1448 : ]);
1449 :
1450 0 : json_response(StatusCode::OK, response)
1451 0 : }
1452 :
1453 0 : async fn update_tenant_config_handler(
1454 0 : mut request: Request<Body>,
1455 0 : _cancel: CancellationToken,
1456 0 : ) -> Result<Response<Body>, ApiError> {
1457 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1458 0 : let tenant_id = request_data.tenant_id;
1459 0 : check_permission(&request, Some(tenant_id))?;
1460 :
1461 0 : let new_tenant_conf =
1462 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1463 :
1464 0 : let state = get_state(&request);
1465 0 :
1466 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1467 :
1468 0 : let tenant = state
1469 0 : .tenant_manager
1470 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1471 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1472 :
1473 : // This is a legacy API that only operates on attached tenants: the preferred
1474 : // API to use is the location_config/ endpoint, which lets the caller provide
1475 : // the full LocationConf.
1476 0 : let location_conf = LocationConf::attached_single(
1477 0 : new_tenant_conf.clone(),
1478 0 : tenant.get_generation(),
1479 0 : &ShardParameters::default(),
1480 0 : );
1481 0 :
1482 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1483 0 : .await
1484 0 : .map_err(ApiError::InternalServerError)?;
1485 0 : tenant.set_new_tenant_config(new_tenant_conf);
1486 0 :
1487 0 : json_response(StatusCode::OK, ())
1488 0 : }
1489 :
1490 0 : async fn put_tenant_location_config_handler(
1491 0 : mut request: Request<Body>,
1492 0 : _cancel: CancellationToken,
1493 0 : ) -> Result<Response<Body>, ApiError> {
1494 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1495 :
1496 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1497 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1498 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1499 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1500 :
1501 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1502 0 : let state = get_state(&request);
1503 0 : let conf = state.conf;
1504 0 :
1505 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1506 0 : // its local disk content and drops it from memory.
1507 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1508 0 : if let Err(e) = state
1509 0 : .tenant_manager
1510 0 : .detach_tenant(conf, tenant_shard_id, true, &state.deletion_queue_client)
1511 0 : .instrument(info_span!("tenant_detach",
1512 : tenant_id = %tenant_shard_id.tenant_id,
1513 0 : shard_id = %tenant_shard_id.shard_slug()
1514 : ))
1515 0 : .await
1516 : {
1517 0 : match e {
1518 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1519 0 : // This API is idempotent: a NotFound on a detach is fine.
1520 0 : }
1521 0 : _ => return Err(e.into()),
1522 : }
1523 0 : }
1524 0 : return json_response(StatusCode::OK, ());
1525 0 : }
1526 :
1527 0 : let location_conf =
1528 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1529 :
1530 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1531 : // similar to at startup ordering.
1532 0 : let spawn_mode = if lazy {
1533 0 : tenant::SpawnMode::Lazy
1534 : } else {
1535 0 : tenant::SpawnMode::Eager
1536 : };
1537 :
1538 0 : let tenant = state
1539 0 : .tenant_manager
1540 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1541 0 : .await?;
1542 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1543 0 : let attached = tenant.is_some();
1544 :
1545 0 : if let Some(_flush_ms) = flush {
1546 0 : match state
1547 0 : .secondary_controller
1548 0 : .upload_tenant(tenant_shard_id)
1549 0 : .await
1550 : {
1551 : Ok(()) => {
1552 0 : tracing::info!("Uploaded heatmap during flush");
1553 : }
1554 0 : Err(e) => {
1555 0 : tracing::warn!("Failed to flush heatmap: {e}");
1556 : }
1557 : }
1558 : } else {
1559 0 : tracing::info!("No flush requested when configuring");
1560 : }
1561 :
1562 : // This API returns a vector of pageservers where the tenant is attached: this is
1563 : // primarily for use in the sharding service. For compatibilty, we also return this
1564 : // when called directly on a pageserver, but the payload is always zero or one shards.
1565 0 : let mut response = TenantLocationConfigResponse {
1566 0 : shards: Vec::new(),
1567 0 : stripe_size: None,
1568 0 : };
1569 0 : if attached {
1570 0 : response.shards.push(TenantShardLocation {
1571 0 : shard_id: tenant_shard_id,
1572 0 : node_id: state.conf.id,
1573 0 : });
1574 0 : if tenant_shard_id.shard_count.count() > 1 {
1575 : // Stripe size should be set if we are attached
1576 0 : debug_assert!(stripe_size.is_some());
1577 0 : response.stripe_size = stripe_size;
1578 0 : }
1579 0 : }
1580 :
1581 0 : json_response(StatusCode::OK, response)
1582 0 : }
1583 :
1584 0 : async fn list_location_config_handler(
1585 0 : request: Request<Body>,
1586 0 : _cancel: CancellationToken,
1587 0 : ) -> Result<Response<Body>, ApiError> {
1588 0 : let state = get_state(&request);
1589 0 : let slots = state.tenant_manager.list();
1590 0 : let result = LocationConfigListResponse {
1591 0 : tenant_shards: slots
1592 0 : .into_iter()
1593 0 : .map(|(tenant_shard_id, slot)| {
1594 0 : let v = match slot {
1595 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1596 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1597 0 : TenantSlot::InProgress(_) => None,
1598 : };
1599 0 : (tenant_shard_id, v)
1600 0 : })
1601 0 : .collect(),
1602 0 : };
1603 0 : json_response(StatusCode::OK, result)
1604 0 : }
1605 :
1606 0 : async fn get_location_config_handler(
1607 0 : request: Request<Body>,
1608 0 : _cancel: CancellationToken,
1609 0 : ) -> Result<Response<Body>, ApiError> {
1610 0 : let state = get_state(&request);
1611 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1612 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1613 :
1614 0 : let Some(slot) = slot else {
1615 0 : return Err(ApiError::NotFound(
1616 0 : anyhow::anyhow!("Tenant shard not found").into(),
1617 0 : ));
1618 : };
1619 :
1620 0 : let result: Option<LocationConfig> = match slot {
1621 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1622 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1623 0 : TenantSlot::InProgress(_) => None,
1624 : };
1625 :
1626 0 : json_response(StatusCode::OK, result)
1627 0 : }
1628 :
1629 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1630 : // (from all pageservers) as it invalidates consistency assumptions.
