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::StreamExt;
14 : use futures::TryFutureExt;
15 : use humantime::format_rfc3339;
16 : use hyper::header;
17 : use hyper::StatusCode;
18 : use hyper::{Body, Request, Response, Uri};
19 : use metrics::launch_timestamp::LaunchTimestamp;
20 : use pageserver_api::models::virtual_file::IoMode;
21 : use pageserver_api::models::DownloadRemoteLayersTaskSpawnRequest;
22 : use pageserver_api::models::IngestAuxFilesRequest;
23 : use pageserver_api::models::ListAuxFilesRequest;
24 : use pageserver_api::models::LocationConfig;
25 : use pageserver_api::models::LocationConfigListResponse;
26 : use pageserver_api::models::LocationConfigMode;
27 : use pageserver_api::models::LsnLease;
28 : use pageserver_api::models::LsnLeaseRequest;
29 : use pageserver_api::models::OffloadedTimelineInfo;
30 : use pageserver_api::models::ShardParameters;
31 : use pageserver_api::models::TenantDetails;
32 : use pageserver_api::models::TenantLocationConfigRequest;
33 : use pageserver_api::models::TenantLocationConfigResponse;
34 : use pageserver_api::models::TenantScanRemoteStorageResponse;
35 : use pageserver_api::models::TenantScanRemoteStorageShard;
36 : use pageserver_api::models::TenantShardLocation;
37 : use pageserver_api::models::TenantShardSplitRequest;
38 : use pageserver_api::models::TenantShardSplitResponse;
39 : use pageserver_api::models::TenantSorting;
40 : use pageserver_api::models::TenantState;
41 : use pageserver_api::models::TimelineArchivalConfigRequest;
42 : use pageserver_api::models::TimelineCreateRequestMode;
43 : use pageserver_api::models::TimelineCreateRequestModeImportPgdata;
44 : use pageserver_api::models::TimelinesInfoAndOffloaded;
45 : use pageserver_api::models::TopTenantShardItem;
46 : use pageserver_api::models::TopTenantShardsRequest;
47 : use pageserver_api::models::TopTenantShardsResponse;
48 : use pageserver_api::shard::ShardCount;
49 : use pageserver_api::shard::TenantShardId;
50 : use remote_storage::DownloadError;
51 : use remote_storage::GenericRemoteStorage;
52 : use remote_storage::TimeTravelError;
53 : use tenant_size_model::{svg::SvgBranchKind, SizeResult, StorageModel};
54 : use tokio_util::io::StreamReader;
55 : use tokio_util::sync::CancellationToken;
56 : use tracing::*;
57 : use utils::auth::JwtAuth;
58 : use utils::failpoint_support::failpoints_handler;
59 : use utils::http::endpoint::profile_cpu_handler;
60 : use utils::http::endpoint::prometheus_metrics_handler;
61 : use utils::http::endpoint::request_span;
62 : use utils::http::request::must_parse_query_param;
63 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
64 :
65 : use crate::config::PageServerConf;
66 : use crate::context::{DownloadBehavior, RequestContext};
67 : use crate::deletion_queue::DeletionQueueClient;
68 : use crate::pgdatadir_mapping::LsnForTimestamp;
69 : use crate::task_mgr::TaskKind;
70 : use crate::tenant::config::{LocationConf, TenantConfOpt};
71 : use crate::tenant::mgr::GetActiveTenantError;
72 : use crate::tenant::mgr::{
73 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
74 : TenantSlotUpsertError, TenantStateError,
75 : };
76 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
77 : use crate::tenant::remote_timeline_client;
78 : use crate::tenant::remote_timeline_client::download_index_part;
79 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
80 : use crate::tenant::remote_timeline_client::list_remote_timelines;
81 : use crate::tenant::secondary::SecondaryController;
82 : use crate::tenant::size::ModelInputs;
83 : use crate::tenant::storage_layer::LayerAccessStatsReset;
84 : use crate::tenant::storage_layer::LayerName;
85 : use crate::tenant::timeline::import_pgdata;
86 : use crate::tenant::timeline::offload::offload_timeline;
87 : use crate::tenant::timeline::offload::OffloadError;
88 : use crate::tenant::timeline::CompactFlags;
89 : use crate::tenant::timeline::CompactOptions;
90 : use crate::tenant::timeline::CompactRange;
91 : use crate::tenant::timeline::CompactionError;
92 : use crate::tenant::timeline::Timeline;
93 : use crate::tenant::GetTimelineError;
94 : use crate::tenant::OffloadedTimeline;
95 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
96 : use crate::DEFAULT_PG_VERSION;
97 : use crate::{disk_usage_eviction_task, tenant};
98 : use pageserver_api::models::{
99 : StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest, TimelineGcRequest,
100 : TimelineInfo,
101 : };
102 : use utils::{
103 : auth::SwappableJwtAuth,
104 : generation::Generation,
105 : http::{
106 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
107 : error::{ApiError, HttpErrorBody},
108 : json::{json_request, json_request_maybe, json_response},
109 : request::parse_request_param,
110 : RequestExt, RouterBuilder,
111 : },
112 : id::{TenantId, TimelineId},
113 : lsn::Lsn,
114 : };
115 :
116 : // For APIs that require an Active tenant, how long should we block waiting for that state?
117 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
118 : // failed API calls while tenants are activating.
119 : #[cfg(not(feature = "testing"))]
120 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
121 :
122 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
123 : // finish attaching, if calls to remote storage are slow.
124 : #[cfg(feature = "testing")]
125 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
126 :
127 : pub struct State {
128 : conf: &'static PageServerConf,
129 : tenant_manager: Arc<TenantManager>,
130 : auth: Option<Arc<SwappableJwtAuth>>,
131 : allowlist_routes: &'static [&'static str],
132 : remote_storage: GenericRemoteStorage,
133 : broker_client: storage_broker::BrokerClientChannel,
134 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
135 : deletion_queue_client: DeletionQueueClient,
136 : secondary_controller: SecondaryController,
137 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
138 : }
139 :
140 : impl State {
141 : #[allow(clippy::too_many_arguments)]
142 0 : pub fn new(
143 0 : conf: &'static PageServerConf,
144 0 : tenant_manager: Arc<TenantManager>,
145 0 : auth: Option<Arc<SwappableJwtAuth>>,
146 0 : remote_storage: GenericRemoteStorage,
147 0 : broker_client: storage_broker::BrokerClientChannel,
148 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
149 0 : deletion_queue_client: DeletionQueueClient,
150 0 : secondary_controller: SecondaryController,
151 0 : ) -> anyhow::Result<Self> {
152 0 : let allowlist_routes = &[
153 0 : "/v1/status",
154 0 : "/v1/doc",
155 0 : "/swagger.yml",
156 0 : "/metrics",
157 0 : "/profile/cpu",
158 0 : ];
159 0 : Ok(Self {
160 0 : conf,
161 0 : tenant_manager,
162 0 : auth,
163 0 : allowlist_routes,
164 0 : remote_storage,
165 0 : broker_client,
166 0 : disk_usage_eviction_state,
167 0 : deletion_queue_client,
168 0 : secondary_controller,
169 0 : latest_utilization: Default::default(),
170 0 : })
171 0 : }
172 : }
173 :
174 : #[inline(always)]
175 0 : fn get_state(request: &Request<Body>) -> &State {
176 0 : request
177 0 : .data::<Arc<State>>()
178 0 : .expect("unknown state type")
179 0 : .as_ref()
180 0 : }
181 :
182 : #[inline(always)]
183 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
184 0 : get_state(request).conf
185 0 : }
186 :
187 : /// Check that the requester is authorized to operate on given tenant
188 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
189 0 : check_permission_with(request, |claims| {
190 0 : crate::auth::check_permission(claims, tenant_id)
191 0 : })
192 0 : }
193 :
194 : impl From<PageReconstructError> for ApiError {
195 0 : fn from(pre: PageReconstructError) -> ApiError {
196 0 : match pre {
197 0 : PageReconstructError::Other(other) => ApiError::InternalServerError(other),
198 0 : PageReconstructError::MissingKey(e) => ApiError::InternalServerError(e.into()),
199 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
200 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
201 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
202 : }
203 0 : }
204 : }
205 :
206 : impl From<TenantMapInsertError> for ApiError {
207 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
208 0 : match tmie {
209 0 : TenantMapInsertError::SlotError(e) => e.into(),
210 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
211 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
212 : }
213 0 : }
214 : }
215 :
216 : impl From<TenantSlotError> for ApiError {
217 0 : fn from(e: TenantSlotError) -> ApiError {
218 : use TenantSlotError::*;
219 0 : match e {
220 0 : NotFound(tenant_id) => {
221 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
222 : }
223 : InProgress => {
224 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
225 : }
226 0 : MapState(e) => e.into(),
227 : }
228 0 : }
229 : }
230 :
231 : impl From<TenantSlotUpsertError> for ApiError {
232 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
233 : use TenantSlotUpsertError::*;
234 0 : match e {
235 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
236 0 : MapState(e) => e.into(),
237 0 : ShuttingDown(_) => ApiError::ShuttingDown,
238 : }
239 0 : }
240 : }
241 :
242 : impl From<UpsertLocationError> for ApiError {
243 0 : fn from(e: UpsertLocationError) -> ApiError {
244 : use UpsertLocationError::*;
245 0 : match e {
246 0 : BadRequest(e) => ApiError::BadRequest(e),
247 0 : Unavailable(_) => ApiError::ShuttingDown,
248 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
249 0 : Flush(e) | InternalError(e) => ApiError::InternalServerError(e),
250 : }
251 0 : }
252 : }
253 :
254 : impl From<TenantMapError> for ApiError {
255 0 : fn from(e: TenantMapError) -> ApiError {
256 : use TenantMapError::*;
257 0 : match e {
258 : StillInitializing | ShuttingDown => {
259 0 : ApiError::ResourceUnavailable(format!("{e}").into())
260 0 : }
261 0 : }
262 0 : }
263 : }
264 :
265 : impl From<TenantStateError> for ApiError {
266 0 : fn from(tse: TenantStateError) -> ApiError {
267 0 : match tse {
268 : TenantStateError::IsStopping(_) => {
269 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
270 : }
271 0 : TenantStateError::SlotError(e) => e.into(),
272 0 : TenantStateError::SlotUpsertError(e) => e.into(),
273 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
274 : }
275 0 : }
276 : }
277 :
278 : impl From<GetTenantError> for ApiError {
279 0 : fn from(tse: GetTenantError) -> ApiError {
280 0 : match tse {
281 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
282 : GetTenantError::NotActive(_) => {
283 : // Why is this not `ApiError::NotFound`?
284 : // Because we must be careful to never return 404 for a tenant if it does
285 : // in fact exist locally. If we did, the caller could draw the conclusion
286 : // that it can attach the tenant to another PS and we'd be in split-brain.
287 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
288 : }
289 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
290 : }
291 0 : }
292 : }
293 :
294 : impl From<GetTimelineError> for ApiError {
295 0 : fn from(gte: GetTimelineError) -> Self {
296 0 : // Rationale: tenant is activated only after eligble timelines activate
297 0 : ApiError::NotFound(gte.into())
298 0 : }
299 : }
300 :
301 : impl From<GetActiveTenantError> for ApiError {
302 0 : fn from(e: GetActiveTenantError) -> ApiError {
303 0 : match e {
304 0 : GetActiveTenantError::Broken(reason) => {
305 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
306 : }
307 : GetActiveTenantError::WillNotBecomeActive(TenantState::Stopping { .. }) => {
308 0 : ApiError::ShuttingDown
309 : }
310 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
311 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
312 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
313 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
314 0 : ApiError::ResourceUnavailable(format!("{}", e).into())
315 : }
316 : GetActiveTenantError::SwitchedTenant => {
317 : // in our HTTP handlers, this error doesn't happen
318 : // TODO: separate error types
319 0 : ApiError::ResourceUnavailable("switched tenant".into())
320 : }
321 : }
322 0 : }
323 : }
324 :
325 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
326 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
327 : use crate::tenant::DeleteTimelineError::*;
328 0 : match value {
329 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
330 0 : HasChildren(children) => ApiError::PreconditionFailed(
331 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
332 0 : .into_boxed_str(),
333 0 : ),
334 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
335 0 : Cancelled => ApiError::ResourceUnavailable("shutting down".into()),
336 0 : Other(e) => ApiError::InternalServerError(e),
337 : }
338 0 : }
339 : }
340 :
341 : impl From<crate::tenant::TimelineArchivalError> for ApiError {
342 0 : fn from(value: crate::tenant::TimelineArchivalError) -> Self {
343 : use crate::tenant::TimelineArchivalError::*;
344 0 : match value {
345 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
346 0 : Timeout => ApiError::Timeout("hit pageserver internal timeout".into()),
347 0 : Cancelled => ApiError::ShuttingDown,
348 0 : e @ HasArchivedParent(_) => {
349 0 : ApiError::PreconditionFailed(e.to_string().into_boxed_str())
350 : }
351 0 : HasUnarchivedChildren(children) => ApiError::PreconditionFailed(
352 0 : format!(
353 0 : "Cannot archive timeline which has non-archived child timelines: {children:?}"
354 0 : )
355 0 : .into_boxed_str(),
356 0 : ),
357 0 : a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
358 0 : Other(e) => ApiError::InternalServerError(e),
359 : }
360 0 : }
361 : }
362 :
363 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
364 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
365 : use crate::tenant::mgr::DeleteTimelineError::*;
366 0 : match value {
367 : // Report Precondition failed so client can distinguish between
368 : // "tenant is missing" case from "timeline is missing"
369 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
370 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
371 0 : ),
372 0 : Tenant(t) => ApiError::from(t),
373 0 : Timeline(t) => ApiError::from(t),
374 : }
375 0 : }
376 : }
377 :
378 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
379 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
380 : use crate::tenant::mgr::DeleteTenantError::*;
381 0 : match value {
382 0 : SlotError(e) => e.into(),
383 0 : Other(o) => ApiError::InternalServerError(o),
384 0 : Cancelled => ApiError::ShuttingDown,
385 : }
386 0 : }
387 : }
388 :
389 : // Helper function to construct a TimelineInfo struct for a timeline
390 0 : async fn build_timeline_info(
391 0 : timeline: &Arc<Timeline>,
392 0 : include_non_incremental_logical_size: bool,
393 0 : force_await_initial_logical_size: bool,
394 0 : ctx: &RequestContext,
395 0 : ) -> anyhow::Result<TimelineInfo> {
396 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
397 0 :
398 0 : if force_await_initial_logical_size {
399 0 : timeline.clone().await_initial_logical_size().await
400 0 : }
401 :
402 0 : let mut info = build_timeline_info_common(
403 0 : timeline,
404 0 : ctx,
405 0 : tenant::timeline::GetLogicalSizePriority::Background,
406 0 : )
407 0 : .await?;
408 0 : if include_non_incremental_logical_size {
409 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
410 : // Otherwise, if someone deletes the timeline / detaches the tenant while
411 : // we're executing this function, we will outlive the timeline on-disk state.
