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::IngestAuxFilesRequest;
20 : use pageserver_api::models::ListAuxFilesRequest;
21 : use pageserver_api::models::LocationConfig;
22 : use pageserver_api::models::LocationConfigListResponse;
23 : use pageserver_api::models::ShardParameters;
24 : use pageserver_api::models::TenantDetails;
25 : use pageserver_api::models::TenantLocationConfigResponse;
26 : use pageserver_api::models::TenantScanRemoteStorageResponse;
27 : use pageserver_api::models::TenantScanRemoteStorageShard;
28 : use pageserver_api::models::TenantShardLocation;
29 : use pageserver_api::models::TenantShardSplitRequest;
30 : use pageserver_api::models::TenantShardSplitResponse;
31 : use pageserver_api::models::TenantSorting;
32 : use pageserver_api::models::TenantState;
33 : use pageserver_api::models::TopTenantShardItem;
34 : use pageserver_api::models::TopTenantShardsRequest;
35 : use pageserver_api::models::TopTenantShardsResponse;
36 : use pageserver_api::models::{
37 : DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
38 : TenantLoadRequest, TenantLocationConfigRequest,
39 : };
40 : use pageserver_api::shard::ShardCount;
41 : use pageserver_api::shard::TenantShardId;
42 : use remote_storage::DownloadError;
43 : use remote_storage::GenericRemoteStorage;
44 : use remote_storage::TimeTravelError;
45 : use tenant_size_model::{SizeResult, StorageModel};
46 : use tokio_util::sync::CancellationToken;
47 : use tracing::*;
48 : use utils::auth::JwtAuth;
49 : use utils::failpoint_support::failpoints_handler;
50 : use utils::http::endpoint::prometheus_metrics_handler;
51 : use utils::http::endpoint::request_span;
52 : use utils::http::json::json_request_or_empty_body;
53 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
54 :
55 : use crate::context::{DownloadBehavior, RequestContext};
56 : use crate::deletion_queue::DeletionQueueClient;
57 : use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
58 : use crate::pgdatadir_mapping::LsnForTimestamp;
59 : use crate::task_mgr::TaskKind;
60 : use crate::tenant::config::{LocationConf, TenantConfOpt};
61 : use crate::tenant::mgr::GetActiveTenantError;
62 : use crate::tenant::mgr::{
63 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
64 : TenantSlotUpsertError, TenantStateError,
65 : };
66 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
67 : use crate::tenant::remote_timeline_client;
68 : use crate::tenant::remote_timeline_client::download_index_part;
69 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
70 : use crate::tenant::remote_timeline_client::list_remote_timelines;
71 : use crate::tenant::secondary::SecondaryController;
72 : use crate::tenant::size::ModelInputs;
73 : use crate::tenant::storage_layer::LayerAccessStatsReset;
74 : use crate::tenant::storage_layer::LayerName;
75 : use crate::tenant::timeline::CompactFlags;
76 : use crate::tenant::timeline::Timeline;
77 : use crate::tenant::SpawnMode;
78 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
79 : use crate::{config::PageServerConf, tenant::mgr};
80 : use crate::{disk_usage_eviction_task, tenant};
81 : use pageserver_api::models::{
82 : StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
83 : TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
84 : };
85 : use utils::{
86 : auth::SwappableJwtAuth,
87 : generation::Generation,
88 : http::{
89 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
90 : error::{ApiError, HttpErrorBody},
91 : json::{json_request, json_response},
92 : request::parse_request_param,
93 : RequestExt, RouterBuilder,
94 : },
95 : id::{TenantId, TimelineId},
96 : lsn::Lsn,
97 : };
98 :
99 : // For APIs that require an Active tenant, how long should we block waiting for that state?
100 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
101 : // failed API calls while tenants are activating.
102 : #[cfg(not(feature = "testing"))]
103 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
104 :
105 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
106 : // finish attaching, if calls to remote storage are slow.
107 : #[cfg(feature = "testing")]
108 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
109 :
110 : pub struct State {
111 : conf: &'static PageServerConf,
112 : tenant_manager: Arc<TenantManager>,
113 : auth: Option<Arc<SwappableJwtAuth>>,
114 : allowlist_routes: Vec<Uri>,
115 : remote_storage: GenericRemoteStorage,
116 : broker_client: storage_broker::BrokerClientChannel,
117 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
118 : deletion_queue_client: DeletionQueueClient,
119 : secondary_controller: SecondaryController,
120 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
121 : }
122 :
123 : impl State {
124 : #[allow(clippy::too_many_arguments)]
125 0 : pub fn new(
126 0 : conf: &'static PageServerConf,
127 0 : tenant_manager: Arc<TenantManager>,
128 0 : auth: Option<Arc<SwappableJwtAuth>>,
129 0 : remote_storage: GenericRemoteStorage,
130 0 : broker_client: storage_broker::BrokerClientChannel,
131 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
132 0 : deletion_queue_client: DeletionQueueClient,
133 0 : secondary_controller: SecondaryController,
134 0 : ) -> anyhow::Result<Self> {
135 0 : let allowlist_routes = ["/v1/status", "/v1/doc", "/swagger.yml", "/metrics"]
136 0 : .iter()
137 0 : .map(|v| v.parse().unwrap())
138 0 : .collect::<Vec<_>>();
139 0 : Ok(Self {
140 0 : conf,
141 0 : tenant_manager,
142 0 : auth,
143 0 : allowlist_routes,
144 0 : remote_storage,
145 0 : broker_client,
146 0 : disk_usage_eviction_state,
147 0 : deletion_queue_client,
148 0 : secondary_controller,
149 0 : latest_utilization: Default::default(),
150 0 : })
151 0 : }
152 : }
153 :
154 : #[inline(always)]
155 0 : fn get_state(request: &Request<Body>) -> &State {
156 0 : request
157 0 : .data::<Arc<State>>()
158 0 : .expect("unknown state type")
159 0 : .as_ref()
160 0 : }
161 :
162 : #[inline(always)]
163 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
164 0 : get_state(request).conf
165 0 : }
166 :
167 : /// Check that the requester is authorized to operate on given tenant
168 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
169 0 : check_permission_with(request, |claims| {
170 0 : crate::auth::check_permission(claims, tenant_id)
171 0 : })
172 0 : }
173 :
174 : impl From<PageReconstructError> for ApiError {
175 0 : fn from(pre: PageReconstructError) -> ApiError {
176 0 : match pre {
177 0 : PageReconstructError::Other(pre) => ApiError::InternalServerError(pre),
178 0 : PageReconstructError::MissingKey(e) => {
179 0 : ApiError::InternalServerError(anyhow::anyhow!("{e}"))
180 : }
181 : PageReconstructError::Cancelled => {
182 0 : ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
183 : }
184 : PageReconstructError::AncestorStopping(_) => {
185 0 : ApiError::ResourceUnavailable(format!("{pre}").into())
186 : }
187 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
188 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
189 : }
190 0 : }
191 : }
192 :
193 : impl From<TenantMapInsertError> for ApiError {
194 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
195 0 : match tmie {
196 0 : TenantMapInsertError::SlotError(e) => e.into(),
197 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
198 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
199 : }
200 0 : }
201 : }
202 :
203 : impl From<TenantSlotError> for ApiError {
204 0 : fn from(e: TenantSlotError) -> ApiError {
205 0 : use TenantSlotError::*;
206 0 : match e {
207 0 : NotFound(tenant_id) => {
208 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
209 : }
210 0 : e @ AlreadyExists(_, _) => ApiError::Conflict(format!("{e}")),
211 : InProgress => {
212 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
213 : }
214 0 : MapState(e) => e.into(),
215 : }
216 0 : }
217 : }
218 :
219 : impl From<TenantSlotUpsertError> for ApiError {
220 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
221 0 : use TenantSlotUpsertError::*;
222 0 : match e {
223 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
224 0 : MapState(e) => e.into(),
225 0 : ShuttingDown(_) => ApiError::ShuttingDown,
226 : }
227 0 : }
228 : }
229 :
230 : impl From<UpsertLocationError> for ApiError {
231 0 : fn from(e: UpsertLocationError) -> ApiError {
232 0 : use UpsertLocationError::*;
233 0 : match e {
234 0 : BadRequest(e) => ApiError::BadRequest(e),
235 0 : Unavailable(_) => ApiError::ShuttingDown,
236 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
237 0 : Flush(e) | Other(e) => ApiError::InternalServerError(e),
238 : }
239 0 : }
240 : }
241 :
242 : impl From<TenantMapError> for ApiError {
243 0 : fn from(e: TenantMapError) -> ApiError {
244 0 : use TenantMapError::*;
245 0 : match e {
246 : StillInitializing | ShuttingDown => {
247 0 : ApiError::ResourceUnavailable(format!("{e}").into())
248 0 : }
249 0 : }
250 0 : }
251 : }
252 :
253 : impl From<TenantStateError> for ApiError {
254 0 : fn from(tse: TenantStateError) -> ApiError {
255 0 : match tse {
256 : TenantStateError::IsStopping(_) => {
257 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
258 : }
259 0 : TenantStateError::SlotError(e) => e.into(),
260 0 : TenantStateError::SlotUpsertError(e) => e.into(),
261 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
262 : }
263 0 : }
264 : }
265 :
266 : impl From<GetTenantError> for ApiError {
267 0 : fn from(tse: GetTenantError) -> ApiError {
268 0 : match tse {
269 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
270 : GetTenantError::NotActive(_) => {
271 : // Why is this not `ApiError::NotFound`?
272 : // Because we must be careful to never return 404 for a tenant if it does
273 : // in fact exist locally. If we did, the caller could draw the conclusion
274 : // that it can attach the tenant to another PS and we'd be in split-brain.
275 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
276 : }
277 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
278 : }
279 0 : }
280 : }
281 :
282 : impl From<GetActiveTenantError> for ApiError {
283 0 : fn from(e: GetActiveTenantError) -> ApiError {
284 0 : match e {
285 0 : GetActiveTenantError::Broken(reason) => {
286 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
287 : }
288 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
289 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
290 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
291 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
292 0 : ApiError::ResourceUnavailable(format!("{}", e).into())
293 : }
294 : }
295 0 : }
296 : }
297 :
298 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
299 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
300 0 : use crate::tenant::DeleteTimelineError::*;
301 0 : match value {
302 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
303 0 : HasChildren(children) => ApiError::PreconditionFailed(
304 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
305 0 : .into_boxed_str(),
306 0 : ),
307 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
308 0 : Other(e) => ApiError::InternalServerError(e),
309 : }
310 0 : }
311 : }
312 :
313 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
314 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
315 : use crate::tenant::mgr::DeleteTimelineError::*;
316 0 : match value {
317 : // Report Precondition failed so client can distinguish between
318 : // "tenant is missing" case from "timeline is missing"
319 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
320 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
321 0 : ),
322 0 : Tenant(t) => ApiError::from(t),
323 0 : Timeline(t) => ApiError::from(t),
324 : }
325 0 : }
326 : }
327 :
328 : impl From<crate::tenant::delete::DeleteTenantError> for ApiError {
329 0 : fn from(value: crate::tenant::delete::DeleteTenantError) -> Self {
330 0 : use crate::tenant::delete::DeleteTenantError::*;
331 0 : match value {
332 0 : Get(g) => ApiError::from(g),
333 0 : e @ AlreadyInProgress => ApiError::Conflict(e.to_string()),
334 0 : Timeline(t) => ApiError::from(t),
335 0 : NotAttached => ApiError::NotFound(anyhow::anyhow!("Tenant is not attached").into()),
336 0 : SlotError(e) => e.into(),
337 0 : SlotUpsertError(e) => e.into(),
338 0 : Other(o) => ApiError::InternalServerError(o),
339 0 : e @ InvalidState(_) => ApiError::PreconditionFailed(e.to_string().into_boxed_str()),
340 0 : Cancelled => ApiError::ShuttingDown,
341 : }
342 0 : }
343 : }
344 :
345 : // Helper function to construct a TimelineInfo struct for a timeline
346 0 : async fn build_timeline_info(
347 0 : timeline: &Arc<Timeline>,
348 0 : include_non_incremental_logical_size: bool,
349 0 : force_await_initial_logical_size: bool,
350 0 : ctx: &RequestContext,
351 0 : ) -> anyhow::Result<TimelineInfo> {
352 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
353 0 :
354 0 : if force_await_initial_logical_size {
355 0 : timeline.clone().await_initial_logical_size().await
356 0 : }
357 :
358 0 : let mut info = build_timeline_info_common(
359 0 : timeline,
360 0 : ctx,
361 0 : tenant::timeline::GetLogicalSizePriority::Background,
362 0 : )
363 0 : .await?;
364 0 : if include_non_incremental_logical_size {
365 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
366 : // Otherwise, if someone deletes the timeline / detaches the tenant while
367 : // we're executing this function, we will outlive the timeline on-disk state.