1631 0 : async fn tenant_time_travel_remote_storage_handler(
1632 0 : request: Request<Body>,
1633 0 : cancel: CancellationToken,
1634 0 : ) -> Result<Response<Body>, ApiError> {
1635 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1636 :
1637 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1638 :
1639 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1640 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1641 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1642 0 : .map_err(ApiError::BadRequest)?;
1643 :
1644 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1645 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1646 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1647 0 : .map_err(ApiError::BadRequest)?;
1648 :
1649 : // This is just a sanity check to fend off naive wrong usages of the API:
1650 : // the tenant needs to be detached *everywhere*
1651 0 : let state = get_state(&request);
1652 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1653 0 : if we_manage_tenant {
1654 0 : return Err(ApiError::BadRequest(anyhow!(
1655 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1656 0 : )));
1657 0 : }
1658 0 :
1659 0 : if timestamp > done_if_after {
1660 0 : return Err(ApiError::BadRequest(anyhow!(
1661 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1662 0 : )));
1663 0 : }
1664 0 :
1665 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1666 :
1667 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1668 0 : &state.remote_storage,
1669 0 : &tenant_shard_id,
1670 0 : timestamp,
1671 0 : done_if_after,
1672 0 : &cancel,
1673 0 : )
1674 0 : .await
1675 0 : .map_err(|e| match e {
1676 0 : TimeTravelError::BadInput(e) => {
1677 0 : warn!("bad input error: {e}");
1678 0 : ApiError::BadRequest(anyhow!("bad input error"))
1679 : }
1680 : TimeTravelError::Unimplemented => {
1681 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1682 : }
1683 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1684 : TimeTravelError::TooManyVersions => {
1685 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1686 : }
1687 0 : TimeTravelError::Other(e) => {
1688 0 : warn!("internal error: {e}");
1689 0 : ApiError::InternalServerError(anyhow!("internal error"))
1690 : }
1691 0 : })?;
1692 :
1693 0 : json_response(StatusCode::OK, ())
1694 0 : }
1695 :
1696 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1697 0 : async fn handle_tenant_break(
1698 0 : r: Request<Body>,
1699 0 : _cancel: CancellationToken,
1700 0 : ) -> Result<Response<Body>, ApiError> {
1701 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1702 :
1703 0 : let state = get_state(&r);
1704 0 : state
1705 0 : .tenant_manager
1706 0 : .get_attached_tenant_shard(tenant_shard_id)?
1707 0 : .set_broken("broken from test".to_owned())
1708 0 : .await;
1709 :
1710 0 : json_response(StatusCode::OK, ())
1711 0 : }
1712 :
1713 : // Obtains an lsn lease on the given timeline.
1714 0 : async fn lsn_lease_handler(
1715 0 : request: Request<Body>,
1716 0 : _cancel: CancellationToken,
1717 0 : ) -> Result<Response<Body>, ApiError> {
1718 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1719 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1720 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1721 :
1722 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1723 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1724 :
1725 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1726 0 :
1727 0 : let state = get_state(&request);
1728 :
1729 0 : let timeline =
1730 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1731 0 : .await?;
1732 0 : let result = timeline
1733 0 : .make_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
1734 0 : .map_err(|e| ApiError::InternalServerError(e.context("lsn lease http handler")))?;
1735 :
1736 0 : json_response(StatusCode::OK, result)
1737 0 : }
1738 :
1739 : // Run GC immediately on given timeline.
1740 0 : async fn timeline_gc_handler(
1741 0 : mut request: Request<Body>,
1742 0 : cancel: CancellationToken,
1743 0 : ) -> Result<Response<Body>, ApiError> {
1744 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1745 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1746 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1747 :
1748 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1749 :
1750 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1751 0 : let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
1752 :
1753 0 : json_response(StatusCode::OK, gc_result)
1754 0 : }
1755 :
1756 : // Run compaction immediately on given timeline.
1757 0 : async fn timeline_compact_handler(
1758 0 : request: Request<Body>,
1759 0 : cancel: CancellationToken,
1760 0 : ) -> Result<Response<Body>, ApiError> {
1761 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1762 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1763 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1764 :
1765 0 : let state = get_state(&request);
1766 0 :
1767 0 : let mut flags = EnumSet::empty();
1768 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1769 0 : flags |= CompactFlags::ForceRepartition;
1770 0 : }
1771 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1772 0 : flags |= CompactFlags::ForceImageLayerCreation;
1773 0 : }
1774 0 : let wait_until_uploaded =
1775 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1776 :
1777 0 : async {
1778 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1779 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1780 0 : timeline
1781 0 : .compact(&cancel, flags, &ctx)
1782 0 : .await
1783 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1784 0 : if wait_until_uploaded {
1785 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1786 0 : }
1787 0 : json_response(StatusCode::OK, ())
1788 0 : }
1789 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1790 0 : .await
1791 0 : }
1792 :
1793 : // Run checkpoint immediately on given timeline.
