Line data Source code
1 : //!
2 : //! Management HTTP API
3 : //!
4 : use std::cmp::Reverse;
5 : use std::collections::{BinaryHeap, HashMap};
6 : use std::str::FromStr;
7 : use std::sync::Arc;
8 : use std::time::Duration;
9 :
10 : use anyhow::{Context, Result, anyhow};
11 : use enumset::EnumSet;
12 : use futures::future::join_all;
13 : use futures::{StreamExt, TryFutureExt};
14 : use http_utils::endpoint::{
15 : self, attach_openapi_ui, auth_middleware, check_permission_with, profile_cpu_handler,
16 : profile_heap_handler, prometheus_metrics_handler, request_span,
17 : };
18 : use http_utils::error::{ApiError, HttpErrorBody};
19 : use http_utils::failpoints::failpoints_handler;
20 : use http_utils::json::{json_request, json_request_maybe, json_response};
21 : use http_utils::request::{
22 : get_request_param, must_get_query_param, must_parse_query_param, parse_query_param,
23 : parse_request_param,
24 : };
25 : use http_utils::{RequestExt, RouterBuilder};
26 : use humantime::format_rfc3339;
27 : use hyper::{Body, Request, Response, StatusCode, Uri, header};
28 : use metrics::launch_timestamp::LaunchTimestamp;
29 : use pageserver_api::models::virtual_file::IoMode;
30 : use pageserver_api::models::{
31 : DetachBehavior, DownloadRemoteLayersTaskSpawnRequest, IngestAuxFilesRequest,
32 : ListAuxFilesRequest, LocationConfig, LocationConfigListResponse, LocationConfigMode, LsnLease,
33 : LsnLeaseRequest, OffloadedTimelineInfo, PageTraceEvent, ShardParameters, StatusResponse,
34 : TenantConfigPatchRequest, TenantConfigRequest, TenantDetails, TenantInfo,
35 : TenantLocationConfigRequest, TenantLocationConfigResponse, TenantScanRemoteStorageResponse,
36 : TenantScanRemoteStorageShard, TenantShardLocation, TenantShardSplitRequest,
37 : TenantShardSplitResponse, TenantSorting, TenantState, TenantWaitLsnRequest,
38 : TimelineArchivalConfigRequest, TimelineCreateRequest, TimelineCreateRequestMode,
39 : TimelineCreateRequestModeImportPgdata, TimelineGcRequest, TimelineInfo,
40 : TimelinePatchIndexPartRequest, TimelineVisibilityState, TimelinesInfoAndOffloaded,
41 : TopTenantShardItem, TopTenantShardsRequest, TopTenantShardsResponse,
42 : };
43 : use pageserver_api::shard::{ShardCount, TenantShardId};
44 : use postgres_ffi::PgMajorVersion;
45 : use remote_storage::{DownloadError, GenericRemoteStorage, TimeTravelError};
46 : use scopeguard::defer;
47 : use serde_json::json;
48 : use tenant_size_model::svg::SvgBranchKind;
49 : use tenant_size_model::{SizeResult, StorageModel};
50 : use tokio::time::Instant;
51 : use tokio_util::io::StreamReader;
52 : use tokio_util::sync::CancellationToken;
53 : use tracing::*;
54 : use utils::auth::SwappableJwtAuth;
55 : use utils::generation::Generation;
56 : use utils::id::{TenantId, TimelineId};
57 : use utils::lsn::Lsn;
58 :
59 : use crate::config::PageServerConf;
60 : use crate::context;
61 : use crate::context::{DownloadBehavior, RequestContext, RequestContextBuilder};
62 : use crate::deletion_queue::DeletionQueueClient;
63 : use crate::feature_resolver::FeatureResolver;
64 : use crate::pgdatadir_mapping::LsnForTimestamp;
65 : use crate::task_mgr::TaskKind;
66 : use crate::tenant::config::LocationConf;
67 : use crate::tenant::mgr::{
68 : GetActiveTenantError, GetTenantError, TenantManager, TenantMapError, TenantMapInsertError,
69 : TenantSlot, TenantSlotError, TenantSlotUpsertError, TenantStateError, UpsertLocationError,
70 : };
71 : use crate::tenant::remote_timeline_client::index::GcCompactionState;
72 : use crate::tenant::remote_timeline_client::{
73 : download_index_part, download_tenant_manifest, list_remote_tenant_shards, list_remote_timelines,
74 : };
75 : use crate::tenant::secondary::SecondaryController;
76 : use crate::tenant::size::ModelInputs;
77 : use crate::tenant::storage_layer::{IoConcurrency, LayerAccessStatsReset, LayerName};
78 : use crate::tenant::timeline::layer_manager::LayerManagerLockHolder;
79 : use crate::tenant::timeline::offload::{OffloadError, offload_timeline};
80 : use crate::tenant::timeline::{
81 : CompactFlags, CompactOptions, CompactRequest, CompactionError, MarkInvisibleRequest, Timeline,
82 : WaitLsnTimeout, WaitLsnWaiter, import_pgdata,
83 : };
84 : use crate::tenant::{
85 : GetTimelineError, LogicalSizeCalculationCause, OffloadedTimeline, PageReconstructError,
86 : remote_timeline_client,
87 : };
88 : use crate::{DEFAULT_PG_VERSION, disk_usage_eviction_task, tenant};
89 :
90 : // For APIs that require an Active tenant, how long should we block waiting for that state?
91 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
92 : // failed API calls while tenants are activating.
93 : #[cfg(not(feature = "testing"))]
94 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
95 :
96 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
97 : // finish attaching, if calls to remote storage are slow.
98 : #[cfg(feature = "testing")]
99 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
100 :
101 : pub struct State {
102 : conf: &'static PageServerConf,
103 : tenant_manager: Arc<TenantManager>,
104 : auth: Option<Arc<SwappableJwtAuth>>,
105 : allowlist_routes: &'static [&'static str],
106 : remote_storage: GenericRemoteStorage,
107 : broker_client: storage_broker::BrokerClientChannel,
108 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
109 : deletion_queue_client: DeletionQueueClient,
110 : secondary_controller: SecondaryController,
111 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
112 : feature_resolver: FeatureResolver,
113 : }
114 :
115 : impl State {
116 : #[allow(clippy::too_many_arguments)]
117 0 : pub fn new(
118 0 : conf: &'static PageServerConf,
119 0 : tenant_manager: Arc<TenantManager>,
120 0 : auth: Option<Arc<SwappableJwtAuth>>,
121 0 : remote_storage: GenericRemoteStorage,
122 0 : broker_client: storage_broker::BrokerClientChannel,
123 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
124 0 : deletion_queue_client: DeletionQueueClient,
125 0 : secondary_controller: SecondaryController,
126 0 : feature_resolver: FeatureResolver,
127 0 : ) -> anyhow::Result<Self> {
128 0 : let allowlist_routes = &[
129 0 : "/v1/status",
130 0 : "/v1/doc",
131 0 : "/swagger.yml",
132 0 : "/metrics",
133 0 : "/profile/cpu",
134 0 : "/profile/heap",
135 0 : ];
136 0 : Ok(Self {
137 0 : conf,
138 0 : tenant_manager,
139 0 : auth,
140 0 : allowlist_routes,
141 0 : remote_storage,
142 0 : broker_client,
143 0 : disk_usage_eviction_state,
144 0 : deletion_queue_client,
145 0 : secondary_controller,
146 0 : latest_utilization: Default::default(),
147 0 : feature_resolver,
148 0 : })
149 0 : }
150 : }
151 :
152 : #[inline(always)]
153 0 : fn get_state(request: &Request<Body>) -> &State {
154 0 : request
155 0 : .data::<Arc<State>>()
156 0 : .expect("unknown state type")
157 0 : .as_ref()
158 0 : }
159 :
160 : #[inline(always)]
161 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
162 0 : get_state(request).conf
163 0 : }
164 :
165 : /// Check that the requester is authorized to operate on given tenant
166 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
167 0 : check_permission_with(request, |claims| {
168 0 : crate::auth::check_permission(claims, tenant_id)
169 0 : })
170 0 : }
171 :
172 : impl From<PageReconstructError> for ApiError {
173 0 : fn from(pre: PageReconstructError) -> ApiError {
174 0 : match pre {
175 0 : PageReconstructError::Other(other) => ApiError::InternalServerError(other),
176 0 : PageReconstructError::MissingKey(e) => ApiError::InternalServerError(e.into()),
177 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
178 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
179 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
180 : }
181 0 : }
182 : }
183 :
184 : impl From<TenantMapInsertError> for ApiError {
185 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
186 0 : match tmie {
187 0 : TenantMapInsertError::SlotError(e) => e.into(),
188 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
189 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
190 : }
191 0 : }
192 : }
193 :
194 : impl From<TenantSlotError> for ApiError {
195 0 : fn from(e: TenantSlotError) -> ApiError {
196 : use TenantSlotError::*;
197 0 : match e {
198 0 : NotFound(tenant_id) => {
199 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
200 : }
201 : InProgress => {
202 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
203 : }
204 0 : MapState(e) => e.into(),
205 : }
206 0 : }
207 : }
208 :
209 : impl From<TenantSlotUpsertError> for ApiError {
210 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
211 : use TenantSlotUpsertError::*;
212 0 : match e {
213 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
214 0 : MapState(e) => e.into(),
215 0 : ShuttingDown(_) => ApiError::ShuttingDown,
216 : }
217 0 : }
218 : }
219 :
220 : impl From<UpsertLocationError> for ApiError {
221 0 : fn from(e: UpsertLocationError) -> ApiError {
222 : use UpsertLocationError::*;
223 0 : match e {
224 0 : BadRequest(e) => ApiError::BadRequest(e),
225 0 : Unavailable(_) => ApiError::ShuttingDown,
226 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
227 0 : Flush(e) | InternalError(e) => ApiError::InternalServerError(e),
228 : }
229 0 : }
230 : }
231 :
232 : impl From<TenantMapError> for ApiError {
233 0 : fn from(e: TenantMapError) -> ApiError {
234 : use TenantMapError::*;
235 0 : match e {
236 : StillInitializing | ShuttingDown => {
237 0 : ApiError::ResourceUnavailable(format!("{e}").into())
238 : }
239 : }
240 0 : }
241 : }
242 :
243 : impl From<TenantStateError> for ApiError {
244 0 : fn from(tse: TenantStateError) -> ApiError {
245 0 : match tse {
246 : TenantStateError::IsStopping(_) => {
247 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
248 : }
249 0 : TenantStateError::SlotError(e) => e.into(),
250 0 : TenantStateError::SlotUpsertError(e) => e.into(),
251 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
252 : }
253 0 : }
254 : }
255 :
256 : impl From<GetTenantError> for ApiError {
257 0 : fn from(tse: GetTenantError) -> ApiError {
258 0 : match tse {
259 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {tid}").into()),
260 0 : GetTenantError::ShardNotFound(tid) => {
261 0 : ApiError::NotFound(anyhow!("tenant {tid}").into())
262 : }
263 : GetTenantError::NotActive(_) => {
264 : // Why is this not `ApiError::NotFound`?
265 : // Because we must be careful to never return 404 for a tenant if it does
266 : // in fact exist locally. If we did, the caller could draw the conclusion
267 : // that it can attach the tenant to another PS and we'd be in split-brain.
