Line data Source code
1 : //!
2 : //! Management HTTP API
3 : //!
4 : use std::cmp::Reverse;
5 : use std::collections::BTreeMap;
6 : use std::collections::BinaryHeap;
7 : use std::collections::HashMap;
8 : use std::str::FromStr;
9 : use std::sync::Arc;
10 : use std::time::Duration;
11 :
12 : use anyhow::{Context, Result, anyhow};
13 : use enumset::EnumSet;
14 : use futures::future::join_all;
15 : use futures::{StreamExt, TryFutureExt};
16 : use http_utils::endpoint::{
17 : self, attach_openapi_ui, auth_middleware, check_permission_with, profile_cpu_handler,
18 : profile_heap_handler, prometheus_metrics_handler, request_span,
19 : };
20 : use http_utils::error::{ApiError, HttpErrorBody};
21 : use http_utils::failpoints::failpoints_handler;
22 : use http_utils::json::{json_request, json_request_maybe, json_response};
23 : use http_utils::request::{
24 : get_request_param, must_get_query_param, must_parse_query_param, parse_query_param,
25 : parse_request_param,
26 : };
27 : use http_utils::{RequestExt, RouterBuilder};
28 : use humantime::format_rfc3339;
29 : use hyper::{Body, Request, Response, StatusCode, Uri, header};
30 : use metrics::launch_timestamp::LaunchTimestamp;
31 : use pageserver_api::models::virtual_file::IoMode;
32 : use pageserver_api::models::{
33 : DetachBehavior, DownloadRemoteLayersTaskSpawnRequest, IngestAuxFilesRequest,
34 : ListAuxFilesRequest, LocationConfig, LocationConfigListResponse, LocationConfigMode, LsnLease,
35 : LsnLeaseRequest, OffloadedTimelineInfo, PageTraceEvent, ShardParameters, StatusResponse,
36 : TenantConfigPatchRequest, TenantConfigRequest, TenantDetails, TenantInfo,
37 : TenantLocationConfigRequest, TenantLocationConfigResponse, TenantScanRemoteStorageResponse,
38 : TenantScanRemoteStorageShard, TenantShardLocation, TenantShardSplitRequest,
39 : TenantShardSplitResponse, TenantSorting, TenantState, TenantWaitLsnRequest,
40 : TimelineArchivalConfigRequest, TimelineCreateRequest, TimelineCreateRequestMode,
41 : TimelineCreateRequestModeImportPgdata, TimelineGcRequest, TimelineInfo,
42 : TimelinePatchIndexPartRequest, TimelineVisibilityState, TimelinesInfoAndOffloaded,
43 : TopTenantShardItem, TopTenantShardsRequest, TopTenantShardsResponse,
44 : };
45 : use pageserver_api::shard::{ShardCount, TenantShardId};
46 : use postgres_ffi::PgMajorVersion;
47 : use remote_storage::{DownloadError, GenericRemoteStorage, TimeTravelError};
48 : use scopeguard::defer;
49 : use serde_json::json;
50 : use tenant_size_model::svg::SvgBranchKind;
51 : use tenant_size_model::{SizeResult, StorageModel};
52 : use tokio::time::Instant;
53 : use tokio_util::io::StreamReader;
54 : use tokio_util::sync::CancellationToken;
55 : use tracing::*;
56 : use utils::auth::SwappableJwtAuth;
57 : use utils::generation::Generation;
58 : use utils::id::{TenantId, TimelineId};
59 : use utils::lsn::Lsn;
60 :
61 : use crate::config::PageServerConf;
62 : use crate::context;
63 : use crate::context::{DownloadBehavior, RequestContext, RequestContextBuilder};
64 : use crate::deletion_queue::DeletionQueueClient;
65 : use crate::feature_resolver::FeatureResolver;
66 : use crate::metrics::LOCAL_DATA_LOSS_SUSPECTED;
67 : use crate::pgdatadir_mapping::LsnForTimestamp;
68 : use crate::task_mgr::TaskKind;
69 : use crate::tenant::config::LocationConf;
70 : use crate::tenant::mgr::{
71 : GetActiveTenantError, GetTenantError, TenantManager, TenantMapError, TenantMapInsertError,
72 : TenantSlot, TenantSlotError, TenantSlotUpsertError, TenantStateError, UpsertLocationError,
73 : };
74 : use crate::tenant::remote_timeline_client::index::GcCompactionState;
75 : use crate::tenant::remote_timeline_client::{
76 : download_index_part, download_tenant_manifest, list_remote_tenant_shards, list_remote_timelines,
77 : };
78 : use crate::tenant::secondary::SecondaryController;
79 : use crate::tenant::size::ModelInputs;
80 : use crate::tenant::storage_layer::{IoConcurrency, LayerAccessStatsReset, LayerName};
81 : use crate::tenant::timeline::layer_manager::LayerManagerLockHolder;
82 : use crate::tenant::timeline::offload::{OffloadError, offload_timeline};
83 : use crate::tenant::timeline::{
84 : CompactFlags, CompactOptions, CompactRequest, MarkInvisibleRequest, Timeline, WaitLsnTimeout,
85 : WaitLsnWaiter, import_pgdata,
86 : };
87 : use crate::tenant::{
88 : GetTimelineError, LogicalSizeCalculationCause, OffloadedTimeline, PageReconstructError,
89 : remote_timeline_client,
90 : };
91 : use crate::{DEFAULT_PG_VERSION, disk_usage_eviction_task, tenant};
92 :
93 : // For APIs that require an Active tenant, how long should we block waiting for that state?
94 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
95 : // failed API calls while tenants are activating.
96 : #[cfg(not(feature = "testing"))]
97 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
98 :
99 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
100 : // finish attaching, if calls to remote storage are slow.
101 : #[cfg(feature = "testing")]
102 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
103 :
104 : pub struct State {
105 : conf: &'static PageServerConf,
106 : tenant_manager: Arc<TenantManager>,
107 : auth: Option<Arc<SwappableJwtAuth>>,
108 : allowlist_routes: &'static [&'static str],
109 : remote_storage: GenericRemoteStorage,
110 : broker_client: storage_broker::BrokerClientChannel,
111 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
112 : deletion_queue_client: DeletionQueueClient,
113 : secondary_controller: SecondaryController,
114 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
115 : feature_resolver: FeatureResolver,
116 : }
117 :
118 : impl State {
119 : #[allow(clippy::too_many_arguments)]
120 0 : pub fn new(
121 0 : conf: &'static PageServerConf,
122 0 : tenant_manager: Arc<TenantManager>,
123 0 : auth: Option<Arc<SwappableJwtAuth>>,
124 0 : remote_storage: GenericRemoteStorage,
125 0 : broker_client: storage_broker::BrokerClientChannel,
126 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
127 0 : deletion_queue_client: DeletionQueueClient,
128 0 : secondary_controller: SecondaryController,
129 0 : feature_resolver: FeatureResolver,
130 0 : ) -> anyhow::Result<Self> {
131 0 : let allowlist_routes = &[
132 0 : "/v1/status",
133 0 : "/v1/doc",
134 0 : "/swagger.yml",
135 0 : "/metrics",
136 0 : "/profile/cpu",
137 0 : "/profile/heap",
138 0 : ];
139 0 : Ok(Self {
140 0 : conf,
141 0 : tenant_manager,
142 0 : auth,
143 0 : allowlist_routes,
144 0 : remote_storage,
145 0 : broker_client,
146 0 : disk_usage_eviction_state,
147 0 : deletion_queue_client,
148 0 : secondary_controller,
149 0 : latest_utilization: Default::default(),
150 0 : feature_resolver,
151 0 : })
152 0 : }
153 : }
154 :
155 : #[inline(always)]
156 0 : fn get_state(request: &Request<Body>) -> &State {
157 0 : request
158 0 : .data::<Arc<State>>()
159 0 : .expect("unknown state type")
160 0 : .as_ref()
161 0 : }
162 :
163 : #[inline(always)]
164 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
165 0 : get_state(request).conf
166 0 : }
167 :
168 : /// Check that the requester is authorized to operate on given tenant
169 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
170 0 : check_permission_with(request, |claims| {
171 0 : crate::auth::check_permission(claims, tenant_id)
172 0 : })
173 0 : }
174 :
175 : impl From<PageReconstructError> for ApiError {
176 0 : fn from(pre: PageReconstructError) -> ApiError {
177 0 : match pre {
178 0 : PageReconstructError::Other(other) => ApiError::InternalServerError(other),
179 0 : PageReconstructError::MissingKey(e) => ApiError::InternalServerError(e.into()),
180 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
181 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
182 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
183 : }
184 0 : }
185 : }
186 :
187 : impl From<TenantMapInsertError> for ApiError {
188 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
189 0 : match tmie {
190 0 : TenantMapInsertError::SlotError(e) => e.into(),
191 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
192 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
193 : }
194 0 : }
195 : }
196 :
197 : impl From<TenantSlotError> for ApiError {
198 0 : fn from(e: TenantSlotError) -> ApiError {
199 : use TenantSlotError::*;
200 0 : match e {
201 0 : NotFound(tenant_id) => {
202 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
203 : }
204 : InProgress => {
205 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
206 : }
207 0 : MapState(e) => e.into(),
208 : }
209 0 : }
210 : }
211 :
212 : impl From<TenantSlotUpsertError> for ApiError {
213 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
214 : use TenantSlotUpsertError::*;
215 0 : match e {
216 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
217 0 : MapState(e) => e.into(),
218 0 : ShuttingDown(_) => ApiError::ShuttingDown,
219 : }
220 0 : }
221 : }
222 :
223 : impl From<UpsertLocationError> for ApiError {
224 0 : fn from(e: UpsertLocationError) -> ApiError {
225 : use UpsertLocationError::*;
226 0 : match e {
227 0 : BadRequest(e) => ApiError::BadRequest(e),
228 0 : Unavailable(_) => ApiError::ShuttingDown,
229 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
230 0 : Flush(e) | InternalError(e) => ApiError::InternalServerError(e),
231 : }
232 0 : }
233 : }
234 :
235 : impl From<TenantMapError> for ApiError {
236 0 : fn from(e: TenantMapError) -> ApiError {
237 : use TenantMapError::*;
238 0 : match e {
239 : StillInitializing | ShuttingDown => {
240 0 : ApiError::ResourceUnavailable(format!("{e}").into())
241 : }
242 : }
243 0 : }
244 : }
245 :
246 : impl From<TenantStateError> for ApiError {
247 0 : fn from(tse: TenantStateError) -> ApiError {
248 0 : match tse {
249 : TenantStateError::IsStopping(_) => {
250 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
251 : }
252 0 : TenantStateError::SlotError(e) => e.into(),
253 0 : TenantStateError::SlotUpsertError(e) => e.into(),
254 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
255 : }
256 0 : }
257 : }
258 :
259 : impl From<GetTenantError> for ApiError {
260 0 : fn from(tse: GetTenantError) -> ApiError {
261 0 : match tse {
262 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {tid}").into()),
263 0 : GetTenantError::ShardNotFound(tid) => {
264 0 : ApiError::NotFound(anyhow!("tenant {tid}").into())
265 : }
266 : GetTenantError::NotActive(_) => {
267 : // Why is this not `ApiError::NotFound`?
268 : // Because we must be careful to never return 404 for a tenant if it does
269 : // in fact exist locally. If we did, the caller could draw the conclusion
270 : // that it can attach the tenant to another PS and we'd be in split-brain.
271 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
272 : }
273 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
274 : }
275 0 : }
276 : }
277 :
278 : impl From<GetTimelineError> for ApiError {
279 0 : fn from(gte: GetTimelineError) -> Self {
280 : // Rationale: tenant is activated only after eligble timelines activate
281 0 : ApiError::NotFound(gte.into())
282 0 : }
283 : }
284 :
285 : impl From<GetActiveTenantError> for ApiError {
286 0 : fn from(e: GetActiveTenantError) -> ApiError {
287 0 : match e {
288 0 : GetActiveTenantError::Broken(reason) => {
289 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
290 : }
291 : GetActiveTenantError::WillNotBecomeActive(TenantState::Stopping { .. }) => {
292 0 : ApiError::ShuttingDown
293 : }
294 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{e}")),
295 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
296 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
297 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
298 0 : ApiError::ResourceUnavailable(format!("{e}").into())
299 : }
300 : GetActiveTenantError::SwitchedTenant => {
301 : // in our HTTP handlers, this error doesn't happen
302 : // TODO: separate error types
303 0 : ApiError::ResourceUnavailable("switched tenant".into())
304 : }
305 : }
306 0 : }
307 : }
308 :
309 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
310 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
311 : use crate::tenant::DeleteTimelineError::*;
312 0 : match value {
313 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
314 0 : HasChildren(children) => ApiError::PreconditionFailed(
315 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
316 0 : .into_boxed_str(),
317 0 : ),
318 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
319 0 : Cancelled => ApiError::ResourceUnavailable("shutting down".into()),
320 0 : Other(e) => ApiError::InternalServerError(e),
321 : }
322 0 : }
323 : }
324 :
325 : impl From<crate::tenant::TimelineArchivalError> for ApiError {
326 0 : fn from(value: crate::tenant::TimelineArchivalError) -> Self {
327 : use crate::tenant::TimelineArchivalError::*;
328 0 : match value {
329 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
330 0 : Timeout => ApiError::Timeout("hit pageserver internal timeout".into()),
331 0 : Cancelled => ApiError::ShuttingDown,
332 0 : e @ HasArchivedParent(_) => {
333 0 : ApiError::PreconditionFailed(e.to_string().into_boxed_str())
334 : }
335 0 : HasUnarchivedChildren(children) => ApiError::PreconditionFailed(
336 0 : format!(
337 0 : "Cannot archive timeline which has non-archived child timelines: {children:?}"
338 0 : )
339 0 : .into_boxed_str(),
340 0 : ),
341 0 : a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
342 0 : Other(e) => ApiError::InternalServerError(e),
343 : }
344 0 : }
345 : }
346 :
347 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
348 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
349 : use crate::tenant::mgr::DeleteTimelineError::*;
350 0 : match value {
351 : // Report Precondition failed so client can distinguish between
352 : // "tenant is missing" case from "timeline is missing"
353 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
354 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
355 0 : ),
356 0 : Tenant(t) => ApiError::from(t),
357 0 : Timeline(t) => ApiError::from(t),
358 : }
359 0 : }
360 : }
361 :
362 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
363 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
364 : use crate::tenant::mgr::DeleteTenantError::*;
365 0 : match value {
366 0 : SlotError(e) => e.into(),
367 0 : Other(o) => ApiError::InternalServerError(o),
368 0 : Cancelled => ApiError::ShuttingDown,
369 : }
370 0 : }
371 : }
372 :
373 : impl From<crate::tenant::secondary::SecondaryTenantError> for ApiError {
374 0 : fn from(ste: crate::tenant::secondary::SecondaryTenantError) -> ApiError {
375 : use crate::tenant::secondary::SecondaryTenantError;
376 0 : match ste {
377 0 : SecondaryTenantError::GetTenant(gte) => gte.into(),
378 0 : SecondaryTenantError::ShuttingDown => ApiError::ShuttingDown,
379 : }
380 0 : }
381 : }
382 :
383 : impl From<crate::tenant::FinalizeTimelineImportError> for ApiError {
384 0 : fn from(err: crate::tenant::FinalizeTimelineImportError) -> ApiError {
385 : use crate::tenant::FinalizeTimelineImportError::*;
386 0 : match err {
387 : ImportTaskStillRunning => {
388 0 : ApiError::ResourceUnavailable("Import task still running".into())
389 : }
390 0 : ShuttingDown => ApiError::ShuttingDown,
391 : }
392 0 : }
393 : }
394 :
395 : // Helper function to construct a TimelineInfo struct for a timeline
396 0 : async fn build_timeline_info(
397 0 : timeline: &Arc<Timeline>,
398 0 : include_non_incremental_logical_size: bool,
399 0 : force_await_initial_logical_size: bool,
400 0 : include_image_consistent_lsn: bool,
401 0 : ctx: &RequestContext,
402 0 : ) -> anyhow::Result<TimelineInfo> {
403 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
404 :
405 0 : if force_await_initial_logical_size {
406 0 : timeline.clone().await_initial_logical_size().await
407 0 : }
408 :
409 0 : let mut info = build_timeline_info_common(
410 0 : timeline,
411 0 : ctx,
412 0 : tenant::timeline::GetLogicalSizePriority::Background,
413 0 : )
414 0 : .await?;
415 0 : if include_non_incremental_logical_size {
416 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
417 : // Otherwise, if someone deletes the timeline / detaches the tenant while
418 : // we're executing this function, we will outlive the timeline on-disk state.
419 : info.current_logical_size_non_incremental = Some(
420 0 : timeline
421 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
422 0 : .await?,
423 : );
424 0 : }
425 : // HADRON
426 0 : if include_image_consistent_lsn {
427 0 : info.image_consistent_lsn = Some(timeline.compute_image_consistent_lsn().await?);
428 0 : }
429 0 : Ok(info)
430 0 : }
431 :
432 0 : async fn build_timeline_info_common(
433 0 : timeline: &Arc<Timeline>,
434 0 : ctx: &RequestContext,
435 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
436 0 : ) -> anyhow::Result<TimelineInfo> {
437 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
438 0 : let initdb_lsn = timeline.initdb_lsn;
439 0 : let last_record_lsn = timeline.get_last_record_lsn();
440 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
441 0 : let guard = timeline.last_received_wal.lock().unwrap();
442 0 : if let Some(info) = guard.as_ref() {
443 0 : (
444 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
445 0 : Some(info.last_received_msg_lsn),
446 0 : Some(info.last_received_msg_ts),
447 0 : )
448 : } else {
449 0 : (None, None, None)
450 : }
451 : };
452 :
453 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
454 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
455 0 : Lsn(0) => None,
456 0 : lsn @ Lsn(_) => Some(lsn),
457 : };
458 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
459 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
460 0 : let state = timeline.current_state();
461 : // Report is_archived = false if the timeline is still loading
462 0 : let is_archived = timeline.is_archived().unwrap_or(false);
463 0 : let remote_consistent_lsn_projected = timeline
464 0 : .get_remote_consistent_lsn_projected()
465 0 : .unwrap_or(Lsn(0));
466 0 : let remote_consistent_lsn_visible = timeline
467 0 : .get_remote_consistent_lsn_visible()
468 0 : .unwrap_or(Lsn(0));
469 0 : let is_invisible = timeline.remote_client.is_invisible().unwrap_or(false);
470 :
471 0 : let walreceiver_status = timeline.walreceiver_status();
472 :
473 0 : let (pitr_history_size, within_ancestor_pitr) = timeline.get_pitr_history_stats();
474 :
475 : // Externally, expose the lowest LSN that can be used to create a branch.
476 : // Internally we distinguish between the planned GC cutoff (PITR point) and the "applied" GC cutoff (where we
477 : // actually trimmed data to), which can pass each other when PITR is changed.