412 : info.current_logical_size_non_incremental = Some(
413 0 : timeline
414 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
415 0 : .await?,
416 : );
417 0 : }
418 0 : Ok(info)
419 0 : }
420 :
421 0 : async fn build_timeline_info_common(
422 0 : timeline: &Arc<Timeline>,
423 0 : ctx: &RequestContext,
424 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
425 0 : ) -> anyhow::Result<TimelineInfo> {
426 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
427 0 : let initdb_lsn = timeline.initdb_lsn;
428 0 : let last_record_lsn = timeline.get_last_record_lsn();
429 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
430 0 : let guard = timeline.last_received_wal.lock().unwrap();
431 0 : if let Some(info) = guard.as_ref() {
432 0 : (
433 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
434 0 : Some(info.last_received_msg_lsn),
435 0 : Some(info.last_received_msg_ts),
436 0 : )
437 : } else {
438 0 : (None, None, None)
439 : }
440 : };
441 :
442 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
443 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
444 0 : Lsn(0) => None,
445 0 : lsn @ Lsn(_) => Some(lsn),
446 : };
447 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
448 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
449 0 : let state = timeline.current_state();
450 0 : // Report is_archived = false if the timeline is still loading
451 0 : let is_archived = timeline.is_archived().unwrap_or(false);
452 0 : let remote_consistent_lsn_projected = timeline
453 0 : .get_remote_consistent_lsn_projected()
454 0 : .unwrap_or(Lsn(0));
455 0 : let remote_consistent_lsn_visible = timeline
456 0 : .get_remote_consistent_lsn_visible()
457 0 : .unwrap_or(Lsn(0));
458 0 :
459 0 : let walreceiver_status = timeline.walreceiver_status();
460 0 :
461 0 : let (pitr_history_size, within_ancestor_pitr) = timeline.get_pitr_history_stats();
462 :
463 0 : let info = TimelineInfo {
464 0 : tenant_id: timeline.tenant_shard_id,
465 0 : timeline_id: timeline.timeline_id,
466 0 : ancestor_timeline_id,
467 0 : ancestor_lsn,
468 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
469 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
470 0 : remote_consistent_lsn_visible,
471 0 : initdb_lsn,
472 0 : last_record_lsn,
473 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
474 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
475 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
476 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
477 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
478 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
479 : },
480 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
481 0 : current_physical_size,
482 0 : current_logical_size_non_incremental: None,
483 0 : pitr_history_size,
484 0 : within_ancestor_pitr,
485 0 : timeline_dir_layer_file_size_sum: None,
486 0 : wal_source_connstr,
487 0 : last_received_msg_lsn,
488 0 : last_received_msg_ts,
489 0 : pg_version: timeline.pg_version,
490 0 :
491 0 : state,
492 0 : is_archived: Some(is_archived),
493 0 :
494 0 : walreceiver_status,
495 0 : };
496 0 : Ok(info)
497 0 : }
498 :
499 0 : fn build_timeline_offloaded_info(offloaded: &Arc<OffloadedTimeline>) -> OffloadedTimelineInfo {
500 0 : let &OffloadedTimeline {
501 0 : tenant_shard_id,
502 0 : timeline_id,
503 0 : ancestor_retain_lsn,
504 0 : ancestor_timeline_id,
505 0 : archived_at,
506 0 : ..
507 0 : } = offloaded.as_ref();
508 0 : OffloadedTimelineInfo {
509 0 : tenant_id: tenant_shard_id,
510 0 : timeline_id,
511 0 : ancestor_retain_lsn,
512 0 : ancestor_timeline_id,
513 0 : archived_at: archived_at.and_utc(),
514 0 : }
515 0 : }
516 :
517 : // healthcheck handler
518 0 : async fn status_handler(
519 0 : request: Request<Body>,
520 0 : _cancel: CancellationToken,
521 0 : ) -> Result<Response<Body>, ApiError> {
522 0 : check_permission(&request, None)?;
523 0 : let config = get_config(&request);
524 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
525 0 : }
526 :
527 0 : async fn reload_auth_validation_keys_handler(
528 0 : request: Request<Body>,
529 0 : _cancel: CancellationToken,
530 0 : ) -> Result<Response<Body>, ApiError> {
531 0 : check_permission(&request, None)?;
532 0 : let config = get_config(&request);
533 0 : let state = get_state(&request);
534 0 : let Some(shared_auth) = &state.auth else {
535 0 : return json_response(StatusCode::BAD_REQUEST, ());
536 : };
537 : // unwrap is ok because check is performed when creating config, so path is set and exists
538 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
539 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
540 :
541 0 : match JwtAuth::from_key_path(key_path) {
542 0 : Ok(new_auth) => {
543 0 : shared_auth.swap(new_auth);
544 0 : json_response(StatusCode::OK, ())
545 : }
546 0 : Err(e) => {
547 0 : let err_msg = "Error reloading public keys";
548 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
549 0 : json_response(
550 0 : StatusCode::INTERNAL_SERVER_ERROR,
551 0 : HttpErrorBody::from_msg(err_msg.to_string()),
552 0 : )
553 : }
554 : }
555 0 : }
556 :
557 0 : async fn timeline_create_handler(
558 0 : mut request: Request<Body>,
559 0 : _cancel: CancellationToken,
560 0 : ) -> Result<Response<Body>, ApiError> {
561 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
562 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
563 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
564 :
565 0 : let new_timeline_id = request_data.new_timeline_id;
566 : // fill in the default pg_version if not provided & convert request into domain model
567 0 : let params: tenant::CreateTimelineParams = match request_data.mode {
568 : TimelineCreateRequestMode::Bootstrap {
569 0 : existing_initdb_timeline_id,
570 0 : pg_version,
571 0 : } => tenant::CreateTimelineParams::Bootstrap(tenant::CreateTimelineParamsBootstrap {
572 0 : new_timeline_id,
573 0 : existing_initdb_timeline_id,
574 0 : pg_version: pg_version.unwrap_or(DEFAULT_PG_VERSION),
575 0 : }),
576 : TimelineCreateRequestMode::Branch {
577 0 : ancestor_timeline_id,
578 0 : ancestor_start_lsn,
579 0 : pg_version: _,
580 0 : } => tenant::CreateTimelineParams::Branch(tenant::CreateTimelineParamsBranch {
581 0 : new_timeline_id,
582 0 : ancestor_timeline_id,
583 0 : ancestor_start_lsn,
584 0 : }),
585 : TimelineCreateRequestMode::ImportPgdata {
586 : import_pgdata:
587 : TimelineCreateRequestModeImportPgdata {
588 0 : location,
589 0 : idempotency_key,
590 0 : },
591 0 : } => tenant::CreateTimelineParams::ImportPgdata(tenant::CreateTimelineParamsImportPgdata {
592 0 : idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new(
593 0 : idempotency_key.0,
594 0 : ),
595 0 : new_timeline_id,
596 : location: {
597 0 : use import_pgdata::index_part_format::Location;
598 0 : use pageserver_api::models::ImportPgdataLocation;
599 0 : match location {
600 : #[cfg(feature = "testing")]
601 0 : ImportPgdataLocation::LocalFs { path } => Location::LocalFs { path },
602 : ImportPgdataLocation::AwsS3 {
603 0 : region,
604 0 : bucket,
605 0 : key,
606 0 : } => Location::AwsS3 {
607 0 : region,
608 0 : bucket,
609 0 : key,
610 0 : },
611 : }
612 : },
613 : }),
614 : };
615 :
616 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
617 0 :
618 0 : let state = get_state(&request);
619 :
620 0 : async {
621 0 : let tenant = state
622 0 : .tenant_manager
623 0 : .get_attached_tenant_shard(tenant_shard_id)?;
624 :
625 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
626 :
627 : // earlier versions of the code had pg_version and ancestor_lsn in the span
628 : // => continue to provide that information, but, through a log message that doesn't require us to destructure
629 0 : tracing::info!(?params, "creating timeline");
630 :
631 0 : match tenant
632 0 : .create_timeline(params, state.broker_client.clone(), &ctx)
633 0 : .await
634 : {
635 0 : Ok(new_timeline) => {
636 : // Created. Construct a TimelineInfo for it.
637 0 : let timeline_info = build_timeline_info_common(
638 0 : &new_timeline,
639 0 : &ctx,
640 0 : tenant::timeline::GetLogicalSizePriority::User,
641 0 : )
642 0 : .await
643 0 : .map_err(ApiError::InternalServerError)?;
644 0 : json_response(StatusCode::CREATED, timeline_info)
645 : }
646 0 : Err(_) if tenant.cancel.is_cancelled() => {
647 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
648 0 : json_response(
649 0 : StatusCode::SERVICE_UNAVAILABLE,
650 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
651 0 : )
652 : }
653 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
654 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
655 : }
656 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
657 0 : StatusCode::TOO_MANY_REQUESTS,
658 0 : HttpErrorBody::from_msg(e.to_string()),
659 0 : ),
660 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
661 0 : StatusCode::NOT_ACCEPTABLE,
662 0 : HttpErrorBody::from_msg(format!("{err:#}")),
663 0 : ),
664 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
665 0 : StatusCode::SERVICE_UNAVAILABLE,
666 0 : HttpErrorBody::from_msg(e.to_string()),
667 0 : ),
668 0 : Err(e @ tenant::CreateTimelineError::AncestorArchived) => json_response(
669 0 : StatusCode::NOT_ACCEPTABLE,
670 0 : HttpErrorBody::from_msg(e.to_string()),
671 0 : ),
672 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
673 0 : StatusCode::SERVICE_UNAVAILABLE,
674 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
675 0 : ),
676 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
677 : }
678 0 : }
679 0 : .instrument(info_span!("timeline_create",
680 : tenant_id = %tenant_shard_id.tenant_id,
681 0 : shard_id = %tenant_shard_id.shard_slug(),
682 : timeline_id = %new_timeline_id,
683 : ))
684 0 : .await
685 0 : }
686 :
687 0 : async fn timeline_list_handler(
688 0 : request: Request<Body>,
689 0 : _cancel: CancellationToken,
690 0 : ) -> Result<Response<Body>, ApiError> {
691 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
692 0 : let include_non_incremental_logical_size: Option<bool> =
693 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
694 0 : let force_await_initial_logical_size: Option<bool> =
695 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
696 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
697 :
698 0 : let state = get_state(&request);
699 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
700 :
701 0 : let response_data = async {
702 0 : let tenant = state
703 0 : .tenant_manager
704 0 : .get_attached_tenant_shard(tenant_shard_id)?;
705 :
706 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
707 :
708 0 : let timelines = tenant.list_timelines();
709 0 :
710 0 : let mut response_data = Vec::with_capacity(timelines.len());
711 0 : for timeline in timelines {
712 0 : let timeline_info = build_timeline_info(
713 0 : &timeline,
714 0 : include_non_incremental_logical_size.unwrap_or(false),
715 0 : force_await_initial_logical_size.unwrap_or(false),
716 0 : &ctx,
717 0 : )
718 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
719 0 : .await
720 0 : .context("Failed to build timeline info")
721 0 : .map_err(ApiError::InternalServerError)?;
722 :
723 0 : response_data.push(timeline_info);
724 : }
725 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
726 0 : }
727 0 : .instrument(info_span!("timeline_list",
728 : tenant_id = %tenant_shard_id.tenant_id,
729 0 : shard_id = %tenant_shard_id.shard_slug()))
730 0 : .await?;
731 :
732 0 : json_response(StatusCode::OK, response_data)
733 0 : }
734 :
735 0 : async fn timeline_and_offloaded_list_handler(
736 0 : request: Request<Body>,
737 0 : _cancel: CancellationToken,
738 0 : ) -> Result<Response<Body>, ApiError> {
739 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
740 0 : let include_non_incremental_logical_size: Option<bool> =
741 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
742 0 : let force_await_initial_logical_size: Option<bool> =
743 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
744 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
745 :
746 0 : let state = get_state(&request);
747 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
748 :
749 0 : let response_data = async {
750 0 : let tenant = state
751 0 : .tenant_manager
752 0 : .get_attached_tenant_shard(tenant_shard_id)?;
753 :
754 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
755 :
756 0 : let (timelines, offloadeds) = tenant.list_timelines_and_offloaded();
757 0 :
758 0 : let mut timeline_infos = Vec::with_capacity(timelines.len());
759 0 : for timeline in timelines {
760 0 : let timeline_info = build_timeline_info(
761 0 : &timeline,
762 0 : include_non_incremental_logical_size.unwrap_or(false),
763 0 : force_await_initial_logical_size.unwrap_or(false),
764 0 : &ctx,
765 0 : )
766 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
767 0 : .await
768 0 : .context("Failed to build timeline info")
769 0 : .map_err(ApiError::InternalServerError)?;
770 :
771 0 : timeline_infos.push(timeline_info);
772 : }
773 0 : let offloaded_infos = offloadeds
774 0 : .into_iter()
775 0 : .map(|offloaded| build_timeline_offloaded_info(&offloaded))
776 0 : .collect::<Vec<_>>();
777 0 : let res = TimelinesInfoAndOffloaded {
778 0 : timelines: timeline_infos,
779 0 : offloaded: offloaded_infos,
780 0 : };
781 0 : Ok::<TimelinesInfoAndOffloaded, ApiError>(res)
782 0 : }
783 0 : .instrument(info_span!("timeline_and_offloaded_list",
784 : tenant_id = %tenant_shard_id.tenant_id,
785 0 : shard_id = %tenant_shard_id.shard_slug()))
786 0 : .await?;
787 :
788 0 : json_response(StatusCode::OK, response_data)
789 0 : }
790 :
791 0 : async fn timeline_preserve_initdb_handler(
792 0 : request: Request<Body>,
793 0 : _cancel: CancellationToken,
794 0 : ) -> Result<Response<Body>, ApiError> {
795 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
796 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
797 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
798 0 : let state = get_state(&request);
799 :
800 : // Part of the process for disaster recovery from safekeeper-stored WAL:
801 : // If we don't recover into a new timeline but want to keep the timeline ID,
802 : // then the initdb archive is deleted. This endpoint copies it to a different
803 : // location where timeline recreation cand find it.