1794 0 : async fn timeline_checkpoint_handler(
1795 0 : request: Request<Body>,
1796 0 : cancel: CancellationToken,
1797 0 : ) -> Result<Response<Body>, ApiError> {
1798 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1799 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1800 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1801 :
1802 0 : let state = get_state(&request);
1803 0 :
1804 0 : let mut flags = EnumSet::empty();
1805 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1806 0 : flags |= CompactFlags::ForceRepartition;
1807 0 : }
1808 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1809 0 : flags |= CompactFlags::ForceImageLayerCreation;
1810 0 : }
1811 0 : let wait_until_uploaded =
1812 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1813 :
1814 0 : async {
1815 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1816 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1817 0 : timeline
1818 0 : .freeze_and_flush()
1819 0 : .await
1820 0 : .map_err(|e| {
1821 0 : match e {
1822 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
1823 0 : other => ApiError::InternalServerError(other.into()),
1824 :
1825 : }
1826 0 : })?;
1827 0 : timeline
1828 0 : .compact(&cancel, flags, &ctx)
1829 0 : .await
1830 0 : .map_err(|e|
1831 0 : match e {
1832 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
1833 0 : CompactionError::Other(e) => ApiError::InternalServerError(e)
1834 0 : }
1835 0 : )?;
1836 :
1837 0 : if wait_until_uploaded {
1838 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1839 0 : }
1840 :
1841 0 : json_response(StatusCode::OK, ())
1842 0 : }
1843 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1844 0 : .await
1845 0 : }
1846 :
1847 0 : async fn timeline_download_remote_layers_handler_post(
1848 0 : mut request: Request<Body>,
1849 0 : _cancel: CancellationToken,
1850 0 : ) -> Result<Response<Body>, ApiError> {
1851 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1852 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1853 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
1854 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1855 :
1856 0 : let state = get_state(&request);
1857 :
1858 0 : let timeline =
1859 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1860 0 : .await?;
1861 0 : match timeline.spawn_download_all_remote_layers(body).await {
1862 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
1863 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
1864 : }
1865 0 : }
1866 :
1867 0 : async fn timeline_download_remote_layers_handler_get(
1868 0 : request: Request<Body>,
1869 0 : _cancel: CancellationToken,
1870 0 : ) -> Result<Response<Body>, ApiError> {
1871 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1872 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1873 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1874 0 : let state = get_state(&request);
1875 :
1876 0 : let timeline =
1877 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1878 0 : .await?;
1879 0 : let info = timeline
1880 0 : .get_download_all_remote_layers_task_info()
1881 0 : .context("task never started since last pageserver process start")
1882 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1883 0 : json_response(StatusCode::OK, info)
1884 0 : }
1885 :
1886 0 : async fn timeline_detach_ancestor_handler(
1887 0 : request: Request<Body>,
1888 0 : _cancel: CancellationToken,
1889 0 : ) -> Result<Response<Body>, ApiError> {
1890 : use crate::tenant::timeline::detach_ancestor::Options;
1891 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1892 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1893 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1894 :
1895 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
1896 :
1897 0 : async move {
1898 0 : let mut options = Options::default();
1899 :
1900 0 : let rewrite_concurrency =
1901 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
1902 0 : let copy_concurrency =
1903 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
1904 :
1905 0 : [
1906 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
1907 0 : (&mut options.copy_concurrency, copy_concurrency),
1908 0 : ]
1909 0 : .into_iter()
1910 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
1911 0 : .for_each(|(target, val)| *target = val);
1912 0 :
1913 0 : let state = get_state(&request);
1914 :
1915 0 : let tenant = state
1916 0 : .tenant_manager
1917 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1918 :
1919 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1920 :
1921 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
1922 0 : let ctx = &ctx;
1923 :
1924 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1925 :
1926 0 : let (_guard, prepared) = timeline
1927 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
1928 0 : .await?;
1929 :
1930 0 : let res = state
1931 0 : .tenant_manager
1932 0 : .complete_detaching_timeline_ancestor(tenant_shard_id, timeline_id, prepared, ctx)
1933 0 : .await;
1934 :
1935 0 : match res {
1936 0 : Ok(reparented_timelines) => {
1937 0 : let resp = pageserver_api::models::detach_ancestor::AncestorDetached {
1938 0 : reparented_timelines,
1939 0 : };
1940 0 :
1941 0 : json_response(StatusCode::OK, resp)
1942 : }
1943 0 : Err(e) => Err(ApiError::InternalServerError(
1944 0 : e.context("timeline detach completion"),
1945 0 : )),
1946 : }
1947 0 : }
1948 0 : .instrument(span)
1949 0 : .await
1950 0 : }
1951 :
1952 0 : async fn deletion_queue_flush(
1953 0 : r: Request<Body>,
1954 0 : cancel: CancellationToken,
1955 0 : ) -> Result<Response<Body>, ApiError> {
1956 0 : let state = get_state(&r);
1957 :
1958 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
1959 0 :
1960 0 : let flush = async {
1961 0 : if execute {
1962 0 : state.deletion_queue_client.flush_execute().await
1963 : } else {
1964 0 : state.deletion_queue_client.flush().await
1965 : }
1966 0 : }
1967 : // DeletionQueueError's only case is shutting down.
1968 0 : .map_err(|_| ApiError::ShuttingDown);
1969 :
1970 : tokio::select! {
1971 : res = flush => {
1972 0 : res.map(|()| json_response(StatusCode::OK, ()))?
1973 : }
1974 : _ = cancel.cancelled() => {
1975 : Err(ApiError::ShuttingDown)
1976 : }
1977 : }
1978 0 : }
1979 :
1980 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
1981 0 : async fn getpage_at_lsn_handler(
1982 0 : request: Request<Body>,
1983 0 : _cancel: CancellationToken,
1984 0 : ) -> Result<Response<Body>, ApiError> {
1985 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1986 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1987 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1988 0 : let state = get_state(&request);
1989 :
1990 : struct Key(crate::repository::Key);
1991 :
1992 : impl std::str::FromStr for Key {
1993 : type Err = anyhow::Error;
1994 :
1995 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
1996 0 : crate::repository::Key::from_hex(s).map(Key)
1997 0 : }
1998 : }
1999 :
2000 0 : let key: Key = parse_query_param(&request, "key")?