268 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
269 : }
270 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
271 : }
272 0 : }
273 : }
274 :
275 : impl From<GetTimelineError> for ApiError {
276 0 : fn from(gte: GetTimelineError) -> Self {
277 : // Rationale: tenant is activated only after eligble timelines activate
278 0 : ApiError::NotFound(gte.into())
279 0 : }
280 : }
281 :
282 : impl From<GetActiveTenantError> for ApiError {
283 0 : fn from(e: GetActiveTenantError) -> ApiError {
284 0 : match e {
285 0 : GetActiveTenantError::Broken(reason) => {
286 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
287 : }
288 : GetActiveTenantError::WillNotBecomeActive(TenantState::Stopping { .. }) => {
289 0 : ApiError::ShuttingDown
290 : }
291 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{e}")),
292 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
293 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
294 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
295 0 : ApiError::ResourceUnavailable(format!("{e}").into())
296 : }
297 : GetActiveTenantError::SwitchedTenant => {
298 : // in our HTTP handlers, this error doesn't happen
299 : // TODO: separate error types
300 0 : ApiError::ResourceUnavailable("switched tenant".into())
301 : }
302 : }
303 0 : }
304 : }
305 :
306 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
307 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
308 : use crate::tenant::DeleteTimelineError::*;
309 0 : match value {
310 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
311 0 : HasChildren(children) => ApiError::PreconditionFailed(
312 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
313 0 : .into_boxed_str(),
314 0 : ),
315 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
316 0 : Cancelled => ApiError::ResourceUnavailable("shutting down".into()),
317 0 : Other(e) => ApiError::InternalServerError(e),
318 : }
319 0 : }
320 : }
321 :
322 : impl From<crate::tenant::TimelineArchivalError> for ApiError {
323 0 : fn from(value: crate::tenant::TimelineArchivalError) -> Self {
324 : use crate::tenant::TimelineArchivalError::*;
325 0 : match value {
326 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
327 0 : Timeout => ApiError::Timeout("hit pageserver internal timeout".into()),
328 0 : Cancelled => ApiError::ShuttingDown,
329 0 : e @ HasArchivedParent(_) => {
330 0 : ApiError::PreconditionFailed(e.to_string().into_boxed_str())
331 : }
332 0 : HasUnarchivedChildren(children) => ApiError::PreconditionFailed(
333 0 : format!(
334 0 : "Cannot archive timeline which has non-archived child timelines: {children:?}"
335 0 : )
336 0 : .into_boxed_str(),
337 0 : ),
338 0 : a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
339 0 : Other(e) => ApiError::InternalServerError(e),
340 : }
341 0 : }
342 : }
343 :
344 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
345 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
346 : use crate::tenant::mgr::DeleteTimelineError::*;
347 0 : match value {
348 : // Report Precondition failed so client can distinguish between
349 : // "tenant is missing" case from "timeline is missing"
350 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
351 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
352 0 : ),
353 0 : Tenant(t) => ApiError::from(t),
354 0 : Timeline(t) => ApiError::from(t),
355 : }
356 0 : }
357 : }
358 :
359 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
360 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
361 : use crate::tenant::mgr::DeleteTenantError::*;
362 0 : match value {
363 0 : SlotError(e) => e.into(),
364 0 : Other(o) => ApiError::InternalServerError(o),
365 0 : Cancelled => ApiError::ShuttingDown,
366 : }
367 0 : }
368 : }
369 :
370 : impl From<crate::tenant::secondary::SecondaryTenantError> for ApiError {
371 0 : fn from(ste: crate::tenant::secondary::SecondaryTenantError) -> ApiError {
372 : use crate::tenant::secondary::SecondaryTenantError;
373 0 : match ste {
374 0 : SecondaryTenantError::GetTenant(gte) => gte.into(),
375 0 : SecondaryTenantError::ShuttingDown => ApiError::ShuttingDown,
376 : }
377 0 : }
378 : }
379 :
380 : impl From<crate::tenant::FinalizeTimelineImportError> for ApiError {
381 0 : fn from(err: crate::tenant::FinalizeTimelineImportError) -> ApiError {
382 : use crate::tenant::FinalizeTimelineImportError::*;
383 0 : match err {
384 : ImportTaskStillRunning => {
385 0 : ApiError::ResourceUnavailable("Import task still running".into())
386 : }
387 0 : ShuttingDown => ApiError::ShuttingDown,
388 : }
389 0 : }
390 : }
391 :
392 : // Helper function to construct a TimelineInfo struct for a timeline
393 0 : async fn build_timeline_info(
394 0 : timeline: &Arc<Timeline>,
395 0 : include_non_incremental_logical_size: bool,
396 0 : force_await_initial_logical_size: bool,
397 0 : ctx: &RequestContext,
398 0 : ) -> anyhow::Result<TimelineInfo> {
399 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
400 :
401 0 : if force_await_initial_logical_size {
402 0 : timeline.clone().await_initial_logical_size().await
403 0 : }
404 :
405 0 : let mut info = build_timeline_info_common(
406 0 : timeline,
407 0 : ctx,
408 0 : tenant::timeline::GetLogicalSizePriority::Background,
409 0 : )
410 0 : .await?;
411 0 : if include_non_incremental_logical_size {
412 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
413 : // Otherwise, if someone deletes the timeline / detaches the tenant while
414 : // we're executing this function, we will outlive the timeline on-disk state.
415 : info.current_logical_size_non_incremental = Some(
416 0 : timeline
417 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
418 0 : .await?,
419 : );
420 0 : }
421 0 : Ok(info)
422 0 : }
423 :
424 0 : async fn build_timeline_info_common(
425 0 : timeline: &Arc<Timeline>,
426 0 : ctx: &RequestContext,
427 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
428 0 : ) -> anyhow::Result<TimelineInfo> {
429 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
430 0 : let initdb_lsn = timeline.initdb_lsn;
431 0 : let last_record_lsn = timeline.get_last_record_lsn();
432 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
433 0 : let guard = timeline.last_received_wal.lock().unwrap();
434 0 : if let Some(info) = guard.as_ref() {
435 0 : (
436 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
437 0 : Some(info.last_received_msg_lsn),
438 0 : Some(info.last_received_msg_ts),
439 0 : )
440 : } else {
441 0 : (None, None, None)
442 : }
443 : };
444 :
445 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
446 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
447 0 : Lsn(0) => None,
448 0 : lsn @ Lsn(_) => Some(lsn),
449 : };
450 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
451 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
452 0 : let state = timeline.current_state();
453 : // Report is_archived = false if the timeline is still loading
454 0 : let is_archived = timeline.is_archived().unwrap_or(false);
455 0 : let remote_consistent_lsn_projected = timeline
456 0 : .get_remote_consistent_lsn_projected()
457 0 : .unwrap_or(Lsn(0));
458 0 : let remote_consistent_lsn_visible = timeline
459 0 : .get_remote_consistent_lsn_visible()
460 0 : .unwrap_or(Lsn(0));
461 0 : let is_invisible = timeline.remote_client.is_invisible().unwrap_or(false);
462 :
463 0 : let walreceiver_status = timeline.walreceiver_status();
464 :
465 0 : let (pitr_history_size, within_ancestor_pitr) = timeline.get_pitr_history_stats();
466 :
467 : // Externally, expose the lowest LSN that can be used to create a branch.
468 : // Internally we distinguish between the planned GC cutoff (PITR point) and the "applied" GC cutoff (where we
469 : // actually trimmed data to), which can pass each other when PITR is changed.
470 0 : let min_readable_lsn = std::cmp::max(
471 0 : timeline.get_gc_cutoff_lsn().unwrap_or_default(),
472 0 : *timeline.get_applied_gc_cutoff_lsn(),
473 : );
474 :
475 0 : let info = TimelineInfo {
476 0 : tenant_id: timeline.tenant_shard_id,
477 0 : timeline_id: timeline.timeline_id,
478 0 : ancestor_timeline_id,
479 0 : ancestor_lsn,
480 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
481 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
482 0 : remote_consistent_lsn_visible,
483 0 : initdb_lsn,
484 0 : last_record_lsn,
485 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
486 0 : min_readable_lsn,
487 0 : applied_gc_cutoff_lsn: *timeline.get_applied_gc_cutoff_lsn(),
488 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
489 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
490 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
491 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
492 : },
493 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
494 0 : current_physical_size,
495 0 : current_logical_size_non_incremental: None,
496 0 : pitr_history_size,
497 0 : within_ancestor_pitr,
498 0 : timeline_dir_layer_file_size_sum: None,
499 0 : wal_source_connstr,
500 0 : last_received_msg_lsn,
501 0 : last_received_msg_ts,
502 0 : pg_version: timeline.pg_version,
503 :
504 0 : state,
505 0 : is_archived: Some(is_archived),
506 0 : rel_size_migration: Some(timeline.get_rel_size_v2_status()),
507 0 : is_invisible: Some(is_invisible),
508 :
509 0 : walreceiver_status,
510 : };
511 0 : Ok(info)
512 0 : }
513 :
514 0 : fn build_timeline_offloaded_info(offloaded: &Arc<OffloadedTimeline>) -> OffloadedTimelineInfo {
515 : let &OffloadedTimeline {
516 0 : tenant_shard_id,
517 0 : timeline_id,
518 0 : ancestor_retain_lsn,
519 0 : ancestor_timeline_id,
520 0 : archived_at,
521 : ..
522 0 : } = offloaded.as_ref();
523 0 : OffloadedTimelineInfo {
524 0 : tenant_id: tenant_shard_id,
525 0 : timeline_id,
526 0 : ancestor_retain_lsn,
527 0 : ancestor_timeline_id,
528 0 : archived_at: archived_at.and_utc(),
529 0 : }
530 0 : }
531 :
532 : // healthcheck handler
533 0 : async fn status_handler(
534 0 : request: Request<Body>,
535 0 : _cancel: CancellationToken,
536 0 : ) -> Result<Response<Body>, ApiError> {
537 0 : check_permission(&request, None)?;
538 0 : let config = get_config(&request);
539 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
540 0 : }
541 :
542 0 : async fn reload_auth_validation_keys_handler(
543 0 : request: Request<Body>,
544 0 : _cancel: CancellationToken,
545 0 : ) -> Result<Response<Body>, ApiError> {
546 0 : check_permission(&request, None)?;
547 0 : let config = get_config(&request);
548 0 : let state = get_state(&request);
549 0 : let Some(shared_auth) = &state.auth else {
550 0 : return json_response(StatusCode::BAD_REQUEST, ());
551 : };
552 : // unwrap is ok because check is performed when creating config, so path is set and exists
553 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
554 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
555 :
556 0 : match utils::auth::JwtAuth::from_key_path(key_path) {
557 0 : Ok(new_auth) => {
558 0 : shared_auth.swap(new_auth);
559 0 : json_response(StatusCode::OK, ())
560 : }
561 0 : Err(e) => {
562 0 : let err_msg = "Error reloading public keys";
563 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
564 0 : json_response(
565 : StatusCode::INTERNAL_SERVER_ERROR,
566 0 : HttpErrorBody::from_msg(err_msg.to_string()),
567 : )
568 : }
569 : }
570 0 : }
571 :
572 0 : async fn timeline_create_handler(
573 0 : mut request: Request<Body>,
574 0 : _cancel: CancellationToken,
575 0 : ) -> Result<Response<Body>, ApiError> {
576 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
577 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
578 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
579 :
580 0 : let new_timeline_id = request_data.new_timeline_id;
581 : // fill in the default pg_version if not provided & convert request into domain model
582 0 : let params: tenant::CreateTimelineParams = match request_data.mode {
583 : TimelineCreateRequestMode::Bootstrap {
584 0 : existing_initdb_timeline_id,
585 0 : pg_version,
586 0 : } => tenant::CreateTimelineParams::Bootstrap(tenant::CreateTimelineParamsBootstrap {
587 0 : new_timeline_id,
588 0 : existing_initdb_timeline_id,
589 0 : pg_version: pg_version.unwrap_or(DEFAULT_PG_VERSION),
590 0 : }),
591 : TimelineCreateRequestMode::Branch {
592 0 : ancestor_timeline_id,
593 0 : ancestor_start_lsn,
594 : read_only: _,
595 : pg_version: _,
596 0 : } => tenant::CreateTimelineParams::Branch(tenant::CreateTimelineParamsBranch {
597 0 : new_timeline_id,
598 0 : ancestor_timeline_id,
599 0 : ancestor_start_lsn,
600 0 : }),
601 : TimelineCreateRequestMode::ImportPgdata {
602 : import_pgdata:
603 : TimelineCreateRequestModeImportPgdata {
604 0 : location,
605 0 : idempotency_key,
606 : },
607 : } => tenant::CreateTimelineParams::ImportPgdata(tenant::CreateTimelineParamsImportPgdata {
608 0 : idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new(
609 0 : idempotency_key.0,
610 : ),
611 0 : new_timeline_id,
612 : location: {
613 : use import_pgdata::index_part_format::Location;
614 : use pageserver_api::models::ImportPgdataLocation;
615 0 : match location {
616 : #[cfg(feature = "testing")]
617 0 : ImportPgdataLocation::LocalFs { path } => Location::LocalFs { path },
618 : ImportPgdataLocation::AwsS3 {
619 0 : region,
620 0 : bucket,
621 0 : key,
622 0 : } => Location::AwsS3 {
623 0 : region,
624 0 : bucket,
625 0 : key,
626 0 : },
627 : }
628 : },
629 : }),
630 : };
631 :
632 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
633 :
634 0 : let state = get_state(&request);
635 :
636 0 : async {
637 0 : let tenant = state
638 0 : .tenant_manager
639 0 : .get_attached_tenant_shard(tenant_shard_id)?;
640 :
641 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
642 :
643 : // earlier versions of the code had pg_version and ancestor_lsn in the span
644 : // => continue to provide that information, but, through a log message that doesn't require us to destructure
645 0 : tracing::info!(?params, "creating timeline");
646 :
647 0 : match tenant
648 0 : .create_timeline(params, state.broker_client.clone(), &ctx)
649 0 : .await
650 : {
651 0 : Ok(new_timeline) => {
652 : // Created. Construct a TimelineInfo for it.