804 :
805 0 : async {
806 0 : let tenant = state
807 0 : .tenant_manager
808 0 : .get_attached_tenant_shard(tenant_shard_id)?;
809 :
810 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
811 :
812 0 : timeline
813 0 : .preserve_initdb_archive()
814 0 : .await
815 0 : .context("preserving initdb archive")
816 0 : .map_err(ApiError::InternalServerError)?;
817 :
818 0 : Ok::<_, ApiError>(())
819 0 : }
820 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
821 : tenant_id = %tenant_shard_id.tenant_id,
822 0 : shard_id = %tenant_shard_id.shard_slug(),
823 : %timeline_id))
824 0 : .await?;
825 :
826 0 : json_response(StatusCode::OK, ())
827 0 : }
828 :
829 0 : async fn timeline_archival_config_handler(
830 0 : mut request: Request<Body>,
831 0 : _cancel: CancellationToken,
832 0 : ) -> Result<Response<Body>, ApiError> {
833 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
834 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
835 :
836 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
837 :
838 0 : let request_data: TimelineArchivalConfigRequest = json_request(&mut request).await?;
839 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
840 0 : let state = get_state(&request);
841 :
842 0 : async {
843 0 : let tenant = state
844 0 : .tenant_manager
845 0 : .get_attached_tenant_shard(tenant_shard_id)?;
846 :
847 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
848 :
849 0 : tenant
850 0 : .apply_timeline_archival_config(
851 0 : timeline_id,
852 0 : request_data.state,
853 0 : state.broker_client.clone(),
854 0 : ctx,
855 0 : )
856 0 : .await?;
857 0 : Ok::<_, ApiError>(())
858 0 : }
859 0 : .instrument(info_span!("timeline_archival_config",
860 : tenant_id = %tenant_shard_id.tenant_id,
861 0 : shard_id = %tenant_shard_id.shard_slug(),
862 : state = ?request_data.state,
863 : %timeline_id))
864 0 : .await?;
865 :
866 0 : json_response(StatusCode::OK, ())
867 0 : }
868 :
869 0 : async fn timeline_detail_handler(
870 0 : request: Request<Body>,
871 0 : _cancel: CancellationToken,
872 0 : ) -> Result<Response<Body>, ApiError> {
873 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
874 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
875 0 : let include_non_incremental_logical_size: Option<bool> =
876 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
877 0 : let force_await_initial_logical_size: Option<bool> =
878 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
879 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
880 :
881 : // Logical size calculation needs downloading.
882 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
883 0 : let state = get_state(&request);
884 :
885 0 : let timeline_info = async {
886 0 : let tenant = state
887 0 : .tenant_manager
888 0 : .get_attached_tenant_shard(tenant_shard_id)?;
889 :
890 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
891 :
892 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
893 :
894 0 : let timeline_info = build_timeline_info(
895 0 : &timeline,
896 0 : include_non_incremental_logical_size.unwrap_or(false),
897 0 : force_await_initial_logical_size.unwrap_or(false),
898 0 : &ctx,
899 0 : )
900 0 : .await
901 0 : .context("get local timeline info")
902 0 : .map_err(ApiError::InternalServerError)?;
903 :
904 0 : Ok::<_, ApiError>(timeline_info)
905 0 : }
906 0 : .instrument(info_span!("timeline_detail",
907 : tenant_id = %tenant_shard_id.tenant_id,
908 0 : shard_id = %tenant_shard_id.shard_slug(),
909 : %timeline_id))
910 0 : .await?;
911 :
912 0 : json_response(StatusCode::OK, timeline_info)
913 0 : }
914 :
915 0 : async fn get_lsn_by_timestamp_handler(
916 0 : request: Request<Body>,
917 0 : cancel: CancellationToken,
918 0 : ) -> Result<Response<Body>, ApiError> {
919 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
920 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
921 0 : let state = get_state(&request);
922 0 :
923 0 : if !tenant_shard_id.is_shard_zero() {
924 : // Requires SLRU contents, which are only stored on shard zero
925 0 : return Err(ApiError::BadRequest(anyhow!(
926 0 : "Size calculations are only available on shard zero"
927 0 : )));
928 0 : }
929 :
930 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
931 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
932 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
933 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
934 0 : .map_err(ApiError::BadRequest)?;
935 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
936 :
937 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
938 0 :
939 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
940 :
941 0 : let timeline =
942 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
943 0 : .await?;
944 0 : let result = timeline
945 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
946 0 : .await?;
947 :
948 : #[derive(serde::Serialize, Debug)]
949 : struct Result {
950 : lsn: Lsn,
951 : kind: &'static str,
952 : #[serde(default)]
953 : #[serde(skip_serializing_if = "Option::is_none")]
954 : #[serde(flatten)]
955 : lease: Option<LsnLease>,
956 : }
957 0 : let (lsn, kind) = match result {
958 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
959 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
960 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
961 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
962 : };
963 :
964 0 : let lease = if with_lease {
965 0 : timeline
966 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
967 0 : .inspect_err(|_| {
968 0 : warn!("fail to grant a lease to {}", lsn);
969 0 : })
970 0 : .ok()
971 : } else {
972 0 : None
973 : };
974 :
975 0 : let result = Result { lsn, kind, lease };
976 0 : let valid_until = result
977 0 : .lease
978 0 : .as_ref()
979 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
980 0 : tracing::info!(
981 : lsn=?result.lsn,
982 : kind=%result.kind,
983 : timestamp=%timestamp_raw,
984 : valid_until=?valid_until,
985 0 : "lsn_by_timestamp finished"
986 : );
987 0 : json_response(StatusCode::OK, result)
988 0 : }
989 :
990 0 : async fn get_timestamp_of_lsn_handler(
991 0 : request: Request<Body>,
992 0 : _cancel: CancellationToken,
993 0 : ) -> Result<Response<Body>, ApiError> {
994 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
995 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
996 0 : let state = get_state(&request);
997 0 :
998 0 : if !tenant_shard_id.is_shard_zero() {
999 : // Requires SLRU contents, which are only stored on shard zero
1000 0 : return Err(ApiError::BadRequest(anyhow!(
1001 0 : "Size calculations are only available on shard zero"
1002 0 : )));
1003 0 : }
1004 :
1005 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1006 :
1007 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
1008 0 : let lsn = Lsn::from_str(&lsn_str)
1009 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
1010 0 : .map_err(ApiError::BadRequest)?;
1011 :
1012 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1013 0 : let timeline =
1014 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1015 0 : .await?;
1016 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
1017 :
1018 0 : match result {
1019 0 : Some(time) => {
1020 0 : let time = format_rfc3339(
1021 0 : postgres_ffi::try_from_pg_timestamp(time).map_err(ApiError::InternalServerError)?,
1022 : )
1023 0 : .to_string();
1024 0 : json_response(StatusCode::OK, time)
1025 : }
1026 0 : None => Err(ApiError::NotFound(
1027 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
1028 0 : )),
1029 : }
1030 0 : }
1031 :
1032 0 : async fn timeline_delete_handler(
1033 0 : request: Request<Body>,
1034 0 : _cancel: CancellationToken,
1035 0 : ) -> Result<Response<Body>, ApiError> {
1036 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1037 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1038 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1039 :
1040 0 : let state = get_state(&request);
1041 :
1042 0 : let tenant = state
1043 0 : .tenant_manager
1044 0 : .get_attached_tenant_shard(tenant_shard_id)
1045 0 : .map_err(|e| {
1046 0 : match e {
1047 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
1048 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
1049 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
1050 0 : "Requested tenant is missing".to_string().into_boxed_str(),
1051 0 : ),
1052 0 : e => e.into(),
1053 : }
1054 0 : })?;
1055 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1056 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))
1057 0 : .await?;
1058 :
1059 0 : json_response(StatusCode::ACCEPTED, ())
1060 0 : }
1061 :
1062 0 : async fn tenant_reset_handler(
1063 0 : request: Request<Body>,
1064 0 : _cancel: CancellationToken,
1065 0 : ) -> Result<Response<Body>, ApiError> {
1066 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1067 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1068 :
1069 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
1070 :
1071 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1072 0 : let state = get_state(&request);
1073 0 : state
1074 0 : .tenant_manager
1075 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
1076 0 : .await
1077 0 : .map_err(ApiError::InternalServerError)?;
1078 :
1079 0 : json_response(StatusCode::OK, ())
1080 0 : }
1081 :
1082 0 : async fn tenant_list_handler(
1083 0 : request: Request<Body>,
1084 0 : _cancel: CancellationToken,
1085 0 : ) -> Result<Response<Body>, ApiError> {
1086 0 : check_permission(&request, None)?;
1087 0 : let state = get_state(&request);
1088 :
1089 0 : let response_data = state
1090 0 : .tenant_manager
1091 0 : .list_tenants()
1092 0 : .map_err(|_| {
1093 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
1094 0 : })?
1095 0 : .iter()
1096 0 : .map(|(id, state, gen)| TenantInfo {
1097 0 : id: *id,
1098 0 : state: state.clone(),
1099 0 : current_physical_size: None,
1100 0 : attachment_status: state.attachment_status(),
1101 0 : generation: (*gen)
1102 0 : .into()
1103 0 : .expect("Tenants are always attached with a generation"),
1104 0 : gc_blocking: None,
1105 0 : })
1106 0 : .collect::<Vec<TenantInfo>>();
1107 0 :
1108 0 : json_response(StatusCode::OK, response_data)
1109 0 : }
1110 :
1111 0 : async fn tenant_status(
1112 0 : request: Request<Body>,
1113 0 : _cancel: CancellationToken,
1114 0 : ) -> Result<Response<Body>, ApiError> {
1115 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1116 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1117 0 : let state = get_state(&request);
1118 0 :
1119 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1120 0 : let activate = true;
1121 : #[cfg(feature = "testing")]
1122 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1123 :
1124 0 : let tenant_info = async {
1125 0 : let tenant = state
1126 0 : .tenant_manager
1127 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1128 :
1129 0 : if activate {
1130 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1131 : // called, but it is still valid to return 200 and describe the current state of the tenant
1132 : // if it doesn't make it into an active state.
1133 0 : tenant
1134 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1135 0 : .await
1136 0 : .ok();
1137 0 : }
1138 :
1139 : // Calculate total physical size of all timelines
1140 0 : let mut current_physical_size = 0;
1141 0 : for timeline in tenant.list_timelines().iter() {
1142 0 : current_physical_size += timeline.layer_size_sum().await;
1143 : }
1144 :
1145 0 : let state = tenant.current_state();
1146 0 : Result::<_, ApiError>::Ok(TenantDetails {
1147 0 : tenant_info: TenantInfo {
1148 0 : id: tenant_shard_id,
1149 0 : state: state.clone(),
1150 0 : current_physical_size: Some(current_physical_size),
1151 0 : attachment_status: state.attachment_status(),
1152 0 : generation: tenant
1153 0 : .generation()
1154 0 : .into()
1155 0 : .expect("Tenants are always attached with a generation"),
1156 0 : gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
1157 0 : },
1158 0 : walredo: tenant.wal_redo_manager_status(),
1159 0 : timelines: tenant.list_timeline_ids(),
1160 0 : })
1161 0 : }
1162 0 : .instrument(info_span!("tenant_status_handler",
1163 : tenant_id = %tenant_shard_id.tenant_id,
1164 0 : shard_id = %tenant_shard_id.shard_slug()))
1165 0 : .await?;
1166 :
1167 0 : json_response(StatusCode::OK, tenant_info)
1168 0 : }
1169 :
1170 0 : async fn tenant_delete_handler(
1171 0 : request: Request<Body>,
1172 0 : _cancel: CancellationToken,
1173 0 : ) -> Result<Response<Body>, ApiError> {
1174 : // TODO openapi spec
1175 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1176 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1177 :
1178 0 : let state = get_state(&request);
1179 0 :
1180 0 : state
1181 0 : .tenant_manager
1182 0 : .delete_tenant(tenant_shard_id)
1183 0 : .instrument(info_span!("tenant_delete_handler",
1184 : tenant_id = %tenant_shard_id.tenant_id,
1185 0 : shard_id = %tenant_shard_id.shard_slug()
1186 : ))
1187 0 : .await?;
1188 :
1189 0 : json_response(StatusCode::OK, ())
1190 0 : }
1191 :
1192 : /// HTTP endpoint to query the current tenant_size of a tenant.
1193 : ///
1194 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1195 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1196 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1197 : /// values.
1198 : ///
1199 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1200 : /// (only if it is shorter than the real cutoff).
1201 : ///
1202 : /// Note: we don't update the cached size and prometheus metric here.
1203 : /// The retention period might be different, and it's nice to have a method to just calculate it
1204 : /// without modifying anything anyway.
1205 0 : async fn tenant_size_handler(
1206 0 : request: Request<Body>,
1207 0 : cancel: CancellationToken,
1208 0 : ) -> Result<Response<Body>, ApiError> {
1209 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1210 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1211 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1212 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1213 0 : let headers = request.headers();
1214 0 : let state = get_state(&request);
1215 0 :
1216 0 : if !tenant_shard_id.is_shard_zero() {
1217 0 : return Err(ApiError::BadRequest(anyhow!(
1218 0 : "Size calculations are only available on shard zero"
1219 0 : )));
1220 0 : }
1221 0 :
1222 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1223 0 : let tenant = state
1224 0 : .tenant_manager
1225 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1226 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1227 :
1228 : // this can be long operation
1229 0 : let inputs = tenant
1230 0 : .gather_size_inputs(
1231 0 : retention_period,
1232 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1233 0 : &cancel,
1234 0 : &ctx,
1235 0 : )
1236 0 : .await
1237 0 : .map_err(|e| match e {
1238 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1239 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1240 0 : })?;
1241 :
1242 0 : let mut sizes = None;
1243 0 : let accepts_html = headers
1244 0 : .get(header::ACCEPT)
1245 0 : .map(|v| v == "text/html")
1246 0 : .unwrap_or_default();
1247 0 : if !inputs_only.unwrap_or(false) {
1248 0 : let storage_model = inputs.calculate_model();
1249 0 : let size = storage_model.calculate();
1250 0 :
1251 0 : // If request header expects html, return html
1252 0 : if accepts_html {
1253 0 : return synthetic_size_html_response(inputs, storage_model, size);
1254 0 : }
1255 0 : sizes = Some(size);
1256 0 : } else if accepts_html {
1257 0 : return Err(ApiError::BadRequest(anyhow!(
1258 0 : "inputs_only parameter is incompatible with html output request"
1259 0 : )));
1260 0 : }
1261 :
1262 : /// The type resides in the pageserver not to expose `ModelInputs`.