478 0 : let min_readable_lsn = std::cmp::max(
479 0 : timeline.get_gc_cutoff_lsn().unwrap_or_default(),
480 0 : *timeline.get_applied_gc_cutoff_lsn(),
481 : );
482 :
483 0 : let info = TimelineInfo {
484 0 : tenant_id: timeline.tenant_shard_id,
485 0 : timeline_id: timeline.timeline_id,
486 0 : ancestor_timeline_id,
487 0 : ancestor_lsn,
488 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
489 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
490 0 : remote_consistent_lsn_visible,
491 0 : initdb_lsn,
492 0 : last_record_lsn,
493 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
494 0 : min_readable_lsn,
495 0 : applied_gc_cutoff_lsn: *timeline.get_applied_gc_cutoff_lsn(),
496 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
497 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
498 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
499 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
500 : },
501 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
502 0 : current_physical_size,
503 0 : current_logical_size_non_incremental: None,
504 0 : pitr_history_size,
505 0 : within_ancestor_pitr,
506 0 : timeline_dir_layer_file_size_sum: None,
507 0 : wal_source_connstr,
508 0 : last_received_msg_lsn,
509 0 : last_received_msg_ts,
510 0 : pg_version: timeline.pg_version,
511 :
512 0 : state,
513 0 : is_archived: Some(is_archived),
514 0 : rel_size_migration: Some(timeline.get_rel_size_v2_status()),
515 0 : is_invisible: Some(is_invisible),
516 :
517 0 : walreceiver_status,
518 : // HADRON
519 0 : image_consistent_lsn: None,
520 : };
521 0 : Ok(info)
522 0 : }
523 :
524 0 : fn build_timeline_offloaded_info(offloaded: &Arc<OffloadedTimeline>) -> OffloadedTimelineInfo {
525 : let &OffloadedTimeline {
526 0 : tenant_shard_id,
527 0 : timeline_id,
528 0 : ancestor_retain_lsn,
529 0 : ancestor_timeline_id,
530 0 : archived_at,
531 : ..
532 0 : } = offloaded.as_ref();
533 0 : OffloadedTimelineInfo {
534 0 : tenant_id: tenant_shard_id,
535 0 : timeline_id,
536 0 : ancestor_retain_lsn,
537 0 : ancestor_timeline_id,
538 0 : archived_at: archived_at.and_utc(),
539 0 : }
540 0 : }
541 :
542 : // healthcheck handler
543 0 : async fn status_handler(
544 0 : request: Request<Body>,
545 0 : _cancel: CancellationToken,
546 0 : ) -> Result<Response<Body>, ApiError> {
547 0 : check_permission(&request, None)?;
548 0 : let config = get_config(&request);
549 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
550 0 : }
551 :
552 0 : async fn reload_auth_validation_keys_handler(
553 0 : request: Request<Body>,
554 0 : _cancel: CancellationToken,
555 0 : ) -> Result<Response<Body>, ApiError> {
556 0 : check_permission(&request, None)?;
557 0 : let config = get_config(&request);
558 0 : let state = get_state(&request);
559 0 : let Some(shared_auth) = &state.auth else {
560 0 : return json_response(StatusCode::BAD_REQUEST, ());
561 : };
562 : // unwrap is ok because check is performed when creating config, so path is set and exists
563 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
564 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
565 :
566 0 : match utils::auth::JwtAuth::from_key_path(key_path) {
567 0 : Ok(new_auth) => {
568 0 : shared_auth.swap(new_auth);
569 0 : json_response(StatusCode::OK, ())
570 : }
571 0 : Err(e) => {
572 0 : let err_msg = "Error reloading public keys";
573 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
574 0 : json_response(
575 : StatusCode::INTERNAL_SERVER_ERROR,
576 0 : HttpErrorBody::from_msg(err_msg.to_string()),
577 : )
578 : }
579 : }
580 0 : }
581 :
582 0 : async fn timeline_create_handler(
583 0 : mut request: Request<Body>,
584 0 : _cancel: CancellationToken,
585 0 : ) -> Result<Response<Body>, ApiError> {
586 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
587 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
588 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
589 :
590 0 : let new_timeline_id = request_data.new_timeline_id;
591 : // fill in the default pg_version if not provided & convert request into domain model
592 0 : let params: tenant::CreateTimelineParams = match request_data.mode {
593 : TimelineCreateRequestMode::Bootstrap {
594 0 : existing_initdb_timeline_id,
595 0 : pg_version,
596 0 : } => tenant::CreateTimelineParams::Bootstrap(tenant::CreateTimelineParamsBootstrap {
597 0 : new_timeline_id,
598 0 : existing_initdb_timeline_id,
599 0 : pg_version: pg_version.unwrap_or(DEFAULT_PG_VERSION),
600 0 : }),
601 : TimelineCreateRequestMode::Branch {
602 0 : ancestor_timeline_id,
603 0 : ancestor_start_lsn,
604 : read_only: _,
605 : pg_version: _,
606 0 : } => tenant::CreateTimelineParams::Branch(tenant::CreateTimelineParamsBranch {
607 0 : new_timeline_id,
608 0 : ancestor_timeline_id,
609 0 : ancestor_start_lsn,
610 0 : }),
611 : TimelineCreateRequestMode::ImportPgdata {
612 : import_pgdata:
613 : TimelineCreateRequestModeImportPgdata {
614 0 : location,
615 0 : idempotency_key,
616 : },
617 : } => tenant::CreateTimelineParams::ImportPgdata(tenant::CreateTimelineParamsImportPgdata {
618 0 : idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new(
619 0 : idempotency_key.0,
620 : ),
621 0 : new_timeline_id,
622 : location: {
623 : use import_pgdata::index_part_format::Location;
624 : use pageserver_api::models::ImportPgdataLocation;
625 0 : match location {
626 : #[cfg(feature = "testing")]
627 0 : ImportPgdataLocation::LocalFs { path } => Location::LocalFs { path },
628 : ImportPgdataLocation::AwsS3 {
629 0 : region,
630 0 : bucket,
631 0 : key,
632 0 : } => Location::AwsS3 {
633 0 : region,
634 0 : bucket,
635 0 : key,
636 0 : },
637 : }
638 : },
639 : }),
640 : };
641 :
642 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
643 :
644 0 : let state = get_state(&request);
645 :
646 0 : async {
647 0 : let tenant = state
648 0 : .tenant_manager
649 0 : .get_attached_tenant_shard(tenant_shard_id)?;
650 :
651 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
652 :
653 : // earlier versions of the code had pg_version and ancestor_lsn in the span
654 : // => continue to provide that information, but, through a log message that doesn't require us to destructure
655 0 : tracing::info!(?params, "creating timeline");
656 :
657 0 : match tenant
658 0 : .create_timeline(params, state.broker_client.clone(), &ctx)
659 0 : .await
660 : {
661 0 : Ok(new_timeline) => {
662 : // Created. Construct a TimelineInfo for it.
663 0 : let timeline_info = build_timeline_info_common(
664 0 : &new_timeline,
665 0 : &ctx,
666 0 : tenant::timeline::GetLogicalSizePriority::User,
667 0 : )
668 0 : .await
669 0 : .map_err(ApiError::InternalServerError)?;
670 0 : json_response(StatusCode::CREATED, timeline_info)
671 : }
672 0 : Err(_) if tenant.cancel.is_cancelled() => {
673 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
674 0 : json_response(
675 : StatusCode::SERVICE_UNAVAILABLE,
676 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
677 : )
678 : }
679 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
680 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
681 : }
682 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
683 : StatusCode::TOO_MANY_REQUESTS,
684 0 : HttpErrorBody::from_msg(e.to_string()),
685 : ),
686 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
687 : StatusCode::NOT_ACCEPTABLE,
688 0 : HttpErrorBody::from_msg(format!("{err:#}")),
689 : ),
690 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
691 : StatusCode::SERVICE_UNAVAILABLE,
692 0 : HttpErrorBody::from_msg(e.to_string()),
693 : ),
694 0 : Err(e @ tenant::CreateTimelineError::AncestorArchived) => json_response(
695 : StatusCode::NOT_ACCEPTABLE,
696 0 : HttpErrorBody::from_msg(e.to_string()),
697 : ),
698 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
699 : StatusCode::SERVICE_UNAVAILABLE,
700 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
701 : ),
702 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
703 : }
704 0 : }
705 0 : .instrument(info_span!("timeline_create",
706 : tenant_id = %tenant_shard_id.tenant_id,
707 0 : shard_id = %tenant_shard_id.shard_slug(),
708 : timeline_id = %new_timeline_id,
709 : ))
710 0 : .await
711 0 : }
712 :
713 0 : async fn timeline_list_handler(
714 0 : request: Request<Body>,
715 0 : _cancel: CancellationToken,
716 0 : ) -> Result<Response<Body>, ApiError> {
717 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
718 0 : let include_non_incremental_logical_size: Option<bool> =
719 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
720 0 : let force_await_initial_logical_size: Option<bool> =
721 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
722 0 : let include_image_consistent_lsn: Option<bool> =
723 0 : parse_query_param(&request, "include-image-consistent-lsn")?;
724 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
725 :
726 0 : let state = get_state(&request);
727 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
728 :
729 0 : let response_data = async {
730 0 : let tenant = state
731 0 : .tenant_manager
732 0 : .get_attached_tenant_shard(tenant_shard_id)?;
733 :
734 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
735 :
736 0 : let timelines = tenant.list_timelines();
737 :
738 0 : let mut response_data = Vec::with_capacity(timelines.len());
739 0 : for timeline in timelines {
740 0 : let timeline_info = build_timeline_info(
741 0 : &timeline,
742 0 : include_non_incremental_logical_size.unwrap_or(false),
743 0 : force_await_initial_logical_size.unwrap_or(false),
744 0 : include_image_consistent_lsn.unwrap_or(false),
745 0 : &ctx,
746 : )
747 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
748 0 : .await
749 0 : .context("Failed to build timeline info")
750 0 : .map_err(ApiError::InternalServerError)?;
751 :
752 0 : response_data.push(timeline_info);
753 : }
754 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
755 0 : }
756 0 : .instrument(info_span!("timeline_list",
757 : tenant_id = %tenant_shard_id.tenant_id,
758 0 : shard_id = %tenant_shard_id.shard_slug()))
759 0 : .await?;
760 :
761 0 : json_response(StatusCode::OK, response_data)
762 0 : }
763 :
764 0 : async fn timeline_and_offloaded_list_handler(
765 0 : request: Request<Body>,
766 0 : _cancel: CancellationToken,
767 0 : ) -> Result<Response<Body>, ApiError> {
768 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
769 0 : let include_non_incremental_logical_size: Option<bool> =
770 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
771 0 : let force_await_initial_logical_size: Option<bool> =
772 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
773 0 : let include_image_consistent_lsn: Option<bool> =
774 0 : parse_query_param(&request, "include-image-consistent-lsn")?;
775 :
776 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
777 :
778 0 : let state = get_state(&request);
779 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
780 :
781 0 : let response_data = async {
782 0 : let tenant = state
783 0 : .tenant_manager
784 0 : .get_attached_tenant_shard(tenant_shard_id)?;
785 :
786 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
787 :
788 0 : let (timelines, offloadeds) = tenant.list_timelines_and_offloaded();
789 :
790 0 : let mut timeline_infos = Vec::with_capacity(timelines.len());
791 0 : for timeline in timelines {
792 0 : let timeline_info = build_timeline_info(
793 0 : &timeline,
794 0 : include_non_incremental_logical_size.unwrap_or(false),
795 0 : force_await_initial_logical_size.unwrap_or(false),
796 0 : include_image_consistent_lsn.unwrap_or(false),
797 0 : &ctx,
798 : )
799 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
800 0 : .await
801 0 : .context("Failed to build timeline info")
802 0 : .map_err(ApiError::InternalServerError)?;
803 :
804 0 : timeline_infos.push(timeline_info);
805 : }
806 0 : let offloaded_infos = offloadeds
807 0 : .into_iter()
808 0 : .map(|offloaded| build_timeline_offloaded_info(&offloaded))
809 0 : .collect::<Vec<_>>();
810 0 : let res = TimelinesInfoAndOffloaded {
811 0 : timelines: timeline_infos,
812 0 : offloaded: offloaded_infos,
813 0 : };
814 0 : Ok::<TimelinesInfoAndOffloaded, ApiError>(res)
815 0 : }
816 0 : .instrument(info_span!("timeline_and_offloaded_list",
817 : tenant_id = %tenant_shard_id.tenant_id,
818 0 : shard_id = %tenant_shard_id.shard_slug()))
819 0 : .await?;
820 :
821 0 : json_response(StatusCode::OK, response_data)
822 0 : }
823 :
824 0 : async fn timeline_preserve_initdb_handler(
825 0 : request: Request<Body>,
826 0 : _cancel: CancellationToken,
827 0 : ) -> Result<Response<Body>, ApiError> {
828 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
829 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
830 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
831 0 : let state = get_state(&request);
832 :
833 : // Part of the process for disaster recovery from safekeeper-stored WAL:
834 : // If we don't recover into a new timeline but want to keep the timeline ID,
835 : // then the initdb archive is deleted. This endpoint copies it to a different
836 : // location where timeline recreation cand find it.
837 :
838 0 : async {
839 0 : let tenant = state
840 0 : .tenant_manager
841 0 : .get_attached_tenant_shard(tenant_shard_id)?;
842 :
843 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
844 :
845 0 : timeline
846 0 : .preserve_initdb_archive()
847 0 : .await
848 0 : .context("preserving initdb archive")
849 0 : .map_err(ApiError::InternalServerError)?;
850 :
851 0 : Ok::<_, ApiError>(())
852 0 : }
853 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
854 : tenant_id = %tenant_shard_id.tenant_id,
855 0 : shard_id = %tenant_shard_id.shard_slug(),
856 : %timeline_id))
857 0 : .await?;
858 :
859 0 : json_response(StatusCode::OK, ())
860 0 : }
861 :
862 0 : async fn timeline_archival_config_handler(
863 0 : mut request: Request<Body>,
864 0 : _cancel: CancellationToken,
865 0 : ) -> Result<Response<Body>, ApiError> {
866 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
867 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
868 :
869 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
870 :
871 0 : let request_data: TimelineArchivalConfigRequest = json_request(&mut request).await?;
872 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
873 0 : let state = get_state(&request);
874 :
875 0 : async {
876 0 : let tenant = state
877 0 : .tenant_manager
878 0 : .get_attached_tenant_shard(tenant_shard_id)?;
879 :
880 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
881 :
882 0 : tenant
883 0 : .apply_timeline_archival_config(
884 0 : timeline_id,
885 0 : request_data.state,
886 0 : state.broker_client.clone(),
887 0 : ctx,
888 0 : )
889 0 : .await?;
890 0 : Ok::<_, ApiError>(())
891 0 : }
892 0 : .instrument(info_span!("timeline_archival_config",
893 : tenant_id = %tenant_shard_id.tenant_id,
894 0 : shard_id = %tenant_shard_id.shard_slug(),
895 : state = ?request_data.state,
896 : %timeline_id))
897 0 : .await?;
898 :
899 0 : json_response(StatusCode::OK, ())
900 0 : }
901 :
902 : /// This API is used to patch the index part of a timeline. You must ensure such patches are safe to apply. Use this API as an emergency
903 : /// measure only.
904 : ///
905 : /// Some examples of safe patches:
906 : /// - Increase the gc_cutoff and gc_compaction_cutoff to a larger value in case of a bug that didn't bump the cutoff and cause read errors.
907 : /// - Force set the index part to use reldir v2 (migrating/migrated).
908 : ///
909 : /// Some examples of unsafe patches:
910 : /// - Force set the index part from v2 to v1 (legacy). This will cause the code path to ignore anything written to the new keyspace and cause
911 : /// errors.
912 : /// - Decrease the gc_cutoff without validating the data really exists. It will cause read errors in the background.
913 0 : async fn timeline_patch_index_part_handler(
914 0 : mut request: Request<Body>,
915 0 : _cancel: CancellationToken,
916 0 : ) -> Result<Response<Body>, ApiError> {
917 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
918 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
919 :
920 0 : let request_data: TimelinePatchIndexPartRequest = json_request(&mut request).await?;
921 0 : check_permission(&request, None)?; // require global permission for this request
922 0 : let state = get_state(&request);
923 :
924 0 : async {
925 0 : let timeline =
926 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
927 0 : .await?;
928 :
929 0 : if let Some(rel_size_migration) = request_data.rel_size_migration {
930 0 : timeline
931 0 : .update_rel_size_v2_status(rel_size_migration)
932 0 : .map_err(ApiError::InternalServerError)?;
933 0 : }
934 :
935 0 : if let Some(gc_compaction_last_completed_lsn) =
936 0 : request_data.gc_compaction_last_completed_lsn
937 : {
938 0 : timeline
939 0 : .update_gc_compaction_state(GcCompactionState {
940 0 : last_completed_lsn: gc_compaction_last_completed_lsn,
941 0 : })
942 0 : .map_err(ApiError::InternalServerError)?;
943 0 : }
944 :
945 0 : if let Some(applied_gc_cutoff_lsn) = request_data.applied_gc_cutoff_lsn {
946 0 : {
947 0 : let guard = timeline.applied_gc_cutoff_lsn.lock_for_write();
948 0 : guard.store_and_unlock(applied_gc_cutoff_lsn);
949 0 : }
950 0 : }
951 :
952 0 : if request_data.force_index_update {
953 0 : timeline
954 0 : .remote_client
955 0 : .force_schedule_index_upload()
956 0 : .context("force schedule index upload")
957 0 : .map_err(ApiError::InternalServerError)?;
958 0 : }
959 :
960 0 : Ok::<_, ApiError>(())
961 0 : }
962 0 : .instrument(info_span!("timeline_patch_index_part",
963 : tenant_id = %tenant_shard_id.tenant_id,
964 0 : shard_id = %tenant_shard_id.shard_slug(),
965 : %timeline_id))
966 0 : .await?;
967 :
968 0 : json_response(StatusCode::OK, ())
969 0 : }
970 :
971 0 : async fn timeline_detail_handler(
972 0 : request: Request<Body>,
973 0 : _cancel: CancellationToken,
974 0 : ) -> Result<Response<Body>, ApiError> {
975 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
976 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
977 0 : let include_non_incremental_logical_size: Option<bool> =
978 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
979 0 : let force_await_initial_logical_size: Option<bool> =
980 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
981 : // HADRON
982 0 : let include_image_consistent_lsn: Option<bool> =
983 0 : parse_query_param(&request, "include-image-consistent-lsn")?;
984 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
985 :
986 : // Logical size calculation needs downloading.