653 0 : let timeline_info = build_timeline_info_common(
654 0 : &new_timeline,
655 0 : &ctx,
656 0 : tenant::timeline::GetLogicalSizePriority::User,
657 0 : )
658 0 : .await
659 0 : .map_err(ApiError::InternalServerError)?;
660 0 : json_response(StatusCode::CREATED, timeline_info)
661 : }
662 0 : Err(_) if tenant.cancel.is_cancelled() => {
663 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
664 0 : json_response(
665 : StatusCode::SERVICE_UNAVAILABLE,
666 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
667 : )
668 : }
669 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
670 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
671 : }
672 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
673 : StatusCode::TOO_MANY_REQUESTS,
674 0 : HttpErrorBody::from_msg(e.to_string()),
675 : ),
676 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
677 : StatusCode::NOT_ACCEPTABLE,
678 0 : HttpErrorBody::from_msg(format!("{err:#}")),
679 : ),
680 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
681 : StatusCode::SERVICE_UNAVAILABLE,
682 0 : HttpErrorBody::from_msg(e.to_string()),
683 : ),
684 0 : Err(e @ tenant::CreateTimelineError::AncestorArchived) => json_response(
685 : StatusCode::NOT_ACCEPTABLE,
686 0 : HttpErrorBody::from_msg(e.to_string()),
687 : ),
688 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
689 : StatusCode::SERVICE_UNAVAILABLE,
690 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
691 : ),
692 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
693 : }
694 0 : }
695 0 : .instrument(info_span!("timeline_create",
696 : tenant_id = %tenant_shard_id.tenant_id,
697 0 : shard_id = %tenant_shard_id.shard_slug(),
698 : timeline_id = %new_timeline_id,
699 : ))
700 0 : .await
701 0 : }
702 :
703 0 : async fn timeline_list_handler(
704 0 : request: Request<Body>,
705 0 : _cancel: CancellationToken,
706 0 : ) -> Result<Response<Body>, ApiError> {
707 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
708 0 : let include_non_incremental_logical_size: Option<bool> =
709 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
710 0 : let force_await_initial_logical_size: Option<bool> =
711 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
712 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
713 :
714 0 : let state = get_state(&request);
715 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
716 :
717 0 : let response_data = async {
718 0 : let tenant = state
719 0 : .tenant_manager
720 0 : .get_attached_tenant_shard(tenant_shard_id)?;
721 :
722 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
723 :
724 0 : let timelines = tenant.list_timelines();
725 :
726 0 : let mut response_data = Vec::with_capacity(timelines.len());
727 0 : for timeline in timelines {
728 0 : let timeline_info = build_timeline_info(
729 0 : &timeline,
730 0 : include_non_incremental_logical_size.unwrap_or(false),
731 0 : force_await_initial_logical_size.unwrap_or(false),
732 0 : &ctx,
733 : )
734 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
735 0 : .await
736 0 : .context("Failed to build timeline info")
737 0 : .map_err(ApiError::InternalServerError)?;
738 :
739 0 : response_data.push(timeline_info);
740 : }
741 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
742 0 : }
743 0 : .instrument(info_span!("timeline_list",
744 : tenant_id = %tenant_shard_id.tenant_id,
745 0 : shard_id = %tenant_shard_id.shard_slug()))
746 0 : .await?;
747 :
748 0 : json_response(StatusCode::OK, response_data)
749 0 : }
750 :
751 0 : async fn timeline_and_offloaded_list_handler(
752 0 : request: Request<Body>,
753 0 : _cancel: CancellationToken,
754 0 : ) -> Result<Response<Body>, ApiError> {
755 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
756 0 : let include_non_incremental_logical_size: Option<bool> =
757 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
758 0 : let force_await_initial_logical_size: Option<bool> =
759 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
760 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
761 :
762 0 : let state = get_state(&request);
763 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
764 :
765 0 : let response_data = async {
766 0 : let tenant = state
767 0 : .tenant_manager
768 0 : .get_attached_tenant_shard(tenant_shard_id)?;
769 :
770 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
771 :
772 0 : let (timelines, offloadeds) = tenant.list_timelines_and_offloaded();
773 :
774 0 : let mut timeline_infos = Vec::with_capacity(timelines.len());
775 0 : for timeline in timelines {
776 0 : let timeline_info = build_timeline_info(
777 0 : &timeline,
778 0 : include_non_incremental_logical_size.unwrap_or(false),
779 0 : force_await_initial_logical_size.unwrap_or(false),
780 0 : &ctx,
781 : )
782 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
783 0 : .await
784 0 : .context("Failed to build timeline info")
785 0 : .map_err(ApiError::InternalServerError)?;
786 :
787 0 : timeline_infos.push(timeline_info);
788 : }
789 0 : let offloaded_infos = offloadeds
790 0 : .into_iter()
791 0 : .map(|offloaded| build_timeline_offloaded_info(&offloaded))
792 0 : .collect::<Vec<_>>();
793 0 : let res = TimelinesInfoAndOffloaded {
794 0 : timelines: timeline_infos,
795 0 : offloaded: offloaded_infos,
796 0 : };
797 0 : Ok::<TimelinesInfoAndOffloaded, ApiError>(res)
798 0 : }
799 0 : .instrument(info_span!("timeline_and_offloaded_list",
800 : tenant_id = %tenant_shard_id.tenant_id,
801 0 : shard_id = %tenant_shard_id.shard_slug()))
802 0 : .await?;
803 :
804 0 : json_response(StatusCode::OK, response_data)
805 0 : }
806 :
807 0 : async fn timeline_preserve_initdb_handler(
808 0 : request: Request<Body>,
809 0 : _cancel: CancellationToken,
810 0 : ) -> Result<Response<Body>, ApiError> {
811 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
812 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
813 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
814 0 : let state = get_state(&request);
815 :
816 : // Part of the process for disaster recovery from safekeeper-stored WAL:
817 : // If we don't recover into a new timeline but want to keep the timeline ID,
818 : // then the initdb archive is deleted. This endpoint copies it to a different
819 : // location where timeline recreation cand find it.
820 :
821 0 : async {
822 0 : let tenant = state
823 0 : .tenant_manager
824 0 : .get_attached_tenant_shard(tenant_shard_id)?;
825 :
826 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
827 :
828 0 : timeline
829 0 : .preserve_initdb_archive()
830 0 : .await
831 0 : .context("preserving initdb archive")
832 0 : .map_err(ApiError::InternalServerError)?;
833 :
834 0 : Ok::<_, ApiError>(())
835 0 : }
836 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
837 : tenant_id = %tenant_shard_id.tenant_id,
838 0 : shard_id = %tenant_shard_id.shard_slug(),
839 : %timeline_id))
840 0 : .await?;
841 :
842 0 : json_response(StatusCode::OK, ())
843 0 : }
844 :
845 0 : async fn timeline_archival_config_handler(
846 0 : mut request: Request<Body>,
847 0 : _cancel: CancellationToken,
848 0 : ) -> Result<Response<Body>, ApiError> {
849 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
850 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
851 :
852 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
853 :
854 0 : let request_data: TimelineArchivalConfigRequest = json_request(&mut request).await?;
855 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
856 0 : let state = get_state(&request);
857 :
858 0 : async {
859 0 : let tenant = state
860 0 : .tenant_manager
861 0 : .get_attached_tenant_shard(tenant_shard_id)?;
862 :
863 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
864 :
865 0 : tenant
866 0 : .apply_timeline_archival_config(
867 0 : timeline_id,
868 0 : request_data.state,
869 0 : state.broker_client.clone(),
870 0 : ctx,
871 0 : )
872 0 : .await?;
873 0 : Ok::<_, ApiError>(())
874 0 : }
875 0 : .instrument(info_span!("timeline_archival_config",
876 : tenant_id = %tenant_shard_id.tenant_id,
877 0 : shard_id = %tenant_shard_id.shard_slug(),
878 : state = ?request_data.state,
879 : %timeline_id))
880 0 : .await?;
881 :
882 0 : json_response(StatusCode::OK, ())
883 0 : }
884 :
885 : /// 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
886 : /// measure only.
887 : ///
888 : /// Some examples of safe patches:
889 : /// - 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.
890 : /// - Force set the index part to use reldir v2 (migrating/migrated).
891 : ///
892 : /// Some examples of unsafe patches:
893 : /// - 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
894 : /// errors.
895 : /// - Decrease the gc_cutoff without validating the data really exists. It will cause read errors in the background.