1263 : #[derive(serde::Serialize)]
1264 : struct TenantHistorySize {
1265 : id: TenantId,
1266 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1267 : ///
1268 : /// Will be none if `?inputs_only=true` was given.
1269 : size: Option<u64>,
1270 : /// Size of each segment used in the model.
1271 : /// Will be null if `?inputs_only=true` was given.
1272 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1273 : inputs: crate::tenant::size::ModelInputs,
1274 : }
1275 :
1276 0 : json_response(
1277 0 : StatusCode::OK,
1278 0 : TenantHistorySize {
1279 0 : id: tenant_shard_id.tenant_id,
1280 0 : size: sizes.as_ref().map(|x| x.total_size),
1281 0 : segment_sizes: sizes.map(|x| x.segments),
1282 0 : inputs,
1283 0 : },
1284 0 : )
1285 0 : }
1286 :
1287 0 : async fn tenant_shard_split_handler(
1288 0 : mut request: Request<Body>,
1289 0 : _cancel: CancellationToken,
1290 0 : ) -> Result<Response<Body>, ApiError> {
1291 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1292 :
1293 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1294 0 : let state = get_state(&request);
1295 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1296 :
1297 0 : let tenant = state
1298 0 : .tenant_manager
1299 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1300 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1301 :
1302 0 : let new_shards = state
1303 0 : .tenant_manager
1304 0 : .shard_split(
1305 0 : tenant,
1306 0 : ShardCount::new(req.new_shard_count),
1307 0 : req.new_stripe_size,
1308 0 : &ctx,
1309 0 : )
1310 0 : .await
1311 0 : .map_err(ApiError::InternalServerError)?;
1312 :
1313 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1314 0 : }
1315 :
1316 0 : async fn layer_map_info_handler(
1317 0 : request: Request<Body>,
1318 0 : _cancel: CancellationToken,
1319 0 : ) -> Result<Response<Body>, ApiError> {
1320 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1321 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1322 0 : let reset: LayerAccessStatsReset =
1323 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1324 0 : let state = get_state(&request);
1325 0 :
1326 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1327 :
1328 0 : let timeline =
1329 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1330 0 : .await?;
1331 0 : let layer_map_info = timeline
1332 0 : .layer_map_info(reset)
1333 0 : .await
1334 0 : .map_err(|_shutdown| ApiError::ShuttingDown)?;
1335 :
1336 0 : json_response(StatusCode::OK, layer_map_info)
1337 0 : }
1338 :
1339 0 : #[instrument(skip_all, fields(tenant_id, shard_id, timeline_id, layer_name))]
1340 : async fn timeline_layer_scan_disposable_keys(
1341 : request: Request<Body>,
1342 : cancel: CancellationToken,
1343 : ) -> Result<Response<Body>, ApiError> {
1344 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1345 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1346 : let layer_name: LayerName = parse_request_param(&request, "layer_name")?;
1347 :
1348 : tracing::Span::current().record(
1349 : "tenant_id",
1350 : tracing::field::display(&tenant_shard_id.tenant_id),
1351 : );
1352 : tracing::Span::current().record(
1353 : "shard_id",
1354 : tracing::field::display(tenant_shard_id.shard_slug()),
1355 : );
1356 : tracing::Span::current().record("timeline_id", tracing::field::display(&timeline_id));
1357 : tracing::Span::current().record("layer_name", tracing::field::display(&layer_name));
1358 :
1359 : let state = get_state(&request);
1360 :
1361 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1362 :
1363 : // technically the timeline need not be active for this scan to complete
1364 : let timeline =
1365 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1366 : .await?;
1367 :
1368 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1369 :
1370 : let guard = timeline.layers.read().await;
1371 : let Some(layer) = guard.try_get_from_key(&layer_name.clone().into()) else {
1372 : return Err(ApiError::NotFound(
1373 : anyhow::anyhow!("Layer {tenant_shard_id}/{timeline_id}/{layer_name} not found").into(),
1374 : ));
1375 : };
1376 :
1377 : let resident_layer = layer
1378 : .download_and_keep_resident()
1379 : .await
1380 0 : .map_err(|err| match err {
1381 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1382 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1383 0 : ApiError::ShuttingDown
1384 : }
1385 : tenant::storage_layer::layer::DownloadError::ContextAndConfigReallyDeniesDownloads
1386 : | tenant::storage_layer::layer::DownloadError::DownloadRequired
1387 : | tenant::storage_layer::layer::DownloadError::NotFile(_)
1388 : | tenant::storage_layer::layer::DownloadError::DownloadFailed
1389 : | tenant::storage_layer::layer::DownloadError::PreStatFailed(_) => {
1390 0 : ApiError::InternalServerError(err.into())
1391 : }
1392 : #[cfg(test)]
1393 : tenant::storage_layer::layer::DownloadError::Failpoint(_) => {
1394 0 : ApiError::InternalServerError(err.into())
1395 : }
1396 0 : })?;
1397 :
1398 : let keys = resident_layer
1399 : .load_keys(&ctx)
1400 : .await
1401 : .map_err(ApiError::InternalServerError)?;
1402 :
1403 : let shard_identity = timeline.get_shard_identity();
1404 :
1405 : let mut disposable_count = 0;
1406 : let mut not_disposable_count = 0;
1407 : let cancel = cancel.clone();
1408 : for (i, key) in keys.into_iter().enumerate() {
1409 : if shard_identity.is_key_disposable(&key) {
1410 : disposable_count += 1;
1411 : tracing::debug!(key = %key, key.dbg=?key, "disposable key");
1412 : } else {
1413 : not_disposable_count += 1;
1414 : }
1415 : #[allow(clippy::collapsible_if)]
1416 : if i % 10000 == 0 {
1417 : if cancel.is_cancelled() || timeline.cancel.is_cancelled() || timeline.is_stopping() {
1418 : return Err(ApiError::ShuttingDown);
1419 : }
1420 : }
1421 : }
1422 :
1423 : json_response(
1424 : StatusCode::OK,
1425 : pageserver_api::models::ScanDisposableKeysResponse {
1426 : disposable_count,
1427 : not_disposable_count,
1428 : },
1429 : )
1430 : }
1431 :
1432 0 : async fn layer_download_handler(
1433 0 : request: Request<Body>,
1434 0 : _cancel: CancellationToken,
1435 0 : ) -> Result<Response<Body>, ApiError> {
1436 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1437 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1438 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1439 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1440 0 : let layer_name = LayerName::from_str(layer_file_name)
1441 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1442 0 : let state = get_state(&request);
1443 :
1444 0 : let timeline =
1445 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1446 0 : .await?;
1447 0 : let downloaded = timeline
1448 0 : .download_layer(&layer_name)
1449 0 : .await
1450 0 : .map_err(ApiError::InternalServerError)?;
1451 :
1452 0 : match downloaded {
1453 0 : Some(true) => json_response(StatusCode::OK, ()),
1454 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1455 0 : None => json_response(
1456 0 : StatusCode::BAD_REQUEST,
1457 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1458 0 : ),
1459 : }
1460 0 : }
1461 :
1462 0 : async fn evict_timeline_layer_handler(
1463 0 : request: Request<Body>,
1464 0 : _cancel: CancellationToken,
1465 0 : ) -> Result<Response<Body>, ApiError> {
1466 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1467 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1468 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1469 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1470 0 : let state = get_state(&request);
1471 :
1472 0 : let layer_name = LayerName::from_str(layer_file_name)
1473 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1474 :
1475 0 : let timeline =
1476 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1477 0 : .await?;
1478 0 : let evicted = timeline
1479 0 : .evict_layer(&layer_name)
1480 0 : .await
1481 0 : .map_err(ApiError::InternalServerError)?;
1482 :
1483 0 : match evicted {
1484 0 : Some(true) => json_response(StatusCode::OK, ()),
1485 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1486 0 : None => json_response(
1487 0 : StatusCode::BAD_REQUEST,
1488 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1489 0 : ),
1490 : }
1491 0 : }
1492 :
1493 0 : async fn timeline_gc_blocking_handler(
1494 0 : request: Request<Body>,
1495 0 : _cancel: CancellationToken,
1496 0 : ) -> Result<Response<Body>, ApiError> {
1497 0 : block_or_unblock_gc(request, true).await
1498 0 : }
1499 :
1500 0 : async fn timeline_gc_unblocking_handler(
1501 0 : request: Request<Body>,
1502 0 : _cancel: CancellationToken,
1503 0 : ) -> Result<Response<Body>, ApiError> {
1504 0 : block_or_unblock_gc(request, false).await
1505 0 : }
1506 :
1507 : /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
1508 : ///
1509 : /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
1510 0 : async fn block_or_unblock_gc(
1511 0 : request: Request<Body>,
1512 0 : block: bool,
1513 0 : ) -> Result<Response<Body>, ApiError> {
1514 : use crate::tenant::{
1515 : remote_timeline_client::WaitCompletionError, upload_queue::NotInitialized,
1516 : };
1517 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1518 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1519 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1520 0 : let state = get_state(&request);
1521 :
1522 0 : let tenant = state
1523 0 : .tenant_manager
1524 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1525 :
1526 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1527 :
1528 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1529 :
1530 0 : let fut = async {
1531 0 : if block {
1532 0 : timeline.block_gc(&tenant).await.map(|_| ())
1533 : } else {
1534 0 : timeline.unblock_gc(&tenant).await
1535 : }
1536 0 : };
1537 :
1538 0 : let span = tracing::info_span!(
1539 : "block_or_unblock_gc",
1540 : tenant_id = %tenant_shard_id.tenant_id,
1541 0 : shard_id = %tenant_shard_id.shard_slug(),
1542 : timeline_id = %timeline_id,
1543 : block = block,
1544 : );
1545 :
1546 0 : let res = fut.instrument(span).await;
1547 :
1548 0 : res.map_err(|e| {
1549 0 : if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
1550 0 : ApiError::ShuttingDown
1551 : } else {
1552 0 : ApiError::InternalServerError(e)
1553 : }
1554 0 : })?;
1555 :
1556 0 : json_response(StatusCode::OK, ())
1557 0 : }
1558 :
1559 : /// Get tenant_size SVG graph along with the JSON data.
1560 0 : fn synthetic_size_html_response(
1561 0 : inputs: ModelInputs,
1562 0 : storage_model: StorageModel,
1563 0 : sizes: SizeResult,
1564 0 : ) -> Result<Response<Body>, ApiError> {
1565 0 : let mut timeline_ids: Vec<String> = Vec::new();
1566 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1567 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1568 0 : timeline_map.insert(ti.timeline_id, index);
1569 0 : timeline_ids.push(ti.timeline_id.to_string());
1570 0 : }
1571 0 : let seg_to_branch: Vec<(usize, SvgBranchKind)> = inputs
1572 0 : .segments
1573 0 : .iter()
1574 0 : .map(|seg| {
1575 0 : (
1576 0 : *timeline_map.get(&seg.timeline_id).unwrap(),
1577 0 : seg.kind.into(),
1578 0 : )
1579 0 : })
1580 0 : .collect();
1581 :
1582 0 : let svg =
1583 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1584 0 : .map_err(ApiError::InternalServerError)?;
1585 :
1586 0 : let mut response = String::new();
1587 :
1588 : use std::fmt::Write;
1589 0 : write!(response, "<html>\n<body>\n").unwrap();
1590 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1591 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1592 0 : writeln!(response, "<pre>").unwrap();
1593 0 : writeln!(
1594 0 : response,
1595 0 : "{}",
1596 0 : serde_json::to_string_pretty(&inputs).unwrap()
1597 0 : )
1598 0 : .unwrap();
1599 0 : writeln!(
1600 0 : response,
1601 0 : "{}",
1602 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1603 0 : )
1604 0 : .unwrap();
1605 0 : writeln!(response, "</pre>").unwrap();
1606 0 : write!(response, "</body>\n</html>\n").unwrap();
1607 0 :
1608 0 : html_response(StatusCode::OK, response)
1609 0 : }
1610 :
1611 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1612 0 : let response = Response::builder()
1613 0 : .status(status)
1614 0 : .header(header::CONTENT_TYPE, "text/html")
1615 0 : .body(Body::from(data.as_bytes().to_vec()))
1616 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1617 0 : Ok(response)
1618 0 : }
1619 :
1620 0 : async fn get_tenant_config_handler(
1621 0 : request: Request<Body>,
1622 0 : _cancel: CancellationToken,
1623 0 : ) -> Result<Response<Body>, ApiError> {
1624 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1625 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1626 0 : let state = get_state(&request);
1627 :
1628 0 : let tenant = state
1629 0 : .tenant_manager
1630 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1631 :
1632 0 : let response = HashMap::from([
1633 : (
1634 : "tenant_specific_overrides",
1635 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1636 0 : .context("serializing tenant specific overrides")
1637 0 : .map_err(ApiError::InternalServerError)?,
1638 : ),
1639 : (
1640 0 : "effective_config",
1641 0 : serde_json::to_value(tenant.effective_config())
1642 0 : .context("serializing effective config")
1643 0 : .map_err(ApiError::InternalServerError)?,
1644 : ),
1645 : ]);
1646 :
1647 0 : json_response(StatusCode::OK, response)
1648 0 : }
1649 :
1650 0 : async fn update_tenant_config_handler(
1651 0 : mut request: Request<Body>,
1652 0 : _cancel: CancellationToken,
1653 0 : ) -> Result<Response<Body>, ApiError> {
1654 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1655 0 : let tenant_id = request_data.tenant_id;
1656 0 : check_permission(&request, Some(tenant_id))?;
1657 :
1658 0 : let new_tenant_conf =
1659 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1660 :
1661 0 : let state = get_state(&request);
1662 0 :
1663 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1664 :
1665 0 : let tenant = state
1666 0 : .tenant_manager
1667 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1668 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1669 :
1670 : // This is a legacy API that only operates on attached tenants: the preferred
1671 : // API to use is the location_config/ endpoint, which lets the caller provide
1672 : // the full LocationConf.