2001 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
2002 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
2003 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
2004 :
2005 0 : async {
2006 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2007 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2008 :
2009 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
2010 :
2011 0 : Result::<_, ApiError>::Ok(
2012 0 : Response::builder()
2013 0 : .status(StatusCode::OK)
2014 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2015 0 : .body(hyper::Body::from(page))
2016 0 : .unwrap(),
2017 0 : )
2018 0 : }
2019 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2020 0 : .await
2021 0 : }
2022 :
2023 0 : async fn timeline_collect_keyspace(
2024 0 : request: Request<Body>,
2025 0 : _cancel: CancellationToken,
2026 0 : ) -> Result<Response<Body>, ApiError> {
2027 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2028 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2029 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2030 0 : let state = get_state(&request);
2031 :
2032 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2033 :
2034 0 : async {
2035 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2036 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2037 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2038 0 : let (dense_ks, sparse_ks) = timeline
2039 0 : .collect_keyspace(at_lsn, &ctx)
2040 0 : .await
2041 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2042 :
2043 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2044 : // Therefore, we split dense/sparse keys in this API.
2045 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2046 0 :
2047 0 : json_response(StatusCode::OK, res)
2048 0 : }
2049 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2050 0 : .await
2051 0 : }
2052 :
2053 0 : async fn active_timeline_of_active_tenant(
2054 0 : tenant_manager: &TenantManager,
2055 0 : tenant_shard_id: TenantShardId,
2056 0 : timeline_id: TimelineId,
2057 0 : ) -> Result<Arc<Timeline>, ApiError> {
2058 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2059 :
2060 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2061 :
2062 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2063 0 : }
2064 :
2065 0 : async fn always_panic_handler(
2066 0 : req: Request<Body>,
2067 0 : _cancel: CancellationToken,
2068 0 : ) -> Result<Response<Body>, ApiError> {
2069 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2070 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2071 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2072 0 : let query = req.uri().query();
2073 0 : let _ = std::panic::catch_unwind(|| {
2074 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2075 0 : });
2076 0 : json_response(StatusCode::NO_CONTENT, ())
2077 0 : }
2078 :
2079 0 : async fn disk_usage_eviction_run(
2080 0 : mut r: Request<Body>,
2081 0 : cancel: CancellationToken,
2082 0 : ) -> Result<Response<Body>, ApiError> {
2083 0 : check_permission(&r, None)?;
2084 :
2085 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2086 : struct Config {
2087 : /// How many bytes to evict before reporting that pressure is relieved.
2088 : evict_bytes: u64,
2089 :
2090 : #[serde(default)]
2091 : eviction_order: crate::disk_usage_eviction_task::EvictionOrder,
2092 : }
2093 :
2094 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2095 : struct Usage {
2096 : // remains unchanged after instantiation of the struct
2097 : evict_bytes: u64,
2098 : // updated by `add_available_bytes`
2099 : freed_bytes: u64,
2100 : }
2101 :
2102 : impl crate::disk_usage_eviction_task::Usage for Usage {
2103 0 : fn has_pressure(&self) -> bool {
2104 0 : self.evict_bytes > self.freed_bytes
2105 0 : }
2106 :
2107 0 : fn add_available_bytes(&mut self, bytes: u64) {
2108 0 : self.freed_bytes += bytes;
2109 0 : }
2110 : }
2111 :
2112 0 : let config = json_request::<Config>(&mut r).await?;
2113 :
2114 0 : let usage = Usage {
2115 0 : evict_bytes: config.evict_bytes,
2116 0 : freed_bytes: 0,
2117 0 : };
2118 0 :
2119 0 : let state = get_state(&r);
2120 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2121 :
2122 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2123 0 : &eviction_state,
2124 0 : &state.remote_storage,
2125 0 : usage,
2126 0 : &state.tenant_manager,
2127 0 : config.eviction_order,
2128 0 : &cancel,
2129 0 : )
2130 0 : .await;
2131 :
2132 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2133 :
2134 0 : let res = res.map_err(ApiError::InternalServerError)?;
2135 :
2136 0 : json_response(StatusCode::OK, res)
2137 0 : }
2138 :
2139 0 : async fn secondary_upload_handler(
2140 0 : request: Request<Body>,
2141 0 : _cancel: CancellationToken,
2142 0 : ) -> Result<Response<Body>, ApiError> {
2143 0 : let state = get_state(&request);
2144 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2145 0 : state
2146 0 : .secondary_controller
2147 0 : .upload_tenant(tenant_shard_id)
2148 0 : .await
2149 0 : .map_err(ApiError::InternalServerError)?;
2150 :
2151 0 : json_response(StatusCode::OK, ())
2152 0 : }
2153 :
2154 0 : async fn tenant_scan_remote_handler(
2155 0 : request: Request<Body>,
2156 0 : cancel: CancellationToken,
2157 0 : ) -> Result<Response<Body>, ApiError> {
2158 0 : let state = get_state(&request);
2159 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2160 :
2161 0 : let mut response = TenantScanRemoteStorageResponse::default();
2162 :
2163 0 : let (shards, _other_keys) =
2164 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2165 0 : .await
2166 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2167 :
2168 0 : for tenant_shard_id in shards {
2169 0 : let (timeline_ids, _other_keys) =
2170 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2171 0 : .await
2172 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2173 :
2174 0 : let mut generation = Generation::none();
2175 0 : for timeline_id in timeline_ids {
2176 0 : match download_index_part(
2177 0 : &state.remote_storage,