987 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
988 0 : let state = get_state(&request);
989 :
990 0 : let timeline_info = async {
991 0 : let tenant = state
992 0 : .tenant_manager
993 0 : .get_attached_tenant_shard(tenant_shard_id)?;
994 :
995 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
996 :
997 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
998 0 : let ctx = &ctx.with_scope_timeline(&timeline);
999 :
1000 0 : let timeline_info = build_timeline_info(
1001 0 : &timeline,
1002 0 : include_non_incremental_logical_size.unwrap_or(false),
1003 0 : force_await_initial_logical_size.unwrap_or(false),
1004 0 : include_image_consistent_lsn.unwrap_or(false),
1005 0 : ctx,
1006 0 : )
1007 0 : .await
1008 0 : .context("get local timeline info")
1009 0 : .map_err(ApiError::InternalServerError)?;
1010 :
1011 0 : Ok::<_, ApiError>(timeline_info)
1012 0 : }
1013 0 : .instrument(info_span!("timeline_detail",
1014 : tenant_id = %tenant_shard_id.tenant_id,
1015 0 : shard_id = %tenant_shard_id.shard_slug(),
1016 : %timeline_id))
1017 0 : .await?;
1018 :
1019 0 : json_response(StatusCode::OK, timeline_info)
1020 0 : }
1021 :
1022 0 : async fn get_lsn_by_timestamp_handler(
1023 0 : request: Request<Body>,
1024 0 : cancel: CancellationToken,
1025 0 : ) -> Result<Response<Body>, ApiError> {
1026 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1027 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1028 0 : let state = get_state(&request);
1029 :
1030 0 : if !tenant_shard_id.is_shard_zero() {
1031 : // Requires SLRU contents, which are only stored on shard zero
1032 0 : return Err(ApiError::BadRequest(anyhow!(
1033 0 : "Lsn calculations by timestamp are only available on shard zero"
1034 0 : )));
1035 0 : }
1036 :
1037 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1038 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
1039 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1040 0 : .with_context(|| format!("Invalid time: {timestamp_raw:?}"))
1041 0 : .map_err(ApiError::BadRequest)?;
1042 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
1043 :
1044 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
1045 :
1046 0 : let timeline =
1047 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1048 0 : .await?;
1049 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1050 0 : .with_scope_timeline(&timeline);
1051 0 : let result = timeline
1052 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
1053 0 : .await?;
1054 :
1055 : #[derive(serde::Serialize, Debug)]
1056 : struct Result {
1057 : lsn: Lsn,
1058 : kind: &'static str,
1059 : #[serde(default)]
1060 : #[serde(skip_serializing_if = "Option::is_none")]
1061 : #[serde(flatten)]
1062 : lease: Option<LsnLease>,
1063 : }
1064 0 : let (lsn, kind) = match result {
1065 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
1066 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
1067 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
1068 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
1069 : };
1070 :
1071 0 : let lease = if with_lease {
1072 0 : timeline
1073 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
1074 0 : .inspect_err(|_| {
1075 0 : warn!("fail to grant a lease to {}", lsn);
1076 0 : })
1077 0 : .ok()
1078 : } else {
1079 0 : None
1080 : };
1081 :
1082 0 : let result = Result { lsn, kind, lease };
1083 0 : let valid_until = result
1084 0 : .lease
1085 0 : .as_ref()
1086 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
1087 0 : tracing::info!(
1088 : lsn=?result.lsn,
1089 : kind=%result.kind,
1090 : timestamp=%timestamp_raw,
1091 : valid_until=?valid_until,
1092 0 : "lsn_by_timestamp finished"
1093 : );
1094 0 : json_response(StatusCode::OK, result)
1095 0 : }
1096 :
1097 0 : async fn get_timestamp_of_lsn_handler(
1098 0 : request: Request<Body>,
1099 0 : _cancel: CancellationToken,
1100 0 : ) -> Result<Response<Body>, ApiError> {
1101 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1102 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1103 0 : let state = get_state(&request);
1104 :
1105 0 : if !tenant_shard_id.is_shard_zero() {
1106 : // Requires SLRU contents, which are only stored on shard zero
1107 0 : return Err(ApiError::BadRequest(anyhow!(
1108 0 : "Timestamp calculations by lsn are only available on shard zero"
1109 0 : )));
1110 0 : }
1111 :
1112 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1113 :
1114 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
1115 0 : let lsn = Lsn::from_str(&lsn_str)
1116 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
1117 0 : .map_err(ApiError::BadRequest)?;
1118 :
1119 0 : let timeline =
1120 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1121 0 : .await?;
1122 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1123 0 : .with_scope_timeline(&timeline);
1124 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
1125 :
1126 0 : match result {
1127 0 : Some(time) => {
1128 0 : let time = format_rfc3339(
1129 0 : postgres_ffi::try_from_pg_timestamp(time).map_err(ApiError::InternalServerError)?,
1130 : )
1131 0 : .to_string();
1132 0 : json_response(StatusCode::OK, time)
1133 : }
1134 0 : None => Err(ApiError::PreconditionFailed(
1135 0 : format!("Timestamp for lsn {lsn} not found").into(),
1136 0 : )),
1137 : }
1138 0 : }
1139 :
1140 0 : async fn timeline_delete_handler(
1141 0 : request: Request<Body>,
1142 0 : _cancel: CancellationToken,
1143 0 : ) -> Result<Response<Body>, ApiError> {
1144 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1145 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1146 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1147 :
1148 0 : let state = get_state(&request);
1149 :
1150 0 : let tenant = state
1151 0 : .tenant_manager
1152 0 : .get_attached_tenant_shard(tenant_shard_id)
1153 0 : .map_err(|e| {
1154 0 : match e {
1155 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
1156 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
1157 : GetTenantError::NotFound(_) | GetTenantError::ShardNotFound(_) => {
1158 0 : ApiError::PreconditionFailed(
1159 0 : "Requested tenant is missing".to_string().into_boxed_str(),
1160 0 : )
1161 : }
1162 0 : e => e.into(),
1163 : }
1164 0 : })?;
1165 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1166 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))
1167 0 : .await?;
1168 :
1169 0 : json_response(StatusCode::ACCEPTED, ())
1170 0 : }
1171 :
1172 0 : async fn tenant_reset_handler(
1173 0 : request: Request<Body>,
1174 0 : _cancel: CancellationToken,
1175 0 : ) -> Result<Response<Body>, ApiError> {
1176 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1177 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1178 :
1179 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
1180 :
1181 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1182 0 : let state = get_state(&request);
1183 0 : state
1184 0 : .tenant_manager
1185 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
1186 0 : .await
1187 0 : .map_err(ApiError::InternalServerError)?;
1188 :
1189 0 : json_response(StatusCode::OK, ())
1190 0 : }
1191 :
1192 0 : async fn tenant_list_handler(
1193 0 : request: Request<Body>,
1194 0 : _cancel: CancellationToken,
1195 0 : ) -> Result<Response<Body>, ApiError> {
1196 0 : check_permission(&request, None)?;
1197 0 : let state = get_state(&request);
1198 :
1199 0 : let response_data = state
1200 0 : .tenant_manager
1201 0 : .list_tenants()
1202 0 : .map_err(|_| {
1203 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
1204 0 : })?
1205 0 : .iter()
1206 0 : .map(|(id, state, gen_)| TenantInfo {
1207 0 : id: *id,
1208 0 : state: state.clone(),
1209 0 : current_physical_size: None,
1210 0 : attachment_status: state.attachment_status(),
1211 0 : generation: (*gen_)
1212 0 : .into()
1213 0 : .expect("Tenants are always attached with a generation"),
1214 0 : gc_blocking: None,
1215 0 : })
1216 0 : .collect::<Vec<TenantInfo>>();
1217 :
1218 0 : json_response(StatusCode::OK, response_data)
1219 0 : }
1220 :
1221 0 : async fn tenant_status(
1222 0 : request: Request<Body>,
1223 0 : _cancel: CancellationToken,
1224 0 : ) -> Result<Response<Body>, ApiError> {
1225 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1226 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1227 0 : let state = get_state(&request);
1228 :
1229 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1230 0 : let activate = true;
1231 : #[cfg(feature = "testing")]
1232 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1233 :
1234 0 : let tenant_info = async {
1235 0 : let tenant = state
1236 0 : .tenant_manager
1237 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1238 :
1239 0 : if activate {
1240 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1241 : // called, but it is still valid to return 200 and describe the current state of the tenant
1242 : // if it doesn't make it into an active state.
1243 0 : tenant
1244 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1245 0 : .await
1246 0 : .ok();
1247 0 : }
1248 :
1249 : // Calculate total physical size of all timelines
1250 0 : let mut current_physical_size = 0;
1251 0 : for timeline in tenant.list_timelines().iter() {
1252 0 : current_physical_size += timeline.layer_size_sum().await;
1253 : }
1254 :
1255 0 : let state = tenant.current_state();
1256 : Result::<_, ApiError>::Ok(TenantDetails {
1257 : tenant_info: TenantInfo {
1258 0 : id: tenant_shard_id,
1259 0 : state: state.clone(),
1260 0 : current_physical_size: Some(current_physical_size),
1261 0 : attachment_status: state.attachment_status(),
1262 0 : generation: tenant
1263 0 : .generation()
1264 0 : .into()
1265 0 : .expect("Tenants are always attached with a generation"),
1266 0 : gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
1267 : },
1268 0 : walredo: tenant.wal_redo_manager_status(),
1269 0 : timelines: tenant.list_timeline_ids(),
1270 : })
1271 0 : }
1272 0 : .instrument(info_span!("tenant_status_handler",
1273 : tenant_id = %tenant_shard_id.tenant_id,
1274 0 : shard_id = %tenant_shard_id.shard_slug()))
1275 0 : .await?;
1276 :
1277 0 : json_response(StatusCode::OK, tenant_info)
1278 0 : }
1279 :
1280 0 : async fn tenant_delete_handler(
1281 0 : request: Request<Body>,
1282 0 : _cancel: CancellationToken,
1283 0 : ) -> Result<Response<Body>, ApiError> {
1284 : // TODO openapi spec
1285 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1286 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1287 :
1288 0 : let state = get_state(&request);
1289 :
1290 0 : state
1291 0 : .tenant_manager
1292 0 : .delete_tenant(tenant_shard_id)
1293 0 : .instrument(info_span!("tenant_delete_handler",
1294 : tenant_id = %tenant_shard_id.tenant_id,
1295 0 : shard_id = %tenant_shard_id.shard_slug()
1296 : ))
1297 0 : .await?;
1298 :
1299 0 : json_response(StatusCode::OK, ())
1300 0 : }
1301 :
1302 : /// HTTP endpoint to query the current tenant_size of a tenant.
1303 : ///
1304 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1305 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1306 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1307 : /// values.
1308 : ///
1309 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1310 : /// (only if it is shorter than the real cutoff).
1311 : ///
1312 : /// Note: we don't update the cached size and prometheus metric here.
1313 : /// The retention period might be different, and it's nice to have a method to just calculate it
1314 : /// without modifying anything anyway.
1315 0 : async fn tenant_size_handler(
1316 0 : request: Request<Body>,
1317 0 : cancel: CancellationToken,
1318 0 : ) -> Result<Response<Body>, ApiError> {
1319 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1320 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1321 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1322 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1323 0 : let headers = request.headers();
1324 0 : let state = get_state(&request);
1325 :
1326 0 : if !tenant_shard_id.is_shard_zero() {
1327 0 : return Err(ApiError::BadRequest(anyhow!(
1328 0 : "Size calculations are only available on shard zero"
1329 0 : )));
1330 0 : }
1331 :
1332 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1333 0 : let tenant = state
1334 0 : .tenant_manager
1335 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1336 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1337 :
1338 : // this can be long operation
1339 0 : let inputs = tenant
1340 0 : .gather_size_inputs(
1341 0 : retention_period,
1342 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1343 0 : &cancel,
1344 0 : &ctx,
1345 0 : )
1346 0 : .await
1347 0 : .map_err(|e| match e {
1348 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1349 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1350 0 : })?;
1351 :
1352 0 : let mut sizes = None;
1353 0 : let accepts_html = headers
1354 0 : .get(header::ACCEPT)
1355 0 : .map(|v| v == "text/html")
1356 0 : .unwrap_or_default();
1357 0 : if !inputs_only.unwrap_or(false) {
1358 0 : let storage_model = inputs.calculate_model();
1359 0 : let size = storage_model.calculate();
1360 :
1361 : // If request header expects html, return html
1362 0 : if accepts_html {
1363 0 : return synthetic_size_html_response(inputs, storage_model, size);
1364 0 : }
1365 0 : sizes = Some(size);
1366 0 : } else if accepts_html {
1367 0 : return Err(ApiError::BadRequest(anyhow!(
1368 0 : "inputs_only parameter is incompatible with html output request"
1369 0 : )));
1370 0 : }
1371 :
1372 : /// The type resides in the pageserver not to expose `ModelInputs`.
1373 : #[derive(serde::Serialize)]
1374 : struct TenantHistorySize {
1375 : id: TenantId,
1376 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1377 : ///
1378 : /// Will be none if `?inputs_only=true` was given.
1379 : size: Option<u64>,
1380 : /// Size of each segment used in the model.
1381 : /// Will be null if `?inputs_only=true` was given.
1382 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1383 : inputs: crate::tenant::size::ModelInputs,
1384 : }
1385 :
1386 0 : json_response(
1387 : StatusCode::OK,
1388 : TenantHistorySize {
1389 0 : id: tenant_shard_id.tenant_id,
1390 0 : size: sizes.as_ref().map(|x| x.total_size),
1391 0 : segment_sizes: sizes.map(|x| x.segments),
1392 0 : inputs,
1393 : },
1394 : )
1395 0 : }
1396 :
1397 0 : async fn tenant_shard_split_handler(
1398 0 : mut request: Request<Body>,
1399 0 : _cancel: CancellationToken,
1400 0 : ) -> Result<Response<Body>, ApiError> {
1401 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1402 :
1403 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1404 0 : let state = get_state(&request);
1405 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1406 :
1407 0 : let tenant = state
1408 0 : .tenant_manager
1409 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1410 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1411 :
1412 0 : let new_shards = state
1413 0 : .tenant_manager
1414 0 : .shard_split(
1415 0 : tenant,
1416 0 : ShardCount::new(req.new_shard_count),
1417 0 : req.new_stripe_size,
1418 0 : &ctx,
1419 0 : )
1420 0 : .await
1421 0 : .map_err(ApiError::InternalServerError)?;
1422 :
1423 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1424 0 : }
1425 :
1426 0 : async fn layer_map_info_handler(
1427 0 : request: Request<Body>,
1428 0 : _cancel: CancellationToken,
1429 0 : ) -> Result<Response<Body>, ApiError> {
1430 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1431 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1432 0 : let reset: LayerAccessStatsReset =
1433 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1434 0 : let state = get_state(&request);
1435 :
1436 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1437 :
1438 0 : let timeline =
1439 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1440 0 : .await?;
1441 0 : let layer_map_info = timeline
1442 0 : .layer_map_info(reset)
1443 0 : .await
1444 0 : .map_err(|_shutdown| ApiError::ShuttingDown)?;
1445 :
1446 0 : json_response(StatusCode::OK, layer_map_info)
1447 0 : }
1448 :
1449 : #[instrument(skip_all, fields(tenant_id, shard_id, timeline_id, layer_name))]
1450 : async fn timeline_layer_scan_disposable_keys(
1451 : request: Request<Body>,
1452 : cancel: CancellationToken,
1453 : ) -> Result<Response<Body>, ApiError> {
1454 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1455 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1456 : let layer_name: LayerName = parse_request_param(&request, "layer_name")?;
1457 :
1458 : tracing::Span::current().record(
1459 : "tenant_id",
1460 : tracing::field::display(&tenant_shard_id.tenant_id),
1461 : );
1462 : tracing::Span::current().record(
1463 : "shard_id",
1464 : tracing::field::display(tenant_shard_id.shard_slug()),
1465 : );
1466 : tracing::Span::current().record("timeline_id", tracing::field::display(&timeline_id));
1467 : tracing::Span::current().record("layer_name", tracing::field::display(&layer_name));
1468 :
1469 : let state = get_state(&request);
1470 :
1471 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1472 :
1473 : // technically the timeline need not be active for this scan to complete
1474 : let timeline =
1475 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1476 : .await?;
1477 :
1478 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1479 : .with_scope_timeline(&timeline);
1480 :
1481 : let guard = timeline
1482 : .layers
1483 : .read(LayerManagerLockHolder::GetLayerMapInfo)
1484 : .await;
1485 : let Some(layer) = guard.try_get_from_key(&layer_name.clone().into()) else {
1486 : return Err(ApiError::NotFound(
1487 : anyhow::anyhow!("Layer {tenant_shard_id}/{timeline_id}/{layer_name} not found").into(),
1488 : ));
1489 : };
1490 :
1491 : let resident_layer = layer
1492 : .download_and_keep_resident(&ctx)
1493 : .await
1494 0 : .map_err(|err| match err {
1495 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1496 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1497 0 : ApiError::ShuttingDown
1498 : }
1499 : tenant::storage_layer::layer::DownloadError::ContextAndConfigReallyDeniesDownloads
1500 : | tenant::storage_layer::layer::DownloadError::DownloadRequired
1501 : | tenant::storage_layer::layer::DownloadError::NotFile(_)
1502 : | tenant::storage_layer::layer::DownloadError::DownloadFailed
1503 : | tenant::storage_layer::layer::DownloadError::PreStatFailed(_) => {
1504 0 : ApiError::InternalServerError(err.into())
1505 : }
1506 : #[cfg(test)]
1507 : tenant::storage_layer::layer::DownloadError::Failpoint(_) => {
1508 0 : ApiError::InternalServerError(err.into())
1509 : }
1510 0 : })?;
1511 :
1512 : let keys = resident_layer
1513 : .load_keys(&ctx)
1514 : .await
1515 : .map_err(ApiError::InternalServerError)?;
1516 :
1517 : let shard_identity = timeline.get_shard_identity();
1518 :
1519 : let mut disposable_count = 0;
1520 : let mut not_disposable_count = 0;
1521 : let cancel = cancel.clone();
1522 : for (i, key) in keys.into_iter().enumerate() {
1523 : if shard_identity.is_key_disposable(&key) {
1524 : disposable_count += 1;
1525 : tracing::debug!(key = %key, key.dbg=?key, "disposable key");
1526 : } else {
1527 : not_disposable_count += 1;
1528 : }
1529 : #[allow(clippy::collapsible_if)]
1530 : if i % 10000 == 0 {
1531 : if cancel.is_cancelled() || timeline.cancel.is_cancelled() || timeline.is_stopping() {
1532 : return Err(ApiError::ShuttingDown);
1533 : }
1534 : }
1535 : }
1536 :
1537 : json_response(
1538 : StatusCode::OK,
1539 : pageserver_api::models::ScanDisposableKeysResponse {
1540 : disposable_count,
1541 : not_disposable_count,
1542 : },
1543 : )
1544 : }
1545 :
1546 0 : async fn timeline_download_heatmap_layers_handler(
1547 0 : request: Request<Body>,
1548 0 : _cancel: CancellationToken,
1549 0 : ) -> Result<Response<Body>, ApiError> {
1550 : // Only used in the case where remote storage is not configured.