896 0 : async fn timeline_patch_index_part_handler(
897 0 : mut request: Request<Body>,
898 0 : _cancel: CancellationToken,
899 0 : ) -> Result<Response<Body>, ApiError> {
900 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
901 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
902 :
903 0 : let request_data: TimelinePatchIndexPartRequest = json_request(&mut request).await?;
904 0 : check_permission(&request, None)?; // require global permission for this request
905 0 : let state = get_state(&request);
906 :
907 0 : async {
908 0 : let timeline =
909 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
910 0 : .await?;
911 :
912 0 : if let Some(rel_size_migration) = request_data.rel_size_migration {
913 0 : timeline
914 0 : .update_rel_size_v2_status(rel_size_migration)
915 0 : .map_err(ApiError::InternalServerError)?;
916 0 : }
917 :
918 0 : if let Some(gc_compaction_last_completed_lsn) =
919 0 : request_data.gc_compaction_last_completed_lsn
920 : {
921 0 : timeline
922 0 : .update_gc_compaction_state(GcCompactionState {
923 0 : last_completed_lsn: gc_compaction_last_completed_lsn,
924 0 : })
925 0 : .map_err(ApiError::InternalServerError)?;
926 0 : }
927 :
928 0 : if let Some(applied_gc_cutoff_lsn) = request_data.applied_gc_cutoff_lsn {
929 0 : {
930 0 : let guard = timeline.applied_gc_cutoff_lsn.lock_for_write();
931 0 : guard.store_and_unlock(applied_gc_cutoff_lsn);
932 0 : }
933 0 : }
934 :
935 0 : if request_data.force_index_update {
936 0 : timeline
937 0 : .remote_client
938 0 : .force_schedule_index_upload()
939 0 : .context("force schedule index upload")
940 0 : .map_err(ApiError::InternalServerError)?;
941 0 : }
942 :
943 0 : Ok::<_, ApiError>(())
944 0 : }
945 0 : .instrument(info_span!("timeline_patch_index_part",
946 : tenant_id = %tenant_shard_id.tenant_id,
947 0 : shard_id = %tenant_shard_id.shard_slug(),
948 : %timeline_id))
949 0 : .await?;
950 :
951 0 : json_response(StatusCode::OK, ())
952 0 : }
953 :
954 0 : async fn timeline_detail_handler(
955 0 : request: Request<Body>,
956 0 : _cancel: CancellationToken,
957 0 : ) -> Result<Response<Body>, ApiError> {
958 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
959 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
960 0 : let include_non_incremental_logical_size: Option<bool> =
961 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
962 0 : let force_await_initial_logical_size: Option<bool> =
963 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
964 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
965 :
966 : // Logical size calculation needs downloading.
967 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
968 0 : let state = get_state(&request);
969 :
970 0 : let timeline_info = async {
971 0 : let tenant = state
972 0 : .tenant_manager
973 0 : .get_attached_tenant_shard(tenant_shard_id)?;
974 :
975 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
976 :
977 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
978 0 : let ctx = &ctx.with_scope_timeline(&timeline);
979 :
980 0 : let timeline_info = build_timeline_info(
981 0 : &timeline,
982 0 : include_non_incremental_logical_size.unwrap_or(false),
983 0 : force_await_initial_logical_size.unwrap_or(false),
984 0 : ctx,
985 0 : )
986 0 : .await
987 0 : .context("get local timeline info")
988 0 : .map_err(ApiError::InternalServerError)?;
989 :
990 0 : Ok::<_, ApiError>(timeline_info)
991 0 : }
992 0 : .instrument(info_span!("timeline_detail",
993 : tenant_id = %tenant_shard_id.tenant_id,
994 0 : shard_id = %tenant_shard_id.shard_slug(),
995 : %timeline_id))
996 0 : .await?;
997 :
998 0 : json_response(StatusCode::OK, timeline_info)
999 0 : }
1000 :
1001 0 : async fn get_lsn_by_timestamp_handler(
1002 0 : request: Request<Body>,
1003 0 : cancel: CancellationToken,
1004 0 : ) -> Result<Response<Body>, ApiError> {
1005 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1006 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1007 0 : let state = get_state(&request);
1008 :
1009 0 : if !tenant_shard_id.is_shard_zero() {
1010 : // Requires SLRU contents, which are only stored on shard zero
1011 0 : return Err(ApiError::BadRequest(anyhow!(
1012 0 : "Lsn calculations by timestamp are only available on shard zero"
1013 0 : )));
1014 0 : }
1015 :
1016 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1017 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
1018 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1019 0 : .with_context(|| format!("Invalid time: {timestamp_raw:?}"))
1020 0 : .map_err(ApiError::BadRequest)?;
1021 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
1022 :
1023 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
1024 :
1025 0 : let timeline =
1026 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1027 0 : .await?;
1028 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1029 0 : .with_scope_timeline(&timeline);
1030 0 : let result = timeline
1031 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
1032 0 : .await?;
1033 :
1034 : #[derive(serde::Serialize, Debug)]
1035 : struct Result {
1036 : lsn: Lsn,
1037 : kind: &'static str,
1038 : #[serde(default)]
1039 : #[serde(skip_serializing_if = "Option::is_none")]
1040 : #[serde(flatten)]
1041 : lease: Option<LsnLease>,
1042 : }
1043 0 : let (lsn, kind) = match result {
1044 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
1045 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
1046 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
1047 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
1048 : };
1049 :
1050 0 : let lease = if with_lease {
1051 0 : timeline
1052 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
1053 0 : .inspect_err(|_| {
1054 0 : warn!("fail to grant a lease to {}", lsn);
1055 0 : })
1056 0 : .ok()
1057 : } else {
1058 0 : None
1059 : };
1060 :
1061 0 : let result = Result { lsn, kind, lease };
1062 0 : let valid_until = result
1063 0 : .lease
1064 0 : .as_ref()
1065 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
1066 0 : tracing::info!(
1067 : lsn=?result.lsn,
1068 : kind=%result.kind,
1069 : timestamp=%timestamp_raw,
1070 : valid_until=?valid_until,
1071 0 : "lsn_by_timestamp finished"
1072 : );
1073 0 : json_response(StatusCode::OK, result)
1074 0 : }
1075 :
1076 0 : async fn get_timestamp_of_lsn_handler(
1077 0 : request: Request<Body>,
1078 0 : _cancel: CancellationToken,
1079 0 : ) -> Result<Response<Body>, ApiError> {
1080 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1081 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1082 0 : let state = get_state(&request);
1083 :
1084 0 : if !tenant_shard_id.is_shard_zero() {
1085 : // Requires SLRU contents, which are only stored on shard zero
1086 0 : return Err(ApiError::BadRequest(anyhow!(
1087 0 : "Timestamp calculations by lsn are only available on shard zero"
1088 0 : )));
1089 0 : }
1090 :
1091 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1092 :
1093 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
1094 0 : let lsn = Lsn::from_str(&lsn_str)
1095 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
1096 0 : .map_err(ApiError::BadRequest)?;
1097 :
1098 0 : let timeline =
1099 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1100 0 : .await?;
1101 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1102 0 : .with_scope_timeline(&timeline);
1103 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
1104 :
1105 0 : match result {
1106 0 : Some(time) => {
1107 0 : let time = format_rfc3339(
1108 0 : postgres_ffi::try_from_pg_timestamp(time).map_err(ApiError::InternalServerError)?,
1109 : )
1110 0 : .to_string();
1111 0 : json_response(StatusCode::OK, time)
1112 : }
1113 0 : None => Err(ApiError::PreconditionFailed(
1114 0 : format!("Timestamp for lsn {lsn} not found").into(),
1115 0 : )),
1116 : }
1117 0 : }
1118 :
1119 0 : async fn timeline_delete_handler(
1120 0 : request: Request<Body>,
1121 0 : _cancel: CancellationToken,
1122 0 : ) -> Result<Response<Body>, ApiError> {
1123 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1124 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1125 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1126 :
1127 0 : let state = get_state(&request);
1128 :
1129 0 : let tenant = state
1130 0 : .tenant_manager
1131 0 : .get_attached_tenant_shard(tenant_shard_id)
1132 0 : .map_err(|e| {
1133 0 : match e {
1134 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
1135 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
1136 : GetTenantError::NotFound(_) | GetTenantError::ShardNotFound(_) => {
1137 0 : ApiError::PreconditionFailed(
1138 0 : "Requested tenant is missing".to_string().into_boxed_str(),
1139 0 : )
1140 : }
1141 0 : e => e.into(),
1142 : }
1143 0 : })?;
1144 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1145 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))
1146 0 : .await?;
1147 :
1148 0 : json_response(StatusCode::ACCEPTED, ())
1149 0 : }
1150 :
1151 0 : async fn tenant_reset_handler(
1152 0 : request: Request<Body>,
1153 0 : _cancel: CancellationToken,
1154 0 : ) -> Result<Response<Body>, ApiError> {
1155 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1156 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1157 :
1158 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
1159 :
1160 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1161 0 : let state = get_state(&request);
1162 0 : state
1163 0 : .tenant_manager
1164 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
1165 0 : .await
1166 0 : .map_err(ApiError::InternalServerError)?;
1167 :
1168 0 : json_response(StatusCode::OK, ())
1169 0 : }
1170 :
1171 0 : async fn tenant_list_handler(
1172 0 : request: Request<Body>,
1173 0 : _cancel: CancellationToken,
1174 0 : ) -> Result<Response<Body>, ApiError> {
1175 0 : check_permission(&request, None)?;
1176 0 : let state = get_state(&request);
1177 :
1178 0 : let response_data = state
1179 0 : .tenant_manager
1180 0 : .list_tenants()
1181 0 : .map_err(|_| {
1182 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
1183 0 : })?
1184 0 : .iter()
1185 0 : .map(|(id, state, gen_)| TenantInfo {
1186 0 : id: *id,
1187 0 : state: state.clone(),
1188 0 : current_physical_size: None,
1189 0 : attachment_status: state.attachment_status(),
1190 0 : generation: (*gen_)
1191 0 : .into()
1192 0 : .expect("Tenants are always attached with a generation"),
1193 0 : gc_blocking: None,
1194 0 : })
1195 0 : .collect::<Vec<TenantInfo>>();
1196 :
1197 0 : json_response(StatusCode::OK, response_data)
1198 0 : }
1199 :
1200 0 : async fn tenant_status(
1201 0 : request: Request<Body>,
1202 0 : _cancel: CancellationToken,
1203 0 : ) -> Result<Response<Body>, ApiError> {
1204 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1205 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1206 0 : let state = get_state(&request);
1207 :
1208 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1209 0 : let activate = true;
1210 : #[cfg(feature = "testing")]
1211 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1212 :
1213 0 : let tenant_info = async {
1214 0 : let tenant = state
1215 0 : .tenant_manager
1216 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1217 :
1218 0 : if activate {
1219 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1220 : // called, but it is still valid to return 200 and describe the current state of the tenant
1221 : // if it doesn't make it into an active state.
1222 0 : tenant
1223 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1224 0 : .await
1225 0 : .ok();
1226 0 : }
1227 :
1228 : // Calculate total physical size of all timelines
1229 0 : let mut current_physical_size = 0;
1230 0 : for timeline in tenant.list_timelines().iter() {
1231 0 : current_physical_size += timeline.layer_size_sum().await;
1232 : }
1233 :
1234 0 : let state = tenant.current_state();
1235 : Result::<_, ApiError>::Ok(TenantDetails {
1236 : tenant_info: TenantInfo {
1237 0 : id: tenant_shard_id,
1238 0 : state: state.clone(),
1239 0 : current_physical_size: Some(current_physical_size),
1240 0 : attachment_status: state.attachment_status(),
1241 0 : generation: tenant
1242 0 : .generation()
1243 0 : .into()
1244 0 : .expect("Tenants are always attached with a generation"),
1245 0 : gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
1246 : },
1247 0 : walredo: tenant.wal_redo_manager_status(),
1248 0 : timelines: tenant.list_timeline_ids(),
1249 : })
1250 0 : }
1251 0 : .instrument(info_span!("tenant_status_handler",
1252 : tenant_id = %tenant_shard_id.tenant_id,
1253 0 : shard_id = %tenant_shard_id.shard_slug()))
1254 0 : .await?;
1255 :
1256 0 : json_response(StatusCode::OK, tenant_info)
1257 0 : }
1258 :
1259 0 : async fn tenant_delete_handler(
1260 0 : request: Request<Body>,
1261 0 : _cancel: CancellationToken,
1262 0 : ) -> Result<Response<Body>, ApiError> {
1263 : // TODO openapi spec
1264 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1265 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1266 :
1267 0 : let state = get_state(&request);
1268 :
1269 0 : state
1270 0 : .tenant_manager
1271 0 : .delete_tenant(tenant_shard_id)
1272 0 : .instrument(info_span!("tenant_delete_handler",
1273 : tenant_id = %tenant_shard_id.tenant_id,
1274 0 : shard_id = %tenant_shard_id.shard_slug()
1275 : ))
1276 0 : .await?;
1277 :
1278 0 : json_response(StatusCode::OK, ())
1279 0 : }
1280 :
1281 : /// HTTP endpoint to query the current tenant_size of a tenant.