1673 0 : let location_conf = LocationConf::attached_single(
1674 0 : new_tenant_conf.clone(),
1675 0 : tenant.get_generation(),
1676 0 : &ShardParameters::default(),
1677 0 : );
1678 0 :
1679 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1680 0 : .await
1681 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1682 0 : tenant.set_new_tenant_config(new_tenant_conf);
1683 0 :
1684 0 : json_response(StatusCode::OK, ())
1685 0 : }
1686 :
1687 0 : async fn put_tenant_location_config_handler(
1688 0 : mut request: Request<Body>,
1689 0 : _cancel: CancellationToken,
1690 0 : ) -> Result<Response<Body>, ApiError> {
1691 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1692 :
1693 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1694 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1695 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1696 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1697 :
1698 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1699 0 : let state = get_state(&request);
1700 0 : let conf = state.conf;
1701 0 :
1702 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1703 0 : // its local disk content and drops it from memory.
1704 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1705 0 : if let Err(e) = state
1706 0 : .tenant_manager
1707 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
1708 0 : .instrument(info_span!("tenant_detach",
1709 : tenant_id = %tenant_shard_id.tenant_id,
1710 0 : shard_id = %tenant_shard_id.shard_slug()
1711 : ))
1712 0 : .await
1713 : {
1714 0 : match e {
1715 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1716 0 : // This API is idempotent: a NotFound on a detach is fine.
1717 0 : }
1718 0 : _ => return Err(e.into()),
1719 : }
1720 0 : }
1721 0 : return json_response(StatusCode::OK, ());
1722 0 : }
1723 :
1724 0 : let location_conf =
1725 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1726 :
1727 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1728 : // similar to at startup ordering.
1729 0 : let spawn_mode = if lazy {
1730 0 : tenant::SpawnMode::Lazy
1731 : } else {
1732 0 : tenant::SpawnMode::Eager
1733 : };
1734 :
1735 0 : let tenant = state
1736 0 : .tenant_manager
1737 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1738 0 : .await?;
1739 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1740 0 : let attached = tenant.is_some();
1741 :
1742 0 : if let Some(_flush_ms) = flush {
1743 0 : match state
1744 0 : .secondary_controller
1745 0 : .upload_tenant(tenant_shard_id)
1746 0 : .await
1747 : {
1748 : Ok(()) => {
1749 0 : tracing::info!("Uploaded heatmap during flush");
1750 : }
1751 0 : Err(e) => {
1752 0 : tracing::warn!("Failed to flush heatmap: {e}");
1753 : }
1754 : }
1755 : } else {
1756 0 : tracing::info!("No flush requested when configuring");
1757 : }
1758 :
1759 : // This API returns a vector of pageservers where the tenant is attached: this is
1760 : // primarily for use in the sharding service. For compatibilty, we also return this
1761 : // when called directly on a pageserver, but the payload is always zero or one shards.
1762 0 : let mut response = TenantLocationConfigResponse {
1763 0 : shards: Vec::new(),
1764 0 : stripe_size: None,
1765 0 : };
1766 0 : if attached {
1767 0 : response.shards.push(TenantShardLocation {
1768 0 : shard_id: tenant_shard_id,
1769 0 : node_id: state.conf.id,
1770 0 : });
1771 0 : if tenant_shard_id.shard_count.count() > 1 {
1772 : // Stripe size should be set if we are attached
1773 0 : debug_assert!(stripe_size.is_some());
1774 0 : response.stripe_size = stripe_size;
1775 0 : }
1776 0 : }
1777 :
1778 0 : json_response(StatusCode::OK, response)
1779 0 : }
1780 :
1781 0 : async fn list_location_config_handler(
1782 0 : request: Request<Body>,
1783 0 : _cancel: CancellationToken,
1784 0 : ) -> Result<Response<Body>, ApiError> {
1785 0 : let state = get_state(&request);
1786 0 : let slots = state.tenant_manager.list();
1787 0 : let result = LocationConfigListResponse {
1788 0 : tenant_shards: slots
1789 0 : .into_iter()
1790 0 : .map(|(tenant_shard_id, slot)| {
1791 0 : let v = match slot {
1792 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1793 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1794 0 : TenantSlot::InProgress(_) => None,
1795 : };
1796 0 : (tenant_shard_id, v)
1797 0 : })
1798 0 : .collect(),
1799 0 : };
1800 0 : json_response(StatusCode::OK, result)
1801 0 : }
1802 :
1803 0 : async fn get_location_config_handler(
1804 0 : request: Request<Body>,
1805 0 : _cancel: CancellationToken,
1806 0 : ) -> Result<Response<Body>, ApiError> {
1807 0 : let state = get_state(&request);
1808 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1809 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1810 :
1811 0 : let Some(slot) = slot else {
1812 0 : return Err(ApiError::NotFound(
1813 0 : anyhow::anyhow!("Tenant shard not found").into(),
1814 0 : ));
1815 : };
1816 :
1817 0 : let result: Option<LocationConfig> = match slot {
1818 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1819 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1820 0 : TenantSlot::InProgress(_) => None,
1821 : };
1822 :
1823 0 : json_response(StatusCode::OK, result)
1824 0 : }
1825 :
1826 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1827 : // (from all pageservers) as it invalidates consistency assumptions.
1828 0 : async fn tenant_time_travel_remote_storage_handler(
1829 0 : request: Request<Body>,
1830 0 : cancel: CancellationToken,
1831 0 : ) -> Result<Response<Body>, ApiError> {
1832 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1833 :
1834 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1835 :
1836 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1837 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1838 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1839 0 : .map_err(ApiError::BadRequest)?;
1840 :
1841 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1842 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1843 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1844 0 : .map_err(ApiError::BadRequest)?;
1845 :
1846 : // This is just a sanity check to fend off naive wrong usages of the API:
1847 : // the tenant needs to be detached *everywhere*
1848 0 : let state = get_state(&request);
1849 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1850 0 : if we_manage_tenant {
1851 0 : return Err(ApiError::BadRequest(anyhow!(
1852 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1853 0 : )));
1854 0 : }
1855 0 :
1856 0 : if timestamp > done_if_after {
1857 0 : return Err(ApiError::BadRequest(anyhow!(
1858 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1859 0 : )));
1860 0 : }
1861 0 :
1862 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1863 :
1864 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1865 0 : &state.remote_storage,
1866 0 : &tenant_shard_id,
1867 0 : timestamp,
1868 0 : done_if_after,
1869 0 : &cancel,
1870 0 : )
1871 0 : .await
1872 0 : .map_err(|e| match e {
1873 0 : TimeTravelError::BadInput(e) => {
1874 0 : warn!("bad input error: {e}");
1875 0 : ApiError::BadRequest(anyhow!("bad input error"))
1876 : }
1877 : TimeTravelError::Unimplemented => {
1878 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1879 : }
1880 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1881 : TimeTravelError::TooManyVersions => {
1882 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1883 : }
1884 0 : TimeTravelError::Other(e) => {
1885 0 : warn!("internal error: {e}");
1886 0 : ApiError::InternalServerError(anyhow!("internal error"))
1887 : }
1888 0 : })?;
1889 :
1890 0 : json_response(StatusCode::OK, ())
1891 0 : }
1892 :
1893 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1894 0 : async fn handle_tenant_break(
1895 0 : r: Request<Body>,
1896 0 : _cancel: CancellationToken,
1897 0 : ) -> Result<Response<Body>, ApiError> {
1898 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1899 :
1900 0 : let state = get_state(&r);
1901 0 : state
1902 0 : .tenant_manager
1903 0 : .get_attached_tenant_shard(tenant_shard_id)?
1904 0 : .set_broken("broken from test".to_owned())
1905 0 : .await;
1906 :
1907 0 : json_response(StatusCode::OK, ())
1908 0 : }
1909 :
1910 : // Obtains an lsn lease on the given timeline.
1911 0 : async fn lsn_lease_handler(
1912 0 : mut request: Request<Body>,
1913 0 : _cancel: CancellationToken,
1914 0 : ) -> Result<Response<Body>, ApiError> {
1915 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1916 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1917 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1918 0 : let lsn = json_request::<LsnLeaseRequest>(&mut request).await?.lsn;
1919 :
1920 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1921 0 :
1922 0 : let state = get_state(&request);
1923 :
1924 0 : let timeline =
1925 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1926 0 : .await?;
1927 :
1928 0 : let result = async {
1929 0 : timeline
1930 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
1931 0 : .map_err(|e| {
1932 0 : ApiError::InternalServerError(
1933 0 : e.context(format!("invalid lsn lease request at {lsn}")),
1934 0 : )
1935 0 : })
1936 0 : }
1937 0 : .instrument(info_span!("init_lsn_lease", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1938 0 : .await?;
1939 :
1940 0 : json_response(StatusCode::OK, result)
1941 0 : }
1942 :
1943 : // Run GC immediately on given timeline.
1944 0 : async fn timeline_gc_handler(
1945 0 : mut request: Request<Body>,
1946 0 : cancel: CancellationToken,
1947 0 : ) -> Result<Response<Body>, ApiError> {
1948 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1949 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1950 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1951 :
1952 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1953 :
1954 0 : let state = get_state(&request);
1955 0 :
1956 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1957 0 : let gc_result = state
1958 0 : .tenant_manager
1959 0 : .immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)
1960 0 : .await?;
1961 :
1962 0 : json_response(StatusCode::OK, gc_result)
1963 0 : }
1964 :
1965 : // Run compaction immediately on given timeline.
1966 0 : async fn timeline_compact_handler(
1967 0 : mut request: Request<Body>,
1968 0 : cancel: CancellationToken,
1969 0 : ) -> Result<Response<Body>, ApiError> {
1970 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1971 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1972 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1973 :
1974 0 : let compact_range = json_request_maybe::<Option<CompactRange>>(&mut request).await?;
1975 :
1976 0 : let state = get_state(&request);
1977 0 :
1978 0 : let mut flags = EnumSet::empty();
1979 0 :
1980 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
1981 0 : flags |= CompactFlags::ForceL0Compaction;
1982 0 : }
1983 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1984 0 : flags |= CompactFlags::ForceRepartition;
1985 0 : }
1986 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1987 0 : flags |= CompactFlags::ForceImageLayerCreation;
1988 0 : }
1989 0 : if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
1990 0 : flags |= CompactFlags::EnhancedGcBottomMostCompaction;
1991 0 : }
1992 0 : if Some(true) == parse_query_param::<_, bool>(&request, "dry_run")? {
1993 0 : flags |= CompactFlags::DryRun;
1994 0 : }
1995 :
1996 0 : let wait_until_uploaded =
1997 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1998 0 :
1999 0 : let options = CompactOptions {
2000 0 : compact_range,
2001 0 : flags,
2002 0 : };
2003 :
2004 0 : async {
2005 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2006 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2007 0 : timeline
2008 0 : .compact_with_options(&cancel, options, &ctx)
2009 0 : .await
2010 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2011 0 : if wait_until_uploaded {
2012 0 : timeline.remote_client.wait_completion().await
2013 : // XXX map to correct ApiError for the cases where it's due to shutdown
2014 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2015 0 : }
2016 0 : json_response(StatusCode::OK, ())
2017 0 : }
2018 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2019 0 : .await
2020 0 : }
2021 :
2022 : // Run offload immediately on given timeline.
2023 0 : async fn timeline_offload_handler(
2024 0 : request: Request<Body>,
2025 0 : _cancel: CancellationToken,
2026 0 : ) -> Result<Response<Body>, ApiError> {
2027 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2028 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2029 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2030 :
2031 0 : let state = get_state(&request);
2032 :
2033 0 : async {
2034 0 : let tenant = state
2035 0 : .tenant_manager
2036 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2037 :
2038 0 : if tenant.get_offloaded_timeline(timeline_id).is_ok() {
2039 0 : return json_response(StatusCode::OK, ());
2040 0 : }
2041 0 : let timeline =
2042 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2043 0 : .await?;
2044 :
2045 0 : if !tenant.timeline_has_no_attached_children(timeline_id) {
2046 0 : return Err(ApiError::PreconditionFailed(
2047 0 : "timeline has attached children".into(),
2048 0 : ));
2049 0 : }
2050 0 : if let (false, reason) = timeline.can_offload() {
2051 0 : return Err(ApiError::PreconditionFailed(
2052 0 : format!("Timeline::can_offload() check failed: {}", reason) .into(),
2053 0 : ));
2054 0 : }
2055 0 : offload_timeline(&tenant, &timeline)
2056 0 : .await
2057 0 : .map_err(|e| {
2058 0 : match e {
2059 0 : OffloadError::Cancelled => ApiError::ResourceUnavailable("Timeline shutting down".into()),
2060 0 : _ => ApiError::InternalServerError(anyhow!(e))
2061 : }
2062 0 : })?;
2063 :
2064 0 : json_response(StatusCode::OK, ())
2065 0 : }
2066 0 : .instrument(info_span!("manual_timeline_offload", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2067 0 : .await
2068 0 : }
2069 :
2070 : // Run checkpoint immediately on given timeline.
2071 0 : async fn timeline_checkpoint_handler(
2072 0 : request: Request<Body>,
2073 0 : cancel: CancellationToken,
2074 0 : ) -> Result<Response<Body>, ApiError> {
2075 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2076 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2077 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2078 :
2079 0 : let state = get_state(&request);
2080 0 :
2081 0 : let mut flags = EnumSet::empty();
2082 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2083 0 : flags |= CompactFlags::ForceL0Compaction;
2084 0 : }
2085 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2086 0 : flags |= CompactFlags::ForceRepartition;
2087 0 : }
2088 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2089 0 : flags |= CompactFlags::ForceImageLayerCreation;
2090 0 : }
2091 :
2092 : // By default, checkpoints come with a compaction, but this may be optionally disabled by tests that just want to flush + upload.