368 : info.current_logical_size_non_incremental = Some(
369 0 : timeline
370 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
371 0 : .await?,
372 : );
373 0 : }
374 0 : Ok(info)
375 0 : }
376 :
377 0 : async fn build_timeline_info_common(
378 0 : timeline: &Arc<Timeline>,
379 0 : ctx: &RequestContext,
380 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
381 0 : ) -> anyhow::Result<TimelineInfo> {
382 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
383 0 : let initdb_lsn = timeline.initdb_lsn;
384 0 : let last_record_lsn = timeline.get_last_record_lsn();
385 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
386 0 : let guard = timeline.last_received_wal.lock().unwrap();
387 0 : if let Some(info) = guard.as_ref() {
388 0 : (
389 0 : Some(format!("{:?}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
390 0 : Some(info.last_received_msg_lsn),
391 0 : Some(info.last_received_msg_ts),
392 0 : )
393 : } else {
394 0 : (None, None, None)
395 : }
396 : };
397 :
398 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
399 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
400 0 : Lsn(0) => None,
401 0 : lsn @ Lsn(_) => Some(lsn),
402 : };
403 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
404 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
405 0 : let state = timeline.current_state();
406 0 : let remote_consistent_lsn_projected = timeline
407 0 : .get_remote_consistent_lsn_projected()
408 0 : .unwrap_or(Lsn(0));
409 0 : let remote_consistent_lsn_visible = timeline
410 0 : .get_remote_consistent_lsn_visible()
411 0 : .unwrap_or(Lsn(0));
412 0 :
413 0 : let walreceiver_status = timeline.walreceiver_status();
414 :
415 0 : let info = TimelineInfo {
416 0 : tenant_id: timeline.tenant_shard_id,
417 0 : timeline_id: timeline.timeline_id,
418 0 : ancestor_timeline_id,
419 0 : ancestor_lsn,
420 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
421 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
422 0 : remote_consistent_lsn_visible,
423 0 : initdb_lsn,
424 0 : last_record_lsn,
425 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
426 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
427 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
428 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
429 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
430 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
431 : },
432 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
433 0 : current_physical_size,
434 0 : current_logical_size_non_incremental: None,
435 0 : timeline_dir_layer_file_size_sum: None,
436 0 : wal_source_connstr,
437 0 : last_received_msg_lsn,
438 0 : last_received_msg_ts,
439 0 : pg_version: timeline.pg_version,
440 0 :
441 0 : state,
442 0 :
443 0 : walreceiver_status,
444 0 :
445 0 : last_aux_file_policy: timeline.last_aux_file_policy.load(),
446 0 : };
447 0 : Ok(info)
448 0 : }
449 :
450 : // healthcheck handler
451 0 : async fn status_handler(
452 0 : request: Request<Body>,
453 0 : _cancel: CancellationToken,
454 0 : ) -> Result<Response<Body>, ApiError> {
455 0 : check_permission(&request, None)?;
456 0 : let config = get_config(&request);
457 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
458 0 : }
459 :
460 0 : async fn reload_auth_validation_keys_handler(
461 0 : request: Request<Body>,
462 0 : _cancel: CancellationToken,
463 0 : ) -> Result<Response<Body>, ApiError> {
464 0 : check_permission(&request, None)?;
465 0 : let config = get_config(&request);
466 0 : let state = get_state(&request);
467 0 : let Some(shared_auth) = &state.auth else {
468 0 : return json_response(StatusCode::BAD_REQUEST, ());
469 : };
470 : // unwrap is ok because check is performed when creating config, so path is set and exists
471 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
472 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
473 :
474 0 : match JwtAuth::from_key_path(key_path) {
475 0 : Ok(new_auth) => {
476 0 : shared_auth.swap(new_auth);
477 0 : json_response(StatusCode::OK, ())
478 : }
479 0 : Err(e) => {
480 0 : let err_msg = "Error reloading public keys";
481 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
482 0 : json_response(
483 0 : StatusCode::INTERNAL_SERVER_ERROR,
484 0 : HttpErrorBody::from_msg(err_msg.to_string()),
485 0 : )
486 : }
487 : }
488 0 : }
489 :
490 0 : async fn timeline_create_handler(
491 0 : mut request: Request<Body>,
492 0 : _cancel: CancellationToken,
493 0 : ) -> Result<Response<Body>, ApiError> {
494 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
495 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
496 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
497 :
498 0 : let new_timeline_id = request_data.new_timeline_id;
499 0 :
500 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
501 0 :
502 0 : let state = get_state(&request);
503 :
504 0 : async {
505 0 : let tenant = state
506 0 : .tenant_manager
507 0 : .get_attached_tenant_shard(tenant_shard_id)?;
508 :
509 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
510 :
511 0 : if let Some(ancestor_id) = request_data.ancestor_timeline_id.as_ref() {
512 0 : tracing::info!(%ancestor_id, "starting to branch");
513 : } else {
514 0 : tracing::info!("bootstrapping");
515 : }
516 :
517 0 : match tenant
518 0 : .create_timeline(
519 0 : new_timeline_id,
520 0 : request_data.ancestor_timeline_id,
521 0 : request_data.ancestor_start_lsn,
522 0 : request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
523 0 : request_data.existing_initdb_timeline_id,
524 0 : state.broker_client.clone(),
525 0 : &ctx,
526 0 : )
527 0 : .await
528 : {
529 0 : Ok(new_timeline) => {
530 : // Created. Construct a TimelineInfo for it.
531 0 : let timeline_info = build_timeline_info_common(
532 0 : &new_timeline,
533 0 : &ctx,
534 0 : tenant::timeline::GetLogicalSizePriority::User,
535 0 : )
536 0 : .await
537 0 : .map_err(ApiError::InternalServerError)?;
538 0 : json_response(StatusCode::CREATED, timeline_info)
539 : }
540 0 : Err(_) if tenant.cancel.is_cancelled() => {
541 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
542 0 : json_response(
543 0 : StatusCode::SERVICE_UNAVAILABLE,
544 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
545 0 : )
546 : }
547 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
548 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
549 : }
550 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
551 0 : StatusCode::TOO_MANY_REQUESTS,
552 0 : HttpErrorBody::from_msg(e.to_string()),
553 0 : ),
554 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
555 0 : StatusCode::NOT_ACCEPTABLE,
556 0 : HttpErrorBody::from_msg(format!("{err:#}")),
557 0 : ),
558 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
559 0 : StatusCode::SERVICE_UNAVAILABLE,
560 0 : HttpErrorBody::from_msg(e.to_string()),
561 0 : ),
562 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
563 0 : StatusCode::SERVICE_UNAVAILABLE,
564 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
565 0 : ),
566 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
567 : }
568 0 : }
569 0 : .instrument(info_span!("timeline_create",
570 : tenant_id = %tenant_shard_id.tenant_id,
571 0 : shard_id = %tenant_shard_id.shard_slug(),
572 : timeline_id = %new_timeline_id,
573 : lsn=?request_data.ancestor_start_lsn,
574 : pg_version=?request_data.pg_version
575 : ))
576 0 : .await
577 0 : }
578 :
579 0 : async fn timeline_list_handler(
580 0 : request: Request<Body>,
581 0 : _cancel: CancellationToken,
582 0 : ) -> Result<Response<Body>, ApiError> {
583 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
584 0 : let include_non_incremental_logical_size: Option<bool> =
585 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
586 0 : let force_await_initial_logical_size: Option<bool> =
587 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
588 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
589 :
590 0 : let state = get_state(&request);
591 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
592 :
593 0 : let response_data = async {
594 0 : let tenant = state
595 0 : .tenant_manager
596 0 : .get_attached_tenant_shard(tenant_shard_id)?;
597 :
598 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
599 :
600 0 : let timelines = tenant.list_timelines();
601 0 :
602 0 : let mut response_data = Vec::with_capacity(timelines.len());
603 0 : for timeline in timelines {
604 0 : let timeline_info = build_timeline_info(
605 0 : &timeline,
606 0 : include_non_incremental_logical_size.unwrap_or(false),
607 0 : force_await_initial_logical_size.unwrap_or(false),
608 0 : &ctx,
609 0 : )
610 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
611 0 : .await
612 0 : .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
613 0 : .map_err(ApiError::InternalServerError)?;
614 :
615 0 : response_data.push(timeline_info);
616 : }
617 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
618 0 : }
619 0 : .instrument(info_span!("timeline_list",
620 : tenant_id = %tenant_shard_id.tenant_id,
621 0 : shard_id = %tenant_shard_id.shard_slug()))
622 0 : .await?;
623 :
624 0 : json_response(StatusCode::OK, response_data)
625 0 : }
626 :
627 0 : async fn timeline_preserve_initdb_handler(
628 0 : request: Request<Body>,
629 0 : _cancel: CancellationToken,
630 0 : ) -> Result<Response<Body>, ApiError> {
631 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
632 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
633 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
634 0 : let state = get_state(&request);
635 :
636 : // Part of the process for disaster recovery from safekeeper-stored WAL:
637 : // If we don't recover into a new timeline but want to keep the timeline ID,
638 : // then the initdb archive is deleted. This endpoint copies it to a different
639 : // location where timeline recreation cand find it.
640 :
641 0 : async {
642 0 : let tenant = state
643 0 : .tenant_manager
644 0 : .get_attached_tenant_shard(tenant_shard_id)?;
645 :
646 0 : let timeline = tenant
647 0 : .get_timeline(timeline_id, false)
648 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
649 :
650 0 : timeline
651 0 : .preserve_initdb_archive()
652 0 : .await
653 0 : .context("preserving initdb archive")
654 0 : .map_err(ApiError::InternalServerError)?;
655 :
656 0 : Ok::<_, ApiError>(())
657 0 : }
658 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
659 : tenant_id = %tenant_shard_id.tenant_id,
660 0 : shard_id = %tenant_shard_id.shard_slug(),
661 : %timeline_id))
662 0 : .await?;
663 :
664 0 : json_response(StatusCode::OK, ())
665 0 : }
666 :
667 0 : async fn timeline_detail_handler(
668 0 : request: Request<Body>,
669 0 : _cancel: CancellationToken,
670 0 : ) -> Result<Response<Body>, ApiError> {
671 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
672 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
673 0 : let include_non_incremental_logical_size: Option<bool> =
674 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
675 0 : let force_await_initial_logical_size: Option<bool> =
676 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
677 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
678 :
679 : // Logical size calculation needs downloading.