1282 : ///
1283 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1284 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1285 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1286 : /// values.
1287 : ///
1288 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1289 : /// (only if it is shorter than the real cutoff).
1290 : ///
1291 : /// Note: we don't update the cached size and prometheus metric here.
1292 : /// The retention period might be different, and it's nice to have a method to just calculate it
1293 : /// without modifying anything anyway.
1294 0 : async fn tenant_size_handler(
1295 0 : request: Request<Body>,
1296 0 : cancel: CancellationToken,
1297 0 : ) -> Result<Response<Body>, ApiError> {
1298 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1299 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1300 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1301 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1302 0 : let headers = request.headers();
1303 0 : let state = get_state(&request);
1304 :
1305 0 : if !tenant_shard_id.is_shard_zero() {
1306 0 : return Err(ApiError::BadRequest(anyhow!(
1307 0 : "Size calculations are only available on shard zero"
1308 0 : )));
1309 0 : }
1310 :
1311 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1312 0 : let tenant = state
1313 0 : .tenant_manager
1314 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1315 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1316 :
1317 : // this can be long operation
1318 0 : let inputs = tenant
1319 0 : .gather_size_inputs(
1320 0 : retention_period,
1321 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1322 0 : &cancel,
1323 0 : &ctx,
1324 0 : )
1325 0 : .await
1326 0 : .map_err(|e| match e {
1327 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1328 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1329 0 : })?;
1330 :
1331 0 : let mut sizes = None;
1332 0 : let accepts_html = headers
1333 0 : .get(header::ACCEPT)
1334 0 : .map(|v| v == "text/html")
1335 0 : .unwrap_or_default();
1336 0 : if !inputs_only.unwrap_or(false) {
1337 0 : let storage_model = inputs.calculate_model();
1338 0 : let size = storage_model.calculate();
1339 :
1340 : // If request header expects html, return html
1341 0 : if accepts_html {
1342 0 : return synthetic_size_html_response(inputs, storage_model, size);
1343 0 : }
1344 0 : sizes = Some(size);
1345 0 : } else if accepts_html {
1346 0 : return Err(ApiError::BadRequest(anyhow!(
1347 0 : "inputs_only parameter is incompatible with html output request"
1348 0 : )));
1349 0 : }
1350 :
1351 : /// The type resides in the pageserver not to expose `ModelInputs`.
1352 : #[derive(serde::Serialize)]
1353 : struct TenantHistorySize {
1354 : id: TenantId,
1355 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1356 : ///
1357 : /// Will be none if `?inputs_only=true` was given.
1358 : size: Option<u64>,
1359 : /// Size of each segment used in the model.
1360 : /// Will be null if `?inputs_only=true` was given.
1361 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1362 : inputs: crate::tenant::size::ModelInputs,
1363 : }
1364 :
1365 0 : json_response(
1366 : StatusCode::OK,
1367 : TenantHistorySize {
1368 0 : id: tenant_shard_id.tenant_id,
1369 0 : size: sizes.as_ref().map(|x| x.total_size),
1370 0 : segment_sizes: sizes.map(|x| x.segments),
1371 0 : inputs,
1372 : },
1373 : )
1374 0 : }
1375 :
1376 0 : async fn tenant_shard_split_handler(
1377 0 : mut request: Request<Body>,
1378 0 : _cancel: CancellationToken,
1379 0 : ) -> Result<Response<Body>, ApiError> {
1380 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1381 :
1382 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1383 0 : let state = get_state(&request);
1384 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1385 :
1386 0 : let tenant = state
1387 0 : .tenant_manager
1388 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1389 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1390 :
1391 0 : let new_shards = state
1392 0 : .tenant_manager
1393 0 : .shard_split(
1394 0 : tenant,
1395 0 : ShardCount::new(req.new_shard_count),
1396 0 : req.new_stripe_size,
1397 0 : &ctx,
1398 0 : )
1399 0 : .await
1400 0 : .map_err(ApiError::InternalServerError)?;
1401 :
1402 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1403 0 : }
1404 :
1405 0 : async fn layer_map_info_handler(
1406 0 : request: Request<Body>,
1407 0 : _cancel: CancellationToken,
1408 0 : ) -> Result<Response<Body>, ApiError> {
1409 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1410 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1411 0 : let reset: LayerAccessStatsReset =
1412 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1413 0 : let state = get_state(&request);
1414 :
1415 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1416 :
1417 0 : let timeline =
1418 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1419 0 : .await?;
1420 0 : let layer_map_info = timeline
1421 0 : .layer_map_info(reset)
1422 0 : .await
1423 0 : .map_err(|_shutdown| ApiError::ShuttingDown)?;
1424 :
1425 0 : json_response(StatusCode::OK, layer_map_info)
1426 0 : }
1427 :
1428 : #[instrument(skip_all, fields(tenant_id, shard_id, timeline_id, layer_name))]
1429 : async fn timeline_layer_scan_disposable_keys(
1430 : request: Request<Body>,
1431 : cancel: CancellationToken,
1432 : ) -> Result<Response<Body>, ApiError> {
1433 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1434 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1435 : let layer_name: LayerName = parse_request_param(&request, "layer_name")?;
1436 :
1437 : tracing::Span::current().record(
1438 : "tenant_id",
1439 : tracing::field::display(&tenant_shard_id.tenant_id),
1440 : );
1441 : tracing::Span::current().record(
1442 : "shard_id",
1443 : tracing::field::display(tenant_shard_id.shard_slug()),
1444 : );
1445 : tracing::Span::current().record("timeline_id", tracing::field::display(&timeline_id));
1446 : tracing::Span::current().record("layer_name", tracing::field::display(&layer_name));
1447 :
1448 : let state = get_state(&request);
1449 :
1450 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1451 :
1452 : // technically the timeline need not be active for this scan to complete
1453 : let timeline =
1454 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1455 : .await?;
1456 :
1457 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1458 : .with_scope_timeline(&timeline);
1459 :
1460 : let guard = timeline
1461 : .layers
1462 : .read(LayerManagerLockHolder::GetLayerMapInfo)
1463 : .await;
1464 : let Some(layer) = guard.try_get_from_key(&layer_name.clone().into()) else {
1465 : return Err(ApiError::NotFound(
1466 : anyhow::anyhow!("Layer {tenant_shard_id}/{timeline_id}/{layer_name} not found").into(),
1467 : ));
1468 : };
1469 :
1470 : let resident_layer = layer
1471 : .download_and_keep_resident(&ctx)
1472 : .await
1473 0 : .map_err(|err| match err {
1474 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1475 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1476 0 : ApiError::ShuttingDown
1477 : }
1478 : tenant::storage_layer::layer::DownloadError::ContextAndConfigReallyDeniesDownloads
1479 : | tenant::storage_layer::layer::DownloadError::DownloadRequired
1480 : | tenant::storage_layer::layer::DownloadError::NotFile(_)
1481 : | tenant::storage_layer::layer::DownloadError::DownloadFailed
1482 : | tenant::storage_layer::layer::DownloadError::PreStatFailed(_) => {
1483 0 : ApiError::InternalServerError(err.into())
1484 : }
1485 : #[cfg(test)]
1486 : tenant::storage_layer::layer::DownloadError::Failpoint(_) => {
1487 0 : ApiError::InternalServerError(err.into())
1488 : }
1489 0 : })?;
1490 :
1491 : let keys = resident_layer
1492 : .load_keys(&ctx)
1493 : .await
1494 : .map_err(ApiError::InternalServerError)?;
1495 :
1496 : let shard_identity = timeline.get_shard_identity();
1497 :
1498 : let mut disposable_count = 0;
1499 : let mut not_disposable_count = 0;
1500 : let cancel = cancel.clone();
1501 : for (i, key) in keys.into_iter().enumerate() {
1502 : if shard_identity.is_key_disposable(&key) {
1503 : disposable_count += 1;
1504 : tracing::debug!(key = %key, key.dbg=?key, "disposable key");
1505 : } else {
1506 : not_disposable_count += 1;
1507 : }
1508 : #[allow(clippy::collapsible_if)]
1509 : if i % 10000 == 0 {
1510 : if cancel.is_cancelled() || timeline.cancel.is_cancelled() || timeline.is_stopping() {
1511 : return Err(ApiError::ShuttingDown);
1512 : }
1513 : }
1514 : }
1515 :
1516 : json_response(
1517 : StatusCode::OK,
1518 : pageserver_api::models::ScanDisposableKeysResponse {
1519 : disposable_count,
1520 : not_disposable_count,
1521 : },
1522 : )
1523 : }
1524 :
1525 0 : async fn timeline_download_heatmap_layers_handler(
1526 0 : request: Request<Body>,
1527 0 : _cancel: CancellationToken,
1528 0 : ) -> Result<Response<Body>, ApiError> {
1529 : // Only used in the case where remote storage is not configured.
1530 : const DEFAULT_MAX_CONCURRENCY: usize = 100;
1531 : // A conservative default.
1532 : const DEFAULT_CONCURRENCY: usize = 16;
1533 :
1534 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1535 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1536 :
1537 0 : let desired_concurrency =
1538 0 : parse_query_param(&request, "concurrency")?.unwrap_or(DEFAULT_CONCURRENCY);
1539 0 : let recurse = parse_query_param(&request, "recurse")?.unwrap_or(false);
1540 :
1541 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1542 :
1543 0 : let state = get_state(&request);
1544 0 : let timeline =
1545 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1546 0 : .await?;
1547 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1548 0 : .with_scope_timeline(&timeline);
1549 :
1550 0 : let max_concurrency = get_config(&request)
1551 0 : .remote_storage_config
1552 0 : .as_ref()
1553 0 : .map(|c| c.concurrency_limit())
1554 0 : .unwrap_or(DEFAULT_MAX_CONCURRENCY);
1555 0 : let concurrency = std::cmp::min(max_concurrency, desired_concurrency);
1556 :
1557 0 : timeline.start_heatmap_layers_download(concurrency, recurse, &ctx)?;
1558 :
1559 0 : json_response(StatusCode::ACCEPTED, ())
1560 0 : }
1561 :
1562 0 : async fn timeline_shutdown_download_heatmap_layers_handler(
1563 0 : request: Request<Body>,
1564 0 : _cancel: CancellationToken,
1565 0 : ) -> Result<Response<Body>, ApiError> {
1566 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1567 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1568 :
1569 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1570 :
1571 0 : let state = get_state(&request);
1572 0 : let timeline =
1573 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1574 0 : .await?;
1575 :
1576 0 : timeline.stop_and_drain_heatmap_layers_download().await;
1577 :
1578 0 : json_response(StatusCode::OK, ())
1579 0 : }
1580 :
1581 0 : async fn layer_download_handler(
1582 0 : request: Request<Body>,
1583 0 : _cancel: CancellationToken,
1584 0 : ) -> Result<Response<Body>, ApiError> {
1585 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1586 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1587 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1588 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1589 0 : let layer_name = LayerName::from_str(layer_file_name)
1590 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1591 0 : let state = get_state(&request);
1592 :
1593 0 : let timeline =
1594 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1595 0 : .await?;
1596 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
1597 0 : .with_scope_timeline(&timeline);
1598 0 : let downloaded = timeline
1599 0 : .download_layer(&layer_name, &ctx)
1600 0 : .await
1601 0 : .map_err(|e| match e {
1602 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1603 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1604 0 : ApiError::ShuttingDown
1605 : }
1606 0 : other => ApiError::InternalServerError(other.into()),
1607 0 : })?;
1608 :
1609 0 : match downloaded {
1610 0 : Some(true) => json_response(StatusCode::OK, ()),
1611 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1612 0 : None => json_response(
1613 : StatusCode::BAD_REQUEST,
1614 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1615 : ),
1616 : }
1617 0 : }
1618 :
1619 0 : async fn evict_timeline_layer_handler(
1620 0 : request: Request<Body>,
1621 0 : _cancel: CancellationToken,
1622 0 : ) -> Result<Response<Body>, ApiError> {
1623 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1624 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1625 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1626 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1627 0 : let state = get_state(&request);
1628 :
1629 0 : let layer_name = LayerName::from_str(layer_file_name)
1630 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1631 :
1632 0 : let timeline =
1633 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1634 0 : .await?;
1635 0 : let evicted = timeline
1636 0 : .evict_layer(&layer_name)
1637 0 : .await
1638 0 : .map_err(ApiError::InternalServerError)?;
1639 :
1640 0 : match evicted {
1641 0 : Some(true) => json_response(StatusCode::OK, ()),
1642 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1643 0 : None => json_response(
1644 : StatusCode::BAD_REQUEST,
1645 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1646 : ),
1647 : }
1648 0 : }
1649 :
1650 0 : async fn timeline_gc_blocking_handler(
1651 0 : request: Request<Body>,
1652 0 : _cancel: CancellationToken,
1653 0 : ) -> Result<Response<Body>, ApiError> {
1654 0 : block_or_unblock_gc(request, true).await
1655 0 : }
1656 :
1657 0 : async fn timeline_gc_unblocking_handler(
1658 0 : request: Request<Body>,
1659 0 : _cancel: CancellationToken,
1660 0 : ) -> Result<Response<Body>, ApiError> {
1661 0 : block_or_unblock_gc(request, false).await
1662 0 : }
1663 :
1664 : /// Traces GetPage@LSN requests for a timeline, and emits metadata in an efficient binary encoding.