2093 0 : let compact = parse_query_param::<_, bool>(&request, "compact")?.unwrap_or(true);
2094 :
2095 0 : let wait_until_uploaded =
2096 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2097 :
2098 0 : async {
2099 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2100 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2101 0 : timeline
2102 0 : .freeze_and_flush()
2103 0 : .await
2104 0 : .map_err(|e| {
2105 0 : match e {
2106 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
2107 0 : other => ApiError::InternalServerError(other.into()),
2108 :
2109 : }
2110 0 : })?;
2111 0 : if compact {
2112 0 : timeline
2113 0 : .compact(&cancel, flags, &ctx)
2114 0 : .await
2115 0 : .map_err(|e|
2116 0 : match e {
2117 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
2118 0 : CompactionError::Offload(e) => ApiError::InternalServerError(anyhow::anyhow!(e)),
2119 0 : CompactionError::Other(e) => ApiError::InternalServerError(e)
2120 0 : }
2121 0 : )?;
2122 0 : }
2123 :
2124 0 : if wait_until_uploaded {
2125 0 : tracing::info!("Waiting for uploads to complete...");
2126 0 : timeline.remote_client.wait_completion().await
2127 : // XXX map to correct ApiError for the cases where it's due to shutdown
2128 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2129 0 : tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
2130 0 : }
2131 :
2132 0 : json_response(StatusCode::OK, ())
2133 0 : }
2134 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2135 0 : .await
2136 0 : }
2137 :
2138 0 : async fn timeline_download_remote_layers_handler_post(
2139 0 : mut request: Request<Body>,
2140 0 : _cancel: CancellationToken,
2141 0 : ) -> Result<Response<Body>, ApiError> {
2142 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2143 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2144 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
2145 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2146 :
2147 0 : let state = get_state(&request);
2148 :
2149 0 : let timeline =
2150 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2151 0 : .await?;
2152 0 : match timeline.spawn_download_all_remote_layers(body).await {
2153 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
2154 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
2155 : }
2156 0 : }
2157 :
2158 0 : async fn timeline_download_remote_layers_handler_get(
2159 0 : request: Request<Body>,
2160 0 : _cancel: CancellationToken,
2161 0 : ) -> Result<Response<Body>, ApiError> {
2162 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2163 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2164 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2165 0 : let state = get_state(&request);
2166 :
2167 0 : let timeline =
2168 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2169 0 : .await?;
2170 0 : let info = timeline
2171 0 : .get_download_all_remote_layers_task_info()
2172 0 : .context("task never started since last pageserver process start")
2173 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
2174 0 : json_response(StatusCode::OK, info)
2175 0 : }
2176 :
2177 0 : async fn timeline_detach_ancestor_handler(
2178 0 : request: Request<Body>,
2179 0 : _cancel: CancellationToken,
2180 0 : ) -> Result<Response<Body>, ApiError> {
2181 : use crate::tenant::timeline::detach_ancestor;
2182 : use pageserver_api::models::detach_ancestor::AncestorDetached;
2183 :
2184 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2185 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2186 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2187 :
2188 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
2189 :
2190 0 : async move {
2191 0 : let mut options = detach_ancestor::Options::default();
2192 :
2193 0 : let rewrite_concurrency =
2194 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
2195 0 : let copy_concurrency =
2196 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
2197 :
2198 0 : [
2199 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
2200 0 : (&mut options.copy_concurrency, copy_concurrency),
2201 0 : ]
2202 0 : .into_iter()
2203 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
2204 0 : .for_each(|(target, val)| *target = val);
2205 0 :
2206 0 : let state = get_state(&request);
2207 :
2208 0 : let tenant = state
2209 0 : .tenant_manager
2210 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2211 :
2212 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2213 :
2214 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
2215 0 : let ctx = &ctx;
2216 :
2217 : // Flush the upload queues of all timelines before detaching ancestor. We do the same thing again
2218 : // during shutdown. This early upload ensures the pageserver does not need to upload too many
2219 : // things and creates downtime during timeline reloads.
2220 0 : for timeline in tenant.list_timelines() {
2221 0 : timeline
2222 0 : .remote_client
2223 0 : .wait_completion()
2224 0 : .await
2225 0 : .map_err(|e| {
2226 0 : ApiError::PreconditionFailed(format!("cannot drain upload queue: {e}").into())
2227 0 : })?;
2228 : }
2229 :
2230 0 : tracing::info!("all timeline upload queues are drained");
2231 :
2232 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2233 :
2234 0 : let progress = timeline
2235 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
2236 0 : .await?;
2237 :
2238 : // uncomment to allow early as possible Tenant::drop
2239 : // drop(tenant);
2240 :
2241 0 : let resp = match progress {
2242 0 : detach_ancestor::Progress::Prepared(attempt, prepared) => {
2243 : // it would be great to tag the guard on to the tenant activation future
2244 0 : let reparented_timelines = state
2245 0 : .tenant_manager
2246 0 : .complete_detaching_timeline_ancestor(
2247 0 : tenant_shard_id,
2248 0 : timeline_id,
2249 0 : prepared,
2250 0 : attempt,
2251 0 : ctx,
2252 0 : )
2253 0 : .await?;
2254 :
2255 0 : AncestorDetached {
2256 0 : reparented_timelines,
2257 0 : }
2258 : }
2259 0 : detach_ancestor::Progress::Done(resp) => resp,
2260 : };
2261 :
2262 0 : json_response(StatusCode::OK, resp)
2263 0 : }
2264 0 : .instrument(span)
2265 0 : .await
2266 0 : }
2267 :
2268 0 : async fn deletion_queue_flush(
2269 0 : r: Request<Body>,
2270 0 : cancel: CancellationToken,
2271 0 : ) -> Result<Response<Body>, ApiError> {
2272 0 : let state = get_state(&r);
2273 :
2274 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
2275 0 :
2276 0 : let flush = async {
2277 0 : if execute {
2278 0 : state.deletion_queue_client.flush_execute().await
2279 : } else {
2280 0 : state.deletion_queue_client.flush().await
2281 : }
2282 0 : }
2283 : // DeletionQueueError's only case is shutting down.
2284 0 : .map_err(|_| ApiError::ShuttingDown);
2285 0 :
2286 0 : tokio::select! {
2287 0 : res = flush => {
2288 0 : res.map(|()| json_response(StatusCode::OK, ()))?
2289 : }
2290 0 : _ = cancel.cancelled() => {
2291 0 : Err(ApiError::ShuttingDown)
2292 : }
2293 : }
2294 0 : }
2295 :
2296 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
2297 0 : async fn getpage_at_lsn_handler(
2298 0 : request: Request<Body>,
2299 0 : _cancel: CancellationToken,
2300 0 : ) -> Result<Response<Body>, ApiError> {
2301 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2302 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2303 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2304 0 : let state = get_state(&request);
2305 :
2306 : struct Key(pageserver_api::key::Key);
2307 :
2308 : impl std::str::FromStr for Key {
2309 : type Err = anyhow::Error;
2310 :
2311 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
2312 0 : pageserver_api::key::Key::from_hex(s).map(Key)
2313 0 : }
2314 : }
2315 :
2316 0 : let key: Key = parse_query_param(&request, "key")?
2317 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
2318 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
2319 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
2320 :
2321 0 : async {
2322 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2323 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2324 :
2325 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
2326 :
2327 0 : Result::<_, ApiError>::Ok(
2328 0 : Response::builder()
2329 0 : .status(StatusCode::OK)
2330 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2331 0 : .body(hyper::Body::from(page))
2332 0 : .unwrap(),
2333 0 : )
2334 0 : }
2335 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2336 0 : .await
2337 0 : }
2338 :
2339 0 : async fn timeline_collect_keyspace(
2340 0 : request: Request<Body>,
2341 0 : _cancel: CancellationToken,
2342 0 : ) -> Result<Response<Body>, ApiError> {
2343 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2344 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2345 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2346 0 : let state = get_state(&request);
2347 :
2348 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2349 :
2350 0 : async {
2351 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2352 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2353 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2354 0 : let (dense_ks, sparse_ks) = timeline
2355 0 : .collect_keyspace(at_lsn, &ctx)
2356 0 : .await
2357 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2358 :
2359 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2360 : // Therefore, we split dense/sparse keys in this API.
2361 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2362 0 :
2363 0 : json_response(StatusCode::OK, res)
2364 0 : }
2365 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2366 0 : .await
2367 0 : }
2368 :
2369 0 : async fn active_timeline_of_active_tenant(
2370 0 : tenant_manager: &TenantManager,
2371 0 : tenant_shard_id: TenantShardId,
2372 0 : timeline_id: TimelineId,
2373 0 : ) -> Result<Arc<Timeline>, ApiError> {
2374 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2375 :
2376 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2377 :
2378 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2379 0 : }
2380 :
2381 0 : async fn always_panic_handler(
2382 0 : req: Request<Body>,
2383 0 : _cancel: CancellationToken,
2384 0 : ) -> Result<Response<Body>, ApiError> {
2385 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2386 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2387 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2388 0 : let query = req.uri().query();
2389 0 : let _ = std::panic::catch_unwind(|| {
2390 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2391 0 : });
2392 0 : json_response(StatusCode::NO_CONTENT, ())
2393 0 : }
2394 :
2395 0 : async fn disk_usage_eviction_run(
2396 0 : mut r: Request<Body>,
2397 0 : cancel: CancellationToken,
2398 0 : ) -> Result<Response<Body>, ApiError> {
2399 0 : check_permission(&r, None)?;
2400 :
2401 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2402 : struct Config {
2403 : /// How many bytes to evict before reporting that pressure is relieved.
2404 : evict_bytes: u64,
2405 :
2406 : #[serde(default)]
2407 : eviction_order: pageserver_api::config::EvictionOrder,
2408 : }
2409 :
2410 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2411 : struct Usage {
2412 : // remains unchanged after instantiation of the struct
2413 : evict_bytes: u64,
2414 : // updated by `add_available_bytes`
2415 : freed_bytes: u64,
2416 : }
2417 :
2418 : impl crate::disk_usage_eviction_task::Usage for Usage {
2419 0 : fn has_pressure(&self) -> bool {
2420 0 : self.evict_bytes > self.freed_bytes
2421 0 : }
2422 :
2423 0 : fn add_available_bytes(&mut self, bytes: u64) {
2424 0 : self.freed_bytes += bytes;
2425 0 : }
2426 : }
2427 :
2428 0 : let config = json_request::<Config>(&mut r).await?;
2429 :
2430 0 : let usage = Usage {
2431 0 : evict_bytes: config.evict_bytes,
2432 0 : freed_bytes: 0,
2433 0 : };
2434 0 :
2435 0 : let state = get_state(&r);
2436 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2437 :
2438 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2439 0 : &eviction_state,
2440 0 : &state.remote_storage,
2441 0 : usage,
2442 0 : &state.tenant_manager,
2443 0 : config.eviction_order.into(),
2444 0 : &cancel,
2445 0 : )
2446 0 : .await;
2447 :
2448 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2449 :
2450 0 : let res = res.map_err(ApiError::InternalServerError)?;
2451 :
2452 0 : json_response(StatusCode::OK, res)
2453 0 : }
2454 :
2455 0 : async fn secondary_upload_handler(
2456 0 : request: Request<Body>,
2457 0 : _cancel: CancellationToken,
2458 0 : ) -> Result<Response<Body>, ApiError> {
2459 0 : let state = get_state(&request);
2460 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2461 0 : state
2462 0 : .secondary_controller
2463 0 : .upload_tenant(tenant_shard_id)
2464 0 : .await
2465 0 : .map_err(ApiError::InternalServerError)?;
2466 :
2467 0 : json_response(StatusCode::OK, ())
2468 0 : }
2469 :
2470 0 : async fn tenant_scan_remote_handler(
2471 0 : request: Request<Body>,
2472 0 : cancel: CancellationToken,
2473 0 : ) -> Result<Response<Body>, ApiError> {
2474 0 : let state = get_state(&request);
2475 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2476 :
2477 0 : let mut response = TenantScanRemoteStorageResponse::default();
2478 :
2479 0 : let (shards, _other_keys) =
2480 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2481 0 : .await
2482 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2483 :
2484 0 : for tenant_shard_id in shards {
2485 0 : let (timeline_ids, _other_keys) =
2486 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2487 0 : .await
2488 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2489 :
2490 0 : let mut generation = Generation::none();
2491 0 : for timeline_id in timeline_ids {
2492 0 : match download_index_part(
2493 0 : &state.remote_storage,
2494 0 : &tenant_shard_id,
2495 0 : &timeline_id,
2496 0 : Generation::MAX,
2497 0 : &cancel,
2498 0 : )
2499 0 : .instrument(info_span!("download_index_part",
2500 : tenant_id=%tenant_shard_id.tenant_id,
2501 0 : shard_id=%tenant_shard_id.shard_slug(),
2502 : %timeline_id))
2503 0 : .await
2504 : {
2505 0 : Ok((index_part, index_generation, _index_mtime)) => {
2506 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2507 0 : index_part.layer_metadata.len(), index_part.metadata.disk_consistent_lsn());
2508 0 : generation = std::cmp::max(generation, index_generation);
2509 : }
2510 : Err(DownloadError::NotFound) => {
2511 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2512 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2513 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2514 0 : continue;
2515 : }
2516 0 : Err(e) => {
2517 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2518 : }
2519 : };
2520 : }
2521 :
2522 0 : response.shards.push(TenantScanRemoteStorageShard {
2523 0 : tenant_shard_id,
2524 0 : generation: generation.into(),
2525 0 : });
2526 : }
2527 :
2528 0 : if response.shards.is_empty() {
2529 0 : return Err(ApiError::NotFound(
2530 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2531 0 : ));
2532 0 : }
2533 0 :
2534 0 : json_response(StatusCode::OK, response)
2535 0 : }
2536 :
2537 0 : async fn secondary_download_handler(
2538 0 : request: Request<Body>,
2539 0 : _cancel: CancellationToken,
2540 0 : ) -> Result<Response<Body>, ApiError> {
2541 0 : let state = get_state(&request);
2542 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2543 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2544 :
2545 : // We don't need this to issue the download request, but:
2546 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2547 : // - we will use this to provide status feedback in the response
2548 0 : let Some(secondary_tenant) = state
2549 0 : .tenant_manager
2550 0 : .get_secondary_tenant_shard(tenant_shard_id)
2551 : else {
2552 0 : return Err(ApiError::NotFound(
2553 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2554 0 : ));
2555 : };
2556 :
2557 0 : let timeout = wait.unwrap_or(Duration::MAX);
2558 :
2559 0 : let result = tokio::time::timeout(
2560 0 : timeout,
2561 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2562 0 : )
2563 0 : .await;
2564 :
2565 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2566 :
2567 0 : let status = match result {
2568 : Ok(Ok(())) => {
2569 0 : if progress.layers_downloaded >= progress.layers_total {
2570 : // Download job ran to completion
2571 0 : StatusCode::OK
2572 : } else {
2573 : // Download dropped out without errors because it ran out of time budget
2574 0 : StatusCode::ACCEPTED
2575 : }
2576 : }
2577 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2578 : // okay. We could get an error here in the unlikely edge case that the tenant
2579 : // was detached between our check above and executing the download job.