680 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
681 0 : let state = get_state(&request);
682 :
683 0 : let timeline_info = async {
684 0 : let tenant = state
685 0 : .tenant_manager
686 0 : .get_attached_tenant_shard(tenant_shard_id)?;
687 :
688 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
689 :
690 0 : let timeline = tenant
691 0 : .get_timeline(timeline_id, false)
692 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
693 :
694 0 : let timeline_info = build_timeline_info(
695 0 : &timeline,
696 0 : include_non_incremental_logical_size.unwrap_or(false),
697 0 : force_await_initial_logical_size.unwrap_or(false),
698 0 : &ctx,
699 0 : )
700 0 : .await
701 0 : .context("get local timeline info")
702 0 : .map_err(ApiError::InternalServerError)?;
703 :
704 0 : Ok::<_, ApiError>(timeline_info)
705 0 : }
706 0 : .instrument(info_span!("timeline_detail",
707 : tenant_id = %tenant_shard_id.tenant_id,
708 0 : shard_id = %tenant_shard_id.shard_slug(),
709 : %timeline_id))
710 0 : .await?;
711 :
712 0 : json_response(StatusCode::OK, timeline_info)
713 0 : }
714 :
715 0 : async fn get_lsn_by_timestamp_handler(
716 0 : request: Request<Body>,
717 0 : cancel: CancellationToken,
718 0 : ) -> Result<Response<Body>, ApiError> {
719 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
720 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
721 0 : let state = get_state(&request);
722 0 :
723 0 : if !tenant_shard_id.is_shard_zero() {
724 : // Requires SLRU contents, which are only stored on shard zero
725 0 : return Err(ApiError::BadRequest(anyhow!(
726 0 : "Size calculations are only available on shard zero"
727 0 : )));
728 0 : }
729 :
730 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
731 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
732 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
733 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
734 0 : .map_err(ApiError::BadRequest)?;
735 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
736 0 :
737 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
738 :
739 0 : let timeline =
740 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
741 0 : .await?;
742 0 : let result = timeline
743 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
744 0 : .await?;
745 : #[derive(serde::Serialize, Debug)]
746 : struct Result {
747 : lsn: Lsn,
748 : kind: &'static str,
749 : }
750 0 : let (lsn, kind) = match result {
751 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
752 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
753 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
754 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
755 : };
756 0 : let result = Result { lsn, kind };
757 0 : tracing::info!(
758 : lsn=?result.lsn,
759 : kind=%result.kind,
760 : timestamp=%timestamp_raw,
761 0 : "lsn_by_timestamp finished"
762 : );
763 0 : json_response(StatusCode::OK, result)
764 0 : }
765 :
766 0 : async fn get_timestamp_of_lsn_handler(
767 0 : request: Request<Body>,
768 0 : _cancel: CancellationToken,
769 0 : ) -> Result<Response<Body>, ApiError> {
770 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
771 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
772 0 : let state = get_state(&request);
773 0 :
774 0 : if !tenant_shard_id.is_shard_zero() {
775 : // Requires SLRU contents, which are only stored on shard zero
776 0 : return Err(ApiError::BadRequest(anyhow!(
777 0 : "Size calculations are only available on shard zero"
778 0 : )));
779 0 : }
780 :
781 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
782 :
783 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
784 0 : let lsn = Lsn::from_str(&lsn_str)
785 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
786 0 : .map_err(ApiError::BadRequest)?;
787 :
788 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
789 0 : let timeline =
790 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
791 0 : .await?;
792 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
793 :
794 0 : match result {
795 0 : Some(time) => {
796 0 : let time = format_rfc3339(postgres_ffi::from_pg_timestamp(time)).to_string();
797 0 : json_response(StatusCode::OK, time)
798 : }
799 0 : None => Err(ApiError::NotFound(
800 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
801 0 : )),
802 : }
803 0 : }
804 :
805 0 : async fn tenant_attach_handler(
806 0 : mut request: Request<Body>,
807 0 : _cancel: CancellationToken,
808 0 : ) -> Result<Response<Body>, ApiError> {
809 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
810 0 : check_permission(&request, Some(tenant_id))?;
811 :
812 0 : let maybe_body: Option<TenantAttachRequest> = json_request_or_empty_body(&mut request).await?;
813 0 : let tenant_conf = match &maybe_body {
814 0 : Some(request) => TenantConfOpt::try_from(&*request.config).map_err(ApiError::BadRequest)?,
815 0 : None => TenantConfOpt::default(),
816 : };
817 :
818 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
819 0 :
820 0 : info!("Handling tenant attach {tenant_id}");
821 :
822 0 : let state = get_state(&request);
823 :
824 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
825 :
826 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
827 0 : let shard_params = ShardParameters::default();
828 0 : let location_conf = LocationConf::attached_single(tenant_conf, generation, &shard_params);
829 :
830 0 : let tenant = state
831 0 : .tenant_manager
832 0 : .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
833 0 : .await?;
834 :
835 0 : let Some(tenant) = tenant else {
836 : // This should never happen: indicates a bug in upsert_location
837 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
838 0 : "Upsert succeeded but didn't return tenant!"
839 0 : )));
840 : };
841 :
842 : // We might have successfully constructed a Tenant, but it could still
843 : // end up in a broken state:
844 : if let TenantState::Broken {
845 0 : reason,
846 : backtrace: _,
847 0 : } = tenant.current_state()
848 : {
849 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
850 0 : "Tenant state is Broken: {reason}"
851 0 : )));
852 0 : }
853 0 :
854 0 : json_response(StatusCode::ACCEPTED, ())
855 0 : }
856 :
857 0 : async fn timeline_delete_handler(
858 0 : request: Request<Body>,
859 0 : _cancel: CancellationToken,
860 0 : ) -> Result<Response<Body>, ApiError> {
861 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
862 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
863 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
864 :
865 0 : let state = get_state(&request);
866 :
867 0 : let tenant = state
868 0 : .tenant_manager
869 0 : .get_attached_tenant_shard(tenant_shard_id)
870 0 : .map_err(|e| {
871 0 : match e {
872 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
873 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
874 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
875 0 : "Requested tenant is missing".to_string().into_boxed_str(),
876 0 : ),
877 0 : e => e.into(),
878 : }
879 0 : })?;
880 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
881 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))
882 0 : .await?;
883 :
884 0 : json_response(StatusCode::ACCEPTED, ())
885 0 : }
886 :
887 0 : async fn tenant_detach_handler(
888 0 : request: Request<Body>,
889 0 : _cancel: CancellationToken,
890 0 : ) -> Result<Response<Body>, ApiError> {
891 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
892 0 : check_permission(&request, Some(tenant_id))?;
893 0 : let detach_ignored: Option<bool> = parse_query_param(&request, "detach_ignored")?;
894 :
895 : // This is a legacy API (`/location_conf` is the replacement). It only supports unsharded tenants
896 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
897 0 :
898 0 : let state = get_state(&request);
899 0 : let conf = state.conf;
900 0 : state
901 0 : .tenant_manager
902 0 : .detach_tenant(
903 0 : conf,
904 0 : tenant_shard_id,
905 0 : detach_ignored.unwrap_or(false),
906 0 : &state.deletion_queue_client,
907 0 : )
908 0 : .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
909 0 : .await?;
910 :
911 0 : json_response(StatusCode::OK, ())
912 0 : }
913 :
914 0 : async fn tenant_reset_handler(
915 0 : request: Request<Body>,
916 0 : _cancel: CancellationToken,
917 0 : ) -> Result<Response<Body>, ApiError> {
918 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
919 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
920 :
921 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
922 :
923 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
924 0 : let state = get_state(&request);
925 0 : state
926 0 : .tenant_manager
927 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
928 0 : .await
929 0 : .map_err(ApiError::InternalServerError)?;
930 :
931 0 : json_response(StatusCode::OK, ())
932 0 : }
933 :
934 0 : async fn tenant_load_handler(
935 0 : mut request: Request<Body>,
936 0 : _cancel: CancellationToken,
937 0 : ) -> Result<Response<Body>, ApiError> {
938 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
939 0 : check_permission(&request, Some(tenant_id))?;
940 :
941 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
942 :
943 0 : let maybe_body: Option<TenantLoadRequest> = json_request_or_empty_body(&mut request).await?;
944 :
945 0 : let state = get_state(&request);
946 :
947 : // The /load request is only usable when control_plane_api is not set. Once it is set, callers
948 : // should always use /attach instead.
949 0 : let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
950 :
951 0 : mgr::load_tenant(
952 0 : state.conf,
953 0 : tenant_id,
954 0 : generation,
955 0 : state.broker_client.clone(),
956 0 : state.remote_storage.clone(),
957 0 : state.deletion_queue_client.clone(),
958 0 : &ctx,
959 0 : )
960 0 : .instrument(info_span!("load", %tenant_id))
961 0 : .await?;
962 :
963 0 : json_response(StatusCode::ACCEPTED, ())
964 0 : }
965 :
966 0 : async fn tenant_ignore_handler(
967 0 : request: Request<Body>,
968 0 : _cancel: CancellationToken,
969 0 : ) -> Result<Response<Body>, ApiError> {
970 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
971 0 : check_permission(&request, Some(tenant_id))?;
972 :
973 0 : let state = get_state(&request);
974 0 : let conf = state.conf;
975 0 : mgr::ignore_tenant(conf, tenant_id)
976 0 : .instrument(info_span!("ignore_tenant", %tenant_id))
977 0 : .await?;
978 :
979 0 : json_response(StatusCode::OK, ())
980 0 : }
981 :
982 0 : async fn tenant_list_handler(
983 0 : request: Request<Body>,
984 0 : _cancel: CancellationToken,
985 0 : ) -> Result<Response<Body>, ApiError> {
986 0 : check_permission(&request, None)?;
987 0 : let state = get_state(&request);
988 :
989 0 : let response_data = state
990 0 : .tenant_manager
991 0 : .list_tenants()
992 0 : .map_err(|_| {
993 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
994 0 : })?
995 0 : .iter()
996 0 : .map(|(id, state, gen)| TenantInfo {
997 0 : id: *id,
998 0 : state: state.clone(),
999 0 : current_physical_size: None,
1000 0 : attachment_status: state.attachment_status(),
1001 0 : generation: (*gen).into(),
1002 0 : })
1003 0 : .collect::<Vec<TenantInfo>>();
1004 0 :
1005 0 : json_response(StatusCode::OK, response_data)
1006 0 : }
1007 :
1008 0 : async fn tenant_status(
1009 0 : request: Request<Body>,
1010 0 : _cancel: CancellationToken,
1011 0 : ) -> Result<Response<Body>, ApiError> {
1012 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1013 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1014 0 : let state = get_state(&request);
1015 0 :
1016 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1017 0 : let activate = true;
1018 : #[cfg(feature = "testing")]
1019 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1020 :
1021 0 : let tenant_info = async {
1022 0 : let tenant = state
1023 0 : .tenant_manager
1024 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1025 :
1026 0 : if activate {
1027 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1028 : // called, but it is still valid to return 200 and describe the current state of the tenant
1029 : // if it doesn't make it into an active state.