1665 : /// Use the `pagectl page-trace` command to decode and analyze the output.
1666 0 : async fn timeline_page_trace_handler(
1667 0 : request: Request<Body>,
1668 0 : cancel: CancellationToken,
1669 0 : ) -> Result<Response<Body>, ApiError> {
1670 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1671 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1672 0 : let state = get_state(&request);
1673 0 : check_permission(&request, None)?;
1674 :
1675 0 : let size_limit: usize = parse_query_param(&request, "size_limit_bytes")?.unwrap_or(1024 * 1024);
1676 0 : let time_limit_secs: u64 = parse_query_param(&request, "time_limit_secs")?.unwrap_or(5);
1677 :
1678 : // Convert size limit to event limit based on the serialized size of an event. The event size is
1679 : // fixed, as the default bincode serializer uses fixed-width integer encoding.
1680 0 : let event_size = bincode::serialize(&PageTraceEvent::default())
1681 0 : .map_err(|err| ApiError::InternalServerError(err.into()))?
1682 0 : .len();
1683 0 : let event_limit = size_limit / event_size;
1684 :
1685 0 : let timeline =
1686 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1687 0 : .await?;
1688 :
1689 : // Install a page trace, unless one is already in progress. We just use a buffered channel,
1690 : // which may 2x the memory usage in the worst case, but it's still bounded.
1691 0 : let (trace_tx, mut trace_rx) = tokio::sync::mpsc::channel(event_limit);
1692 0 : let cur = timeline.page_trace.load();
1693 0 : let installed = cur.is_none()
1694 0 : && timeline
1695 0 : .page_trace
1696 0 : .compare_and_swap(cur, Some(Arc::new(trace_tx)))
1697 0 : .is_none();
1698 0 : if !installed {
1699 0 : return Err(ApiError::Conflict("page trace already active".to_string()));
1700 0 : }
1701 0 : defer!(timeline.page_trace.store(None)); // uninstall on return
1702 :
1703 : // Collect the trace and return it to the client. We could stream the response, but this is
1704 : // simple and fine.
1705 0 : let mut body = Vec::with_capacity(size_limit);
1706 0 : let deadline = Instant::now() + Duration::from_secs(time_limit_secs);
1707 :
1708 0 : while body.len() < size_limit {
1709 0 : tokio::select! {
1710 0 : event = trace_rx.recv() => {
1711 0 : let Some(event) = event else {
1712 0 : break; // shouldn't happen (sender doesn't close, unless timeline dropped)
1713 : };
1714 0 : bincode::serialize_into(&mut body, &event)
1715 0 : .map_err(|err| ApiError::InternalServerError(err.into()))?;
1716 : }
1717 0 : _ = tokio::time::sleep_until(deadline) => break, // time limit reached
1718 0 : _ = cancel.cancelled() => return Err(ApiError::Cancelled),
1719 : }
1720 : }
1721 :
1722 0 : Ok(Response::builder()
1723 0 : .status(StatusCode::OK)
1724 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
1725 0 : .body(hyper::Body::from(body))
1726 0 : .unwrap())
1727 0 : }
1728 :
1729 : /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
1730 : ///
1731 : /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
1732 0 : async fn block_or_unblock_gc(
1733 0 : request: Request<Body>,
1734 0 : block: bool,
1735 0 : ) -> Result<Response<Body>, ApiError> {
1736 : use crate::tenant::remote_timeline_client::WaitCompletionError;
1737 : use crate::tenant::upload_queue::NotInitialized;
1738 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1739 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1740 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1741 0 : let state = get_state(&request);
1742 :
1743 0 : let tenant = state
1744 0 : .tenant_manager
1745 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1746 :
1747 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1748 :
1749 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1750 :
1751 0 : let fut = async {
1752 0 : if block {
1753 0 : timeline.block_gc(&tenant).await.map(|_| ())
1754 : } else {
1755 0 : timeline.unblock_gc(&tenant).await
1756 : }
1757 0 : };
1758 :
1759 0 : let span = tracing::info_span!(
1760 : "block_or_unblock_gc",
1761 : tenant_id = %tenant_shard_id.tenant_id,
1762 0 : shard_id = %tenant_shard_id.shard_slug(),
1763 : timeline_id = %timeline_id,
1764 : block = block,
1765 : );
1766 :
1767 0 : let res = fut.instrument(span).await;
1768 :
1769 0 : res.map_err(|e| {
1770 0 : if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
1771 0 : ApiError::ShuttingDown
1772 : } else {
1773 0 : ApiError::InternalServerError(e)
1774 : }
1775 0 : })?;
1776 :
1777 0 : json_response(StatusCode::OK, ())
1778 0 : }
1779 :
1780 : /// Get tenant_size SVG graph along with the JSON data.
1781 0 : fn synthetic_size_html_response(
1782 0 : inputs: ModelInputs,
1783 0 : storage_model: StorageModel,
1784 0 : sizes: SizeResult,
1785 0 : ) -> Result<Response<Body>, ApiError> {
1786 0 : let mut timeline_ids: Vec<String> = Vec::new();
1787 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1788 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1789 0 : timeline_map.insert(ti.timeline_id, index);
1790 0 : timeline_ids.push(ti.timeline_id.to_string());
1791 0 : }
1792 0 : let seg_to_branch: Vec<(usize, SvgBranchKind)> = inputs
1793 0 : .segments
1794 0 : .iter()
1795 0 : .map(|seg| {
1796 0 : (
1797 0 : *timeline_map.get(&seg.timeline_id).unwrap(),
1798 0 : seg.kind.into(),
1799 0 : )
1800 0 : })
1801 0 : .collect();
1802 :
1803 0 : let svg =
1804 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1805 0 : .map_err(ApiError::InternalServerError)?;
1806 :
1807 0 : let mut response = String::new();
1808 :
1809 : use std::fmt::Write;
1810 0 : write!(response, "<html>\n<body>\n").unwrap();
1811 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1812 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1813 0 : writeln!(response, "<pre>").unwrap();
1814 0 : writeln!(
1815 0 : response,
1816 0 : "{}",
1817 0 : serde_json::to_string_pretty(&inputs).unwrap()
1818 : )
1819 0 : .unwrap();
1820 0 : writeln!(
1821 0 : response,
1822 0 : "{}",
1823 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1824 : )
1825 0 : .unwrap();
1826 0 : writeln!(response, "</pre>").unwrap();
1827 0 : write!(response, "</body>\n</html>\n").unwrap();
1828 :
1829 0 : html_response(StatusCode::OK, response)
1830 0 : }
1831 :
1832 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1833 0 : let response = Response::builder()
1834 0 : .status(status)
1835 0 : .header(header::CONTENT_TYPE, "text/html")
1836 0 : .body(Body::from(data.as_bytes().to_vec()))
1837 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1838 0 : Ok(response)
1839 0 : }
1840 :
1841 0 : async fn get_tenant_config_handler(
1842 0 : request: Request<Body>,
1843 0 : _cancel: CancellationToken,
1844 0 : ) -> Result<Response<Body>, ApiError> {
1845 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1846 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1847 0 : let state = get_state(&request);
1848 :
1849 0 : let tenant = state
1850 0 : .tenant_manager
1851 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1852 :
1853 0 : let response = HashMap::from([
1854 : (
1855 : "tenant_specific_overrides",
1856 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1857 0 : .context("serializing tenant specific overrides")
1858 0 : .map_err(ApiError::InternalServerError)?,
1859 : ),
1860 : (
1861 0 : "effective_config",
1862 0 : serde_json::to_value(tenant.effective_config())
1863 0 : .context("serializing effective config")
1864 0 : .map_err(ApiError::InternalServerError)?,
1865 : ),
1866 : ]);
1867 :
1868 0 : json_response(StatusCode::OK, response)
1869 0 : }
1870 :
1871 0 : async fn update_tenant_config_handler(
1872 0 : mut request: Request<Body>,
1873 0 : _cancel: CancellationToken,
1874 0 : ) -> Result<Response<Body>, ApiError> {
1875 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1876 0 : let tenant_id = request_data.tenant_id;
1877 0 : check_permission(&request, Some(tenant_id))?;
1878 :
1879 0 : let new_tenant_conf = request_data.config;
1880 :
1881 0 : let state = get_state(&request);
1882 :
1883 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1884 :
1885 0 : let tenant = state
1886 0 : .tenant_manager
1887 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1888 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1889 :
1890 : // This is a legacy API that only operates on attached tenants: the preferred
1891 : // API to use is the location_config/ endpoint, which lets the caller provide
1892 : // the full LocationConf.
1893 0 : let location_conf = LocationConf::attached_single(
1894 0 : new_tenant_conf.clone(),
1895 0 : tenant.get_generation(),
1896 0 : ShardParameters::from(tenant.get_shard_identity()),
1897 : );
1898 :
1899 0 : tenant
1900 0 : .get_shard_identity()
1901 0 : .assert_equal(location_conf.shard); // not strictly necessary since we construct it above
1902 :
1903 0 : crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1904 0 : .await
1905 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1906 :
1907 0 : let _ = tenant
1908 0 : .update_tenant_config(|_crnt| Ok(new_tenant_conf.clone()))
1909 0 : .expect("Closure returns Ok()");
1910 :
1911 0 : json_response(StatusCode::OK, ())
1912 0 : }
1913 :
1914 0 : async fn patch_tenant_config_handler(
1915 0 : mut request: Request<Body>,
1916 0 : _cancel: CancellationToken,
1917 0 : ) -> Result<Response<Body>, ApiError> {
1918 0 : let request_data: TenantConfigPatchRequest = json_request(&mut request).await?;
1919 0 : let tenant_id = request_data.tenant_id;
1920 0 : check_permission(&request, Some(tenant_id))?;
1921 :
1922 0 : let state = get_state(&request);
1923 :
1924 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1925 :
1926 0 : let tenant = state
1927 0 : .tenant_manager
1928 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1929 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1930 :
1931 0 : let updated = tenant
1932 0 : .update_tenant_config(|crnt| {
1933 0 : crnt.apply_patch(request_data.config.clone())
1934 0 : .map_err(anyhow::Error::new)
1935 0 : })
1936 0 : .map_err(ApiError::BadRequest)?;
1937 :
1938 : // This is a legacy API that only operates on attached tenants: the preferred
1939 : // API to use is the location_config/ endpoint, which lets the caller provide
1940 : // the full LocationConf.