1551 : const DEFAULT_MAX_CONCURRENCY: usize = 100;
1552 : // A conservative default.
1553 : const DEFAULT_CONCURRENCY: usize = 16;
1554 :
1555 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1556 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1557 :
1558 0 : let desired_concurrency =
1559 0 : parse_query_param(&request, "concurrency")?.unwrap_or(DEFAULT_CONCURRENCY);
1560 0 : let recurse = parse_query_param(&request, "recurse")?.unwrap_or(false);
1561 :
1562 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1563 :
1564 0 : let state = get_state(&request);
1565 0 : let timeline =
1566 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1567 0 : .await?;
1568 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1569 0 : .with_scope_timeline(&timeline);
1570 :
1571 0 : let max_concurrency = get_config(&request)
1572 0 : .remote_storage_config
1573 0 : .as_ref()
1574 0 : .map(|c| c.concurrency_limit())
1575 0 : .unwrap_or(DEFAULT_MAX_CONCURRENCY);
1576 0 : let concurrency = std::cmp::min(max_concurrency, desired_concurrency);
1577 :
1578 0 : timeline.start_heatmap_layers_download(concurrency, recurse, &ctx)?;
1579 :
1580 0 : json_response(StatusCode::ACCEPTED, ())
1581 0 : }
1582 :
1583 0 : async fn timeline_shutdown_download_heatmap_layers_handler(
1584 0 : request: Request<Body>,
1585 0 : _cancel: CancellationToken,
1586 0 : ) -> Result<Response<Body>, ApiError> {
1587 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1588 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1589 :
1590 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1591 :
1592 0 : let state = get_state(&request);
1593 0 : let timeline =
1594 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1595 0 : .await?;
1596 :
1597 0 : timeline.stop_and_drain_heatmap_layers_download().await;
1598 :
1599 0 : json_response(StatusCode::OK, ())
1600 0 : }
1601 :
1602 0 : async fn layer_download_handler(
1603 0 : request: Request<Body>,
1604 0 : _cancel: CancellationToken,
1605 0 : ) -> Result<Response<Body>, ApiError> {
1606 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1607 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1608 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1609 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1610 0 : let layer_name = LayerName::from_str(layer_file_name)
1611 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1612 0 : let state = get_state(&request);
1613 :
1614 0 : let timeline =
1615 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1616 0 : .await?;
1617 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1618 0 : .with_scope_timeline(&timeline);
1619 0 : let downloaded = timeline
1620 0 : .download_layer(&layer_name, &ctx)
1621 0 : .await
1622 0 : .map_err(|e| match e {
1623 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1624 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1625 0 : ApiError::ShuttingDown
1626 : }
1627 0 : other => ApiError::InternalServerError(other.into()),
1628 0 : })?;
1629 :
1630 0 : match downloaded {
1631 0 : Some(true) => json_response(StatusCode::OK, ()),
1632 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1633 0 : None => json_response(
1634 : StatusCode::BAD_REQUEST,
1635 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1636 : ),
1637 : }
1638 0 : }
1639 :
1640 0 : async fn evict_timeline_layer_handler(
1641 0 : request: Request<Body>,
1642 0 : _cancel: CancellationToken,
1643 0 : ) -> Result<Response<Body>, ApiError> {
1644 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1645 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1646 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1647 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1648 0 : let state = get_state(&request);
1649 :
1650 0 : let layer_name = LayerName::from_str(layer_file_name)
1651 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1652 :
1653 0 : let timeline =
1654 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1655 0 : .await?;
1656 0 : let evicted = timeline
1657 0 : .evict_layer(&layer_name)
1658 0 : .await
1659 0 : .map_err(ApiError::InternalServerError)?;
1660 :
1661 0 : match evicted {
1662 0 : Some(true) => json_response(StatusCode::OK, ()),
1663 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1664 0 : None => json_response(
1665 : StatusCode::BAD_REQUEST,
1666 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1667 : ),
1668 : }
1669 0 : }
1670 :
1671 0 : async fn timeline_gc_blocking_handler(
1672 0 : request: Request<Body>,
1673 0 : _cancel: CancellationToken,
1674 0 : ) -> Result<Response<Body>, ApiError> {
1675 0 : block_or_unblock_gc(request, true).await
1676 0 : }
1677 :
1678 0 : async fn timeline_gc_unblocking_handler(
1679 0 : request: Request<Body>,
1680 0 : _cancel: CancellationToken,
1681 0 : ) -> Result<Response<Body>, ApiError> {
1682 0 : block_or_unblock_gc(request, false).await
1683 0 : }
1684 :
1685 : /// Traces GetPage@LSN requests for a timeline, and emits metadata in an efficient binary encoding.
1686 : /// Use the `pagectl page-trace` command to decode and analyze the output.
1687 0 : async fn timeline_page_trace_handler(
1688 0 : 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 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1693 0 : let state = get_state(&request);
1694 0 : check_permission(&request, None)?;
1695 :
1696 0 : let size_limit: usize = parse_query_param(&request, "size_limit_bytes")?.unwrap_or(1024 * 1024);
1697 0 : let time_limit_secs: u64 = parse_query_param(&request, "time_limit_secs")?.unwrap_or(5);
1698 :
1699 : // Convert size limit to event limit based on the serialized size of an event. The event size is
1700 : // fixed, as the default bincode serializer uses fixed-width integer encoding.
1701 0 : let event_size = bincode::serialize(&PageTraceEvent::default())
1702 0 : .map_err(|err| ApiError::InternalServerError(err.into()))?
1703 0 : .len();
1704 0 : let event_limit = size_limit / event_size;
1705 :
1706 0 : let timeline =
1707 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1708 0 : .await?;
1709 :
1710 : // Install a page trace, unless one is already in progress. We just use a buffered channel,
1711 : // which may 2x the memory usage in the worst case, but it's still bounded.
1712 0 : let (trace_tx, mut trace_rx) = tokio::sync::mpsc::channel(event_limit);
1713 0 : let cur = timeline.page_trace.load();
1714 0 : let installed = cur.is_none()
1715 0 : && timeline
1716 0 : .page_trace
1717 0 : .compare_and_swap(cur, Some(Arc::new(trace_tx)))
1718 0 : .is_none();
1719 0 : if !installed {
1720 0 : return Err(ApiError::Conflict("page trace already active".to_string()));
1721 0 : }
1722 0 : defer!(timeline.page_trace.store(None)); // uninstall on return
1723 :
1724 : // Collect the trace and return it to the client. We could stream the response, but this is
1725 : // simple and fine.
1726 0 : let mut body = Vec::with_capacity(size_limit);
1727 0 : let deadline = Instant::now() + Duration::from_secs(time_limit_secs);
1728 :
1729 0 : while body.len() < size_limit {
1730 0 : tokio::select! {
1731 0 : event = trace_rx.recv() => {
1732 0 : let Some(event) = event else {
1733 0 : break; // shouldn't happen (sender doesn't close, unless timeline dropped)
1734 : };
1735 0 : bincode::serialize_into(&mut body, &event)
1736 0 : .map_err(|err| ApiError::InternalServerError(err.into()))?;
1737 : }
1738 0 : _ = tokio::time::sleep_until(deadline) => break, // time limit reached
1739 0 : _ = cancel.cancelled() => return Err(ApiError::Cancelled),
1740 : }
1741 : }
1742 :
1743 0 : Ok(Response::builder()
1744 0 : .status(StatusCode::OK)
1745 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
1746 0 : .body(hyper::Body::from(body))
1747 0 : .unwrap())
1748 0 : }
1749 :
1750 : /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
1751 : ///
1752 : /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
1753 0 : async fn block_or_unblock_gc(
1754 0 : request: Request<Body>,
1755 0 : block: bool,
1756 0 : ) -> Result<Response<Body>, ApiError> {
1757 : use crate::tenant::remote_timeline_client::WaitCompletionError;
1758 : use crate::tenant::upload_queue::NotInitialized;
1759 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1760 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1761 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1762 0 : let state = get_state(&request);
1763 :
1764 0 : let tenant = state
1765 0 : .tenant_manager
1766 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1767 :
1768 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1769 :
1770 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1771 :
1772 0 : let fut = async {
1773 0 : if block {
1774 0 : timeline.block_gc(&tenant).await.map(|_| ())
1775 : } else {
1776 0 : timeline.unblock_gc(&tenant).await
1777 : }
1778 0 : };
1779 :
1780 0 : let span = tracing::info_span!(
1781 : "block_or_unblock_gc",
1782 : tenant_id = %tenant_shard_id.tenant_id,
1783 0 : shard_id = %tenant_shard_id.shard_slug(),
1784 : timeline_id = %timeline_id,
1785 : block = block,
1786 : );
1787 :
1788 0 : let res = fut.instrument(span).await;
1789 :
1790 0 : res.map_err(|e| {
1791 0 : if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
1792 0 : ApiError::ShuttingDown
1793 : } else {
1794 0 : ApiError::InternalServerError(e)
1795 : }
1796 0 : })?;
1797 :
1798 0 : json_response(StatusCode::OK, ())
1799 0 : }
1800 :
1801 : /// Get tenant_size SVG graph along with the JSON data.
1802 0 : fn synthetic_size_html_response(
1803 0 : inputs: ModelInputs,
1804 0 : storage_model: StorageModel,
1805 0 : sizes: SizeResult,
1806 0 : ) -> Result<Response<Body>, ApiError> {
1807 0 : let mut timeline_ids: Vec<String> = Vec::new();
1808 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1809 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1810 0 : timeline_map.insert(ti.timeline_id, index);
1811 0 : timeline_ids.push(ti.timeline_id.to_string());
1812 0 : }
1813 0 : let seg_to_branch: Vec<(usize, SvgBranchKind)> = inputs
1814 0 : .segments
1815 0 : .iter()
1816 0 : .map(|seg| {
1817 0 : (
1818 0 : *timeline_map.get(&seg.timeline_id).unwrap(),
1819 0 : seg.kind.into(),
1820 0 : )
1821 0 : })
1822 0 : .collect();
1823 :
1824 0 : let svg =
1825 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1826 0 : .map_err(ApiError::InternalServerError)?;
1827 :
1828 0 : let mut response = String::new();
1829 :
1830 : use std::fmt::Write;
1831 0 : write!(response, "<html>\n<body>\n").unwrap();
1832 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1833 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1834 0 : writeln!(response, "<pre>").unwrap();
1835 0 : writeln!(
1836 0 : response,
1837 0 : "{}",
1838 0 : serde_json::to_string_pretty(&inputs).unwrap()
1839 : )
1840 0 : .unwrap();
1841 0 : writeln!(
1842 0 : response,
1843 0 : "{}",
1844 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1845 : )
1846 0 : .unwrap();
1847 0 : writeln!(response, "</pre>").unwrap();
1848 0 : write!(response, "</body>\n</html>\n").unwrap();
1849 :
1850 0 : html_response(StatusCode::OK, response)
1851 0 : }
1852 :
1853 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1854 0 : let response = Response::builder()
1855 0 : .status(status)
1856 0 : .header(header::CONTENT_TYPE, "text/html")
1857 0 : .body(Body::from(data.as_bytes().to_vec()))
1858 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1859 0 : Ok(response)
1860 0 : }
1861 :
1862 0 : async fn get_tenant_config_handler(
1863 0 : request: Request<Body>,
1864 0 : _cancel: CancellationToken,
1865 0 : ) -> Result<Response<Body>, ApiError> {
1866 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1867 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1868 0 : let state = get_state(&request);
1869 :
1870 0 : let tenant = state
1871 0 : .tenant_manager
1872 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1873 :
1874 0 : let response = HashMap::from([
1875 : (
1876 : "tenant_specific_overrides",
1877 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1878 0 : .context("serializing tenant specific overrides")
1879 0 : .map_err(ApiError::InternalServerError)?,
1880 : ),
1881 : (
1882 0 : "effective_config",
1883 0 : serde_json::to_value(tenant.effective_config())
1884 0 : .context("serializing effective config")
1885 0 : .map_err(ApiError::InternalServerError)?,
1886 : ),
1887 : ]);
1888 :
1889 0 : json_response(StatusCode::OK, response)
1890 0 : }
1891 :
1892 0 : async fn update_tenant_config_handler(
1893 0 : mut request: Request<Body>,
1894 0 : _cancel: CancellationToken,
1895 0 : ) -> Result<Response<Body>, ApiError> {
1896 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1897 0 : let tenant_id = request_data.tenant_id;
1898 0 : check_permission(&request, Some(tenant_id))?;
1899 :
1900 0 : let new_tenant_conf = request_data.config;
1901 :
1902 0 : let state = get_state(&request);
1903 :
1904 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1905 :
1906 0 : let tenant = state
1907 0 : .tenant_manager
1908 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1909 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1910 :
1911 : // This is a legacy API that only operates on attached tenants: the preferred
1912 : // API to use is the location_config/ endpoint, which lets the caller provide
1913 : // the full LocationConf.
1914 0 : let location_conf = LocationConf::attached_single(
1915 0 : new_tenant_conf.clone(),
1916 0 : tenant.get_generation(),
1917 0 : ShardParameters::from(tenant.get_shard_identity()),
1918 : );
1919 :
1920 0 : tenant
1921 0 : .get_shard_identity()
1922 0 : .assert_equal(location_conf.shard); // not strictly necessary since we construct it above
1923 :
1924 0 : crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1925 0 : .await
1926 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1927 :
1928 0 : let _ = tenant
1929 0 : .update_tenant_config(|_crnt| Ok(new_tenant_conf.clone()))
1930 0 : .expect("Closure returns Ok()");
1931 :
1932 0 : json_response(StatusCode::OK, ())
1933 0 : }
1934 :
1935 0 : async fn patch_tenant_config_handler(
1936 0 : mut request: Request<Body>,
1937 0 : _cancel: CancellationToken,
1938 0 : ) -> Result<Response<Body>, ApiError> {
1939 0 : let request_data: TenantConfigPatchRequest = json_request(&mut request).await?;
1940 0 : let tenant_id = request_data.tenant_id;
1941 0 : check_permission(&request, Some(tenant_id))?;
1942 :
1943 0 : let state = get_state(&request);
1944 :
1945 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1946 :
1947 0 : let tenant = state
1948 0 : .tenant_manager
1949 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1950 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1951 :
1952 0 : let updated = tenant
1953 0 : .update_tenant_config(|crnt| {
1954 0 : crnt.apply_patch(request_data.config.clone())
1955 0 : .map_err(anyhow::Error::new)
1956 0 : })
1957 0 : .map_err(ApiError::BadRequest)?;
1958 :
1959 : // This is a legacy API that only operates on attached tenants: the preferred
1960 : // API to use is the location_config/ endpoint, which lets the caller provide
1961 : // the full LocationConf.
1962 0 : let location_conf = LocationConf::attached_single(
1963 0 : updated,
1964 0 : tenant.get_generation(),
1965 0 : ShardParameters::from(tenant.get_shard_identity()),
1966 : );
1967 :
1968 0 : tenant
1969 0 : .get_shard_identity()
1970 0 : .assert_equal(location_conf.shard); // not strictly necessary since we construct it above
1971 :
1972 0 : crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1973 0 : .await
1974 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1975 :
1976 0 : json_response(StatusCode::OK, ())
1977 0 : }
1978 :
1979 0 : async fn put_tenant_location_config_handler(
1980 0 : mut request: Request<Body>,
1981 0 : _cancel: CancellationToken,
1982 0 : ) -> Result<Response<Body>, ApiError> {
1983 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1984 :
1985 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1986 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1987 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1988 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1989 :
1990 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1991 0 : let state = get_state(&request);
1992 0 : let conf = state.conf;
1993 :
1994 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1995 : // its local disk content and drops it from memory.
1996 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1997 0 : if let Err(e) = state
1998 0 : .tenant_manager
1999 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
2000 0 : .instrument(info_span!("tenant_detach",
2001 : tenant_id = %tenant_shard_id.tenant_id,
2002 0 : shard_id = %tenant_shard_id.shard_slug()
2003 : ))
2004 0 : .await
2005 : {
2006 0 : match e {
2007 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
2008 0 : // This API is idempotent: a NotFound on a detach is fine.
2009 0 : }
2010 0 : _ => return Err(e.into()),
2011 : }
2012 0 : }
2013 0 : return json_response(StatusCode::OK, ());
2014 0 : }
2015 :
2016 0 : let location_conf =
2017 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
2018 :
2019 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
2020 : // similar to at startup ordering.
2021 0 : let spawn_mode = if lazy {
2022 0 : tenant::SpawnMode::Lazy
2023 : } else {
2024 0 : tenant::SpawnMode::Eager
2025 : };
2026 :
2027 0 : let tenant = state
2028 0 : .tenant_manager
2029 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
2030 0 : .await?;
2031 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
2032 0 : let attached = tenant.is_some();
2033 :
2034 0 : if let Some(_flush_ms) = flush {
2035 0 : match state
2036 0 : .secondary_controller
2037 0 : .upload_tenant(tenant_shard_id)
2038 0 : .await
2039 : {
2040 : Ok(()) => {
2041 0 : tracing::info!("Uploaded heatmap during flush");
2042 : }
2043 0 : Err(e) => {
2044 0 : tracing::warn!("Failed to flush heatmap: {e}");
2045 : }
2046 : }
2047 : } else {
2048 0 : tracing::info!("No flush requested when configuring");
2049 : }
2050 :
2051 : // This API returns a vector of pageservers where the tenant is attached: this is
2052 : // primarily for use in the sharding service. For compatibilty, we also return this
2053 : // when called directly on a pageserver, but the payload is always zero or one shards.
2054 0 : let mut response = TenantLocationConfigResponse {
2055 0 : shards: Vec::new(),
2056 0 : stripe_size: None,
2057 0 : };
2058 0 : if attached {
2059 0 : response.shards.push(TenantShardLocation {
2060 0 : shard_id: tenant_shard_id,
2061 0 : node_id: state.conf.id,
2062 0 : });
2063 0 : if tenant_shard_id.shard_count.count() > 1 {
2064 : // Stripe size should be set if we are attached
2065 0 : debug_assert!(stripe_size.is_some());
2066 0 : response.stripe_size = stripe_size;
2067 0 : }
2068 0 : }
2069 :
2070 0 : json_response(StatusCode::OK, response)
2071 0 : }
2072 :
2073 0 : async fn list_location_config_handler(
2074 0 : request: Request<Body>,
2075 0 : _cancel: CancellationToken,
2076 0 : ) -> Result<Response<Body>, ApiError> {
2077 0 : let state = get_state(&request);
2078 0 : let slots = state.tenant_manager.list();
2079 0 : let result = LocationConfigListResponse {
2080 0 : tenant_shards: slots
2081 0 : .into_iter()
2082 0 : .map(|(tenant_shard_id, slot)| {
2083 0 : let v = match slot {
2084 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
2085 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
2086 0 : TenantSlot::InProgress(_) => None,
2087 : };
2088 0 : (tenant_shard_id, v)
2089 0 : })
2090 0 : .collect(),
2091 : };
2092 0 : json_response(StatusCode::OK, result)
2093 0 : }
2094 :
2095 0 : async fn get_location_config_handler(
2096 0 : request: Request<Body>,
2097 0 : _cancel: CancellationToken,
2098 0 : ) -> Result<Response<Body>, ApiError> {
2099 0 : let state = get_state(&request);
2100 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2101 0 : let slot = state.tenant_manager.get(tenant_shard_id);
2102 :
2103 0 : let Some(slot) = slot else {
2104 0 : return Err(ApiError::NotFound(
2105 0 : anyhow::anyhow!("Tenant shard not found").into(),
2106 0 : ));
2107 : };
2108 :
2109 0 : let result: Option<LocationConfig> = match slot {
2110 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
2111 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
2112 0 : TenantSlot::InProgress(_) => None,
2113 : };
2114 :
2115 0 : json_response(StatusCode::OK, result)
2116 0 : }
2117 :
2118 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
2119 : // (from all pageservers) as it invalidates consistency assumptions.