1030 0 : tenant
1031 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1032 0 : .await
1033 0 : .ok();
1034 0 : }
1035 :
1036 : // Calculate total physical size of all timelines
1037 0 : let mut current_physical_size = 0;
1038 0 : for timeline in tenant.list_timelines().iter() {
1039 0 : current_physical_size += timeline.layer_size_sum().await;
1040 : }
1041 :
1042 0 : let state = tenant.current_state();
1043 0 : Result::<_, ApiError>::Ok(TenantDetails {
1044 0 : tenant_info: TenantInfo {
1045 0 : id: tenant_shard_id,
1046 0 : state: state.clone(),
1047 0 : current_physical_size: Some(current_physical_size),
1048 0 : attachment_status: state.attachment_status(),
1049 0 : generation: tenant.generation().into(),
1050 0 : },
1051 0 : walredo: tenant.wal_redo_manager_status(),
1052 0 : timelines: tenant.list_timeline_ids(),
1053 0 : })
1054 0 : }
1055 0 : .instrument(info_span!("tenant_status_handler",
1056 : tenant_id = %tenant_shard_id.tenant_id,
1057 0 : shard_id = %tenant_shard_id.shard_slug()))
1058 0 : .await?;
1059 :
1060 0 : json_response(StatusCode::OK, tenant_info)
1061 0 : }
1062 :
1063 0 : async fn tenant_delete_handler(
1064 0 : request: Request<Body>,
1065 0 : _cancel: CancellationToken,
1066 0 : ) -> Result<Response<Body>, ApiError> {
1067 : // TODO openapi spec
1068 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1069 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1070 :
1071 0 : let state = get_state(&request);
1072 0 :
1073 0 : state
1074 0 : .tenant_manager
1075 0 : .delete_tenant(tenant_shard_id, ACTIVE_TENANT_TIMEOUT)
1076 0 : .instrument(info_span!("tenant_delete_handler",
1077 : tenant_id = %tenant_shard_id.tenant_id,
1078 0 : shard_id = %tenant_shard_id.shard_slug()
1079 : ))
1080 0 : .await?;
1081 :
1082 0 : json_response(StatusCode::ACCEPTED, ())
1083 0 : }
1084 :
1085 : /// HTTP endpoint to query the current tenant_size of a tenant.
1086 : ///
1087 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1088 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1089 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1090 : /// values.
1091 : ///
1092 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1093 : /// (only if it is shorter than the real cutoff).
1094 : ///
1095 : /// Note: we don't update the cached size and prometheus metric here.
1096 : /// The retention period might be different, and it's nice to have a method to just calculate it
1097 : /// without modifying anything anyway.
1098 0 : async fn tenant_size_handler(
1099 0 : request: Request<Body>,
1100 0 : cancel: CancellationToken,
1101 0 : ) -> Result<Response<Body>, ApiError> {
1102 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1103 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1104 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1105 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1106 0 : let headers = request.headers();
1107 0 : let state = get_state(&request);
1108 0 :
1109 0 : if !tenant_shard_id.is_shard_zero() {
1110 0 : return Err(ApiError::BadRequest(anyhow!(
1111 0 : "Size calculations are only available on shard zero"
1112 0 : )));
1113 0 : }
1114 0 :
1115 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1116 0 : let tenant = state
1117 0 : .tenant_manager
1118 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1119 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1120 :
1121 : // this can be long operation
1122 0 : let inputs = tenant
1123 0 : .gather_size_inputs(
1124 0 : retention_period,
1125 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1126 0 : &cancel,
1127 0 : &ctx,
1128 0 : )
1129 0 : .await
1130 0 : .map_err(ApiError::InternalServerError)?;
1131 :
1132 0 : let mut sizes = None;
1133 0 : let accepts_html = headers
1134 0 : .get(header::ACCEPT)
1135 0 : .map(|v| v == "text/html")
1136 0 : .unwrap_or_default();
1137 0 : if !inputs_only.unwrap_or(false) {
1138 0 : let storage_model = inputs
1139 0 : .calculate_model()
1140 0 : .map_err(ApiError::InternalServerError)?;
1141 0 : let size = storage_model.calculate();
1142 0 :
1143 0 : // If request header expects html, return html
1144 0 : if accepts_html {
1145 0 : return synthetic_size_html_response(inputs, storage_model, size);
1146 0 : }
1147 0 : sizes = Some(size);
1148 0 : } else if accepts_html {
1149 0 : return Err(ApiError::BadRequest(anyhow!(
1150 0 : "inputs_only parameter is incompatible with html output request"
1151 0 : )));
1152 0 : }
1153 :
1154 : /// The type resides in the pageserver not to expose `ModelInputs`.
1155 : #[derive(serde::Serialize)]
1156 : struct TenantHistorySize {
1157 : id: TenantId,
1158 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1159 : ///
1160 : /// Will be none if `?inputs_only=true` was given.
1161 : size: Option<u64>,
1162 : /// Size of each segment used in the model.
1163 : /// Will be null if `?inputs_only=true` was given.
1164 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1165 : inputs: crate::tenant::size::ModelInputs,
1166 : }
1167 :
1168 0 : json_response(
1169 0 : StatusCode::OK,
1170 0 : TenantHistorySize {
1171 0 : id: tenant_shard_id.tenant_id,
1172 0 : size: sizes.as_ref().map(|x| x.total_size),
1173 0 : segment_sizes: sizes.map(|x| x.segments),
1174 0 : inputs,
1175 0 : },
1176 0 : )
1177 0 : }
1178 :
1179 0 : async fn tenant_shard_split_handler(
1180 0 : mut request: Request<Body>,
1181 0 : _cancel: CancellationToken,
1182 0 : ) -> Result<Response<Body>, ApiError> {
1183 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1184 :
1185 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1186 0 : let state = get_state(&request);
1187 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1188 :
1189 0 : let tenant = state
1190 0 : .tenant_manager
1191 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1192 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1193 :
1194 0 : let new_shards = state
1195 0 : .tenant_manager
1196 0 : .shard_split(
1197 0 : tenant,
1198 0 : ShardCount::new(req.new_shard_count),
1199 0 : req.new_stripe_size,
1200 0 : &ctx,
1201 0 : )
1202 0 : .await
1203 0 : .map_err(ApiError::InternalServerError)?;
1204 :
1205 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1206 0 : }
1207 :
1208 0 : async fn layer_map_info_handler(
1209 0 : request: Request<Body>,
1210 0 : _cancel: CancellationToken,
1211 0 : ) -> Result<Response<Body>, ApiError> {
1212 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1213 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1214 0 : let reset: LayerAccessStatsReset =
1215 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1216 0 : let state = get_state(&request);
1217 0 :
1218 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1219 :
1220 0 : let timeline =
1221 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1222 0 : .await?;
1223 0 : let layer_map_info = timeline.layer_map_info(reset).await;
1224 :
1225 0 : json_response(StatusCode::OK, layer_map_info)
1226 0 : }
1227 :
1228 0 : async fn layer_download_handler(
1229 0 : request: Request<Body>,
1230 0 : _cancel: CancellationToken,
1231 0 : ) -> Result<Response<Body>, ApiError> {
1232 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1233 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1234 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1235 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1236 0 : let layer_name = LayerName::from_str(layer_file_name)
1237 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1238 0 : let state = get_state(&request);
1239 :
1240 0 : let timeline =
1241 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1242 0 : .await?;
1243 0 : let downloaded = timeline
1244 0 : .download_layer(&layer_name)
1245 0 : .await
1246 0 : .map_err(ApiError::InternalServerError)?;
1247 :
1248 0 : match downloaded {
1249 0 : Some(true) => json_response(StatusCode::OK, ()),
1250 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1251 0 : None => json_response(
1252 0 : StatusCode::BAD_REQUEST,
1253 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1254 0 : ),
1255 : }
1256 0 : }
1257 :
1258 0 : async fn evict_timeline_layer_handler(
1259 0 : request: Request<Body>,
1260 0 : _cancel: CancellationToken,
1261 0 : ) -> Result<Response<Body>, ApiError> {
1262 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1263 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1264 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1265 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1266 0 : let state = get_state(&request);
1267 :
1268 0 : let layer_name = LayerName::from_str(layer_file_name)
1269 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1270 :
1271 0 : let timeline =
1272 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1273 0 : .await?;
1274 0 : let evicted = timeline
1275 0 : .evict_layer(&layer_name)
1276 0 : .await
1277 0 : .map_err(ApiError::InternalServerError)?;
1278 :
1279 0 : match evicted {
1280 0 : Some(true) => json_response(StatusCode::OK, ()),
1281 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1282 0 : None => json_response(
1283 0 : StatusCode::BAD_REQUEST,
1284 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1285 0 : ),
1286 : }
1287 0 : }
1288 :
1289 : /// Get tenant_size SVG graph along with the JSON data.
1290 0 : fn synthetic_size_html_response(
1291 0 : inputs: ModelInputs,
1292 0 : storage_model: StorageModel,
1293 0 : sizes: SizeResult,
1294 0 : ) -> Result<Response<Body>, ApiError> {
1295 0 : let mut timeline_ids: Vec<String> = Vec::new();
1296 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1297 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1298 0 : timeline_map.insert(ti.timeline_id, index);
1299 0 : timeline_ids.push(ti.timeline_id.to_string());
1300 0 : }
1301 0 : let seg_to_branch: Vec<usize> = inputs
1302 0 : .segments
1303 0 : .iter()
1304 0 : .map(|seg| *timeline_map.get(&seg.timeline_id).unwrap())
1305 0 : .collect();
1306 :
1307 0 : let svg =
1308 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1309 0 : .map_err(ApiError::InternalServerError)?;
1310 :
1311 0 : let mut response = String::new();
1312 0 :
1313 0 : use std::fmt::Write;
1314 0 : write!(response, "<html>\n<body>\n").unwrap();
1315 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1316 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1317 0 : writeln!(response, "<pre>").unwrap();
1318 0 : writeln!(
1319 0 : response,
1320 0 : "{}",
1321 0 : serde_json::to_string_pretty(&inputs).unwrap()
1322 0 : )
1323 0 : .unwrap();
1324 0 : writeln!(
1325 0 : response,
1326 0 : "{}",
1327 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1328 0 : )
1329 0 : .unwrap();
1330 0 : writeln!(response, "</pre>").unwrap();
1331 0 : write!(response, "</body>\n</html>\n").unwrap();
1332 0 :
1333 0 : html_response(StatusCode::OK, response)
1334 0 : }
1335 :
1336 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1337 0 : let response = Response::builder()
1338 0 : .status(status)
1339 0 : .header(header::CONTENT_TYPE, "text/html")
1340 0 : .body(Body::from(data.as_bytes().to_vec()))
1341 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1342 0 : Ok(response)
1343 0 : }
1344 :
1345 : /// Helper for requests that may take a generation, which is mandatory
1346 : /// when control_plane_api is set, but otherwise defaults to Generation::none()
1347 0 : fn get_request_generation(state: &State, req_gen: Option<u32>) -> Result<Generation, ApiError> {
1348 0 : if state.conf.control_plane_api.is_some() {
1349 0 : req_gen
1350 0 : .map(Generation::new)
1351 0 : .ok_or(ApiError::BadRequest(anyhow!(
1352 0 : "generation attribute missing"
1353 0 : )))
1354 : } else {
1355 : // Legacy mode: all tenants operate with no generation
1356 0 : Ok(Generation::none())
1357 : }
1358 0 : }
1359 :
1360 0 : async fn tenant_create_handler(
1361 0 : mut request: Request<Body>,
1362 0 : _cancel: CancellationToken,
1363 0 : ) -> Result<Response<Body>, ApiError> {
1364 0 : let request_data: TenantCreateRequest = json_request(&mut request).await?;
1365 0 : let target_tenant_id = request_data.new_tenant_id;
1366 0 : check_permission(&request, None)?;
1367 :
1368 0 : let _timer = STORAGE_TIME_GLOBAL
1369 0 : .get_metric_with_label_values(&[StorageTimeOperation::CreateTenant.into()])
1370 0 : .expect("bug")
1371 0 : .start_timer();
1372 :
1373 0 : let tenant_conf =
1374 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1375 :
1376 0 : let state = get_state(&request);
1377 :
1378 0 : let generation = get_request_generation(state, request_data.generation)?;
1379 :
1380 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1381 0 :
1382 0 : let location_conf =
1383 0 : LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
1384 :
1385 0 : let new_tenant = state
1386 0 : .tenant_manager
1387 0 : .upsert_location(
1388 0 : target_tenant_id,
1389 0 : location_conf,
1390 0 : None,
1391 0 : SpawnMode::Create,
1392 0 : &ctx,
1393 0 : )
1394 0 : .await?;
1395 :
1396 0 : let Some(new_tenant) = new_tenant else {
1397 : // This should never happen: indicates a bug in upsert_location
1398 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
1399 0 : "Upsert succeeded but didn't return tenant!"