1941 0 : let location_conf = LocationConf::attached_single(
1942 0 : updated,
1943 0 : tenant.get_generation(),
1944 0 : ShardParameters::from(tenant.get_shard_identity()),
1945 : );
1946 :
1947 0 : tenant
1948 0 : .get_shard_identity()
1949 0 : .assert_equal(location_conf.shard); // not strictly necessary since we construct it above
1950 :
1951 0 : crate::tenant::TenantShard::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1952 0 : .await
1953 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1954 :
1955 0 : json_response(StatusCode::OK, ())
1956 0 : }
1957 :
1958 0 : async fn put_tenant_location_config_handler(
1959 0 : mut request: Request<Body>,
1960 0 : _cancel: CancellationToken,
1961 0 : ) -> Result<Response<Body>, ApiError> {
1962 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1963 :
1964 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1965 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1966 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1967 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1968 :
1969 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1970 0 : let state = get_state(&request);
1971 0 : let conf = state.conf;
1972 :
1973 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1974 : // its local disk content and drops it from memory.
1975 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1976 0 : if let Err(e) = state
1977 0 : .tenant_manager
1978 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
1979 0 : .instrument(info_span!("tenant_detach",
1980 : tenant_id = %tenant_shard_id.tenant_id,
1981 0 : shard_id = %tenant_shard_id.shard_slug()
1982 : ))
1983 0 : .await
1984 : {
1985 0 : match e {
1986 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1987 0 : // This API is idempotent: a NotFound on a detach is fine.
1988 0 : }
1989 0 : _ => return Err(e.into()),
1990 : }
1991 0 : }
1992 0 : return json_response(StatusCode::OK, ());
1993 0 : }
1994 :
1995 0 : let location_conf =
1996 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1997 :
1998 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1999 : // similar to at startup ordering.
2000 0 : let spawn_mode = if lazy {
2001 0 : tenant::SpawnMode::Lazy
2002 : } else {
2003 0 : tenant::SpawnMode::Eager
2004 : };
2005 :
2006 0 : let tenant = state
2007 0 : .tenant_manager
2008 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
2009 0 : .await?;
2010 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
2011 0 : let attached = tenant.is_some();
2012 :
2013 0 : if let Some(_flush_ms) = flush {
2014 0 : match state
2015 0 : .secondary_controller
2016 0 : .upload_tenant(tenant_shard_id)
2017 0 : .await
2018 : {
2019 : Ok(()) => {
2020 0 : tracing::info!("Uploaded heatmap during flush");
2021 : }
2022 0 : Err(e) => {
2023 0 : tracing::warn!("Failed to flush heatmap: {e}");
2024 : }
2025 : }
2026 : } else {
2027 0 : tracing::info!("No flush requested when configuring");
2028 : }
2029 :
2030 : // This API returns a vector of pageservers where the tenant is attached: this is
2031 : // primarily for use in the sharding service. For compatibilty, we also return this
2032 : // when called directly on a pageserver, but the payload is always zero or one shards.
2033 0 : let mut response = TenantLocationConfigResponse {
2034 0 : shards: Vec::new(),
2035 0 : stripe_size: None,
2036 0 : };
2037 0 : if attached {
2038 0 : response.shards.push(TenantShardLocation {
2039 0 : shard_id: tenant_shard_id,
2040 0 : node_id: state.conf.id,
2041 0 : });
2042 0 : if tenant_shard_id.shard_count.count() > 1 {
2043 : // Stripe size should be set if we are attached
2044 0 : debug_assert!(stripe_size.is_some());
2045 0 : response.stripe_size = stripe_size;
2046 0 : }
2047 0 : }
2048 :
2049 0 : json_response(StatusCode::OK, response)
2050 0 : }
2051 :
2052 0 : async fn list_location_config_handler(
2053 0 : request: Request<Body>,
2054 0 : _cancel: CancellationToken,
2055 0 : ) -> Result<Response<Body>, ApiError> {
2056 0 : let state = get_state(&request);
2057 0 : let slots = state.tenant_manager.list();
2058 0 : let result = LocationConfigListResponse {
2059 0 : tenant_shards: slots
2060 0 : .into_iter()
2061 0 : .map(|(tenant_shard_id, slot)| {
2062 0 : let v = match slot {
2063 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
2064 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
2065 0 : TenantSlot::InProgress(_) => None,
2066 : };
2067 0 : (tenant_shard_id, v)
2068 0 : })
2069 0 : .collect(),
2070 : };
2071 0 : json_response(StatusCode::OK, result)
2072 0 : }
2073 :
2074 0 : async fn get_location_config_handler(
2075 0 : request: Request<Body>,
2076 0 : _cancel: CancellationToken,
2077 0 : ) -> Result<Response<Body>, ApiError> {
2078 0 : let state = get_state(&request);
2079 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2080 0 : let slot = state.tenant_manager.get(tenant_shard_id);
2081 :
2082 0 : let Some(slot) = slot else {
2083 0 : return Err(ApiError::NotFound(
2084 0 : anyhow::anyhow!("Tenant shard not found").into(),
2085 0 : ));
2086 : };
2087 :
2088 0 : let result: Option<LocationConfig> = match slot {
2089 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
2090 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
2091 0 : TenantSlot::InProgress(_) => None,
2092 : };
2093 :
2094 0 : json_response(StatusCode::OK, result)
2095 0 : }
2096 :
2097 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
2098 : // (from all pageservers) as it invalidates consistency assumptions.
2099 0 : async fn tenant_time_travel_remote_storage_handler(
2100 0 : request: Request<Body>,
2101 0 : cancel: CancellationToken,
2102 0 : ) -> Result<Response<Body>, ApiError> {
2103 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2104 :
2105 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2106 :
2107 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
2108 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
2109 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
2110 0 : .map_err(ApiError::BadRequest)?;
2111 :
2112 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
2113 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
2114 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
2115 0 : .map_err(ApiError::BadRequest)?;
2116 :
2117 : // This is just a sanity check to fend off naive wrong usages of the API:
2118 : // the tenant needs to be detached *everywhere*
2119 0 : let state = get_state(&request);
2120 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
2121 0 : if we_manage_tenant {
2122 0 : return Err(ApiError::BadRequest(anyhow!(
2123 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
2124 0 : )));
2125 0 : }
2126 :
2127 0 : if timestamp > done_if_after {
2128 0 : return Err(ApiError::BadRequest(anyhow!(
2129 0 : "The done_if_after timestamp comes before the timestamp to recover to"
2130 0 : )));
2131 0 : }
2132 :
2133 0 : tracing::info!(
2134 0 : "Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}"
2135 : );
2136 :
2137 0 : remote_timeline_client::upload::time_travel_recover_tenant(
2138 0 : &state.remote_storage,
2139 0 : &tenant_shard_id,
2140 0 : timestamp,
2141 0 : done_if_after,
2142 0 : &cancel,
2143 0 : )
2144 0 : .await
2145 0 : .map_err(|e| match e {
2146 0 : TimeTravelError::BadInput(e) => {
2147 0 : warn!("bad input error: {e}");
2148 0 : ApiError::BadRequest(anyhow!("bad input error"))
2149 : }
2150 : TimeTravelError::Unimplemented => {
2151 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
2152 : }
2153 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
2154 : TimeTravelError::TooManyVersions => {
2155 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
2156 : }
2157 0 : TimeTravelError::Other(e) => {
2158 0 : warn!("internal error: {e}");
2159 0 : ApiError::InternalServerError(anyhow!("internal error"))
2160 : }
2161 0 : })?;
2162 :
2163 0 : json_response(StatusCode::OK, ())
2164 0 : }
2165 :
2166 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
2167 0 : async fn handle_tenant_break(
2168 0 : r: Request<Body>,
2169 0 : _cancel: CancellationToken,
2170 0 : ) -> Result<Response<Body>, ApiError> {
2171 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
2172 :
2173 0 : let state = get_state(&r);
2174 0 : state
2175 0 : .tenant_manager
2176 0 : .get_attached_tenant_shard(tenant_shard_id)?
2177 0 : .set_broken("broken from test".to_owned())
2178 0 : .await;
2179 :
2180 0 : json_response(StatusCode::OK, ())
2181 0 : }
2182 :
2183 : // Obtains an lsn lease on the given timeline.
2184 0 : async fn lsn_lease_handler(
2185 0 : mut request: Request<Body>,
2186 0 : _cancel: CancellationToken,
2187 0 : ) -> Result<Response<Body>, ApiError> {
2188 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2189 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2190 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2191 0 : let lsn = json_request::<LsnLeaseRequest>(&mut request).await?.lsn;
2192 :
2193 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2194 :
2195 0 : let state = get_state(&request);
2196 :
2197 0 : let timeline =
2198 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2199 0 : .await?;
2200 :
2201 0 : let result = async {
2202 0 : timeline
2203 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
2204 0 : .map_err(|e| {
2205 0 : ApiError::InternalServerError(
2206 0 : e.context(format!("invalid lsn lease request at {lsn}")),
2207 0 : )
2208 0 : })
2209 0 : }
2210 0 : .instrument(info_span!("init_lsn_lease", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2211 0 : .await?;
2212 :
2213 0 : json_response(StatusCode::OK, result)
2214 0 : }
2215 :
2216 : // Run GC immediately on given timeline.
2217 0 : async fn timeline_gc_handler(
2218 0 : mut request: Request<Body>,
2219 0 : cancel: CancellationToken,
2220 0 : ) -> Result<Response<Body>, ApiError> {
2221 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2222 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2223 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2224 :
2225 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
2226 :
2227 0 : let state = get_state(&request);
2228 :
2229 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2230 0 : let gc_result = state
2231 0 : .tenant_manager
2232 0 : .immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)
2233 0 : .await?;
2234 :
2235 0 : json_response(StatusCode::OK, gc_result)
2236 0 : }
2237 :
2238 : // Cancel scheduled compaction tasks
2239 0 : async fn timeline_cancel_compact_handler(
2240 0 : request: Request<Body>,
2241 0 : _cancel: CancellationToken,
2242 0 : ) -> Result<Response<Body>, ApiError> {
2243 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2244 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2245 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2246 0 : let state = get_state(&request);
2247 0 : async {
2248 0 : let tenant = state
2249 0 : .tenant_manager
2250 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2251 0 : tenant.cancel_scheduled_compaction(timeline_id);
2252 0 : json_response(StatusCode::OK, ())
2253 0 : }
2254 0 : .instrument(info_span!("timeline_cancel_compact", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2255 0 : .await
2256 0 : }
2257 :
2258 : // Get compact info of a timeline
2259 0 : async fn timeline_compact_info_handler(
2260 0 : request: Request<Body>,
2261 0 : _cancel: CancellationToken,
2262 0 : ) -> Result<Response<Body>, ApiError> {
2263 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2264 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2265 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2266 0 : let state = get_state(&request);
2267 0 : async {
2268 0 : let tenant = state
2269 0 : .tenant_manager
2270 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2271 0 : let resp = tenant.get_scheduled_compaction_tasks(timeline_id);
2272 0 : json_response(StatusCode::OK, resp)
2273 0 : }
2274 0 : .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2275 0 : .await
2276 0 : }
2277 :
2278 : // Run compaction immediately on given timeline.