2580 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2581 : // A timeout is not an error: we have started the download, we're just not done
2582 : // yet. The caller will get a response body indicating status.
2583 0 : Err(_) => StatusCode::ACCEPTED,
2584 : };
2585 :
2586 0 : json_response(status, progress)
2587 0 : }
2588 :
2589 0 : async fn secondary_status_handler(
2590 0 : request: Request<Body>,
2591 0 : _cancel: CancellationToken,
2592 0 : ) -> Result<Response<Body>, ApiError> {
2593 0 : let state = get_state(&request);
2594 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2595 :
2596 0 : let Some(secondary_tenant) = state
2597 0 : .tenant_manager
2598 0 : .get_secondary_tenant_shard(tenant_shard_id)
2599 : else {
2600 0 : return Err(ApiError::NotFound(
2601 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2602 0 : ));
2603 : };
2604 :
2605 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2606 0 :
2607 0 : json_response(StatusCode::OK, progress)
2608 0 : }
2609 :
2610 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2611 0 : json_response(
2612 0 : StatusCode::NOT_FOUND,
2613 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2614 0 : )
2615 0 : }
2616 :
2617 0 : async fn post_tracing_event_handler(
2618 0 : mut r: Request<Body>,
2619 0 : _cancel: CancellationToken,
2620 0 : ) -> Result<Response<Body>, ApiError> {
2621 0 : #[derive(Debug, serde::Deserialize)]
2622 : #[serde(rename_all = "lowercase")]
2623 : enum Level {
2624 : Error,
2625 : Warn,
2626 : Info,
2627 : Debug,
2628 : Trace,
2629 : }
2630 0 : #[derive(Debug, serde::Deserialize)]
2631 : struct Request {
2632 : level: Level,
2633 : message: String,
2634 : }
2635 0 : let body: Request = json_request(&mut r)
2636 0 : .await
2637 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2638 :
2639 0 : match body.level {
2640 0 : Level::Error => tracing::error!(?body.message),
2641 0 : Level::Warn => tracing::warn!(?body.message),
2642 0 : Level::Info => tracing::info!(?body.message),
2643 0 : Level::Debug => tracing::debug!(?body.message),
2644 0 : Level::Trace => tracing::trace!(?body.message),
2645 : }
2646 :
2647 0 : json_response(StatusCode::OK, ())
2648 0 : }
2649 :
2650 0 : async fn put_io_engine_handler(
2651 0 : mut r: Request<Body>,
2652 0 : _cancel: CancellationToken,
2653 0 : ) -> Result<Response<Body>, ApiError> {
2654 0 : check_permission(&r, None)?;
2655 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2656 0 : crate::virtual_file::io_engine::set(kind);
2657 0 : json_response(StatusCode::OK, ())
2658 0 : }
2659 :
2660 0 : async fn put_io_mode_handler(
2661 0 : mut r: Request<Body>,
2662 0 : _cancel: CancellationToken,
2663 0 : ) -> Result<Response<Body>, ApiError> {
2664 0 : check_permission(&r, None)?;
2665 0 : let mode: IoMode = json_request(&mut r).await?;
2666 0 : crate::virtual_file::set_io_mode(mode);
2667 0 : json_response(StatusCode::OK, ())
2668 0 : }
2669 :
2670 : /// Polled by control plane.
2671 : ///
2672 : /// See [`crate::utilization`].
2673 0 : async fn get_utilization(
2674 0 : r: Request<Body>,
2675 0 : _cancel: CancellationToken,
2676 0 : ) -> Result<Response<Body>, ApiError> {
2677 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2678 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2679 0 : });
2680 :
2681 : // this probably could be completely public, but lets make that change later.
2682 0 : check_permission(&r, None)?;
2683 :
2684 0 : let state = get_state(&r);
2685 0 : let mut g = state.latest_utilization.lock().await;
2686 :
2687 0 : let regenerate_every = Duration::from_secs(1);
2688 0 : let still_valid = g
2689 0 : .as_ref()
2690 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2691 0 :
2692 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2693 0 : // regenerate at most 1Hz to allow polling at any rate.
2694 0 : if !still_valid {
2695 0 : let path = state.conf.tenants_path();
2696 0 : let doc =
2697 0 : crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
2698 0 : .map_err(ApiError::InternalServerError)?;
2699 :
2700 0 : let mut buf = Vec::new();
2701 0 : serde_json::to_writer(&mut buf, &doc)
2702 0 : .context("serialize")
2703 0 : .map_err(ApiError::InternalServerError)?;
2704 :
2705 0 : let body = bytes::Bytes::from(buf);
2706 0 :
2707 0 : *g = Some((std::time::Instant::now(), body));
2708 0 : }
2709 :
2710 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2711 0 : let cached = g.as_ref().expect("just set").1.clone();
2712 0 :
2713 0 : Response::builder()
2714 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2715 0 : // thought of using http date header, but that is second precision which does not give any
2716 0 : // debugging aid
2717 0 : .status(StatusCode::OK)
2718 0 : .body(hyper::Body::from(cached))
2719 0 : .context("build response")
2720 0 : .map_err(ApiError::InternalServerError)
2721 0 : }
2722 :
2723 0 : async fn list_aux_files(
2724 0 : mut request: Request<Body>,
2725 0 : _cancel: CancellationToken,
2726 0 : ) -> Result<Response<Body>, ApiError> {
2727 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2728 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2729 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2730 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2731 :
2732 0 : let state = get_state(&request);
2733 :
2734 0 : let timeline =
2735 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2736 0 : .await?;
2737 :
2738 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2739 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2740 0 : json_response(StatusCode::OK, files)
2741 0 : }
2742 :
2743 0 : async fn perf_info(
2744 0 : request: Request<Body>,
2745 0 : _cancel: CancellationToken,
2746 0 : ) -> Result<Response<Body>, ApiError> {
2747 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2748 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2749 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2750 :
2751 0 : let state = get_state(&request);
2752 :
2753 0 : let timeline =
2754 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2755 0 : .await?;
2756 :
2757 0 : let result = timeline.perf_info().await;
2758 :
2759 0 : json_response(StatusCode::OK, result)
2760 0 : }
2761 :
2762 0 : async fn ingest_aux_files(
2763 0 : mut request: Request<Body>,
2764 0 : _cancel: CancellationToken,
2765 0 : ) -> Result<Response<Body>, ApiError> {
2766 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2767 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2768 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2769 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2770 :
2771 0 : let state = get_state(&request);
2772 :
2773 0 : let timeline =
2774 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2775 0 : .await?;
2776 :
2777 0 : let mut modification = timeline.begin_modification(
2778 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
2779 0 : );
2780 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2781 0 : for (fname, content) in body.aux_files {
2782 0 : modification
2783 0 : .put_file(&fname, content.as_bytes(), &ctx)
2784 0 : .await
2785 0 : .map_err(ApiError::InternalServerError)?;
2786 : }
2787 0 : modification
2788 0 : .commit(&ctx)
2789 0 : .await
2790 0 : .map_err(ApiError::InternalServerError)?;
2791 :
2792 0 : json_response(StatusCode::OK, ())
2793 0 : }
2794 :
2795 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2796 : /// candidates for splitting
2797 0 : async fn post_top_tenants(
2798 0 : mut r: Request<Body>,
2799 0 : _cancel: CancellationToken,
2800 0 : ) -> Result<Response<Body>, ApiError> {
2801 0 : check_permission(&r, None)?;
2802 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2803 0 : let state = get_state(&r);
2804 :
2805 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2806 0 : match order_by {
2807 0 : TenantSorting::ResidentSize => sizes.resident_size,
2808 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2809 : }
2810 0 : }
2811 :
2812 : #[derive(Eq, PartialEq)]
2813 : struct HeapItem {
2814 : metric: u64,
2815 : sizes: TopTenantShardItem,
2816 : }
2817 :
2818 : impl PartialOrd for HeapItem {
2819 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2820 0 : Some(self.cmp(other))
2821 0 : }
2822 : }
2823 :
2824 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2825 : /// supports popping the greatest item but not the smallest.
2826 : impl Ord for HeapItem {
2827 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2828 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2829 0 : }
2830 : }
2831 :
2832 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2833 :
2834 : // FIXME: this is a lot of clones to take this tenant list
2835 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2836 0 : if let Some(shards_lt) = request.where_shards_lt {
2837 : // Ignore tenants which already have >= this many shards
2838 0 : if tenant_shard_id.shard_count >= shards_lt {
2839 0 : continue;
2840 0 : }
2841 0 : }
2842 :
2843 0 : let sizes = match tenant_slot {
2844 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2845 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2846 0 : continue;
2847 : }
2848 : };
2849 0 : let metric = get_size_metric(&sizes, &request.order_by);
2850 :
2851 0 : if let Some(gt) = request.where_gt {
2852 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2853 0 : if metric <= gt {
2854 0 : continue;
2855 0 : }
2856 0 : };
2857 :
2858 0 : match top_n.peek() {
2859 0 : None => {
2860 0 : // Top N list is empty: candidate becomes first member
2861 0 : top_n.push(HeapItem { metric, sizes });
2862 0 : }
2863 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
2864 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
2865 0 : top_n.push(HeapItem { metric, sizes });
2866 0 : }
2867 0 : Some(i) if i.metric > metric => {
2868 0 : // List is at limit and lowest value is greater than our candidate, drop it.
2869 0 : }
2870 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
2871 : }
2872 :
2873 0 : while top_n.len() > request.limit {
2874 0 : top_n.pop();
2875 0 : }
2876 : }
2877 :
2878 0 : json_response(
2879 0 : StatusCode::OK,
2880 0 : TopTenantShardsResponse {
2881 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
2882 0 : },
2883 0 : )
2884 0 : }
2885 :
2886 0 : async fn put_tenant_timeline_import_basebackup(
2887 0 : request: Request<Body>,
2888 0 : _cancel: CancellationToken,
2889 0 : ) -> Result<Response<Body>, ApiError> {
2890 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2891 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2892 0 : let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
2893 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
2894 0 : let pg_version: u32 = must_parse_query_param(&request, "pg_version")?;
2895 :
2896 0 : check_permission(&request, Some(tenant_id))?;
2897 :
2898 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
2899 :
2900 0 : let span = info_span!("import_basebackup", tenant_id=%tenant_id, timeline_id=%timeline_id, base_lsn=%base_lsn, end_lsn=%end_lsn, pg_version=%pg_version);
2901 0 : async move {
2902 0 : let state = get_state(&request);
2903 0 : let tenant = state
2904 0 : .tenant_manager
2905 0 : .get_attached_tenant_shard(TenantShardId::unsharded(tenant_id))?;
2906 :
2907 0 : let broker_client = state.broker_client.clone();
2908 0 :
2909 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
2910 0 : res.map_err(|error| {
2911 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
2912 0 : })
2913 0 : }));
2914 0 :
2915 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2916 :
2917 0 : let timeline = tenant
2918 0 : .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
2919 0 : .map_err(ApiError::InternalServerError)
2920 0 : .await?;
2921 :
2922 : // TODO mark timeline as not ready until it reaches end_lsn.
2923 : // We might have some wal to import as well, and we should prevent compute
2924 : // from connecting before that and writing conflicting wal.
2925 : //
2926 : // This is not relevant for pageserver->pageserver migrations, since there's
2927 : // no wal to import. But should be fixed if we want to import from postgres.
2928 :
2929 : // TODO leave clean state on error. For now you can use detach to clean
2930 : // up broken state from a failed import.
2931 :
2932 : // Import basebackup provided via CopyData
2933 0 : info!("importing basebackup");
2934 :
2935 0 : timeline
2936 0 : .import_basebackup_from_tar(tenant.clone(), &mut body, base_lsn, broker_client, &ctx)
2937 0 : .await
2938 0 : .map_err(ApiError::InternalServerError)?;
2939 :
2940 : // Read the end of the tar archive.
2941 0 : read_tar_eof(body)
2942 0 : .await
2943 0 : .map_err(ApiError::InternalServerError)?;
2944 :
2945 : // TODO check checksum
2946 : // Meanwhile you can verify client-side by taking fullbackup
2947 : // and checking that it matches in size with what was imported.
2948 : // It wouldn't work if base came from vanilla postgres though,
2949 : // since we discard some log files.
2950 :
2951 0 : info!("done");
2952 0 : json_response(StatusCode::OK, ())
2953 0 : }
2954 0 : .instrument(span)
2955 0 : .await
2956 0 : }
2957 :
2958 0 : async fn put_tenant_timeline_import_wal(
2959 0 : request: Request<Body>,
2960 0 : _cancel: CancellationToken,
2961 0 : ) -> Result<Response<Body>, ApiError> {
2962 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2963 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2964 0 : let start_lsn: Lsn = must_parse_query_param(&request, "start_lsn")?;
2965 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
2966 :
2967 0 : check_permission(&request, Some(tenant_id))?;
2968 :
2969 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
2970 :
2971 0 : let span = info_span!("import_wal", tenant_id=%tenant_id, timeline_id=%timeline_id, start_lsn=%start_lsn, end_lsn=%end_lsn);
2972 0 : async move {
2973 0 : let state = get_state(&request);
2974 :
2975 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, TenantShardId::unsharded(tenant_id), timeline_id).await?;
2976 :
2977 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
2978 0 : res.map_err(|error| {
2979 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
2980 0 : })
2981 0 : }));
2982 0 :
2983 0 : let last_record_lsn = timeline.get_last_record_lsn();
2984 0 : if last_record_lsn != start_lsn {
2985 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!("Cannot import WAL from Lsn {start_lsn} because timeline does not start from the same lsn: {last_record_lsn}")));
2986 0 : }
2987 0 :
2988 0 : // TODO leave clean state on error. For now you can use detach to clean
2989 0 : // up broken state from a failed import.