2178 0 : &tenant_shard_id,
2179 0 : &timeline_id,
2180 0 : Generation::MAX,
2181 0 : &cancel,
2182 0 : )
2183 0 : .instrument(info_span!("download_index_part",
2184 : tenant_id=%tenant_shard_id.tenant_id,
2185 0 : shard_id=%tenant_shard_id.shard_slug(),
2186 : %timeline_id))
2187 0 : .await
2188 : {
2189 0 : Ok((index_part, index_generation)) => {
2190 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2191 0 : index_part.layer_metadata.len(), index_part.metadata.disk_consistent_lsn());
2192 0 : generation = std::cmp::max(generation, index_generation);
2193 : }
2194 : Err(DownloadError::NotFound) => {
2195 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2196 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2197 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2198 0 : continue;
2199 : }
2200 0 : Err(e) => {
2201 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2202 : }
2203 : };
2204 : }
2205 :
2206 0 : response.shards.push(TenantScanRemoteStorageShard {
2207 0 : tenant_shard_id,
2208 0 : generation: generation.into(),
2209 0 : });
2210 : }
2211 :
2212 0 : if response.shards.is_empty() {
2213 0 : return Err(ApiError::NotFound(
2214 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2215 0 : ));
2216 0 : }
2217 0 :
2218 0 : json_response(StatusCode::OK, response)
2219 0 : }
2220 :
2221 0 : async fn secondary_download_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 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2228 :
2229 : // We don't need this to issue the download request, but:
2230 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2231 : // - we will use this to provide status feedback in the response
2232 0 : let Some(secondary_tenant) = state
2233 0 : .tenant_manager
2234 0 : .get_secondary_tenant_shard(tenant_shard_id)
2235 : else {
2236 0 : return Err(ApiError::NotFound(
2237 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2238 0 : ));
2239 : };
2240 :
2241 0 : let timeout = wait.unwrap_or(Duration::MAX);
2242 :
2243 0 : let status = match tokio::time::timeout(
2244 0 : timeout,
2245 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2246 0 : )
2247 0 : .await
2248 : {
2249 : // Download job ran to completion.
2250 0 : Ok(Ok(())) => StatusCode::OK,
2251 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2252 : // okay. We could get an error here in the unlikely edge case that the tenant
2253 : // was detached between our check above and executing the download job.
2254 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2255 : // A timeout is not an error: we have started the download, we're just not done
2256 : // yet. The caller will get a response body indicating status.
2257 0 : Err(_) => StatusCode::ACCEPTED,
2258 : };
2259 :
2260 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2261 0 :
2262 0 : json_response(status, progress)
2263 0 : }
2264 :
2265 0 : async fn secondary_status_handler(
2266 0 : request: Request<Body>,
2267 0 : _cancel: CancellationToken,
2268 0 : ) -> Result<Response<Body>, ApiError> {
2269 0 : let state = get_state(&request);
2270 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2271 :
2272 0 : let Some(secondary_tenant) = state
2273 0 : .tenant_manager
2274 0 : .get_secondary_tenant_shard(tenant_shard_id)
2275 : else {
2276 0 : return Err(ApiError::NotFound(
2277 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2278 0 : ));
2279 : };
2280 :
2281 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2282 0 :
2283 0 : json_response(StatusCode::OK, progress)
2284 0 : }
2285 :
2286 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2287 0 : json_response(
2288 0 : StatusCode::NOT_FOUND,
2289 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2290 0 : )
2291 0 : }
2292 :
2293 0 : async fn post_tracing_event_handler(
2294 0 : mut r: Request<Body>,
2295 0 : _cancel: CancellationToken,
2296 0 : ) -> Result<Response<Body>, ApiError> {
2297 0 : #[derive(Debug, serde::Deserialize)]
2298 : #[serde(rename_all = "lowercase")]
2299 : enum Level {
2300 : Error,
2301 : Warn,
2302 : Info,
2303 : Debug,
2304 : Trace,
2305 : }
2306 0 : #[derive(Debug, serde::Deserialize)]
2307 : struct Request {
2308 : level: Level,
2309 : message: String,
2310 : }
2311 0 : let body: Request = json_request(&mut r)
2312 0 : .await
2313 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2314 :
2315 0 : match body.level {
2316 0 : Level::Error => tracing::error!(?body.message),
2317 0 : Level::Warn => tracing::warn!(?body.message),
2318 0 : Level::Info => tracing::info!(?body.message),
2319 0 : Level::Debug => tracing::debug!(?body.message),
2320 0 : Level::Trace => tracing::trace!(?body.message),
2321 : }
2322 :
2323 0 : json_response(StatusCode::OK, ())
2324 0 : }
2325 :
2326 0 : async fn force_aux_policy_switch_handler(
2327 0 : mut r: Request<Body>,
2328 0 : _cancel: CancellationToken,
2329 0 : ) -> Result<Response<Body>, ApiError> {
2330 0 : check_permission(&r, None)?;
2331 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
2332 0 : let timeline_id: TimelineId = parse_request_param(&r, "timeline_id")?;
2333 0 : let policy: AuxFilePolicy = json_request(&mut r).await?;
2334 :
2335 0 : let state = get_state(&r);
2336 :
2337 0 : let tenant = state
2338 0 : .tenant_manager
2339 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2340 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2341 0 : let timeline =
2342 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2343 0 : .await?;
2344 0 : timeline
2345 0 : .do_switch_aux_policy(policy)
2346 0 : .map_err(ApiError::InternalServerError)?;
2347 :
2348 0 : json_response(StatusCode::OK, ())
2349 0 : }
2350 :
2351 0 : async fn put_io_engine_handler(
2352 0 : mut r: Request<Body>,
2353 0 : _cancel: CancellationToken,
2354 0 : ) -> Result<Response<Body>, ApiError> {
2355 0 : check_permission(&r, None)?;
2356 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2357 0 : crate::virtual_file::io_engine::set(kind);
2358 0 : json_response(StatusCode::OK, ())
2359 0 : }
2360 :
2361 : /// Polled by control plane.