1400 0 : )));
1401 : };
1402 : // We created the tenant. Existing API semantics are that the tenant
1403 : // is Active when this function returns.
1404 0 : new_tenant
1405 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1406 0 : .await?;
1407 :
1408 0 : json_response(
1409 0 : StatusCode::CREATED,
1410 0 : TenantCreateResponse(new_tenant.tenant_shard_id().tenant_id),
1411 0 : )
1412 0 : }
1413 :
1414 0 : async fn get_tenant_config_handler(
1415 0 : request: Request<Body>,
1416 0 : _cancel: CancellationToken,
1417 0 : ) -> Result<Response<Body>, ApiError> {
1418 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1419 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1420 0 : let state = get_state(&request);
1421 :
1422 0 : let tenant = state
1423 0 : .tenant_manager
1424 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1425 :
1426 0 : let response = HashMap::from([
1427 : (
1428 : "tenant_specific_overrides",
1429 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1430 0 : .context("serializing tenant specific overrides")
1431 0 : .map_err(ApiError::InternalServerError)?,
1432 : ),
1433 : (
1434 0 : "effective_config",
1435 0 : serde_json::to_value(tenant.effective_config())
1436 0 : .context("serializing effective config")
1437 0 : .map_err(ApiError::InternalServerError)?,
1438 : ),
1439 : ]);
1440 :
1441 0 : json_response(StatusCode::OK, response)
1442 0 : }
1443 :
1444 0 : async fn update_tenant_config_handler(
1445 0 : mut request: Request<Body>,
1446 0 : _cancel: CancellationToken,
1447 0 : ) -> Result<Response<Body>, ApiError> {
1448 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1449 0 : let tenant_id = request_data.tenant_id;
1450 0 : check_permission(&request, Some(tenant_id))?;
1451 :
1452 0 : let new_tenant_conf =
1453 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1454 :
1455 0 : let state = get_state(&request);
1456 0 :
1457 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1458 :
1459 0 : let tenant = state
1460 0 : .tenant_manager
1461 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1462 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1463 :
1464 : // This is a legacy API that only operates on attached tenants: the preferred
1465 : // API to use is the location_config/ endpoint, which lets the caller provide
1466 : // the full LocationConf.
1467 0 : let location_conf = LocationConf::attached_single(
1468 0 : new_tenant_conf.clone(),
1469 0 : tenant.get_generation(),
1470 0 : &ShardParameters::default(),
1471 0 : );
1472 0 :
1473 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1474 0 : .await
1475 0 : .map_err(ApiError::InternalServerError)?;
1476 0 : tenant.set_new_tenant_config(new_tenant_conf);
1477 0 :
1478 0 : json_response(StatusCode::OK, ())
1479 0 : }
1480 :
1481 0 : async fn put_tenant_location_config_handler(
1482 0 : mut request: Request<Body>,
1483 0 : _cancel: CancellationToken,
1484 0 : ) -> Result<Response<Body>, ApiError> {
1485 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1486 :
1487 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1488 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1489 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1490 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1491 :
1492 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1493 0 : let state = get_state(&request);
1494 0 : let conf = state.conf;
1495 0 :
1496 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1497 0 : // its local disk content and drops it from memory.
1498 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1499 0 : if let Err(e) = state
1500 0 : .tenant_manager
1501 0 : .detach_tenant(conf, tenant_shard_id, true, &state.deletion_queue_client)
1502 0 : .instrument(info_span!("tenant_detach",
1503 : tenant_id = %tenant_shard_id.tenant_id,
1504 0 : shard_id = %tenant_shard_id.shard_slug()
1505 : ))
1506 0 : .await
1507 : {
1508 0 : match e {
1509 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1510 0 : // This API is idempotent: a NotFound on a detach is fine.
1511 0 : }
1512 0 : _ => return Err(e.into()),
1513 : }
1514 0 : }
1515 0 : return json_response(StatusCode::OK, ());
1516 0 : }
1517 :
1518 0 : let location_conf =
1519 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1520 :
1521 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1522 : // similar to at startup ordering.
1523 0 : let spawn_mode = if lazy {
1524 0 : tenant::SpawnMode::Lazy
1525 : } else {
1526 0 : tenant::SpawnMode::Eager
1527 : };
1528 :
1529 0 : let tenant = state
1530 0 : .tenant_manager
1531 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1532 0 : .await?;
1533 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1534 0 : let attached = tenant.is_some();
1535 :
1536 0 : if let Some(_flush_ms) = flush {
1537 0 : match state
1538 0 : .secondary_controller
1539 0 : .upload_tenant(tenant_shard_id)
1540 0 : .await
1541 : {
1542 : Ok(()) => {
1543 0 : tracing::info!("Uploaded heatmap during flush");
1544 : }
1545 0 : Err(e) => {
1546 0 : tracing::warn!("Failed to flush heatmap: {e}");
1547 : }
1548 : }
1549 : } else {
1550 0 : tracing::info!("No flush requested when configuring");
1551 : }
1552 :
1553 : // This API returns a vector of pageservers where the tenant is attached: this is
1554 : // primarily for use in the sharding service. For compatibilty, we also return this
1555 : // when called directly on a pageserver, but the payload is always zero or one shards.
1556 0 : let mut response = TenantLocationConfigResponse {
1557 0 : shards: Vec::new(),
1558 0 : stripe_size: None,
1559 0 : };
1560 0 : if attached {
1561 0 : response.shards.push(TenantShardLocation {
1562 0 : shard_id: tenant_shard_id,
1563 0 : node_id: state.conf.id,
1564 0 : });
1565 0 : if tenant_shard_id.shard_count.count() > 1 {
1566 : // Stripe size should be set if we are attached
1567 0 : debug_assert!(stripe_size.is_some());
1568 0 : response.stripe_size = stripe_size;
1569 0 : }
1570 0 : }
1571 :
1572 0 : json_response(StatusCode::OK, response)
1573 0 : }
1574 :
1575 0 : async fn list_location_config_handler(
1576 0 : request: Request<Body>,
1577 0 : _cancel: CancellationToken,
1578 0 : ) -> Result<Response<Body>, ApiError> {
1579 0 : let state = get_state(&request);
1580 0 : let slots = state.tenant_manager.list();
1581 0 : let result = LocationConfigListResponse {
1582 0 : tenant_shards: slots
1583 0 : .into_iter()
1584 0 : .map(|(tenant_shard_id, slot)| {
1585 0 : let v = match slot {
1586 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1587 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1588 0 : TenantSlot::InProgress(_) => None,
1589 : };
1590 0 : (tenant_shard_id, v)
1591 0 : })
1592 0 : .collect(),
1593 0 : };
1594 0 : json_response(StatusCode::OK, result)
1595 0 : }
1596 :
1597 0 : async fn get_location_config_handler(
1598 0 : request: Request<Body>,
1599 0 : _cancel: CancellationToken,
1600 0 : ) -> Result<Response<Body>, ApiError> {
1601 0 : let state = get_state(&request);
1602 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1603 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1604 :
1605 0 : let Some(slot) = slot else {
1606 0 : return Err(ApiError::NotFound(
1607 0 : anyhow::anyhow!("Tenant shard not found").into(),
1608 0 : ));
1609 : };
1610 :
1611 0 : let result: Option<LocationConfig> = match slot {
1612 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1613 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1614 0 : TenantSlot::InProgress(_) => None,
1615 : };
1616 :
1617 0 : json_response(StatusCode::OK, result)
1618 0 : }
1619 :
1620 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1621 : // (from all pageservers) as it invalidates consistency assumptions.
1622 0 : async fn tenant_time_travel_remote_storage_handler(
1623 0 : request: Request<Body>,
1624 0 : cancel: CancellationToken,
1625 0 : ) -> Result<Response<Body>, ApiError> {
1626 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1627 :
1628 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1629 :
1630 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1631 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1632 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1633 0 : .map_err(ApiError::BadRequest)?;
1634 :
1635 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1636 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1637 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1638 0 : .map_err(ApiError::BadRequest)?;
1639 :
1640 : // This is just a sanity check to fend off naive wrong usages of the API:
1641 : // the tenant needs to be detached *everywhere*
1642 0 : let state = get_state(&request);
1643 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1644 0 : if we_manage_tenant {
1645 0 : return Err(ApiError::BadRequest(anyhow!(
1646 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1647 0 : )));
1648 0 : }
1649 0 :
1650 0 : if timestamp > done_if_after {
1651 0 : return Err(ApiError::BadRequest(anyhow!(
1652 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1653 0 : )));
1654 0 : }
1655 0 :
1656 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1657 :
1658 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1659 0 : &state.remote_storage,
1660 0 : &tenant_shard_id,
1661 0 : timestamp,
1662 0 : done_if_after,
1663 0 : &cancel,
1664 0 : )
1665 0 : .await
1666 0 : .map_err(|e| match e {
1667 0 : TimeTravelError::BadInput(e) => {
1668 0 : warn!("bad input error: {e}");
1669 0 : ApiError::BadRequest(anyhow!("bad input error"))
1670 : }
1671 : TimeTravelError::Unimplemented => {
1672 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1673 : }
1674 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1675 : TimeTravelError::TooManyVersions => {
1676 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1677 : }
1678 0 : TimeTravelError::Other(e) => {
1679 0 : warn!("internal error: {e}");
1680 0 : ApiError::InternalServerError(anyhow!("internal error"))
1681 : }
1682 0 : })?;
1683 :
1684 0 : json_response(StatusCode::OK, ())
1685 0 : }
1686 :
1687 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1688 0 : async fn handle_tenant_break(
1689 0 : r: Request<Body>,
1690 0 : _cancel: CancellationToken,
1691 0 : ) -> Result<Response<Body>, ApiError> {
1692 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1693 :
1694 0 : let state = get_state(&r);
1695 0 : state
1696 0 : .tenant_manager
1697 0 : .get_attached_tenant_shard(tenant_shard_id)?
1698 0 : .set_broken("broken from test".to_owned())
1699 0 : .await;
1700 :
1701 0 : json_response(StatusCode::OK, ())
1702 0 : }
1703 :
1704 : // Run GC immediately on given timeline.