2120 0 : async fn tenant_time_travel_remote_storage_handler(
2121 0 : request: Request<Body>,
2122 0 : cancel: CancellationToken,
2123 0 : ) -> Result<Response<Body>, ApiError> {
2124 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2125 :
2126 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2127 :
2128 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
2129 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
2130 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
2131 0 : .map_err(ApiError::BadRequest)?;
2132 :
2133 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
2134 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
2135 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
2136 0 : .map_err(ApiError::BadRequest)?;
2137 :
2138 : // This is just a sanity check to fend off naive wrong usages of the API:
2139 : // the tenant needs to be detached *everywhere*
2140 0 : let state = get_state(&request);
2141 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
2142 0 : if we_manage_tenant {
2143 0 : return Err(ApiError::BadRequest(anyhow!(
2144 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
2145 0 : )));
2146 0 : }
2147 :
2148 0 : if timestamp > done_if_after {
2149 0 : return Err(ApiError::BadRequest(anyhow!(
2150 0 : "The done_if_after timestamp comes before the timestamp to recover to"
2151 0 : )));
2152 0 : }
2153 :
2154 0 : tracing::info!(
2155 0 : "Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}"
2156 : );
2157 :
2158 0 : remote_timeline_client::upload::time_travel_recover_tenant(
2159 0 : &state.remote_storage,
2160 0 : &tenant_shard_id,
2161 0 : timestamp,
2162 0 : done_if_after,
2163 0 : &cancel,
2164 0 : )
2165 0 : .await
2166 0 : .map_err(|e| match e {
2167 0 : TimeTravelError::BadInput(e) => {
2168 0 : warn!("bad input error: {e}");
2169 0 : ApiError::BadRequest(anyhow!("bad input error"))
2170 : }
2171 : TimeTravelError::Unimplemented => {
2172 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
2173 : }
2174 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
2175 : TimeTravelError::TooManyVersions => {
2176 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
2177 : }
2178 0 : TimeTravelError::Other(e) => {
2179 0 : warn!("internal error: {e}");
2180 0 : ApiError::InternalServerError(anyhow!("internal error"))
2181 : }
2182 0 : })?;
2183 :
2184 0 : json_response(StatusCode::OK, ())
2185 0 : }
2186 :
2187 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
2188 0 : async fn handle_tenant_break(
2189 0 : r: Request<Body>,
2190 0 : _cancel: CancellationToken,
2191 0 : ) -> Result<Response<Body>, ApiError> {
2192 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
2193 :
2194 0 : let state = get_state(&r);
2195 0 : state
2196 0 : .tenant_manager
2197 0 : .get_attached_tenant_shard(tenant_shard_id)?
2198 0 : .set_broken("broken from test".to_owned())
2199 0 : .await;
2200 :
2201 0 : json_response(StatusCode::OK, ())
2202 0 : }
2203 :
2204 : // Obtains an lsn lease on the given timeline.
2205 0 : async fn lsn_lease_handler(
2206 0 : mut request: Request<Body>,
2207 0 : _cancel: CancellationToken,
2208 0 : ) -> Result<Response<Body>, ApiError> {
2209 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2210 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2211 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2212 0 : let lsn = json_request::<LsnLeaseRequest>(&mut request).await?.lsn;
2213 :
2214 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2215 :
2216 0 : let state = get_state(&request);
2217 :
2218 0 : let timeline =
2219 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2220 0 : .await?;
2221 :
2222 0 : let result = async {
2223 0 : timeline
2224 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
2225 0 : .map_err(|e| {
2226 0 : ApiError::InternalServerError(
2227 0 : e.context(format!("invalid lsn lease request at {lsn}")),
2228 0 : )
2229 0 : })
2230 0 : }
2231 0 : .instrument(info_span!("init_lsn_lease", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2232 0 : .await?;
2233 :
2234 0 : json_response(StatusCode::OK, result)
2235 0 : }
2236 :
2237 : // Run GC immediately on given timeline.
2238 0 : async fn timeline_gc_handler(
2239 0 : mut request: Request<Body>,
2240 0 : cancel: CancellationToken,
2241 0 : ) -> Result<Response<Body>, ApiError> {
2242 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2243 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2244 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2245 :
2246 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
2247 :
2248 0 : let state = get_state(&request);
2249 :
2250 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2251 0 : let gc_result = state
2252 0 : .tenant_manager
2253 0 : .immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)
2254 0 : .await?;
2255 :
2256 0 : json_response(StatusCode::OK, gc_result)
2257 0 : }
2258 :
2259 : // Cancel scheduled compaction tasks
2260 0 : async fn timeline_cancel_compact_handler(
2261 0 : request: Request<Body>,
2262 0 : _cancel: CancellationToken,
2263 0 : ) -> Result<Response<Body>, ApiError> {
2264 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2265 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2266 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2267 0 : let state = get_state(&request);
2268 0 : async {
2269 0 : let tenant = state
2270 0 : .tenant_manager
2271 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2272 0 : tenant.cancel_scheduled_compaction(timeline_id);
2273 0 : json_response(StatusCode::OK, ())
2274 0 : }
2275 0 : .instrument(info_span!("timeline_cancel_compact", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2276 0 : .await
2277 0 : }
2278 :
2279 : // Get compact info of a timeline
2280 0 : async fn timeline_compact_info_handler(
2281 0 : request: Request<Body>,
2282 0 : _cancel: CancellationToken,
2283 0 : ) -> Result<Response<Body>, ApiError> {
2284 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2285 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2286 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2287 0 : let state = get_state(&request);
2288 0 : async {
2289 0 : let tenant = state
2290 0 : .tenant_manager
2291 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2292 0 : let resp = tenant.get_scheduled_compaction_tasks(timeline_id);
2293 0 : json_response(StatusCode::OK, resp)
2294 0 : }
2295 0 : .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2296 0 : .await
2297 0 : }
2298 :
2299 : // Run compaction immediately on given timeline.
2300 0 : async fn timeline_compact_handler(
2301 0 : mut request: Request<Body>,
2302 0 : cancel: CancellationToken,
2303 0 : ) -> Result<Response<Body>, ApiError> {
2304 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2305 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2306 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2307 :
2308 0 : let compact_request = json_request_maybe::<Option<CompactRequest>>(&mut request).await?;
2309 :
2310 0 : let state = get_state(&request);
2311 :
2312 0 : let mut flags = EnumSet::empty();
2313 :
2314 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2315 0 : flags |= CompactFlags::ForceL0Compaction;
2316 0 : }
2317 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2318 0 : flags |= CompactFlags::ForceRepartition;
2319 0 : }
2320 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2321 0 : flags |= CompactFlags::ForceImageLayerCreation;
2322 0 : }
2323 0 : if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
2324 0 : flags |= CompactFlags::EnhancedGcBottomMostCompaction;
2325 0 : }
2326 0 : if Some(true) == parse_query_param::<_, bool>(&request, "dry_run")? {
2327 0 : flags |= CompactFlags::DryRun;
2328 0 : }
2329 : // Manual compaction does not yield for L0.
2330 :
2331 0 : let wait_until_uploaded =
2332 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2333 :
2334 0 : let wait_until_scheduled_compaction_done =
2335 0 : parse_query_param::<_, bool>(&request, "wait_until_scheduled_compaction_done")?
2336 0 : .unwrap_or(false);
2337 :
2338 0 : let sub_compaction = compact_request
2339 0 : .as_ref()
2340 0 : .map(|r| r.sub_compaction)
2341 0 : .unwrap_or(false);
2342 0 : let sub_compaction_max_job_size_mb = compact_request
2343 0 : .as_ref()
2344 0 : .and_then(|r| r.sub_compaction_max_job_size_mb);
2345 :
2346 0 : let options = CompactOptions {
2347 0 : compact_key_range: compact_request
2348 0 : .as_ref()
2349 0 : .and_then(|r| r.compact_key_range.clone()),
2350 0 : compact_lsn_range: compact_request
2351 0 : .as_ref()
2352 0 : .and_then(|r| r.compact_lsn_range.clone()),
2353 0 : flags,
2354 0 : sub_compaction,
2355 0 : sub_compaction_max_job_size_mb,
2356 : };
2357 :
2358 0 : let scheduled = compact_request
2359 0 : .as_ref()
2360 0 : .map(|r| r.scheduled)
2361 0 : .unwrap_or(false);
2362 :
2363 0 : async {
2364 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2365 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2366 0 : if scheduled {
2367 0 : let tenant = state
2368 0 : .tenant_manager
2369 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2370 0 : let rx = tenant.schedule_compaction(timeline_id, options).await.map_err(ApiError::InternalServerError)?;
2371 0 : if wait_until_scheduled_compaction_done {
2372 : // It is possible that this will take a long time, dropping the HTTP request will not cancel the compaction.
2373 0 : rx.await.ok();
2374 0 : }
2375 : } else {
2376 0 : timeline
2377 0 : .compact_with_options(&cancel, options, &ctx)
2378 0 : .await
2379 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2380 0 : if wait_until_uploaded {
2381 0 : timeline.remote_client.wait_completion().await
2382 : // XXX map to correct ApiError for the cases where it's due to shutdown
2383 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2384 0 : }
2385 : }
2386 0 : json_response(StatusCode::OK, ())
2387 0 : }
2388 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2389 0 : .await
2390 0 : }
2391 :
2392 0 : async fn timeline_mark_invisible_handler(
2393 0 : mut request: Request<Body>,
2394 0 : _cancel: CancellationToken,
2395 0 : ) -> Result<Response<Body>, ApiError> {
2396 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2397 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2398 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2399 :
2400 0 : let compact_request = json_request_maybe::<Option<MarkInvisibleRequest>>(&mut request).await?;
2401 :
2402 0 : let state = get_state(&request);
2403 :
2404 0 : let visibility = match compact_request {
2405 0 : Some(req) => match req.is_visible {
2406 0 : Some(true) => TimelineVisibilityState::Visible,
2407 0 : Some(false) | None => TimelineVisibilityState::Invisible,
2408 : },
2409 0 : None => TimelineVisibilityState::Invisible,
2410 : };
2411 :
2412 0 : async {
2413 0 : let tenant = state
2414 0 : .tenant_manager
2415 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2416 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2417 0 : timeline.remote_client.schedule_index_upload_for_timeline_invisible_state(visibility).map_err(ApiError::InternalServerError)?;
2418 0 : json_response(StatusCode::OK, ())
2419 0 : }
2420 0 : .instrument(info_span!("manual_timeline_mark_invisible", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2421 0 : .await
2422 0 : }
2423 :
2424 : // Run offload immediately on given timeline.
2425 0 : async fn timeline_offload_handler(
2426 0 : request: Request<Body>,
2427 0 : _cancel: CancellationToken,
2428 0 : ) -> Result<Response<Body>, ApiError> {
2429 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2430 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2431 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2432 :
2433 0 : let state = get_state(&request);
2434 :
2435 0 : async {
2436 0 : let tenant = state
2437 0 : .tenant_manager
2438 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2439 :
2440 0 : if tenant.get_offloaded_timeline(timeline_id).is_ok() {
2441 0 : return json_response(StatusCode::OK, ());
2442 0 : }
2443 0 : let timeline =
2444 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2445 0 : .await?;
2446 :
2447 0 : if !tenant.timeline_has_no_attached_children(timeline_id) {
2448 0 : return Err(ApiError::PreconditionFailed(
2449 0 : "timeline has attached children".into(),
2450 0 : ));
2451 0 : }
2452 0 : if let (false, reason) = timeline.can_offload() {
2453 0 : return Err(ApiError::PreconditionFailed(
2454 0 : format!("Timeline::can_offload() check failed: {reason}") .into(),
2455 0 : ));
2456 0 : }
2457 0 : offload_timeline(&tenant, &timeline)
2458 0 : .await
2459 0 : .map_err(|e| {
2460 0 : match e {
2461 0 : OffloadError::Cancelled => ApiError::ResourceUnavailable("Timeline shutting down".into()),
2462 0 : OffloadError::AlreadyInProgress => ApiError::Conflict("Timeline already being offloaded or deleted".into()),
2463 0 : _ => ApiError::InternalServerError(anyhow!(e))
2464 : }
2465 0 : })?;
2466 :
2467 0 : json_response(StatusCode::OK, ())
2468 0 : }
2469 0 : .instrument(info_span!("manual_timeline_offload", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2470 0 : .await
2471 0 : }
2472 :
2473 : // Run checkpoint immediately on given timeline.
2474 0 : async fn timeline_checkpoint_handler(
2475 0 : request: Request<Body>,
2476 0 : cancel: CancellationToken,
2477 0 : ) -> Result<Response<Body>, ApiError> {
2478 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2479 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2480 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2481 :
2482 0 : let state = get_state(&request);
2483 :
2484 0 : let mut flags = EnumSet::empty();
2485 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2486 0 : flags |= CompactFlags::ForceL0Compaction;
2487 0 : }
2488 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2489 0 : flags |= CompactFlags::ForceRepartition;
2490 0 : }
2491 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2492 0 : flags |= CompactFlags::ForceImageLayerCreation;
2493 0 : }
2494 :
2495 : // By default, checkpoints come with a compaction, but this may be optionally disabled by tests that just want to flush + upload.
2496 0 : let compact = parse_query_param::<_, bool>(&request, "compact")?.unwrap_or(true);
2497 :
2498 0 : let wait_until_flushed: bool =
2499 0 : parse_query_param(&request, "wait_until_flushed")?.unwrap_or(true);
2500 :
2501 0 : let wait_until_uploaded =
2502 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2503 :
2504 0 : async {
2505 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2506 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2507 0 : if wait_until_flushed {
2508 0 : timeline.freeze_and_flush().await
2509 : } else {
2510 0 : timeline.freeze().await.and(Ok(()))
2511 0 : }.map_err(|e| {
2512 0 : match e {
2513 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
2514 0 : other => ApiError::InternalServerError(other.into()),
2515 :
2516 : }
2517 0 : })?;
2518 0 : if compact {
2519 0 : timeline
2520 0 : .compact(&cancel, flags, &ctx)
2521 0 : .await
2522 0 : .map_err(|e|
2523 0 : if e.is_cancel() {
2524 0 : ApiError::ShuttingDown
2525 : } else {
2526 0 : ApiError::InternalServerError(e.into_anyhow())
2527 0 : }
2528 0 : )?;
2529 0 : }
2530 :
2531 0 : if wait_until_uploaded {
2532 0 : tracing::info!("Waiting for uploads to complete...");
2533 0 : timeline.remote_client.wait_completion().await
2534 : // XXX map to correct ApiError for the cases where it's due to shutdown
2535 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2536 0 : tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
2537 0 : }
2538 :
2539 0 : json_response(StatusCode::OK, ())
2540 0 : }
2541 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2542 0 : .await
2543 0 : }
2544 :
2545 0 : async fn timeline_download_remote_layers_handler_post(
2546 0 : mut request: Request<Body>,
2547 0 : _cancel: CancellationToken,
2548 0 : ) -> Result<Response<Body>, ApiError> {
2549 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2550 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2551 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
2552 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2553 :
2554 0 : let state = get_state(&request);
2555 :
2556 0 : let timeline =
2557 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2558 0 : .await?;
2559 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
2560 0 : .with_scope_timeline(&timeline);
2561 0 : match timeline.spawn_download_all_remote_layers(body, &ctx).await {
2562 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
2563 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
2564 : }
2565 0 : }
2566 :
2567 0 : async fn timeline_download_remote_layers_handler_get(
2568 0 : request: Request<Body>,
2569 0 : _cancel: CancellationToken,
2570 0 : ) -> Result<Response<Body>, ApiError> {
2571 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2572 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2573 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2574 0 : let state = get_state(&request);
2575 :
2576 0 : let timeline =
2577 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2578 0 : .await?;
2579 0 : let info = timeline
2580 0 : .get_download_all_remote_layers_task_info()
2581 0 : .context("task never started since last pageserver process start")
2582 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
2583 0 : json_response(StatusCode::OK, info)
2584 0 : }
2585 :
2586 0 : async fn timeline_detach_ancestor_handler(
2587 0 : request: Request<Body>,
2588 0 : _cancel: CancellationToken,
2589 0 : ) -> Result<Response<Body>, ApiError> {
2590 : use pageserver_api::models::detach_ancestor::AncestorDetached;
2591 :
2592 : use crate::tenant::timeline::detach_ancestor;
2593 :
2594 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2595 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2596 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2597 0 : let behavior: Option<DetachBehavior> = parse_query_param(&request, "detach_behavior")?;
2598 :
2599 0 : let behavior = behavior.unwrap_or_default();
2600 :
2601 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
2602 :
2603 0 : async move {
2604 0 : let mut options = detach_ancestor::Options::default();
2605 :
2606 0 : let rewrite_concurrency =
2607 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
2608 0 : let copy_concurrency =
2609 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
2610 :
2611 0 : [
2612 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
2613 0 : (&mut options.copy_concurrency, copy_concurrency),
2614 0 : ]
2615 0 : .into_iter()
2616 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
2617 0 : .for_each(|(target, val)| *target = val);
2618 :
2619 0 : let state = get_state(&request);
2620 :
2621 0 : let tenant = state
2622 0 : .tenant_manager
2623 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2624 :
2625 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2626 :
2627 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
2628 0 : let ctx = &ctx;
2629 :
2630 : // Flush the upload queues of all timelines before detaching ancestor. We do the same thing again
2631 : // during shutdown. This early upload ensures the pageserver does not need to upload too many
2632 : // things and creates downtime during timeline reloads.