2362 : ///
2363 : /// See [`crate::utilization`].
2364 0 : async fn get_utilization(
2365 0 : r: Request<Body>,
2366 0 : _cancel: CancellationToken,
2367 0 : ) -> Result<Response<Body>, ApiError> {
2368 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2369 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2370 0 : });
2371 :
2372 : // this probably could be completely public, but lets make that change later.
2373 0 : check_permission(&r, None)?;
2374 :
2375 0 : let state = get_state(&r);
2376 0 : let mut g = state.latest_utilization.lock().await;
2377 :
2378 0 : let regenerate_every = Duration::from_secs(1);
2379 0 : let still_valid = g
2380 0 : .as_ref()
2381 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2382 0 :
2383 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2384 0 : // regenerate at most 1Hz to allow polling at any rate.
2385 0 : if !still_valid {
2386 0 : let path = state.conf.tenants_path();
2387 0 : let doc = crate::utilization::regenerate(path.as_std_path())
2388 0 : .map_err(ApiError::InternalServerError)?;
2389 :
2390 0 : let mut buf = Vec::new();
2391 0 : serde_json::to_writer(&mut buf, &doc)
2392 0 : .context("serialize")
2393 0 : .map_err(ApiError::InternalServerError)?;
2394 :
2395 0 : let body = bytes::Bytes::from(buf);
2396 0 :
2397 0 : *g = Some((std::time::Instant::now(), body));
2398 0 : }
2399 :
2400 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2401 0 : let cached = g.as_ref().expect("just set").1.clone();
2402 0 :
2403 0 : Response::builder()
2404 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2405 0 : // thought of using http date header, but that is second precision which does not give any
2406 0 : // debugging aid
2407 0 : .status(StatusCode::OK)
2408 0 : .body(hyper::Body::from(cached))
2409 0 : .context("build response")
2410 0 : .map_err(ApiError::InternalServerError)
2411 0 : }
2412 :
2413 0 : async fn list_aux_files(
2414 0 : mut request: Request<Body>,
2415 0 : _cancel: CancellationToken,
2416 0 : ) -> Result<Response<Body>, ApiError> {
2417 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2418 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2419 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2420 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2421 :
2422 0 : let state = get_state(&request);
2423 :
2424 0 : let timeline =
2425 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2426 0 : .await?;
2427 :
2428 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2429 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2430 0 : json_response(StatusCode::OK, files)
2431 0 : }
2432 :
2433 0 : async fn perf_info(
2434 0 : request: Request<Body>,
2435 0 : _cancel: CancellationToken,
2436 0 : ) -> Result<Response<Body>, ApiError> {
2437 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2438 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2439 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2440 :
2441 0 : let state = get_state(&request);
2442 :
2443 0 : let timeline =
2444 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2445 0 : .await?;
2446 :
2447 0 : let result = timeline.perf_info().await;
2448 :
2449 0 : json_response(StatusCode::OK, result)
2450 0 : }
2451 :
2452 0 : async fn ingest_aux_files(
2453 0 : mut request: Request<Body>,
2454 0 : _cancel: CancellationToken,
2455 0 : ) -> Result<Response<Body>, ApiError> {
2456 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2457 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2458 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2459 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2460 :
2461 0 : let state = get_state(&request);
2462 :
2463 0 : let timeline =
2464 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2465 0 : .await?;
2466 :
2467 0 : let mut modification = timeline.begin_modification(
2468 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
2469 0 : );
2470 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2471 0 : for (fname, content) in body.aux_files {
2472 0 : modification
2473 0 : .put_file(&fname, content.as_bytes(), &ctx)
2474 0 : .await
2475 0 : .map_err(ApiError::InternalServerError)?;
2476 : }
2477 0 : modification
2478 0 : .commit(&ctx)
2479 0 : .await
2480 0 : .map_err(ApiError::InternalServerError)?;
2481 :
2482 0 : json_response(StatusCode::OK, ())
2483 0 : }
2484 :
2485 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2486 : /// candidates for splitting
2487 0 : async fn post_top_tenants(
2488 0 : mut r: Request<Body>,
2489 0 : _cancel: CancellationToken,
2490 0 : ) -> Result<Response<Body>, ApiError> {
2491 0 : check_permission(&r, None)?;
2492 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2493 0 : let state = get_state(&r);
2494 0 :
2495 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2496 0 : match order_by {
2497 0 : TenantSorting::ResidentSize => sizes.resident_size,
2498 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2499 0 : }
2500 0 : }
2501 0 :
2502 0 : #[derive(Eq, PartialEq)]
2503 0 : struct HeapItem {
2504 0 : metric: u64,
2505 0 : sizes: TopTenantShardItem,
2506 0 : }
2507 0 :
2508 0 : impl PartialOrd for HeapItem {
2509 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2510 0 : Some(self.cmp(other))
2511 0 : }
2512 0 : }
2513 0 :
2514 0 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2515 0 : /// supports popping the greatest item but not the smallest.