1705 0 : async fn timeline_gc_handler(
1706 0 : mut request: Request<Body>,
1707 0 : cancel: CancellationToken,
1708 0 : ) -> Result<Response<Body>, ApiError> {
1709 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1710 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1711 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1712 :
1713 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1714 :
1715 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1716 0 : let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
1717 :
1718 0 : json_response(StatusCode::OK, gc_result)
1719 0 : }
1720 :
1721 : // Run compaction immediately on given timeline.
1722 0 : async fn timeline_compact_handler(
1723 0 : request: Request<Body>,
1724 0 : cancel: CancellationToken,
1725 0 : ) -> Result<Response<Body>, ApiError> {
1726 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1727 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1728 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1729 :
1730 0 : let state = get_state(&request);
1731 0 :
1732 0 : let mut flags = EnumSet::empty();
1733 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1734 0 : flags |= CompactFlags::ForceRepartition;
1735 0 : }
1736 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1737 0 : flags |= CompactFlags::ForceImageLayerCreation;
1738 0 : }
1739 0 : let wait_until_uploaded =
1740 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1741 :
1742 0 : async {
1743 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1744 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1745 0 : timeline
1746 0 : .compact(&cancel, flags, &ctx)
1747 0 : .await
1748 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1749 0 : if wait_until_uploaded {
1750 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1751 0 : }
1752 0 : json_response(StatusCode::OK, ())
1753 0 : }
1754 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1755 0 : .await
1756 0 : }
1757 :
1758 : // Run checkpoint immediately on given timeline.
1759 0 : async fn timeline_checkpoint_handler(
1760 0 : request: Request<Body>,
1761 0 : cancel: CancellationToken,
1762 0 : ) -> Result<Response<Body>, ApiError> {
1763 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1764 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1765 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1766 :
1767 0 : let state = get_state(&request);
1768 0 :
1769 0 : let mut flags = EnumSet::empty();
1770 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1771 0 : flags |= CompactFlags::ForceRepartition;
1772 0 : }
1773 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1774 0 : flags |= CompactFlags::ForceImageLayerCreation;
1775 0 : }
1776 0 : let wait_until_uploaded =
1777 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1778 :
1779 0 : async {
1780 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1781 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1782 0 : timeline
1783 0 : .freeze_and_flush()
1784 0 : .await
1785 0 : .map_err(ApiError::InternalServerError)?;
1786 0 : timeline
1787 0 : .compact(&cancel, flags, &ctx)
1788 0 : .await
1789 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1790 :
1791 0 : if wait_until_uploaded {
1792 0 : timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
1793 0 : }
1794 :
1795 0 : json_response(StatusCode::OK, ())
1796 0 : }
1797 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1798 0 : .await
1799 0 : }
1800 :
1801 0 : async fn timeline_download_remote_layers_handler_post(
1802 0 : mut request: Request<Body>,
1803 0 : _cancel: CancellationToken,
1804 0 : ) -> Result<Response<Body>, ApiError> {
1805 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1806 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1807 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
1808 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1809 :
1810 0 : let state = get_state(&request);
1811 :
1812 0 : let timeline =
1813 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1814 0 : .await?;
1815 0 : match timeline.spawn_download_all_remote_layers(body).await {
1816 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
1817 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
1818 : }
1819 0 : }
1820 :
1821 0 : async fn timeline_download_remote_layers_handler_get(
1822 0 : request: Request<Body>,
1823 0 : _cancel: CancellationToken,
1824 0 : ) -> Result<Response<Body>, ApiError> {
1825 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1826 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1827 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1828 0 : let state = get_state(&request);
1829 :
1830 0 : let timeline =
1831 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1832 0 : .await?;
1833 0 : let info = timeline
1834 0 : .get_download_all_remote_layers_task_info()
1835 0 : .context("task never started since last pageserver process start")
1836 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1837 0 : json_response(StatusCode::OK, info)
1838 0 : }
1839 :
1840 0 : async fn timeline_detach_ancestor_handler(
1841 0 : request: Request<Body>,
1842 0 : _cancel: CancellationToken,
1843 0 : ) -> Result<Response<Body>, ApiError> {
1844 : use crate::tenant::timeline::detach_ancestor::Options;
1845 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1846 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1847 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1848 :
1849 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
1850 :
1851 0 : async move {
1852 0 : let mut options = Options::default();
1853 :
1854 0 : let rewrite_concurrency =
1855 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
1856 0 : let copy_concurrency =
1857 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
1858 :
1859 0 : [
1860 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
1861 0 : (&mut options.copy_concurrency, copy_concurrency),
1862 0 : ]
1863 0 : .into_iter()
1864 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
1865 0 : .for_each(|(target, val)| *target = val);
1866 0 :
1867 0 : let state = get_state(&request);
1868 :
1869 0 : let tenant = state
1870 0 : .tenant_manager
1871 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1872 :
1873 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1874 :
1875 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
1876 0 : let ctx = &ctx;
1877 :
1878 0 : let timeline = tenant
1879 0 : .get_timeline(timeline_id, true)
1880 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1881 :
1882 0 : let (_guard, prepared) = timeline
1883 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
1884 0 : .await
1885 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1886 :
1887 0 : let res = state
1888 0 : .tenant_manager
1889 0 : .complete_detaching_timeline_ancestor(tenant_shard_id, timeline_id, prepared, ctx)
1890 0 : .await;
1891 :
1892 0 : match res {
1893 0 : Ok(reparented_timelines) => {
1894 0 : let resp = pageserver_api::models::detach_ancestor::AncestorDetached {
1895 0 : reparented_timelines,
1896 0 : };
1897 0 :
1898 0 : json_response(StatusCode::OK, resp)
1899 : }
1900 0 : Err(e) => Err(ApiError::InternalServerError(
1901 0 : e.context("timeline detach completion"),
1902 0 : )),
1903 : }
1904 0 : }
1905 0 : .instrument(span)
1906 0 : .await
1907 0 : }
1908 :
1909 0 : async fn deletion_queue_flush(
1910 0 : r: Request<Body>,
1911 0 : cancel: CancellationToken,
1912 0 : ) -> Result<Response<Body>, ApiError> {
1913 0 : let state = get_state(&r);
1914 :
1915 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
1916 0 :
1917 0 : let flush = async {
1918 0 : if execute {
1919 0 : state.deletion_queue_client.flush_execute().await
1920 : } else {
1921 0 : state.deletion_queue_client.flush().await
1922 : }
1923 0 : }
1924 : // DeletionQueueError's only case is shutting down.
1925 0 : .map_err(|_| ApiError::ShuttingDown);
1926 :
1927 : tokio::select! {
1928 : res = flush => {
1929 0 : res.map(|()| json_response(StatusCode::OK, ()))?
1930 : }
1931 : _ = cancel.cancelled() => {
1932 : Err(ApiError::ShuttingDown)
1933 : }
1934 : }
1935 0 : }
1936 :
1937 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
1938 0 : async fn getpage_at_lsn_handler(
1939 0 : request: Request<Body>,
1940 0 : _cancel: CancellationToken,
1941 0 : ) -> Result<Response<Body>, ApiError> {
1942 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1943 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1944 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1945 0 : let state = get_state(&request);
1946 :
1947 : struct Key(crate::repository::Key);
1948 :
1949 : impl std::str::FromStr for Key {
1950 : type Err = anyhow::Error;
1951 :
1952 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
1953 0 : crate::repository::Key::from_hex(s).map(Key)
1954 0 : }
1955 : }
1956 :
1957 0 : let key: Key = parse_query_param(&request, "key")?
1958 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
1959 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1960 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1961 :
1962 0 : async {
1963 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1964 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1965 :
1966 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
1967 :
1968 0 : Result::<_, ApiError>::Ok(
1969 0 : Response::builder()
1970 0 : .status(StatusCode::OK)
1971 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
1972 0 : .body(hyper::Body::from(page))
1973 0 : .unwrap(),
1974 0 : )
1975 0 : }
1976 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1977 0 : .await
1978 0 : }
1979 :
1980 0 : async fn timeline_collect_keyspace(
1981 0 : request: Request<Body>,
1982 0 : _cancel: CancellationToken,
1983 0 : ) -> Result<Response<Body>, ApiError> {
1984 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1985 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1986 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1987 0 : let state = get_state(&request);
1988 :
1989 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
1990 :
1991 0 : async {
1992 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1993 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1994 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
1995 0 : let (dense_ks, sparse_ks) = timeline
1996 0 : .collect_keyspace(at_lsn, &ctx)
1997 0 : .await
1998 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1999 :
2000 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2001 : // Therefore, we split dense/sparse keys in this API.
2002 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2003 0 :
2004 0 : json_response(StatusCode::OK, res)
2005 0 : }
2006 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2007 0 : .await
2008 0 : }
2009 :
2010 0 : async fn active_timeline_of_active_tenant(
2011 0 : tenant_manager: &TenantManager,
2012 0 : tenant_shard_id: TenantShardId,
2013 0 : timeline_id: TimelineId,
2014 0 : ) -> Result<Arc<Timeline>, ApiError> {
2015 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2016 :
2017 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2018 :
2019 0 : tenant
2020 0 : .get_timeline(timeline_id, true)
2021 0 : .map_err(|e| ApiError::NotFound(e.into()))
2022 0 : }
2023 :
2024 0 : async fn always_panic_handler(
2025 0 : req: Request<Body>,
2026 0 : _cancel: CancellationToken,
2027 0 : ) -> Result<Response<Body>, ApiError> {
2028 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2029 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2030 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2031 0 : let query = req.uri().query();
2032 0 : let _ = std::panic::catch_unwind(|| {
2033 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2034 0 : });
2035 0 : json_response(StatusCode::NO_CONTENT, ())
2036 0 : }
2037 :
2038 0 : async fn disk_usage_eviction_run(
2039 0 : mut r: Request<Body>,
2040 0 : cancel: CancellationToken,
2041 0 : ) -> Result<Response<Body>, ApiError> {
2042 0 : check_permission(&r, None)?;
2043 :
2044 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2045 : struct Config {
2046 : /// How many bytes to evict before reporting that pressure is relieved.