2633 0 : for timeline in tenant.list_timelines() {
2634 0 : timeline
2635 0 : .remote_client
2636 0 : .wait_completion()
2637 0 : .await
2638 0 : .map_err(|e| {
2639 0 : ApiError::PreconditionFailed(format!("cannot drain upload queue: {e}").into())
2640 0 : })?;
2641 : }
2642 :
2643 0 : tracing::info!("all timeline upload queues are drained");
2644 :
2645 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2646 0 : let ctx = &ctx.with_scope_timeline(&timeline);
2647 :
2648 0 : let progress = timeline
2649 0 : .prepare_to_detach_from_ancestor(&tenant, options, behavior, ctx)
2650 0 : .await?;
2651 :
2652 : // uncomment to allow early as possible Tenant::drop
2653 : // drop(tenant);
2654 :
2655 0 : let resp = match progress {
2656 0 : detach_ancestor::Progress::Prepared(attempt, prepared) => {
2657 : // it would be great to tag the guard on to the tenant activation future
2658 0 : let reparented_timelines = state
2659 0 : .tenant_manager
2660 0 : .complete_detaching_timeline_ancestor(
2661 0 : tenant_shard_id,
2662 0 : timeline_id,
2663 0 : prepared,
2664 0 : behavior,
2665 0 : attempt,
2666 0 : ctx,
2667 0 : )
2668 0 : .await?;
2669 :
2670 0 : AncestorDetached {
2671 0 : reparented_timelines,
2672 0 : }
2673 : }
2674 0 : detach_ancestor::Progress::Done(resp) => resp,
2675 : };
2676 :
2677 0 : json_response(StatusCode::OK, resp)
2678 0 : }
2679 0 : .instrument(span)
2680 0 : .await
2681 0 : }
2682 :
2683 0 : async fn deletion_queue_flush(
2684 0 : r: Request<Body>,
2685 0 : cancel: CancellationToken,
2686 0 : ) -> Result<Response<Body>, ApiError> {
2687 0 : let state = get_state(&r);
2688 :
2689 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
2690 :
2691 0 : let flush = async {
2692 0 : if execute {
2693 0 : state.deletion_queue_client.flush_execute().await
2694 : } else {
2695 0 : state.deletion_queue_client.flush().await
2696 : }
2697 0 : }
2698 : // DeletionQueueError's only case is shutting down.
2699 0 : .map_err(|_| ApiError::ShuttingDown);
2700 :
2701 0 : tokio::select! {
2702 0 : res = flush => {
2703 0 : res.map(|()| json_response(StatusCode::OK, ()))?
2704 : }
2705 0 : _ = cancel.cancelled() => {
2706 0 : Err(ApiError::ShuttingDown)
2707 : }
2708 : }
2709 0 : }
2710 :
2711 0 : async fn getpage_at_lsn_handler(
2712 0 : request: Request<Body>,
2713 0 : cancel: CancellationToken,
2714 0 : ) -> Result<Response<Body>, ApiError> {
2715 0 : getpage_at_lsn_handler_inner(false, request, cancel).await
2716 0 : }
2717 :
2718 0 : async fn touchpage_at_lsn_handler(
2719 0 : request: Request<Body>,
2720 0 : cancel: CancellationToken,
2721 0 : ) -> Result<Response<Body>, ApiError> {
2722 0 : getpage_at_lsn_handler_inner(true, request, cancel).await
2723 0 : }
2724 :
2725 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
2726 0 : async fn getpage_at_lsn_handler_inner(
2727 0 : touch: bool,
2728 0 : request: Request<Body>,
2729 0 : _cancel: CancellationToken,
2730 0 : ) -> Result<Response<Body>, ApiError> {
2731 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2732 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2733 : // Require pageserver admin permission for this API instead of only tenant-level token.
2734 0 : check_permission(&request, None)?;
2735 0 : let state = get_state(&request);
2736 :
2737 : struct Key(pageserver_api::key::Key);
2738 :
2739 : impl std::str::FromStr for Key {
2740 : type Err = anyhow::Error;
2741 :
2742 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
2743 0 : pageserver_api::key::Key::from_hex(s).map(Key)
2744 0 : }
2745 : }
2746 :
2747 0 : let key: Key = parse_query_param(&request, "key")?
2748 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
2749 0 : let lsn: Option<Lsn> = parse_query_param(&request, "lsn")?;
2750 :
2751 0 : async {
2752 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2753 0 : let ctx = RequestContextBuilder::new(TaskKind::MgmtRequest)
2754 0 : .download_behavior(DownloadBehavior::Download)
2755 0 : .scope(context::Scope::new_timeline(&timeline))
2756 0 : .read_path_debug(true)
2757 0 : .root();
2758 :
2759 : // Use last_record_lsn if no lsn is provided
2760 0 : let lsn = lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2761 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
2762 :
2763 0 : if touch {
2764 0 : json_response(StatusCode::OK, ())
2765 : } else {
2766 0 : Result::<_, ApiError>::Ok(
2767 0 : Response::builder()
2768 0 : .status(StatusCode::OK)
2769 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2770 0 : .body(hyper::Body::from(page))
2771 0 : .unwrap(),
2772 0 : )
2773 : }
2774 0 : }
2775 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2776 0 : .await
2777 0 : }
2778 :
2779 0 : async fn timeline_collect_keyspace(
2780 0 : request: Request<Body>,
2781 0 : _cancel: CancellationToken,
2782 0 : ) -> Result<Response<Body>, ApiError> {
2783 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2784 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2785 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2786 0 : let state = get_state(&request);
2787 :
2788 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2789 :
2790 0 : async {
2791 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2792 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2793 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2794 0 : let (dense_ks, sparse_ks) = timeline
2795 0 : .collect_keyspace(at_lsn, &ctx)
2796 0 : .await
2797 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2798 :
2799 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2800 : // Therefore, we split dense/sparse keys in this API.
2801 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2802 :
2803 0 : json_response(StatusCode::OK, res)
2804 0 : }
2805 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2806 0 : .await
2807 0 : }
2808 :
2809 0 : async fn active_timeline_of_active_tenant(
2810 0 : tenant_manager: &TenantManager,
2811 0 : tenant_shard_id: TenantShardId,
2812 0 : timeline_id: TimelineId,
2813 0 : ) -> Result<Arc<Timeline>, ApiError> {
2814 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2815 :
2816 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2817 :
2818 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2819 0 : }
2820 :
2821 0 : async fn always_panic_handler(
2822 0 : req: Request<Body>,
2823 0 : _cancel: CancellationToken,
2824 0 : ) -> Result<Response<Body>, ApiError> {
2825 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2826 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2827 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2828 0 : let query = req.uri().query();
2829 0 : let _ = std::panic::catch_unwind(|| {
2830 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2831 : });
2832 0 : json_response(StatusCode::NO_CONTENT, ())
2833 0 : }
2834 :
2835 0 : async fn disk_usage_eviction_run(
2836 0 : mut r: Request<Body>,
2837 0 : cancel: CancellationToken,
2838 0 : ) -> Result<Response<Body>, ApiError> {
2839 0 : check_permission(&r, None)?;
2840 :
2841 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2842 : struct Config {
2843 : /// How many bytes to evict before reporting that pressure is relieved.
2844 : evict_bytes: u64,
2845 :
2846 : #[serde(default)]
2847 : eviction_order: pageserver_api::config::EvictionOrder,
2848 : }
2849 :
2850 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2851 : struct Usage {
2852 : // remains unchanged after instantiation of the struct
2853 : evict_bytes: u64,
2854 : // updated by `add_available_bytes`
2855 : freed_bytes: u64,
2856 : }
2857 :
2858 : impl crate::disk_usage_eviction_task::Usage for Usage {
2859 0 : fn has_pressure(&self) -> bool {
2860 0 : self.evict_bytes > self.freed_bytes
2861 0 : }
2862 :
2863 0 : fn add_available_bytes(&mut self, bytes: u64) {
2864 0 : self.freed_bytes += bytes;
2865 0 : }
2866 : }
2867 :
2868 0 : let config = json_request::<Config>(&mut r).await?;
2869 :
2870 0 : let usage = Usage {
2871 0 : evict_bytes: config.evict_bytes,
2872 0 : freed_bytes: 0,
2873 0 : };
2874 :
2875 0 : let state = get_state(&r);
2876 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2877 :
2878 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2879 0 : &eviction_state,
2880 0 : &state.remote_storage,
2881 0 : usage,
2882 0 : &state.tenant_manager,
2883 0 : config.eviction_order.into(),
2884 0 : &cancel,
2885 0 : )
2886 0 : .await;
2887 :
2888 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2889 :
2890 0 : let res = res.map_err(ApiError::InternalServerError)?;
2891 :
2892 0 : json_response(StatusCode::OK, res)
2893 0 : }
2894 :
2895 0 : async fn secondary_upload_handler(
2896 0 : request: Request<Body>,
2897 0 : _cancel: CancellationToken,
2898 0 : ) -> Result<Response<Body>, ApiError> {
2899 0 : let state = get_state(&request);
2900 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2901 0 : state
2902 0 : .secondary_controller
2903 0 : .upload_tenant(tenant_shard_id)
2904 0 : .await?;
2905 :
2906 0 : json_response(StatusCode::OK, ())
2907 0 : }
2908 :
2909 0 : async fn tenant_scan_remote_handler(
2910 0 : request: Request<Body>,
2911 0 : cancel: CancellationToken,
2912 0 : ) -> Result<Response<Body>, ApiError> {
2913 0 : let state = get_state(&request);
2914 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2915 :
2916 0 : let mut response = TenantScanRemoteStorageResponse::default();
2917 :
2918 0 : let (shards, _other_keys) =
2919 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2920 0 : .await
2921 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2922 :
2923 0 : for tenant_shard_id in shards {
2924 0 : let (timeline_ids, _other_keys) =
2925 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2926 0 : .await
2927 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2928 :
2929 0 : let mut generation = Generation::none();
2930 0 : for timeline_id in timeline_ids {
2931 0 : match download_index_part(
2932 0 : &state.remote_storage,
2933 0 : &tenant_shard_id,
2934 0 : &timeline_id,
2935 : Generation::MAX,
2936 0 : &cancel,
2937 : )
2938 0 : .instrument(info_span!("download_index_part",
2939 : tenant_id=%tenant_shard_id.tenant_id,
2940 0 : shard_id=%tenant_shard_id.shard_slug(),
2941 : %timeline_id))
2942 0 : .await
2943 : {
2944 0 : Ok((index_part, index_generation, _index_mtime)) => {
2945 0 : tracing::info!(
2946 0 : "Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2947 0 : index_part.layer_metadata.len(),
2948 0 : index_part.metadata.disk_consistent_lsn()
2949 : );
2950 0 : generation = std::cmp::max(generation, index_generation);
2951 : }
2952 : Err(DownloadError::NotFound) => {
2953 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2954 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2955 0 : tracing::info!(
2956 0 : "Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping"
2957 : );
2958 0 : continue;
2959 : }
2960 0 : Err(e) => {
2961 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2962 : }
2963 : };
2964 : }
2965 :
2966 0 : let result =
2967 0 : download_tenant_manifest(&state.remote_storage, &tenant_shard_id, generation, &cancel)
2968 0 : .instrument(info_span!("download_tenant_manifest",
2969 : tenant_id=%tenant_shard_id.tenant_id,
2970 0 : shard_id=%tenant_shard_id.shard_slug()))
2971 0 : .await;
2972 0 : let stripe_size = match result {
2973 0 : Ok((manifest, _, _)) => manifest.stripe_size,
2974 0 : Err(DownloadError::NotFound) => None,
2975 0 : Err(err) => return Err(ApiError::InternalServerError(anyhow!(err))),
2976 : };
2977 :
2978 0 : response.shards.push(TenantScanRemoteStorageShard {
2979 0 : tenant_shard_id,
2980 0 : generation: generation.into(),
2981 0 : stripe_size,
2982 0 : });
2983 : }
2984 :
2985 0 : if response.shards.is_empty() {
2986 0 : return Err(ApiError::NotFound(
2987 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2988 0 : ));
2989 0 : }
2990 :
2991 0 : json_response(StatusCode::OK, response)
2992 0 : }
2993 :
2994 0 : async fn secondary_download_handler(
2995 0 : request: Request<Body>,
2996 0 : _cancel: CancellationToken,
2997 0 : ) -> Result<Response<Body>, ApiError> {
2998 0 : let state = get_state(&request);
2999 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3000 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
3001 :
3002 : // We don't need this to issue the download request, but:
3003 : // - it enables us to cleanly return 404 if we get a request for an absent shard
3004 : // - we will use this to provide status feedback in the response
3005 0 : let Some(secondary_tenant) = state
3006 0 : .tenant_manager
3007 0 : .get_secondary_tenant_shard(tenant_shard_id)
3008 : else {
3009 0 : return Err(ApiError::NotFound(
3010 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
3011 0 : ));
3012 : };
3013 :
3014 0 : let timeout = wait.unwrap_or(Duration::MAX);
3015 :
3016 0 : let result = tokio::time::timeout(
3017 0 : timeout,
3018 0 : state.secondary_controller.download_tenant(tenant_shard_id),
3019 0 : )
3020 0 : .await;
3021 :
3022 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
3023 :
3024 0 : let status = match result {
3025 : Ok(Ok(())) => {
3026 0 : if progress.layers_downloaded >= progress.layers_total {
3027 : // Download job ran to completion
3028 0 : StatusCode::OK
3029 : } else {
3030 : // Download dropped out without errors because it ran out of time budget
3031 0 : StatusCode::ACCEPTED
3032 : }
3033 : }
3034 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
3035 : // okay. We could get an error here in the unlikely edge case that the tenant
3036 : // was detached between our check above and executing the download job.
3037 0 : Ok(Err(e)) => return Err(e.into()),
3038 : // A timeout is not an error: we have started the download, we're just not done
3039 : // yet. The caller will get a response body indicating status.
3040 0 : Err(_) => StatusCode::ACCEPTED,
3041 : };
3042 :
3043 0 : json_response(status, progress)
3044 0 : }
3045 :
3046 0 : async fn wait_lsn_handler(
3047 0 : mut request: Request<Body>,
3048 0 : cancel: CancellationToken,
3049 0 : ) -> Result<Response<Body>, ApiError> {
3050 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3051 0 : let wait_lsn_request: TenantWaitLsnRequest = json_request(&mut request).await?;
3052 :
3053 0 : let state = get_state(&request);
3054 0 : let tenant = state
3055 0 : .tenant_manager
3056 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3057 :
3058 0 : let mut wait_futures = Vec::default();
3059 0 : for timeline in tenant.list_timelines() {
3060 0 : let Some(lsn) = wait_lsn_request.timelines.get(&timeline.timeline_id) else {
3061 0 : continue;
3062 : };
3063 :
3064 0 : let fut = {
3065 0 : let timeline = timeline.clone();
3066 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
3067 0 : async move {
3068 0 : timeline
3069 0 : .wait_lsn(
3070 0 : *lsn,
3071 0 : WaitLsnWaiter::HttpEndpoint,
3072 0 : WaitLsnTimeout::Custom(wait_lsn_request.timeout),
3073 0 : &ctx,
3074 0 : )
3075 0 : .await
3076 0 : }
3077 : };
3078 0 : wait_futures.push(fut);
3079 : }
3080 :
3081 0 : if wait_futures.is_empty() {
3082 0 : return json_response(StatusCode::NOT_FOUND, ());
3083 0 : }
3084 :
3085 0 : let all_done = tokio::select! {
3086 0 : results = join_all(wait_futures) => {
3087 0 : results.iter().all(|res| res.is_ok())
3088 : },
3089 0 : _ = cancel.cancelled() => {
3090 0 : return Err(ApiError::Cancelled);
3091 : }
3092 : };
3093 :
3094 0 : let status = if all_done {
3095 0 : StatusCode::OK
3096 : } else {
3097 0 : StatusCode::ACCEPTED
3098 : };
3099 :
3100 0 : json_response(status, ())
3101 0 : }
3102 :
3103 0 : async fn secondary_status_handler(
3104 0 : request: Request<Body>,
3105 0 : _cancel: CancellationToken,
3106 0 : ) -> Result<Response<Body>, ApiError> {
3107 0 : let state = get_state(&request);
3108 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3109 :
3110 0 : let Some(secondary_tenant) = state
3111 0 : .tenant_manager
3112 0 : .get_secondary_tenant_shard(tenant_shard_id)
3113 : else {
3114 0 : return Err(ApiError::NotFound(
3115 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
3116 0 : ));
3117 : };
3118 :
3119 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
3120 :
3121 0 : json_response(StatusCode::OK, progress)
3122 0 : }
3123 :
3124 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
3125 0 : json_response(
3126 : StatusCode::NOT_FOUND,
3127 0 : HttpErrorBody::from_msg("page not found".to_owned()),
3128 : )
3129 0 : }
3130 :
3131 0 : async fn post_tracing_event_handler(
3132 0 : mut r: Request<Body>,
3133 0 : _cancel: CancellationToken,
3134 0 : ) -> Result<Response<Body>, ApiError> {
3135 0 : #[derive(Debug, serde::Deserialize)]
3136 : #[serde(rename_all = "lowercase")]
3137 : enum Level {
3138 : Error,
3139 : Warn,
3140 : Info,
3141 : Debug,
3142 : Trace,
3143 : }
3144 0 : #[derive(Debug, serde::Deserialize)]
3145 : struct Request {
3146 : level: Level,
3147 : message: String,
3148 : }
3149 0 : let body: Request = json_request(&mut r)
3150 0 : .await
3151 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
3152 :
3153 0 : match body.level {
3154 0 : Level::Error => tracing::error!(?body.message),
3155 0 : Level::Warn => tracing::warn!(?body.message),
3156 0 : Level::Info => tracing::info!(?body.message),
3157 0 : Level::Debug => tracing::debug!(?body.message),
3158 0 : Level::Trace => tracing::trace!(?body.message),
3159 : }
3160 :
3161 0 : json_response(StatusCode::OK, ())
3162 0 : }
3163 :
3164 0 : async fn put_io_engine_handler(
3165 0 : mut r: Request<Body>,
3166 0 : _cancel: CancellationToken,
3167 0 : ) -> Result<Response<Body>, ApiError> {
3168 0 : check_permission(&r, None)?;
3169 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
3170 0 : crate::virtual_file::io_engine::set(kind);
3171 0 : json_response(StatusCode::OK, ())
3172 0 : }
3173 :
3174 0 : async fn put_io_mode_handler(
3175 0 : mut r: Request<Body>,
3176 0 : _cancel: CancellationToken,
3177 0 : ) -> Result<Response<Body>, ApiError> {
3178 0 : check_permission(&r, None)?;
3179 0 : let mode: IoMode = json_request(&mut r).await?;
3180 0 : crate::virtual_file::set_io_mode(mode);
3181 0 : json_response(StatusCode::OK, ())
3182 0 : }
3183 :
3184 : /// Polled by control plane.
3185 : ///
3186 : /// See [`crate::utilization`].
3187 0 : async fn get_utilization(
3188 0 : r: Request<Body>,
3189 0 : _cancel: CancellationToken,
3190 0 : ) -> Result<Response<Body>, ApiError> {
3191 0 : fail::fail_point!("get-utilization-http-handler", |_| {
3192 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
3193 0 : });
3194 :
3195 : // this probably could be completely public, but lets make that change later.