2516 0 : impl Ord for HeapItem {
2517 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2518 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2519 0 : }
2520 0 : }
2521 0 :
2522 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2523 :
2524 : // FIXME: this is a lot of clones to take this tenant list
2525 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2526 0 : if let Some(shards_lt) = request.where_shards_lt {
2527 : // Ignore tenants which already have >= this many shards
2528 0 : if tenant_shard_id.shard_count >= shards_lt {
2529 0 : continue;
2530 0 : }
2531 0 : }
2532 :
2533 0 : let sizes = match tenant_slot {
2534 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2535 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2536 0 : continue;
2537 : }
2538 : };
2539 0 : let metric = get_size_metric(&sizes, &request.order_by);
2540 :
2541 0 : if let Some(gt) = request.where_gt {
2542 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2543 0 : if metric <= gt {
2544 0 : continue;
2545 0 : }
2546 0 : };
2547 :
2548 0 : match top_n.peek() {
2549 0 : None => {
2550 0 : // Top N list is empty: candidate becomes first member
2551 0 : top_n.push(HeapItem { metric, sizes });
2552 0 : }
2553 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
2554 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
2555 0 : top_n.push(HeapItem { metric, sizes });
2556 0 : }
2557 0 : Some(i) if i.metric > metric => {
2558 0 : // List is at limit and lowest value is greater than our candidate, drop it.
2559 0 : }
2560 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
2561 : }
2562 :
2563 0 : while top_n.len() > request.limit {
2564 0 : top_n.pop();
2565 0 : }
2566 : }
2567 :
2568 0 : json_response(
2569 0 : StatusCode::OK,
2570 0 : TopTenantShardsResponse {
2571 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
2572 0 : },
2573 0 : )
2574 0 : }
2575 :
2576 : /// Common functionality of all the HTTP API handlers.
2577 : ///
2578 : /// - Adds a tracing span to each request (by `request_span`)
2579 : /// - Logs the request depending on the request method (by `request_span`)
2580 : /// - Logs the response if it was not successful (by `request_span`
2581 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
2582 : /// Future if the connection to the client is lost, but most of the pageserver code is
2583 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
2584 : /// request with a CancellationToken.
2585 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
2586 0 : where
2587 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2588 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2589 0 : {
2590 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
2591 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
2592 0 : "failpoint".into()
2593 0 : )));
2594 :
2595 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
2596 0 : anyhow::anyhow!("failpoint")
2597 0 : )));
2598 0 : }
2599 :
2600 : // Spawn a new task to handle the request, to protect the handler from unexpected
2601 : // async cancellations. Most pageserver functions are not async cancellation safe.
2602 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
2603 : // with the cancellation token.
2604 0 : let token = CancellationToken::new();
2605 0 : let cancel_guard = token.clone().drop_guard();
2606 0 : let result = request_span(request, move |r| async {
2607 0 : let handle = tokio::spawn(
2608 0 : async {
2609 0 : let token_cloned = token.clone();
2610 0 : let result = handler(r, token).await;
2611 0 : if token_cloned.is_cancelled() {
2612 0 : // dropguard has executed: we will never turn this result into response.
2613 0 : //
2614 0 : // at least temporarily do {:?} logging; these failures are rare enough but
2615 0 : // could hide difficult errors.
2616 0 : match &result {
2617 0 : Ok(response) => {
2618 0 : let status = response.status();
2619 0 : info!(%status, "Cancelled request finished successfully")
2620 0 : }
2621 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
2622 0 : }
2623 0 : }
2624 0 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
2625 0 : // which should suffice.
2626 0 : //
2627 0 : // there is still a chance to lose the result due to race between
2628 0 : // returning from here and the actual connection closing happening
2629 0 : // before outer task gets to execute. leaving that up for #5815.
2630 0 : result
2631 0 : }
2632 0 : .in_current_span(),
2633 0 : );
2634 0 :
2635 0 : match handle.await {
2636 0 : // TODO: never actually return Err from here, always Ok(...) so that we can log
2637 0 : // spanned errors. Call api_error_handler instead and return appropriate Body.
2638 0 : Ok(result) => result,
2639 0 : Err(e) => {
2640 0 : // The handler task panicked. We have a global panic handler that logs the
2641 0 : // panic with its backtrace, so no need to log that here. Only log a brief
2642 0 : // message to make it clear that we returned the error to the client.
2643 0 : error!("HTTP request handler task panicked: {e:#}");
2644 0 :
2645 0 : // Don't return an Error here, because then fallback error handler that was
2646 0 : // installed in make_router() will print the error. Instead, construct the
2647 0 : // HTTP error response and return that.
2648 0 : Ok(
2649 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
2650 0 : .into_response(),
2651 0 : )
2652 0 : }
2653 0 : }
2654 0 : })
2655 0 : .await;
2656 :
2657 0 : cancel_guard.disarm();
2658 0 :
2659 0 : result
2660 0 : }
2661 :
2662 : /// Like api_handler, but returns an error response if the server is built without
2663 : /// the 'testing' feature.