2990 0 :
2991 0 : // Import wal provided via CopyData
2992 0 : info!("importing wal");
2993 0 : crate::import_datadir::import_wal_from_tar(&timeline, &mut body, start_lsn, end_lsn, &ctx).await.map_err(ApiError::InternalServerError)?;
2994 0 : info!("wal import complete");
2995 :
2996 : // Read the end of the tar archive.
2997 0 : read_tar_eof(body).await.map_err(ApiError::InternalServerError)?;
2998 :
2999 : // TODO Does it make sense to overshoot?
3000 0 : if timeline.get_last_record_lsn() < end_lsn {
3001 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!("Cannot import WAL from Lsn {start_lsn} because timeline does not start from the same lsn: {last_record_lsn}")));
3002 0 : }
3003 0 :
3004 0 : // Flush data to disk, then upload to s3. No need for a forced checkpoint.
3005 0 : // We only want to persist the data, and it doesn't matter if it's in the
3006 0 : // shape of deltas or images.
3007 0 : info!("flushing layers");
3008 0 : timeline.freeze_and_flush().await.map_err(|e| match e {
3009 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
3010 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
3011 0 : })?;
3012 :
3013 0 : info!("done");
3014 :
3015 0 : json_response(StatusCode::OK, ())
3016 0 : }.instrument(span).await
3017 0 : }
3018 :
3019 : /// Read the end of a tar archive.
3020 : ///
3021 : /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
3022 : /// `tokio_tar` already read the first such block. Read the second all-zeros block,
3023 : /// and check that there is no more data after the EOF marker.
3024 : ///
3025 : /// 'tar' command can also write extra blocks of zeros, up to a record
3026 : /// size, controlled by the --record-size argument. Ignore them too.
3027 0 : async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow::Result<()> {
3028 : use tokio::io::AsyncReadExt;
3029 0 : let mut buf = [0u8; 512];
3030 0 :
3031 0 : // Read the all-zeros block, and verify it
3032 0 : let mut total_bytes = 0;
3033 0 : while total_bytes < 512 {
3034 0 : let nbytes = reader.read(&mut buf[total_bytes..]).await?;
3035 0 : total_bytes += nbytes;
3036 0 : if nbytes == 0 {
3037 0 : break;
3038 0 : }
3039 : }
3040 0 : if total_bytes < 512 {
3041 0 : anyhow::bail!("incomplete or invalid tar EOF marker");
3042 0 : }
3043 0 : if !buf.iter().all(|&x| x == 0) {
3044 0 : anyhow::bail!("invalid tar EOF marker");
3045 0 : }
3046 0 :
3047 0 : // Drain any extra zero-blocks after the EOF marker
3048 0 : let mut trailing_bytes = 0;
3049 0 : let mut seen_nonzero_bytes = false;
3050 : loop {
3051 0 : let nbytes = reader.read(&mut buf).await?;
3052 0 : trailing_bytes += nbytes;
3053 0 : if !buf.iter().all(|&x| x == 0) {
3054 0 : seen_nonzero_bytes = true;
3055 0 : }
3056 0 : if nbytes == 0 {
3057 0 : break;
3058 0 : }
3059 : }
3060 0 : if seen_nonzero_bytes {
3061 0 : anyhow::bail!("unexpected non-zero bytes after the tar archive");
3062 0 : }
3063 0 : if trailing_bytes % 512 != 0 {
3064 0 : anyhow::bail!("unexpected number of zeros ({trailing_bytes}), not divisible by tar block size (512 bytes), after the tar archive");
3065 0 : }
3066 0 : Ok(())
3067 0 : }
3068 :
3069 : /// Common functionality of all the HTTP API handlers.
3070 : ///
3071 : /// - Adds a tracing span to each request (by `request_span`)
3072 : /// - Logs the request depending on the request method (by `request_span`)
3073 : /// - Logs the response if it was not successful (by `request_span`
3074 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
3075 : /// Future if the connection to the client is lost, but most of the pageserver code is
3076 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
3077 : /// request with a CancellationToken.
3078 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
3079 0 : where
3080 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3081 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3082 0 : {
3083 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
3084 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
3085 0 : "failpoint".into()
3086 0 : )));
3087 :
3088 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
3089 0 : anyhow::anyhow!("failpoint")
3090 0 : )));
3091 0 : }
3092 :
3093 : // Spawn a new task to handle the request, to protect the handler from unexpected
3094 : // async cancellations. Most pageserver functions are not async cancellation safe.
3095 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
3096 : // with the cancellation token.
3097 0 : let token = CancellationToken::new();
3098 0 : let cancel_guard = token.clone().drop_guard();
3099 0 : let result = request_span(request, move |r| async {
3100 0 : let handle = tokio::spawn(
3101 0 : async {
3102 0 : let token_cloned = token.clone();
3103 0 : let result = handler(r, token).await;
3104 0 : if token_cloned.is_cancelled() {
3105 : // dropguard has executed: we will never turn this result into response.
3106 : //
3107 : // at least temporarily do {:?} logging; these failures are rare enough but
3108 : // could hide difficult errors.
3109 0 : match &result {
3110 0 : Ok(response) => {
3111 0 : let status = response.status();
3112 0 : info!(%status, "Cancelled request finished successfully")
3113 : }
3114 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
3115 : }
3116 0 : }
3117 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
3118 : // which should suffice.
3119 : //
3120 : // there is still a chance to lose the result due to race between
3121 : // returning from here and the actual connection closing happening
3122 : // before outer task gets to execute. leaving that up for #5815.
3123 0 : result
3124 0 : }
3125 0 : .in_current_span(),
3126 0 : );
3127 0 :
3128 0 : match handle.await {
3129 : // TODO: never actually return Err from here, always Ok(...) so that we can log
3130 : // spanned errors. Call api_error_handler instead and return appropriate Body.
3131 0 : Ok(result) => result,
3132 0 : Err(e) => {
3133 0 : // The handler task panicked. We have a global panic handler that logs the
3134 0 : // panic with its backtrace, so no need to log that here. Only log a brief
3135 0 : // message to make it clear that we returned the error to the client.
3136 0 : error!("HTTP request handler task panicked: {e:#}");
3137 :
3138 : // Don't return an Error here, because then fallback error handler that was
3139 : // installed in make_router() will print the error. Instead, construct the
3140 : // HTTP error response and return that.
3141 0 : Ok(
3142 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
3143 0 : .into_response(),
3144 0 : )
3145 : }
3146 : }
3147 0 : })
3148 0 : .await;
3149 :
3150 0 : cancel_guard.disarm();
3151 0 :
3152 0 : result
3153 0 : }
3154 :
3155 : /// Like api_handler, but returns an error response if the server is built without
3156 : /// the 'testing' feature.
3157 0 : async fn testing_api_handler<R, H>(
3158 0 : desc: &str,
3159 0 : request: Request<Body>,
3160 0 : handler: H,
3161 0 : ) -> Result<Response<Body>, ApiError>
3162 0 : where
3163 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3164 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3165 0 : {
3166 0 : if cfg!(feature = "testing") {
3167 0 : api_handler(request, handler).await
3168 : } else {
3169 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
3170 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
3171 0 : ))))
3172 0 : .await
3173 : }
3174 0 : }
3175 :
3176 0 : pub fn make_router(
3177 0 : state: Arc<State>,
3178 0 : launch_ts: &'static LaunchTimestamp,
3179 0 : auth: Option<Arc<SwappableJwtAuth>>,
3180 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
3181 0 : let spec = include_bytes!("openapi_spec.yml");
3182 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
3183 0 : if auth.is_some() {
3184 0 : router = router.middleware(auth_middleware(|request| {
3185 0 : let state = get_state(request);
3186 0 : if state.allowlist_routes.contains(&request.uri().path()) {
3187 0 : None
3188 : } else {
3189 0 : state.auth.as_deref()
3190 : }
3191 0 : }))
3192 0 : }
3193 :
3194 0 : router = router.middleware(
3195 0 : endpoint::add_response_header_middleware(
3196 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
3197 0 : &launch_ts.to_string(),
3198 0 : )
3199 0 : .expect("construct launch timestamp header middleware"),
3200 0 : );
3201 0 :
3202 0 : Ok(router
3203 0 : .data(state)
3204 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
3205 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
3206 0 : .get("/v1/status", |r| api_handler(r, status_handler))
3207 0 : .put("/v1/failpoints", |r| {
3208 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
3209 0 : })
3210 0 : .post("/v1/reload_auth_validation_keys", |r| {
3211 0 : api_handler(r, reload_auth_validation_keys_handler)
3212 0 : })
3213 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
3214 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
3215 0 : api_handler(r, tenant_status)
3216 0 : })
3217 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
3218 0 : api_handler(r, tenant_delete_handler)
3219 0 : })
3220 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
3221 0 : api_handler(r, tenant_size_handler)
3222 0 : })
3223 0 : .put("/v1/tenant/config", |r| {
3224 0 : api_handler(r, update_tenant_config_handler)
3225 0 : })
3226 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
3227 0 : api_handler(r, tenant_shard_split_handler)
3228 0 : })
3229 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
3230 0 : api_handler(r, get_tenant_config_handler)
3231 0 : })
3232 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
3233 0 : api_handler(r, put_tenant_location_config_handler)
3234 0 : })
3235 0 : .get("/v1/location_config", |r| {
3236 0 : api_handler(r, list_location_config_handler)
3237 0 : })
3238 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
3239 0 : api_handler(r, get_location_config_handler)
3240 0 : })
3241 0 : .put(
3242 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
3243 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
3244 0 : )
3245 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
3246 0 : api_handler(r, timeline_list_handler)
3247 0 : })
3248 0 : .get("/v1/tenant/:tenant_shard_id/timeline_and_offloaded", |r| {
3249 0 : api_handler(r, timeline_and_offloaded_list_handler)
3250 0 : })
3251 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
3252 0 : api_handler(r, timeline_create_handler)
3253 0 : })
3254 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
3255 0 : api_handler(r, tenant_reset_handler)
3256 0 : })
3257 0 : .post(
3258 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
3259 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
3260 0 : )
3261 0 : .put(
3262 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/archival_config",
3263 0 : |r| api_handler(r, timeline_archival_config_handler),
3264 0 : )
3265 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3266 0 : api_handler(r, timeline_detail_handler)
3267 0 : })
3268 0 : .get(
3269 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
3270 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
3271 0 : )
3272 0 : .get(
3273 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
3274 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
3275 0 : )
3276 0 : .post(
3277 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
3278 0 : |r| api_handler(r, lsn_lease_handler),
3279 0 : )
3280 0 : .put(
3281 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
3282 0 : |r| api_handler(r, timeline_gc_handler),
3283 0 : )
3284 0 : .put(
3285 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
3286 0 : |r| api_handler(r, timeline_compact_handler),
3287 0 : )
3288 0 : .put(
3289 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/offload",
3290 0 : |r| testing_api_handler("attempt timeline offload", r, timeline_offload_handler),
3291 0 : )
3292 0 : .put(
3293 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
3294 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
3295 0 : )
3296 0 : .post(
3297 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
3298 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
3299 0 : )
3300 0 : .get(
3301 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
3302 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
3303 0 : )
3304 0 : .put(
3305 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
3306 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
3307 0 : )
3308 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3309 0 : api_handler(r, timeline_delete_handler)
3310 0 : })
3311 0 : .get(
3312 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
3313 0 : |r| api_handler(r, layer_map_info_handler),
3314 0 : )
3315 0 : .get(
3316 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3317 0 : |r| api_handler(r, layer_download_handler),
3318 0 : )
3319 0 : .delete(
3320 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3321 0 : |r| api_handler(r, evict_timeline_layer_handler),
3322 0 : )
3323 0 : .post(
3324 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_name/scan_disposable_keys",
3325 0 : |r| testing_api_handler("timeline_layer_scan_disposable_keys", r, timeline_layer_scan_disposable_keys),
3326 0 : )
3327 0 : .post(
3328 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
3329 0 : |r| api_handler(r, timeline_gc_blocking_handler),
3330 0 : )
3331 0 : .post(
3332 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
3333 0 : |r| api_handler(r, timeline_gc_unblocking_handler),
3334 0 : )
3335 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
3336 0 : api_handler(r, secondary_upload_handler)
3337 0 : })
3338 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
3339 0 : api_handler(r, tenant_scan_remote_handler)
3340 0 : })
3341 0 : .put("/v1/disk_usage_eviction/run", |r| {
3342 0 : api_handler(r, disk_usage_eviction_run)
3343 0 : })
3344 0 : .put("/v1/deletion_queue/flush", |r| {
3345 0 : api_handler(r, deletion_queue_flush)
3346 0 : })
3347 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
3348 0 : api_handler(r, secondary_status_handler)
3349 0 : })
3350 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
3351 0 : api_handler(r, secondary_download_handler)
3352 0 : })
3353 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
3354 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
3355 0 : })
3356 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
3357 0 : .post("/v1/tracing/event", |r| {
3358 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
3359 0 : })
3360 0 : .get(
3361 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
3362 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
3363 0 : )
3364 0 : .get(
3365 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
3366 0 : |r| api_handler(r, timeline_collect_keyspace),
3367 0 : )
3368 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
3369 0 : .put("/v1/io_mode", |r| api_handler(r, put_io_mode_handler))
3370 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
3371 0 : .post(
3372 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
3373 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
3374 0 : )
3375 0 : .post(
3376 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
3377 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
3378 0 : )
3379 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
3380 0 : .post(
3381 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
3382 0 : |r| testing_api_handler("perf_info", r, perf_info),
3383 0 : )
3384 0 : .put(
3385 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_basebackup",
3386 0 : |r| api_handler(r, put_tenant_timeline_import_basebackup),
3387 0 : )
3388 0 : .put(
3389 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_wal",
3390 0 : |r| api_handler(r, put_tenant_timeline_import_wal),
3391 0 : )
3392 0 : .any(handler_404))
3393 0 : }
|