3196 0 : check_permission(&r, None)?;
3197 :
3198 0 : let state = get_state(&r);
3199 0 : let mut g = state.latest_utilization.lock().await;
3200 :
3201 0 : let regenerate_every = Duration::from_secs(1);
3202 0 : let still_valid = g
3203 0 : .as_ref()
3204 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
3205 :
3206 : // avoid needless statvfs calls even though those should be non-blocking fast.
3207 : // regenerate at most 1Hz to allow polling at any rate.
3208 0 : if !still_valid {
3209 0 : let path = state.conf.tenants_path();
3210 0 : let doc =
3211 0 : crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
3212 0 : .map_err(ApiError::InternalServerError)?;
3213 :
3214 0 : let mut buf = Vec::new();
3215 0 : serde_json::to_writer(&mut buf, &doc)
3216 0 : .context("serialize")
3217 0 : .map_err(ApiError::InternalServerError)?;
3218 :
3219 0 : let body = bytes::Bytes::from(buf);
3220 :
3221 0 : *g = Some((std::time::Instant::now(), body));
3222 0 : }
3223 :
3224 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
3225 0 : let cached = g.as_ref().expect("just set").1.clone();
3226 :
3227 0 : Response::builder()
3228 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
3229 : // thought of using http date header, but that is second precision which does not give any
3230 : // debugging aid
3231 0 : .status(StatusCode::OK)
3232 0 : .body(hyper::Body::from(cached))
3233 0 : .context("build response")
3234 0 : .map_err(ApiError::InternalServerError)
3235 0 : }
3236 :
3237 : /// HADRON
3238 0 : async fn list_tenant_visible_size_handler(
3239 0 : request: Request<Body>,
3240 0 : _cancel: CancellationToken,
3241 0 : ) -> Result<Response<Body>, ApiError> {
3242 0 : check_permission(&request, None)?;
3243 0 : let state = get_state(&request);
3244 :
3245 0 : let mut map = BTreeMap::new();
3246 0 : for (tenant_shard_id, slot) in state.tenant_manager.list() {
3247 0 : match slot {
3248 0 : TenantSlot::Attached(tenant) => {
3249 0 : let visible_size = tenant.get_visible_size();
3250 0 : map.insert(tenant_shard_id, visible_size);
3251 0 : }
3252 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
3253 0 : continue;
3254 : }
3255 : }
3256 : }
3257 :
3258 0 : json_response(StatusCode::OK, map)
3259 0 : }
3260 :
3261 0 : async fn list_aux_files(
3262 0 : mut request: Request<Body>,
3263 0 : _cancel: CancellationToken,
3264 0 : ) -> Result<Response<Body>, ApiError> {
3265 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3266 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3267 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
3268 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3269 :
3270 0 : let state = get_state(&request);
3271 :
3272 0 : let timeline =
3273 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3274 0 : .await?;
3275 :
3276 0 : let io_concurrency = IoConcurrency::spawn_from_conf(
3277 0 : state.conf.get_vectored_concurrent_io,
3278 0 : timeline.gate.enter().map_err(|_| ApiError::Cancelled)?,
3279 : );
3280 :
3281 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
3282 0 : .with_scope_timeline(&timeline);
3283 0 : let files = timeline
3284 0 : .list_aux_files(body.lsn, &ctx, io_concurrency)
3285 0 : .await?;
3286 0 : json_response(StatusCode::OK, files)
3287 0 : }
3288 :
3289 0 : async fn perf_info(
3290 0 : request: Request<Body>,
3291 0 : _cancel: CancellationToken,
3292 0 : ) -> Result<Response<Body>, ApiError> {
3293 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3294 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3295 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3296 :
3297 0 : let state = get_state(&request);
3298 :
3299 0 : let timeline =
3300 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3301 0 : .await?;
3302 :
3303 0 : let result = timeline.perf_info().await;
3304 :
3305 0 : json_response(StatusCode::OK, result)
3306 0 : }
3307 :
3308 0 : async fn ingest_aux_files(
3309 0 : mut request: Request<Body>,
3310 0 : _cancel: CancellationToken,
3311 0 : ) -> Result<Response<Body>, ApiError> {
3312 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3313 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3314 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
3315 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3316 :
3317 0 : let state = get_state(&request);
3318 :
3319 0 : let timeline =
3320 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3321 0 : .await?;
3322 :
3323 0 : let mut modification = timeline.begin_modification(
3324 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
3325 0 : );
3326 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
3327 0 : for (fname, content) in body.aux_files {
3328 0 : modification
3329 0 : .put_file(&fname, content.as_bytes(), &ctx)
3330 0 : .await
3331 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
3332 : }
3333 0 : modification
3334 0 : .commit(&ctx)
3335 0 : .await
3336 0 : .map_err(ApiError::InternalServerError)?;
3337 :
3338 0 : json_response(StatusCode::OK, ())
3339 0 : }
3340 :
3341 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
3342 : /// candidates for splitting
3343 0 : async fn post_top_tenants(
3344 0 : mut r: Request<Body>,
3345 0 : _cancel: CancellationToken,
3346 0 : ) -> Result<Response<Body>, ApiError> {
3347 0 : check_permission(&r, None)?;
3348 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
3349 0 : let state = get_state(&r);
3350 :
3351 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
3352 0 : match order_by {
3353 0 : TenantSorting::ResidentSize => sizes.resident_size,
3354 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
3355 0 : TenantSorting::MaxLogicalSizePerShard => sizes.max_logical_size_per_shard,
3356 : }
3357 0 : }
3358 :
3359 : #[derive(Eq, PartialEq)]
3360 : struct HeapItem {
3361 : metric: u64,
3362 : sizes: TopTenantShardItem,
3363 : }
3364 :
3365 : impl PartialOrd for HeapItem {
3366 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
3367 0 : Some(self.cmp(other))
3368 0 : }
3369 : }
3370 :
3371 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
3372 : /// supports popping the greatest item but not the smallest.
3373 : impl Ord for HeapItem {
3374 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
3375 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
3376 0 : }
3377 : }
3378 :
3379 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
3380 :
3381 : // FIXME: this is a lot of clones to take this tenant list
3382 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
3383 0 : if let Some(shards_lt) = request.where_shards_lt {
3384 : // Ignore tenants which already have >= this many shards
3385 0 : if tenant_shard_id.shard_count >= shards_lt {
3386 0 : continue;
3387 0 : }
3388 0 : }
3389 :
3390 0 : let sizes = match tenant_slot {
3391 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
3392 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
3393 0 : continue;
3394 : }
3395 : };
3396 0 : let metric = get_size_metric(&sizes, &request.order_by);
3397 :
3398 0 : if let Some(gt) = request.where_gt {
3399 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
3400 0 : if metric <= gt {
3401 0 : continue;
3402 0 : }
3403 0 : };
3404 :
3405 0 : match top_n.peek() {
3406 0 : None => {
3407 0 : // Top N list is empty: candidate becomes first member
3408 0 : top_n.push(HeapItem { metric, sizes });
3409 0 : }
3410 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
3411 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
3412 0 : top_n.push(HeapItem { metric, sizes });
3413 0 : }
3414 0 : Some(i) if i.metric > metric => {
3415 0 : // List is at limit and lowest value is greater than our candidate, drop it.
3416 0 : }
3417 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
3418 : }
3419 :
3420 0 : while top_n.len() > request.limit {
3421 0 : top_n.pop();
3422 0 : }
3423 : }
3424 :
3425 0 : json_response(
3426 : StatusCode::OK,
3427 : TopTenantShardsResponse {
3428 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
3429 : },
3430 : )
3431 0 : }
3432 :
3433 0 : async fn put_tenant_timeline_import_basebackup(
3434 0 : request: Request<Body>,
3435 0 : _cancel: CancellationToken,
3436 0 : ) -> Result<Response<Body>, ApiError> {
3437 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3438 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3439 0 : let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
3440 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3441 0 : let pg_version: PgMajorVersion = must_parse_query_param(&request, "pg_version")?;
3442 :
3443 0 : check_permission(&request, Some(tenant_id))?;
3444 :
3445 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3446 :
3447 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
3448 :
3449 0 : let span = info_span!("import_basebackup",
3450 0 : tenant_id=%tenant_id, timeline_id=%timeline_id, shard_id=%tenant_shard_id.shard_slug(),
3451 : base_lsn=%base_lsn, end_lsn=%end_lsn, pg_version=%pg_version);
3452 0 : async move {
3453 0 : let state = get_state(&request);
3454 0 : let tenant = state
3455 0 : .tenant_manager
3456 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3457 :
3458 0 : let broker_client = state.broker_client.clone();
3459 :
3460 0 : let mut body = StreamReader::new(
3461 0 : request
3462 0 : .into_body()
3463 0 : .map(|res| res.map_err(|error| std::io::Error::other(anyhow::anyhow!(error)))),
3464 : );
3465 :
3466 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
3467 :
3468 0 : let (timeline, timeline_ctx) = tenant
3469 0 : .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
3470 0 : .map_err(ApiError::InternalServerError)
3471 0 : .await?;
3472 :
3473 : // TODO mark timeline as not ready until it reaches end_lsn.
3474 : // We might have some wal to import as well, and we should prevent compute
3475 : // from connecting before that and writing conflicting wal.
3476 : //
3477 : // This is not relevant for pageserver->pageserver migrations, since there's
3478 : // no wal to import. But should be fixed if we want to import from postgres.
3479 :
3480 : // TODO leave clean state on error. For now you can use detach to clean
3481 : // up broken state from a failed import.
3482 :
3483 : // Import basebackup provided via CopyData
3484 0 : info!("importing basebackup");
3485 :
3486 0 : timeline
3487 0 : .import_basebackup_from_tar(
3488 0 : tenant.clone(),
3489 0 : &mut body,
3490 0 : base_lsn,
3491 0 : broker_client,
3492 0 : &timeline_ctx,
3493 0 : )
3494 0 : .await
3495 0 : .map_err(ApiError::InternalServerError)?;
3496 :
3497 : // Read the end of the tar archive.
3498 0 : read_tar_eof(body)
3499 0 : .await
3500 0 : .map_err(ApiError::InternalServerError)?;
3501 :
3502 : // TODO check checksum
3503 : // Meanwhile you can verify client-side by taking fullbackup
3504 : // and checking that it matches in size with what was imported.
3505 : // It wouldn't work if base came from vanilla postgres though,
3506 : // since we discard some log files.
3507 :
3508 0 : info!("done");
3509 0 : json_response(StatusCode::OK, ())
3510 0 : }
3511 0 : .instrument(span)
3512 0 : .await
3513 0 : }
3514 :
3515 0 : async fn put_tenant_timeline_import_wal(
3516 0 : request: Request<Body>,
3517 0 : _cancel: CancellationToken,
3518 0 : ) -> Result<Response<Body>, ApiError> {
3519 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3520 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3521 0 : let start_lsn: Lsn = must_parse_query_param(&request, "start_lsn")?;
3522 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3523 :
3524 0 : check_permission(&request, Some(tenant_id))?;
3525 :
3526 0 : let span = info_span!("import_wal", tenant_id=%tenant_id, timeline_id=%timeline_id, start_lsn=%start_lsn, end_lsn=%end_lsn);
3527 0 : async move {
3528 0 : let state = get_state(&request);
3529 :
3530 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, TenantShardId::unsharded(tenant_id), timeline_id).await?;
3531 0 : let ctx = RequestContextBuilder::new(TaskKind::MgmtRequest)
3532 0 : .download_behavior(DownloadBehavior::Warn)
3533 0 : .scope(context::Scope::new_timeline(&timeline))
3534 0 : .root();
3535 :
3536 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
3537 0 : res.map_err(|error| {
3538 0 : std::io::Error::other( anyhow::anyhow!(error))
3539 0 : })
3540 0 : }));
3541 :
3542 0 : let last_record_lsn = timeline.get_last_record_lsn();
3543 0 : if last_record_lsn != start_lsn {
3544 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}")));
3545 0 : }
3546 :
3547 : // TODO leave clean state on error. For now you can use detach to clean
3548 : // up broken state from a failed import.
3549 :
3550 : // Import wal provided via CopyData
3551 0 : info!("importing wal");
3552 0 : crate::import_datadir::import_wal_from_tar(&timeline, &mut body, start_lsn, end_lsn, &ctx).await.map_err(ApiError::InternalServerError)?;
3553 0 : info!("wal import complete");
3554 :
3555 : // Read the end of the tar archive.
3556 0 : read_tar_eof(body).await.map_err(ApiError::InternalServerError)?;
3557 :
3558 : // TODO Does it make sense to overshoot?
3559 0 : if timeline.get_last_record_lsn() < end_lsn {
3560 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}")));
3561 0 : }
3562 :
3563 : // Flush data to disk, then upload to s3. No need for a forced checkpoint.
3564 : // We only want to persist the data, and it doesn't matter if it's in the
3565 : // shape of deltas or images.
3566 0 : info!("flushing layers");
3567 0 : timeline.freeze_and_flush().await.map_err(|e| match e {
3568 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
3569 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
3570 0 : })?;
3571 :
3572 0 : info!("done");
3573 :
3574 0 : json_response(StatusCode::OK, ())
3575 0 : }.instrument(span).await
3576 0 : }
3577 :
3578 : /// Activate a timeline after its import has completed
3579 : ///
3580 : /// The endpoint is idempotent and callers are expected to retry all
3581 : /// errors until a successful response.
3582 0 : async fn activate_post_import_handler(
3583 0 : request: Request<Body>,
3584 0 : _cancel: CancellationToken,
3585 0 : ) -> Result<Response<Body>, ApiError> {
3586 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3587 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3588 :
3589 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3590 : const DEFAULT_ACTIVATE_TIMEOUT: Duration = Duration::from_secs(1);
3591 0 : let activate_timeout = parse_query_param(&request, "timeline_activate_timeout_ms")?
3592 0 : .map(Duration::from_millis)
3593 0 : .unwrap_or(DEFAULT_ACTIVATE_TIMEOUT);
3594 :
3595 0 : let span = info_span!(
3596 : "activate_post_import_handler",
3597 : tenant_id=%tenant_shard_id.tenant_id,
3598 : timeline_id=%timeline_id,
3599 0 : shard_id=%tenant_shard_id.shard_slug()
3600 : );
3601 :
3602 0 : async move {
3603 0 : let state = get_state(&request);
3604 0 : let tenant = state
3605 0 : .tenant_manager
3606 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3607 :
3608 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
3609 :
3610 0 : tenant.finalize_importing_timeline(timeline_id).await?;
3611 :
3612 0 : match tenant.get_timeline(timeline_id, false) {
3613 0 : Ok(_timeline) => {
3614 0 : // Timeline is already visible. Reset not required: fall through.
3615 0 : }
3616 : Err(GetTimelineError::NotFound { .. }) => {
3617 : // This is crude: we reset the whole tenant such that the new timeline is detected
3618 : // and activated. We can come up with something more granular in the future.
3619 : //
3620 : // Note that we only reset the tenant if required: when the timeline is
3621 : // not present in [`Tenant::timelines`].
3622 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3623 0 : state
3624 0 : .tenant_manager
3625 0 : .reset_tenant(tenant_shard_id, false, &ctx)
3626 0 : .await
3627 0 : .map_err(ApiError::InternalServerError)?;
3628 : }
3629 : Err(GetTimelineError::ShuttingDown) => {
3630 0 : return Err(ApiError::ShuttingDown);
3631 : }
3632 : Err(GetTimelineError::NotActive { .. }) => {
3633 0 : unreachable!("Called get_timeline with active_only=false");
3634 : }
3635 : }
3636 :
3637 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
3638 :
3639 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn)
3640 0 : .with_scope_timeline(&timeline);
3641 :
3642 0 : let result =
3643 0 : tokio::time::timeout(activate_timeout, timeline.wait_to_become_active(&ctx)).await;
3644 0 : match result {
3645 0 : Ok(Ok(())) => {
3646 0 : // fallthrough
3647 0 : }
3648 : // Timeline reached some other state that's not active
3649 : // TODO(vlad): if the tenant is broken, return a permananet error
3650 0 : Ok(Err(_timeline_state)) => {
3651 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
3652 0 : "Timeline activation failed"
3653 0 : )));
3654 : }
3655 : // Activation timed out
3656 : Err(_) => {
3657 0 : return Err(ApiError::Timeout("Timeline activation timed out".into()));
3658 : }
3659 : }
3660 :
3661 0 : let timeline_info = build_timeline_info(
3662 0 : &timeline, false, // include_non_incremental_logical_size,
3663 0 : false, // force_await_initial_logical_size
3664 0 : false, // include_image_consistent_lsn
3665 0 : &ctx,
3666 0 : )
3667 0 : .await
3668 0 : .context("get local timeline info")
3669 0 : .map_err(ApiError::InternalServerError)?;
3670 :
3671 0 : json_response(StatusCode::OK, timeline_info)
3672 0 : }
3673 0 : .instrument(span)
3674 0 : .await
3675 0 : }
3676 :
3677 : // [Hadron] Reset gauge metrics that are used to raised alerts. We need this API as a stop-gap measure to reset alerts
3678 : // after we manually rectify situations such as local SSD data loss. We will eventually automate this.
3679 0 : async fn hadron_reset_alert_gauges(
3680 0 : request: Request<Body>,
3681 0 : _cancel: CancellationToken,
3682 0 : ) -> Result<Response<Body>, ApiError> {
3683 0 : check_permission(&request, None)?;
3684 0 : LOCAL_DATA_LOSS_SUSPECTED.set(0);
3685 0 : json_response(StatusCode::OK, ())
3686 0 : }
3687 :
3688 : /// Read the end of a tar archive.
3689 : ///
3690 : /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
3691 : /// `tokio_tar` already read the first such block. Read the second all-zeros block,
3692 : /// and check that there is no more data after the EOF marker.
3693 : ///
3694 : /// 'tar' command can also write extra blocks of zeros, up to a record
3695 : /// size, controlled by the --record-size argument. Ignore them too.