2047 : evict_bytes: u64,
2048 :
2049 : #[serde(default)]
2050 : eviction_order: crate::disk_usage_eviction_task::EvictionOrder,
2051 : }
2052 :
2053 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2054 : struct Usage {
2055 : // remains unchanged after instantiation of the struct
2056 : evict_bytes: u64,
2057 : // updated by `add_available_bytes`
2058 : freed_bytes: u64,
2059 : }
2060 :
2061 : impl crate::disk_usage_eviction_task::Usage for Usage {
2062 0 : fn has_pressure(&self) -> bool {
2063 0 : self.evict_bytes > self.freed_bytes
2064 0 : }
2065 :
2066 0 : fn add_available_bytes(&mut self, bytes: u64) {
2067 0 : self.freed_bytes += bytes;
2068 0 : }
2069 : }
2070 :
2071 0 : let config = json_request::<Config>(&mut r).await?;
2072 :
2073 0 : let usage = Usage {
2074 0 : evict_bytes: config.evict_bytes,
2075 0 : freed_bytes: 0,
2076 0 : };
2077 0 :
2078 0 : let state = get_state(&r);
2079 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2080 :
2081 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2082 0 : &eviction_state,
2083 0 : &state.remote_storage,
2084 0 : usage,
2085 0 : &state.tenant_manager,
2086 0 : config.eviction_order,
2087 0 : &cancel,
2088 0 : )
2089 0 : .await;
2090 :
2091 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2092 :
2093 0 : let res = res.map_err(ApiError::InternalServerError)?;
2094 :
2095 0 : json_response(StatusCode::OK, res)
2096 0 : }
2097 :
2098 0 : async fn secondary_upload_handler(
2099 0 : request: Request<Body>,
2100 0 : _cancel: CancellationToken,
2101 0 : ) -> Result<Response<Body>, ApiError> {
2102 0 : let state = get_state(&request);
2103 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2104 0 : state
2105 0 : .secondary_controller
2106 0 : .upload_tenant(tenant_shard_id)
2107 0 : .await
2108 0 : .map_err(ApiError::InternalServerError)?;
2109 :
2110 0 : json_response(StatusCode::OK, ())
2111 0 : }
2112 :
2113 0 : async fn tenant_scan_remote_handler(
2114 0 : request: Request<Body>,
2115 0 : cancel: CancellationToken,
2116 0 : ) -> Result<Response<Body>, ApiError> {
2117 0 : let state = get_state(&request);
2118 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2119 :
2120 0 : let mut response = TenantScanRemoteStorageResponse::default();
2121 :
2122 0 : let (shards, _other_keys) =
2123 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2124 0 : .await
2125 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2126 :
2127 0 : for tenant_shard_id in shards {
2128 0 : let (timeline_ids, _other_keys) =
2129 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2130 0 : .await
2131 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2132 :
2133 0 : let mut generation = Generation::none();
2134 0 : for timeline_id in timeline_ids {
2135 0 : match download_index_part(
2136 0 : &state.remote_storage,
2137 0 : &tenant_shard_id,
2138 0 : &timeline_id,
2139 0 : Generation::MAX,
2140 0 : &cancel,
2141 0 : )
2142 0 : .instrument(info_span!("download_index_part",
2143 : tenant_id=%tenant_shard_id.tenant_id,
2144 0 : shard_id=%tenant_shard_id.shard_slug(),
2145 : %timeline_id))
2146 0 : .await
2147 : {
2148 0 : Ok((index_part, index_generation)) => {
2149 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2150 0 : index_part.layer_metadata.len(), index_part.get_disk_consistent_lsn());
2151 0 : generation = std::cmp::max(generation, index_generation);
2152 : }
2153 : Err(DownloadError::NotFound) => {
2154 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2155 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2156 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2157 0 : continue;
2158 : }
2159 0 : Err(e) => {
2160 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2161 : }
2162 : };
2163 : }
2164 :
2165 0 : response.shards.push(TenantScanRemoteStorageShard {
2166 0 : tenant_shard_id,
2167 0 : generation: generation.into(),
2168 0 : });
2169 : }
2170 :
2171 0 : if response.shards.is_empty() {
2172 0 : return Err(ApiError::NotFound(
2173 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2174 0 : ));
2175 0 : }
2176 0 :
2177 0 : json_response(StatusCode::OK, response)
2178 0 : }
2179 :
2180 0 : async fn secondary_download_handler(
2181 0 : request: Request<Body>,
2182 0 : _cancel: CancellationToken,
2183 0 : ) -> Result<Response<Body>, ApiError> {
2184 0 : let state = get_state(&request);
2185 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2186 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2187 :
2188 : // We don't need this to issue the download request, but:
2189 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2190 : // - we will use this to provide status feedback in the response
2191 0 : let Some(secondary_tenant) = state
2192 0 : .tenant_manager
2193 0 : .get_secondary_tenant_shard(tenant_shard_id)
2194 : else {
2195 0 : return Err(ApiError::NotFound(
2196 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2197 0 : ));
2198 : };
2199 :
2200 0 : let timeout = wait.unwrap_or(Duration::MAX);
2201 :
2202 0 : let status = match tokio::time::timeout(
2203 0 : timeout,
2204 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2205 0 : )
2206 0 : .await
2207 : {
2208 : // Download job ran to completion.
2209 0 : Ok(Ok(())) => StatusCode::OK,
2210 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2211 : // okay. We could get an error here in the unlikely edge case that the tenant
2212 : // was detached between our check above and executing the download job.
2213 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2214 : // A timeout is not an error: we have started the download, we're just not done
2215 : // yet. The caller will get a response body indicating status.
2216 0 : Err(_) => StatusCode::ACCEPTED,
2217 : };
2218 :
2219 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2220 0 :
2221 0 : json_response(status, progress)
2222 0 : }
2223 :
2224 0 : async fn secondary_status_handler(
2225 0 : request: Request<Body>,
2226 0 : _cancel: CancellationToken,
2227 0 : ) -> Result<Response<Body>, ApiError> {
2228 0 : let state = get_state(&request);
2229 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2230 :
2231 0 : let Some(secondary_tenant) = state
2232 0 : .tenant_manager
2233 0 : .get_secondary_tenant_shard(tenant_shard_id)
2234 : else {
2235 0 : return Err(ApiError::NotFound(
2236 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2237 0 : ));
2238 : };
2239 :
2240 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2241 0 :
2242 0 : json_response(StatusCode::OK, progress)
2243 0 : }
2244 :
2245 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2246 0 : json_response(
2247 0 : StatusCode::NOT_FOUND,
2248 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2249 0 : )
2250 0 : }
2251 :
2252 0 : async fn post_tracing_event_handler(
2253 0 : mut r: Request<Body>,
2254 0 : _cancel: CancellationToken,
2255 0 : ) -> Result<Response<Body>, ApiError> {
2256 0 : #[derive(Debug, serde::Deserialize)]
2257 : #[serde(rename_all = "lowercase")]
2258 : enum Level {
2259 : Error,
2260 : Warn,
2261 : Info,
2262 : Debug,
2263 : Trace,
2264 : }
2265 0 : #[derive(Debug, serde::Deserialize)]
2266 : struct Request {
2267 : level: Level,
2268 : message: String,
2269 : }
2270 0 : let body: Request = json_request(&mut r)
2271 0 : .await
2272 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2273 :
2274 0 : match body.level {
2275 0 : Level::Error => tracing::error!(?body.message),
2276 0 : Level::Warn => tracing::warn!(?body.message),
2277 0 : Level::Info => tracing::info!(?body.message),
2278 0 : Level::Debug => tracing::debug!(?body.message),
2279 0 : Level::Trace => tracing::trace!(?body.message),
2280 : }
2281 :
2282 0 : json_response(StatusCode::OK, ())
2283 0 : }
2284 :
2285 0 : async fn put_io_engine_handler(
2286 0 : mut r: Request<Body>,
2287 0 : _cancel: CancellationToken,
2288 0 : ) -> Result<Response<Body>, ApiError> {
2289 0 : check_permission(&r, None)?;
2290 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2291 0 : crate::virtual_file::io_engine::set(kind);
2292 0 : json_response(StatusCode::OK, ())
2293 0 : }
2294 :
2295 : /// Polled by control plane.
2296 : ///
2297 : /// See [`crate::utilization`].
2298 0 : async fn get_utilization(
2299 0 : r: Request<Body>,
2300 0 : _cancel: CancellationToken,
2301 0 : ) -> Result<Response<Body>, ApiError> {
2302 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2303 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2304 0 : });
2305 :
2306 : // this probably could be completely public, but lets make that change later.
2307 0 : check_permission(&r, None)?;
2308 :
2309 0 : let state = get_state(&r);
2310 0 : let mut g = state.latest_utilization.lock().await;
2311 :
2312 0 : let regenerate_every = Duration::from_secs(1);
2313 0 : let still_valid = g
2314 0 : .as_ref()
2315 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2316 0 :
2317 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2318 0 : // regenerate at most 1Hz to allow polling at any rate.
2319 0 : if !still_valid {
2320 0 : let path = state.conf.tenants_path();
2321 0 : let doc = crate::utilization::regenerate(path.as_std_path())
2322 0 : .map_err(ApiError::InternalServerError)?;
2323 :
2324 0 : let mut buf = Vec::new();
2325 0 : serde_json::to_writer(&mut buf, &doc)
2326 0 : .context("serialize")
2327 0 : .map_err(ApiError::InternalServerError)?;
2328 :
2329 0 : let body = bytes::Bytes::from(buf);
2330 0 :
2331 0 : *g = Some((std::time::Instant::now(), body));
2332 0 : }
2333 :
2334 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2335 0 : let cached = g.as_ref().expect("just set").1.clone();
2336 0 :
2337 0 : Response::builder()
2338 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2339 0 : // thought of using http date header, but that is second precision which does not give any
2340 0 : // debugging aid
2341 0 : .status(StatusCode::OK)
2342 0 : .body(hyper::Body::from(cached))
2343 0 : .context("build response")
2344 0 : .map_err(ApiError::InternalServerError)
2345 0 : }
2346 :
2347 0 : async fn list_aux_files(
2348 0 : mut request: Request<Body>,
2349 0 : _cancel: CancellationToken,
2350 0 : ) -> Result<Response<Body>, ApiError> {
2351 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2352 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2353 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2354 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2355 :
2356 0 : let state = get_state(&request);
2357 :
2358 0 : let timeline =
2359 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2360 0 : .await?;
2361 :
2362 0 : let process = || async move {
2363 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2364 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2365 0 : Ok::<_, anyhow::Error>(files)
2366 0 : };
2367 :
2368 0 : match process().await {
2369 0 : Ok(st) => json_response(StatusCode::OK, st),
2370 0 : Err(err) => json_response(
2371 0 : StatusCode::INTERNAL_SERVER_ERROR,
2372 0 : ApiError::InternalServerError(err).to_string(),
2373 0 : ),
2374 : }
2375 0 : }
2376 :
2377 0 : async fn ingest_aux_files(
2378 0 : mut request: Request<Body>,
2379 0 : _cancel: CancellationToken,
2380 0 : ) -> Result<Response<Body>, ApiError> {
2381 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2382 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2383 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2384 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2385 :
2386 0 : let state = get_state(&request);
2387 :
2388 0 : let timeline =
2389 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2390 0 : .await?;
2391 :
2392 0 : let process = || async move {
2393 0 : let mut modification = timeline.begin_modification(Lsn(
2394 0 : timeline.get_last_record_lsn().0 + 8
2395 0 : ) /* advance LSN by 8 */);
2396 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2397 0 : for (fname, content) in body.aux_files {
2398 0 : modification
2399 0 : .put_file(&fname, content.as_bytes(), &ctx)
2400 0 : .await?;
2401 0 : }
2402 0 : modification.commit(&ctx).await?;
2403 0 : Ok::<_, anyhow::Error>(())
2404 0 : };
2405 :
2406 0 : match process().await {
2407 0 : Ok(st) => json_response(StatusCode::OK, st),
2408 0 : Err(err) => Err(ApiError::InternalServerError(err)),
2409 : }
2410 0 : }
2411 :
2412 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2413 : /// candidates for splitting
2414 0 : async fn post_top_tenants(
2415 0 : mut r: Request<Body>,
2416 0 : _cancel: CancellationToken,
2417 0 : ) -> Result<Response<Body>, ApiError> {
2418 0 : check_permission(&r, None)?;
2419 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2420 0 : let state = get_state(&r);
2421 0 :
2422 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2423 0 : match order_by {
2424 0 : TenantSorting::ResidentSize => sizes.resident_size,
2425 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2426 0 : }
2427 0 : }
2428 0 :
2429 0 : #[derive(Eq, PartialEq)]
2430 0 : struct HeapItem {
2431 0 : metric: u64,
2432 0 : sizes: TopTenantShardItem,
2433 0 : }
2434 0 :
2435 0 : impl PartialOrd for HeapItem {
2436 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2437 0 : Some(self.cmp(other))
2438 0 : }
2439 0 : }
2440 0 :
2441 0 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2442 0 : /// supports popping the greatest item but not the smallest.