2664 0 : async fn testing_api_handler<R, H>(
2665 0 : desc: &str,
2666 0 : request: Request<Body>,
2667 0 : handler: H,
2668 0 : ) -> Result<Response<Body>, ApiError>
2669 0 : where
2670 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2671 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2672 0 : {
2673 0 : if cfg!(feature = "testing") {
2674 0 : api_handler(request, handler).await
2675 : } else {
2676 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
2677 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
2678 0 : ))))
2679 0 : .await
2680 : }
2681 0 : }
2682 :
2683 0 : pub fn make_router(
2684 0 : state: Arc<State>,
2685 0 : launch_ts: &'static LaunchTimestamp,
2686 0 : auth: Option<Arc<SwappableJwtAuth>>,
2687 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
2688 0 : let spec = include_bytes!("openapi_spec.yml");
2689 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
2690 0 : if auth.is_some() {
2691 0 : router = router.middleware(auth_middleware(|request| {
2692 0 : let state = get_state(request);
2693 0 : if state.allowlist_routes.contains(request.uri()) {
2694 0 : None
2695 : } else {
2696 0 : state.auth.as_deref()
2697 : }
2698 0 : }))
2699 0 : }
2700 :
2701 0 : router = router.middleware(
2702 0 : endpoint::add_response_header_middleware(
2703 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
2704 0 : &launch_ts.to_string(),
2705 0 : )
2706 0 : .expect("construct launch timestamp header middleware"),
2707 0 : );
2708 0 :
2709 0 : Ok(router
2710 0 : .data(state)
2711 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
2712 0 : .get("/v1/status", |r| api_handler(r, status_handler))
2713 0 : .put("/v1/failpoints", |r| {
2714 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
2715 0 : })
2716 0 : .post("/v1/reload_auth_validation_keys", |r| {
2717 0 : api_handler(r, reload_auth_validation_keys_handler)
2718 0 : })
2719 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
2720 0 : .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
2721 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
2722 0 : api_handler(r, tenant_status)
2723 0 : })
2724 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
2725 0 : api_handler(r, tenant_delete_handler)
2726 0 : })
2727 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
2728 0 : api_handler(r, tenant_size_handler)
2729 0 : })
2730 0 : .put("/v1/tenant/config", |r| {
2731 0 : api_handler(r, update_tenant_config_handler)
2732 0 : })
2733 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
2734 0 : api_handler(r, tenant_shard_split_handler)
2735 0 : })
2736 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
2737 0 : api_handler(r, get_tenant_config_handler)
2738 0 : })
2739 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2740 0 : api_handler(r, put_tenant_location_config_handler)
2741 0 : })
2742 0 : .get("/v1/location_config", |r| {
2743 0 : api_handler(r, list_location_config_handler)
2744 0 : })
2745 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
2746 0 : api_handler(r, get_location_config_handler)
2747 0 : })
2748 0 : .put(
2749 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
2750 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
2751 0 : )
2752 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
2753 0 : api_handler(r, timeline_list_handler)
2754 0 : })
2755 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
2756 0 : api_handler(r, timeline_create_handler)
2757 0 : })
2758 0 : .post("/v1/tenant/:tenant_id/attach", |r| {
2759 0 : api_handler(r, tenant_attach_handler)
2760 0 : })
2761 0 : .post("/v1/tenant/:tenant_id/detach", |r| {
2762 0 : api_handler(r, tenant_detach_handler)
2763 0 : })
2764 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
2765 0 : api_handler(r, tenant_reset_handler)
2766 0 : })
2767 0 : .post("/v1/tenant/:tenant_id/load", |r| {
2768 0 : api_handler(r, tenant_load_handler)
2769 0 : })
2770 0 : .post("/v1/tenant/:tenant_id/ignore", |r| {
2771 0 : api_handler(r, tenant_ignore_handler)
2772 0 : })
2773 0 : .post(
2774 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
2775 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
2776 0 : )
2777 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2778 0 : api_handler(r, timeline_detail_handler)
2779 0 : })
2780 0 : .get(
2781 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
2782 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
2783 0 : )
2784 0 : .get(
2785 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
2786 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
2787 0 : )
2788 0 : .post(
2789 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
2790 0 : |r| api_handler(r, lsn_lease_handler),
2791 0 : )
2792 0 : .put(
2793 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
2794 0 : |r| api_handler(r, timeline_gc_handler),
2795 0 : )
2796 0 : .put(
2797 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
2798 0 : |r| testing_api_handler("run timeline compaction", r, timeline_compact_handler),
2799 0 : )
2800 0 : .put(
2801 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
2802 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
2803 0 : )
2804 0 : .post(
2805 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2806 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
2807 0 : )
2808 0 : .get(
2809 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2810 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
2811 0 : )
2812 0 : .put(
2813 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
2814 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
2815 0 : )
2816 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2817 0 : api_handler(r, timeline_delete_handler)
2818 0 : })
2819 0 : .get(
2820 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
2821 0 : |r| api_handler(r, layer_map_info_handler),
2822 0 : )
2823 0 : .get(
2824 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2825 0 : |r| api_handler(r, layer_download_handler),
2826 0 : )
2827 0 : .delete(
2828 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2829 0 : |r| api_handler(r, evict_timeline_layer_handler),
2830 0 : )
2831 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
2832 0 : api_handler(r, secondary_upload_handler)
2833 0 : })
2834 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
2835 0 : api_handler(r, tenant_scan_remote_handler)
2836 0 : })
2837 0 : .put("/v1/disk_usage_eviction/run", |r| {
2838 0 : api_handler(r, disk_usage_eviction_run)
2839 0 : })
2840 0 : .put("/v1/deletion_queue/flush", |r| {
2841 0 : api_handler(r, deletion_queue_flush)
2842 0 : })
2843 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
2844 0 : api_handler(r, secondary_status_handler)
2845 0 : })
2846 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
2847 0 : api_handler(r, secondary_download_handler)
2848 0 : })
2849 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
2850 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
2851 0 : })
2852 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
2853 0 : .post("/v1/tracing/event", |r| {
2854 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
2855 0 : })
2856 0 : .get(
2857 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
2858 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
2859 0 : )
2860 0 : .get(
2861 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
2862 0 : |r| api_handler(r, timeline_collect_keyspace),
2863 0 : )
2864 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
2865 0 : .put(
2866 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/force_aux_policy_switch",
2867 0 : |r| api_handler(r, force_aux_policy_switch_handler),
2868 0 : )
2869 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
2870 0 : .post(
2871 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
2872 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
2873 0 : )
2874 0 : .post(
2875 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
2876 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
2877 0 : )
2878 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
2879 0 : .post(
2880 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
2881 0 : |r| testing_api_handler("perf_info", r, perf_info),
2882 0 : )
2883 0 : .any(handler_404))
2884 0 : }
|