3696 0 : async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow::Result<()> {
3697 : use tokio::io::AsyncReadExt;
3698 0 : let mut buf = [0u8; 512];
3699 :
3700 : // Read the all-zeros block, and verify it
3701 0 : let mut total_bytes = 0;
3702 0 : while total_bytes < 512 {
3703 0 : let nbytes = reader.read(&mut buf[total_bytes..]).await?;
3704 0 : total_bytes += nbytes;
3705 0 : if nbytes == 0 {
3706 0 : break;
3707 0 : }
3708 : }
3709 0 : if total_bytes < 512 {
3710 0 : anyhow::bail!("incomplete or invalid tar EOF marker");
3711 0 : }
3712 0 : if !buf.iter().all(|&x| x == 0) {
3713 0 : anyhow::bail!("invalid tar EOF marker");
3714 0 : }
3715 :
3716 : // Drain any extra zero-blocks after the EOF marker
3717 0 : let mut trailing_bytes = 0;
3718 0 : let mut seen_nonzero_bytes = false;
3719 : loop {
3720 0 : let nbytes = reader.read(&mut buf).await?;
3721 0 : trailing_bytes += nbytes;
3722 0 : if !buf.iter().all(|&x| x == 0) {
3723 0 : seen_nonzero_bytes = true;
3724 0 : }
3725 0 : if nbytes == 0 {
3726 0 : break;
3727 0 : }
3728 : }
3729 0 : if seen_nonzero_bytes {
3730 0 : anyhow::bail!("unexpected non-zero bytes after the tar archive");
3731 0 : }
3732 0 : if trailing_bytes % 512 != 0 {
3733 0 : anyhow::bail!(
3734 0 : "unexpected number of zeros ({trailing_bytes}), not divisible by tar block size (512 bytes), after the tar archive"
3735 : );
3736 0 : }
3737 0 : Ok(())
3738 0 : }
3739 :
3740 0 : async fn force_refresh_feature_flag(
3741 0 : request: Request<Body>,
3742 0 : _cancel: CancellationToken,
3743 0 : ) -> Result<Response<Body>, ApiError> {
3744 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3745 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3746 :
3747 0 : let state = get_state(&request);
3748 0 : let tenant = state
3749 0 : .tenant_manager
3750 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3751 0 : tenant
3752 0 : .feature_resolver
3753 0 : .refresh_properties_and_flags(&tenant);
3754 0 : json_response(StatusCode::OK, ())
3755 0 : }
3756 :
3757 0 : async fn tenant_evaluate_feature_flag(
3758 0 : request: Request<Body>,
3759 0 : _cancel: CancellationToken,
3760 0 : ) -> Result<Response<Body>, ApiError> {
3761 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3762 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3763 :
3764 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3765 0 : let as_type: Option<String> = parse_query_param(&request, "as")?;
3766 :
3767 0 : let state = get_state(&request);
3768 :
3769 0 : async {
3770 0 : let tenant = state
3771 0 : .tenant_manager
3772 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3773 : // TODO: the properties we get here might be stale right after it is collected. But such races are rare (updated every 10s)
3774 : // and we don't need to worry about it for now.
3775 0 : let properties = tenant.feature_resolver.collect_properties();
3776 0 : if as_type.as_deref() == Some("boolean") {
3777 0 : let result = tenant.feature_resolver.evaluate_boolean(&flag);
3778 0 : let result = result.map(|_| true).map_err(|e| e.to_string());
3779 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3780 0 : } else if as_type.as_deref() == Some("multivariate") {
3781 0 : let result = tenant.feature_resolver.evaluate_multivariate(&flag).map_err(|e| e.to_string());
3782 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3783 : } else {
3784 : // Auto infer the type of the feature flag.
3785 0 : let is_boolean = tenant.feature_resolver.is_feature_flag_boolean(&flag).map_err(|e| ApiError::InternalServerError(anyhow::anyhow!("{e}")))?;
3786 0 : if is_boolean {
3787 0 : let result = tenant.feature_resolver.evaluate_boolean(&flag);
3788 0 : let result = result.map(|_| true).map_err(|e| e.to_string());
3789 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3790 : } else {
3791 0 : let result = tenant.feature_resolver.evaluate_multivariate(&flag).map_err(|e| e.to_string());
3792 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3793 : }
3794 : }
3795 0 : }
3796 0 : .instrument(info_span!("tenant_evaluate_feature_flag", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
3797 0 : .await
3798 0 : }
3799 :
3800 0 : async fn force_override_feature_flag_for_testing_put(
3801 0 : request: Request<Body>,
3802 0 : _cancel: CancellationToken,
3803 0 : ) -> Result<Response<Body>, ApiError> {
3804 0 : check_permission(&request, None)?;
3805 :
3806 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3807 0 : let value: String = must_parse_query_param(&request, "value")?;
3808 0 : let state = get_state(&request);
3809 0 : state
3810 0 : .feature_resolver
3811 0 : .force_override_for_testing(&flag, Some(&value));
3812 0 : json_response(StatusCode::OK, ())
3813 0 : }
3814 :
3815 0 : async fn force_override_feature_flag_for_testing_delete(
3816 0 : request: Request<Body>,
3817 0 : _cancel: CancellationToken,
3818 0 : ) -> Result<Response<Body>, ApiError> {
3819 0 : check_permission(&request, None)?;
3820 :
3821 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3822 0 : let state = get_state(&request);
3823 0 : state
3824 0 : .feature_resolver
3825 0 : .force_override_for_testing(&flag, None);
3826 0 : json_response(StatusCode::OK, ())
3827 0 : }
3828 :
3829 0 : async fn update_feature_flag_spec(
3830 0 : mut request: Request<Body>,
3831 0 : _cancel: CancellationToken,
3832 0 : ) -> Result<Response<Body>, ApiError> {
3833 0 : check_permission(&request, None)?;
3834 0 : let body = json_request(&mut request).await?;
3835 0 : let state = get_state(&request);
3836 0 : state
3837 0 : .feature_resolver
3838 0 : .update(body)
3839 0 : .map_err(ApiError::InternalServerError)?;
3840 0 : json_response(StatusCode::OK, ())
3841 0 : }
3842 :
3843 : /// Common functionality of all the HTTP API handlers.
3844 : ///
3845 : /// - Adds a tracing span to each request (by `request_span`)
3846 : /// - Logs the request depending on the request method (by `request_span`)
3847 : /// - Logs the response if it was not successful (by `request_span`
3848 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
3849 : /// Future if the connection to the client is lost, but most of the pageserver code is
3850 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
3851 : /// request with a CancellationToken.
3852 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
3853 0 : where
3854 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3855 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3856 0 : {
3857 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
3858 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
3859 0 : "failpoint".into()
3860 0 : )));
3861 :
3862 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
3863 0 : anyhow::anyhow!("failpoint")
3864 0 : )));
3865 0 : }
3866 :
3867 : // Spawn a new task to handle the request, to protect the handler from unexpected
3868 : // async cancellations. Most pageserver functions are not async cancellation safe.
3869 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
3870 : // with the cancellation token.
3871 0 : let token = CancellationToken::new();
3872 0 : let cancel_guard = token.clone().drop_guard();
3873 0 : let result = request_span(request, move |r| async {
3874 0 : let handle = tokio::spawn(
3875 0 : async {
3876 0 : let token_cloned = token.clone();
3877 0 : let result = handler(r, token).await;
3878 0 : if token_cloned.is_cancelled() {
3879 : // dropguard has executed: we will never turn this result into response.
3880 : //
3881 : // at least temporarily do {:?} logging; these failures are rare enough but
3882 : // could hide difficult errors.
3883 0 : match &result {
3884 0 : Ok(response) => {
3885 0 : let status = response.status();
3886 0 : info!(%status, "Cancelled request finished successfully")
3887 : }
3888 0 : Err(e) => match e {
3889 : ApiError::ShuttingDown | ApiError::ResourceUnavailable(_) => {
3890 : // Don't log this at error severity: they are normal during lifecycle of tenants/process
3891 0 : info!("Cancelled request aborted for shutdown")
3892 : }
3893 : _ => {
3894 : // Log these in a highly visible way, because we have no client to send the response to, but
3895 : // would like to know that something went wrong.
3896 0 : error!("Cancelled request finished with an error: {e:?}")
3897 : }
3898 : },
3899 : }
3900 0 : }
3901 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
3902 : // which should suffice.
3903 : //
3904 : // there is still a chance to lose the result due to race between
3905 : // returning from here and the actual connection closing happening
3906 : // before outer task gets to execute. leaving that up for #5815.
3907 0 : result
3908 0 : }
3909 0 : .in_current_span(),
3910 : );
3911 :
3912 0 : match handle.await {
3913 : // TODO: never actually return Err from here, always Ok(...) so that we can log
3914 : // spanned errors. Call api_error_handler instead and return appropriate Body.
3915 0 : Ok(result) => result,
3916 0 : Err(e) => {
3917 : // The handler task panicked. We have a global panic handler that logs the
3918 : // panic with its backtrace, so no need to log that here. Only log a brief
3919 : // message to make it clear that we returned the error to the client.
3920 0 : error!("HTTP request handler task panicked: {e:#}");
3921 :
3922 : // Don't return an Error here, because then fallback error handler that was
3923 : // installed in make_router() will print the error. Instead, construct the
3924 : // HTTP error response and return that.
3925 0 : Ok(
3926 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
3927 0 : .into_response(),
3928 0 : )
3929 : }
3930 : }
3931 0 : })
3932 0 : .await;
3933 :
3934 0 : cancel_guard.disarm();
3935 :
3936 0 : result
3937 0 : }
3938 :
3939 : /// Like api_handler, but returns an error response if the server is built without
3940 : /// the 'testing' feature.
3941 0 : async fn testing_api_handler<R, H>(
3942 0 : desc: &str,
3943 0 : request: Request<Body>,
3944 0 : handler: H,
3945 0 : ) -> Result<Response<Body>, ApiError>
3946 0 : where
3947 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3948 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3949 0 : {
3950 0 : if cfg!(feature = "testing") {
3951 0 : api_handler(request, handler).await
3952 : } else {
3953 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
3954 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
3955 0 : ))))
3956 0 : .await
3957 : }
3958 0 : }
3959 :
3960 0 : pub fn make_router(
3961 0 : state: Arc<State>,
3962 0 : launch_ts: &'static LaunchTimestamp,
3963 0 : auth: Option<Arc<SwappableJwtAuth>>,
3964 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
3965 0 : let spec = include_bytes!("openapi_spec.yml");
3966 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
3967 0 : if auth.is_some() {
3968 0 : router = router.middleware(auth_middleware(|request| {
3969 0 : let state = get_state(request);
3970 0 : if state.allowlist_routes.contains(&request.uri().path()) {
3971 0 : None
3972 : } else {
3973 0 : state.auth.as_deref()
3974 : }
3975 0 : }))
3976 0 : }
3977 :
3978 0 : router = router.middleware(
3979 0 : endpoint::add_response_header_middleware(
3980 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
3981 0 : &launch_ts.to_string(),
3982 : )
3983 0 : .expect("construct launch timestamp header middleware"),
3984 : );
3985 :
3986 0 : let force_metric_collection_on_scrape = state.conf.force_metric_collection_on_scrape;
3987 :
3988 0 : let prometheus_metrics_handler_wrapper =
3989 0 : move |req| prometheus_metrics_handler(req, force_metric_collection_on_scrape);
3990 :
3991 0 : Ok(router
3992 0 : .data(state)
3993 0 : .get("/metrics", move |r| request_span(r, prometheus_metrics_handler_wrapper))
3994 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
3995 0 : .get("/profile/heap", |r| request_span(r, profile_heap_handler))
3996 0 : .get("/v1/status", |r| api_handler(r, status_handler))
3997 0 : .put("/v1/failpoints", |r| {
3998 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
3999 0 : })
4000 0 : .post("/v1/reload_auth_validation_keys", |r| {
4001 0 : api_handler(r, reload_auth_validation_keys_handler)
4002 0 : })
4003 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
4004 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
4005 0 : api_handler(r, tenant_status)
4006 0 : })
4007 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
4008 0 : api_handler(r, tenant_delete_handler)
4009 0 : })
4010 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
4011 0 : api_handler(r, tenant_size_handler)
4012 0 : })
4013 0 : .patch("/v1/tenant/config", |r| {
4014 0 : api_handler(r, patch_tenant_config_handler)
4015 0 : })
4016 0 : .put("/v1/tenant/config", |r| {
4017 0 : api_handler(r, update_tenant_config_handler)
4018 0 : })
4019 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
4020 0 : api_handler(r, tenant_shard_split_handler)
4021 0 : })
4022 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
4023 0 : api_handler(r, get_tenant_config_handler)
4024 0 : })
4025 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
4026 0 : api_handler(r, put_tenant_location_config_handler)
4027 0 : })
4028 0 : .get("/v1/location_config", |r| {
4029 0 : api_handler(r, list_location_config_handler)
4030 0 : })
4031 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
4032 0 : api_handler(r, get_location_config_handler)
4033 0 : })
4034 0 : .put(
4035 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
4036 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
4037 : )
4038 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
4039 0 : api_handler(r, timeline_list_handler)
4040 0 : })
4041 0 : .get("/v1/tenant/:tenant_shard_id/timeline_and_offloaded", |r| {
4042 0 : api_handler(r, timeline_and_offloaded_list_handler)
4043 0 : })
4044 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
4045 0 : api_handler(r, timeline_create_handler)
4046 0 : })
4047 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
4048 0 : api_handler(r, tenant_reset_handler)
4049 0 : })
4050 0 : .post(
4051 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
4052 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
4053 : )
4054 0 : .put(
4055 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/archival_config",
4056 0 : |r| api_handler(r, timeline_archival_config_handler),
4057 : )
4058 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
4059 0 : api_handler(r, timeline_detail_handler)
4060 0 : })
4061 0 : .get(
4062 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
4063 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
4064 : )
4065 0 : .get(
4066 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
4067 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
4068 : )
4069 0 : .post(
4070 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/patch_index_part",
4071 0 : |r| api_handler(r, timeline_patch_index_part_handler),
4072 : )
4073 0 : .post(
4074 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
4075 0 : |r| api_handler(r, lsn_lease_handler),
4076 : )
4077 0 : .put(
4078 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
4079 0 : |r| api_handler(r, timeline_gc_handler),
4080 : )
4081 0 : .get(
4082 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4083 0 : |r| api_handler(r, timeline_compact_info_handler),
4084 : )
4085 0 : .put(
4086 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4087 0 : |r| api_handler(r, timeline_compact_handler),
4088 : )
4089 0 : .delete(
4090 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4091 0 : |r| api_handler(r, timeline_cancel_compact_handler),
4092 : )
4093 0 : .put(
4094 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/offload",
4095 0 : |r| testing_api_handler("attempt timeline offload", r, timeline_offload_handler),
4096 : )
4097 0 : .put(
4098 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/mark_invisible",
4099 0 : |r| api_handler( r, timeline_mark_invisible_handler),
4100 : )
4101 0 : .put(
4102 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
4103 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
4104 : )
4105 0 : .post(
4106 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
4107 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
4108 : )
4109 0 : .get(
4110 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
4111 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
4112 : )
4113 0 : .put(
4114 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
4115 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
4116 : )
4117 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
4118 0 : api_handler(r, timeline_delete_handler)
4119 0 : })
4120 0 : .get(
4121 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
4122 0 : |r| api_handler(r, layer_map_info_handler),
4123 : )
4124 0 : .post(
4125 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
4126 0 : |r| api_handler(r, timeline_download_heatmap_layers_handler),
4127 : )
4128 0 : .delete(
4129 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
4130 0 : |r| api_handler(r, timeline_shutdown_download_heatmap_layers_handler),
4131 : )
4132 0 : .get(
4133 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
4134 0 : |r| api_handler(r, layer_download_handler),
4135 : )
4136 0 : .delete(
4137 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
4138 0 : |r| api_handler(r, evict_timeline_layer_handler),
4139 : )
4140 0 : .post(
4141 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_name/scan_disposable_keys",
4142 0 : |r| testing_api_handler("timeline_layer_scan_disposable_keys", r, timeline_layer_scan_disposable_keys),
4143 : )
4144 0 : .post(
4145 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
4146 0 : |r| api_handler(r, timeline_gc_blocking_handler),
4147 : )
4148 0 : .post(
4149 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
4150 0 : |r| api_handler(r, timeline_gc_unblocking_handler),
4151 : )
4152 0 : .get(
4153 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/page_trace",
4154 0 : |r| api_handler(r, timeline_page_trace_handler),
4155 : )
4156 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
4157 0 : api_handler(r, secondary_upload_handler)
4158 0 : })
4159 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
4160 0 : api_handler(r, tenant_scan_remote_handler)
4161 0 : })
4162 0 : .put("/v1/disk_usage_eviction/run", |r| {
4163 0 : api_handler(r, disk_usage_eviction_run)
4164 0 : })
4165 0 : .put("/v1/deletion_queue/flush", |r| {
4166 0 : api_handler(r, deletion_queue_flush)
4167 0 : })
4168 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
4169 0 : api_handler(r, secondary_status_handler)
4170 0 : })
4171 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
4172 0 : api_handler(r, secondary_download_handler)
4173 0 : })
4174 0 : .post("/v1/tenant/:tenant_shard_id/wait_lsn", |r| {
4175 0 : api_handler(r, wait_lsn_handler)
4176 0 : })
4177 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
4178 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
4179 0 : })
4180 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
4181 0 : .post("/v1/tracing/event", |r| {
4182 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
4183 0 : })
4184 0 : .get(
4185 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
4186 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
4187 : )
4188 0 : .get(
4189 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/touchpage",
4190 0 : |r| api_handler(r, touchpage_at_lsn_handler),
4191 : )
4192 0 : .get(
4193 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
4194 0 : |r| api_handler(r, timeline_collect_keyspace),
4195 : )
4196 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
4197 0 : .put("/v1/io_mode", |r| api_handler(r, put_io_mode_handler))
4198 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
4199 0 : .get("/v1/list_tenant_visible_size", |r| api_handler(r, list_tenant_visible_size_handler))
4200 0 : .post(
4201 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
4202 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
4203 : )
4204 0 : .post(
4205 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
4206 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
4207 : )
4208 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
4209 0 : .post(
4210 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
4211 0 : |r| testing_api_handler("perf_info", r, perf_info),
4212 : )
4213 0 : .put(
4214 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_basebackup",
4215 0 : |r| api_handler(r, put_tenant_timeline_import_basebackup),
4216 : )
4217 0 : .put(
4218 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_wal",
4219 0 : |r| api_handler(r, put_tenant_timeline_import_wal),
4220 : )
4221 0 : .put(
4222 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/activate_post_import",
4223 0 : |r| api_handler(r, activate_post_import_handler),
4224 : )
4225 0 : .get("/v1/tenant/:tenant_shard_id/feature_flag/:flag_key", |r| {
4226 0 : api_handler(r, tenant_evaluate_feature_flag)
4227 0 : })
4228 0 : .post("/v1/tenant/:tenant_shard_id/force_refresh_feature_flag", |r| {
4229 0 : api_handler(r, force_refresh_feature_flag)
4230 0 : })
4231 0 : .put("/v1/feature_flag/:flag_key", |r| {
4232 0 : testing_api_handler("force override feature flag - put", r, force_override_feature_flag_for_testing_put)
4233 0 : })
4234 0 : .delete("/v1/feature_flag/:flag_key", |r| {
4235 0 : testing_api_handler("force override feature flag - delete", r, force_override_feature_flag_for_testing_delete)
4236 0 : })
4237 0 : .post("/v1/feature_flag_spec", |r| {
4238 0 : api_handler(r, update_feature_flag_spec)
4239 0 : })
4240 0 : .post("/hadron-internal/reset_alert_gauges", |r| {
4241 0 : api_handler(r, hadron_reset_alert_gauges)
4242 0 : })
4243 0 : .any(handler_404))
4244 0 : }
|