2443 0 : impl Ord for HeapItem {
2444 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2445 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2446 0 : }
2447 0 : }
2448 0 :
2449 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2450 :
2451 : // FIXME: this is a lot of clones to take this tenant list
2452 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2453 0 : if let Some(shards_lt) = request.where_shards_lt {
2454 : // Ignore tenants which already have >= this many shards
2455 0 : if tenant_shard_id.shard_count >= shards_lt {
2456 0 : continue;
2457 0 : }
2458 0 : }
2459 :
2460 0 : let sizes = match tenant_slot {
2461 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2462 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2463 0 : continue;
2464 : }
2465 : };
2466 0 : let metric = get_size_metric(&sizes, &request.order_by);
2467 :
2468 0 : if let Some(gt) = request.where_gt {
2469 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2470 0 : if metric <= gt {
2471 0 : continue;
2472 0 : }
2473 0 : };
2474 :
2475 0 : match top_n.peek() {
2476 0 : None => {
2477 0 : // Top N list is empty: candidate becomes first member
2478 0 : top_n.push(HeapItem { metric, sizes });
2479 0 : }
2480 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
2481 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
2482 0 : top_n.push(HeapItem { metric, sizes });
2483 0 : }
2484 0 : Some(i) if i.metric > metric => {
2485 0 : // List is at limit and lowest value is greater than our candidate, drop it.
2486 0 : }
2487 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
2488 : }
2489 :
2490 0 : while top_n.len() > request.limit {
2491 0 : top_n.pop();
2492 0 : }
2493 : }
2494 :
2495 0 : json_response(
2496 0 : StatusCode::OK,
2497 0 : TopTenantShardsResponse {
2498 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
2499 0 : },
2500 0 : )
2501 0 : }
2502 :
2503 : /// Common functionality of all the HTTP API handlers.
2504 : ///
2505 : /// - Adds a tracing span to each request (by `request_span`)
2506 : /// - Logs the request depending on the request method (by `request_span`)
2507 : /// - Logs the response if it was not successful (by `request_span`
2508 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
2509 : /// Future if the connection to the client is lost, but most of the pageserver code is
2510 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
2511 : /// request with a CancellationToken.
2512 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
2513 0 : where
2514 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2515 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2516 0 : {
2517 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
2518 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
2519 0 : "failpoint".into()
2520 0 : )));
2521 :
2522 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
2523 0 : anyhow::anyhow!("failpoint")
2524 0 : )));
2525 0 : }
2526 :
2527 : // Spawn a new task to handle the request, to protect the handler from unexpected
2528 : // async cancellations. Most pageserver functions are not async cancellation safe.
2529 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
2530 : // with the cancellation token.
2531 0 : let token = CancellationToken::new();
2532 0 : let cancel_guard = token.clone().drop_guard();
2533 0 : let result = request_span(request, move |r| async {
2534 0 : let handle = tokio::spawn(
2535 0 : async {
2536 0 : let token_cloned = token.clone();
2537 0 : let result = handler(r, token).await;
2538 0 : if token_cloned.is_cancelled() {
2539 0 : // dropguard has executed: we will never turn this result into response.
2540 0 : //
2541 0 : // at least temporarily do {:?} logging; these failures are rare enough but
2542 0 : // could hide difficult errors.
2543 0 : match &result {
2544 0 : Ok(response) => {
2545 0 : let status = response.status();
2546 0 : info!(%status, "Cancelled request finished successfully")
2547 0 : }
2548 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
2549 0 : }
2550 0 : }
2551 0 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
2552 0 : // which should suffice.
2553 0 : //
2554 0 : // there is still a chance to lose the result due to race between
2555 0 : // returning from here and the actual connection closing happening
2556 0 : // before outer task gets to execute. leaving that up for #5815.
2557 0 : result
2558 0 : }
2559 0 : .in_current_span(),
2560 0 : );
2561 0 :
2562 0 : match handle.await {
2563 0 : // TODO: never actually return Err from here, always Ok(...) so that we can log
2564 0 : // spanned errors. Call api_error_handler instead and return appropriate Body.
2565 0 : Ok(result) => result,
2566 0 : Err(e) => {
2567 0 : // The handler task panicked. We have a global panic handler that logs the
2568 0 : // panic with its backtrace, so no need to log that here. Only log a brief
2569 0 : // message to make it clear that we returned the error to the client.
2570 0 : error!("HTTP request handler task panicked: {e:#}");
2571 0 :
2572 0 : // Don't return an Error here, because then fallback error handler that was
2573 0 : // installed in make_router() will print the error. Instead, construct the
2574 0 : // HTTP error response and return that.
2575 0 : Ok(
2576 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
2577 0 : .into_response(),
2578 0 : )
2579 0 : }
2580 0 : }
2581 0 : })
2582 0 : .await;
2583 :
2584 0 : cancel_guard.disarm();
2585 0 :
2586 0 : result
2587 0 : }
2588 :
2589 : /// Like api_handler, but returns an error response if the server is built without
2590 : /// the 'testing' feature.
2591 0 : async fn testing_api_handler<R, H>(
2592 0 : desc: &str,
2593 0 : request: Request<Body>,
2594 0 : handler: H,
2595 0 : ) -> Result<Response<Body>, ApiError>
2596 0 : where
2597 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2598 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2599 0 : {
2600 0 : if cfg!(feature = "testing") {
2601 0 : api_handler(request, handler).await
2602 : } else {
2603 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
2604 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
2605 0 : ))))
2606 0 : .await
2607 : }
2608 0 : }
2609 :
2610 0 : pub fn make_router(
2611 0 : state: Arc<State>,
2612 0 : launch_ts: &'static LaunchTimestamp,
2613 0 : auth: Option<Arc<SwappableJwtAuth>>,
2614 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
2615 0 : let spec = include_bytes!("openapi_spec.yml");
2616 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
2617 0 : if auth.is_some() {
2618 0 : router = router.middleware(auth_middleware(|request| {
2619 0 : let state = get_state(request);
2620 0 : if state.allowlist_routes.contains(request.uri()) {
2621 0 : None
2622 : } else {
2623 0 : state.auth.as_deref()
2624 : }
2625 0 : }))
2626 0 : }
2627 :
2628 0 : router = router.middleware(
2629 0 : endpoint::add_response_header_middleware(
2630 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
2631 0 : &launch_ts.to_string(),
2632 0 : )
2633 0 : .expect("construct launch timestamp header middleware"),
2634 0 : );
2635 0 :
2636 0 : Ok(router
2637 0 : .data(state)
2638 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
2639 0 : .get("/v1/status", |r| api_handler(r, status_handler))
2640 0 : .put("/v1/failpoints", |r| {
2641 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
2642 0 : })
2643 0 : .post("/v1/reload_auth_validation_keys", |r| {
2644 0 : api_handler(r, reload_auth_validation_keys_handler)
2645 0 : })
2646 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
2647 0 : .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
2648 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
2649 0 : api_handler(r, tenant_status)
2650 0 : })
2651 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
2652 0 : api_handler(r, tenant_delete_handler)
2653 0 : })
2654 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
2655 0 : api_handler(r, tenant_size_handler)
2656 0 : })
2657 0 : .put("/v1/tenant/config", |r| {
2658 0 : api_handler(r, update_tenant_config_handler)
2659 0 : })
2660 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
2661 0 : api_handler(r, tenant_shard_split_handler)
2662 0 : })
2663 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
2664 0 : api_handler(r, get_tenant_config_handler)
2665 0 : })
2666 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2667 0 : api_handler(r, put_tenant_location_config_handler)
2668 0 : })
2669 0 : .get("/v1/location_config", |r| {
2670 0 : api_handler(r, list_location_config_handler)
2671 0 : })
2672 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
2673 0 : api_handler(r, get_location_config_handler)
2674 0 : })
2675 0 : .put(
2676 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
2677 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
2678 0 : )
2679 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
2680 0 : api_handler(r, timeline_list_handler)
2681 0 : })
2682 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
2683 0 : api_handler(r, timeline_create_handler)
2684 0 : })
2685 0 : .post("/v1/tenant/:tenant_id/attach", |r| {
2686 0 : api_handler(r, tenant_attach_handler)
2687 0 : })
2688 0 : .post("/v1/tenant/:tenant_id/detach", |r| {
2689 0 : api_handler(r, tenant_detach_handler)
2690 0 : })
2691 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
2692 0 : api_handler(r, tenant_reset_handler)
2693 0 : })
2694 0 : .post("/v1/tenant/:tenant_id/load", |r| {
2695 0 : api_handler(r, tenant_load_handler)
2696 0 : })
2697 0 : .post("/v1/tenant/:tenant_id/ignore", |r| {
2698 0 : api_handler(r, tenant_ignore_handler)
2699 0 : })
2700 0 : .post(
2701 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
2702 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
2703 0 : )
2704 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2705 0 : api_handler(r, timeline_detail_handler)
2706 0 : })
2707 0 : .get(
2708 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
2709 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
2710 0 : )
2711 0 : .get(
2712 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
2713 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
2714 0 : )
2715 0 : .put(
2716 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
2717 0 : |r| api_handler(r, timeline_gc_handler),
2718 0 : )
2719 0 : .put(
2720 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
2721 0 : |r| testing_api_handler("run timeline compaction", r, timeline_compact_handler),
2722 0 : )
2723 0 : .put(
2724 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
2725 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
2726 0 : )
2727 0 : .post(
2728 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2729 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
2730 0 : )
2731 0 : .get(
2732 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2733 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
2734 0 : )
2735 0 : .put(
2736 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
2737 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
2738 0 : )
2739 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2740 0 : api_handler(r, timeline_delete_handler)
2741 0 : })
2742 0 : .get(
2743 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
2744 0 : |r| api_handler(r, layer_map_info_handler),
2745 0 : )
2746 0 : .get(
2747 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2748 0 : |r| api_handler(r, layer_download_handler),
2749 0 : )
2750 0 : .delete(
2751 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
2752 0 : |r| api_handler(r, evict_timeline_layer_handler),
2753 0 : )
2754 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
2755 0 : api_handler(r, secondary_upload_handler)
2756 0 : })
2757 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
2758 0 : api_handler(r, tenant_scan_remote_handler)
2759 0 : })
2760 0 : .put("/v1/disk_usage_eviction/run", |r| {
2761 0 : api_handler(r, disk_usage_eviction_run)
2762 0 : })
2763 0 : .put("/v1/deletion_queue/flush", |r| {
2764 0 : api_handler(r, deletion_queue_flush)
2765 0 : })
2766 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
2767 0 : api_handler(r, secondary_status_handler)
2768 0 : })
2769 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
2770 0 : api_handler(r, secondary_download_handler)
2771 0 : })
2772 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
2773 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
2774 0 : })
2775 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
2776 0 : .post("/v1/tracing/event", |r| {
2777 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
2778 0 : })
2779 0 : .get(
2780 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
2781 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
2782 0 : )
2783 0 : .get(
2784 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
2785 0 : |r| api_handler(r, timeline_collect_keyspace),
2786 0 : )
2787 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
2788 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
2789 0 : .post(
2790 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
2791 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
2792 0 : )
2793 0 : .post(
2794 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
2795 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
2796 0 : )
2797 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
2798 0 : .any(handler_404))
2799 0 : }
|