2279 0 : async fn timeline_compact_handler(
2280 0 : mut request: Request<Body>,
2281 0 : cancel: CancellationToken,
2282 0 : ) -> Result<Response<Body>, ApiError> {
2283 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2284 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2285 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2286 :
2287 0 : let compact_request = json_request_maybe::<Option<CompactRequest>>(&mut request).await?;
2288 :
2289 0 : let state = get_state(&request);
2290 :
2291 0 : let mut flags = EnumSet::empty();
2292 :
2293 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2294 0 : flags |= CompactFlags::ForceL0Compaction;
2295 0 : }
2296 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2297 0 : flags |= CompactFlags::ForceRepartition;
2298 0 : }
2299 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2300 0 : flags |= CompactFlags::ForceImageLayerCreation;
2301 0 : }
2302 0 : if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
2303 0 : flags |= CompactFlags::EnhancedGcBottomMostCompaction;
2304 0 : }
2305 0 : if Some(true) == parse_query_param::<_, bool>(&request, "dry_run")? {
2306 0 : flags |= CompactFlags::DryRun;
2307 0 : }
2308 : // Manual compaction does not yield for L0.
2309 :
2310 0 : let wait_until_uploaded =
2311 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2312 :
2313 0 : let wait_until_scheduled_compaction_done =
2314 0 : parse_query_param::<_, bool>(&request, "wait_until_scheduled_compaction_done")?
2315 0 : .unwrap_or(false);
2316 :
2317 0 : let sub_compaction = compact_request
2318 0 : .as_ref()
2319 0 : .map(|r| r.sub_compaction)
2320 0 : .unwrap_or(false);
2321 0 : let sub_compaction_max_job_size_mb = compact_request
2322 0 : .as_ref()
2323 0 : .and_then(|r| r.sub_compaction_max_job_size_mb);
2324 :
2325 0 : let options = CompactOptions {
2326 0 : compact_key_range: compact_request
2327 0 : .as_ref()
2328 0 : .and_then(|r| r.compact_key_range.clone()),
2329 0 : compact_lsn_range: compact_request
2330 0 : .as_ref()
2331 0 : .and_then(|r| r.compact_lsn_range.clone()),
2332 0 : flags,
2333 0 : sub_compaction,
2334 0 : sub_compaction_max_job_size_mb,
2335 : };
2336 :
2337 0 : let scheduled = compact_request
2338 0 : .as_ref()
2339 0 : .map(|r| r.scheduled)
2340 0 : .unwrap_or(false);
2341 :
2342 0 : async {
2343 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2344 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2345 0 : if scheduled {
2346 0 : let tenant = state
2347 0 : .tenant_manager
2348 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2349 0 : let rx = tenant.schedule_compaction(timeline_id, options).await.map_err(ApiError::InternalServerError)?;
2350 0 : if wait_until_scheduled_compaction_done {
2351 : // It is possible that this will take a long time, dropping the HTTP request will not cancel the compaction.
2352 0 : rx.await.ok();
2353 0 : }
2354 : } else {
2355 0 : timeline
2356 0 : .compact_with_options(&cancel, options, &ctx)
2357 0 : .await
2358 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2359 0 : if wait_until_uploaded {
2360 0 : timeline.remote_client.wait_completion().await
2361 : // XXX map to correct ApiError for the cases where it's due to shutdown
2362 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2363 0 : }
2364 : }
2365 0 : json_response(StatusCode::OK, ())
2366 0 : }
2367 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2368 0 : .await
2369 0 : }
2370 :
2371 0 : async fn timeline_mark_invisible_handler(
2372 0 : mut request: Request<Body>,
2373 0 : _cancel: CancellationToken,
2374 0 : ) -> Result<Response<Body>, ApiError> {
2375 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2376 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2377 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2378 :
2379 0 : let compact_request = json_request_maybe::<Option<MarkInvisibleRequest>>(&mut request).await?;
2380 :
2381 0 : let state = get_state(&request);
2382 :
2383 0 : let visibility = match compact_request {
2384 0 : Some(req) => match req.is_visible {
2385 0 : Some(true) => TimelineVisibilityState::Visible,
2386 0 : Some(false) | None => TimelineVisibilityState::Invisible,
2387 : },
2388 0 : None => TimelineVisibilityState::Invisible,
2389 : };
2390 :
2391 0 : async {
2392 0 : let tenant = state
2393 0 : .tenant_manager
2394 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2395 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2396 0 : timeline.remote_client.schedule_index_upload_for_timeline_invisible_state(visibility).map_err(ApiError::InternalServerError)?;
2397 0 : json_response(StatusCode::OK, ())
2398 0 : }
2399 0 : .instrument(info_span!("manual_timeline_mark_invisible", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2400 0 : .await
2401 0 : }
2402 :
2403 : // Run offload immediately on given timeline.
2404 0 : async fn timeline_offload_handler(
2405 0 : request: Request<Body>,
2406 0 : _cancel: CancellationToken,
2407 0 : ) -> Result<Response<Body>, ApiError> {
2408 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2409 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2410 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2411 :
2412 0 : let state = get_state(&request);
2413 :
2414 0 : async {
2415 0 : let tenant = state
2416 0 : .tenant_manager
2417 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2418 :
2419 0 : if tenant.get_offloaded_timeline(timeline_id).is_ok() {
2420 0 : return json_response(StatusCode::OK, ());
2421 0 : }
2422 0 : let timeline =
2423 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2424 0 : .await?;
2425 :
2426 0 : if !tenant.timeline_has_no_attached_children(timeline_id) {
2427 0 : return Err(ApiError::PreconditionFailed(
2428 0 : "timeline has attached children".into(),
2429 0 : ));
2430 0 : }
2431 0 : if let (false, reason) = timeline.can_offload() {
2432 0 : return Err(ApiError::PreconditionFailed(
2433 0 : format!("Timeline::can_offload() check failed: {reason}") .into(),
2434 0 : ));
2435 0 : }
2436 0 : offload_timeline(&tenant, &timeline)
2437 0 : .await
2438 0 : .map_err(|e| {
2439 0 : match e {
2440 0 : OffloadError::Cancelled => ApiError::ResourceUnavailable("Timeline shutting down".into()),
2441 0 : OffloadError::AlreadyInProgress => ApiError::Conflict("Timeline already being offloaded or deleted".into()),
2442 0 : _ => ApiError::InternalServerError(anyhow!(e))
2443 : }
2444 0 : })?;
2445 :
2446 0 : json_response(StatusCode::OK, ())
2447 0 : }
2448 0 : .instrument(info_span!("manual_timeline_offload", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2449 0 : .await
2450 0 : }
2451 :
2452 : // Run checkpoint immediately on given timeline.
2453 0 : async fn timeline_checkpoint_handler(
2454 0 : request: Request<Body>,
2455 0 : cancel: CancellationToken,
2456 0 : ) -> Result<Response<Body>, ApiError> {
2457 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2458 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2459 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2460 :
2461 0 : let state = get_state(&request);
2462 :
2463 0 : let mut flags = EnumSet::empty();
2464 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2465 0 : flags |= CompactFlags::ForceL0Compaction;
2466 0 : }
2467 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2468 0 : flags |= CompactFlags::ForceRepartition;
2469 0 : }
2470 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2471 0 : flags |= CompactFlags::ForceImageLayerCreation;
2472 0 : }
2473 :
2474 : // By default, checkpoints come with a compaction, but this may be optionally disabled by tests that just want to flush + upload.
2475 0 : let compact = parse_query_param::<_, bool>(&request, "compact")?.unwrap_or(true);
2476 :
2477 0 : let wait_until_flushed: bool =
2478 0 : parse_query_param(&request, "wait_until_flushed")?.unwrap_or(true);
2479 :
2480 0 : let wait_until_uploaded =
2481 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2482 :
2483 0 : async {
2484 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2485 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2486 0 : if wait_until_flushed {
2487 0 : timeline.freeze_and_flush().await
2488 : } else {
2489 0 : timeline.freeze().await.and(Ok(()))
2490 0 : }.map_err(|e| {
2491 0 : match e {
2492 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
2493 0 : other => ApiError::InternalServerError(other.into()),
2494 :
2495 : }
2496 0 : })?;
2497 0 : if compact {
2498 0 : timeline
2499 0 : .compact(&cancel, flags, &ctx)
2500 0 : .await
2501 0 : .map_err(|e|
2502 0 : match e {
2503 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
2504 0 : CompactionError::CollectKeySpaceError(e) => ApiError::InternalServerError(anyhow::anyhow!(e)),
2505 0 : CompactionError::Other(e) => ApiError::InternalServerError(e),
2506 0 : }
2507 0 : )?;
2508 0 : }
2509 :
2510 0 : if wait_until_uploaded {
2511 0 : tracing::info!("Waiting for uploads to complete...");
2512 0 : timeline.remote_client.wait_completion().await
2513 : // XXX map to correct ApiError for the cases where it's due to shutdown
2514 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2515 0 : tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
2516 0 : }
2517 :
2518 0 : json_response(StatusCode::OK, ())
2519 0 : }
2520 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2521 0 : .await
2522 0 : }
2523 :
2524 0 : async fn timeline_download_remote_layers_handler_post(
2525 0 : mut request: Request<Body>,
2526 0 : _cancel: CancellationToken,
2527 0 : ) -> Result<Response<Body>, ApiError> {
2528 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2529 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2530 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
2531 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2532 :
2533 0 : let state = get_state(&request);
2534 :
2535 0 : let timeline =
2536 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2537 0 : .await?;
2538 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
2539 0 : .with_scope_timeline(&timeline);
2540 0 : match timeline.spawn_download_all_remote_layers(body, &ctx).await {
2541 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
2542 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
2543 : }
2544 0 : }
2545 :
2546 0 : async fn timeline_download_remote_layers_handler_get(
2547 0 : request: Request<Body>,
2548 0 : _cancel: CancellationToken,
2549 0 : ) -> Result<Response<Body>, ApiError> {
2550 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2551 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2552 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2553 0 : let state = get_state(&request);
2554 :
2555 0 : let timeline =
2556 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2557 0 : .await?;
2558 0 : let info = timeline
2559 0 : .get_download_all_remote_layers_task_info()
2560 0 : .context("task never started since last pageserver process start")
2561 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
2562 0 : json_response(StatusCode::OK, info)
2563 0 : }
2564 :
2565 0 : async fn timeline_detach_ancestor_handler(
2566 0 : request: Request<Body>,
2567 0 : _cancel: CancellationToken,
2568 0 : ) -> Result<Response<Body>, ApiError> {
2569 : use pageserver_api::models::detach_ancestor::AncestorDetached;
2570 :
2571 : use crate::tenant::timeline::detach_ancestor;
2572 :
2573 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2574 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2575 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2576 0 : let behavior: Option<DetachBehavior> = parse_query_param(&request, "detach_behavior")?;
2577 :
2578 0 : let behavior = behavior.unwrap_or_default();
2579 :
2580 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
2581 :
2582 0 : async move {
2583 0 : let mut options = detach_ancestor::Options::default();
2584 :
2585 0 : let rewrite_concurrency =
2586 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
2587 0 : let copy_concurrency =
2588 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
2589 :
2590 0 : [
2591 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
2592 0 : (&mut options.copy_concurrency, copy_concurrency),
2593 0 : ]
2594 0 : .into_iter()
2595 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
2596 0 : .for_each(|(target, val)| *target = val);
2597 :
2598 0 : let state = get_state(&request);
2599 :
2600 0 : let tenant = state
2601 0 : .tenant_manager
2602 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2603 :
2604 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2605 :
2606 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
2607 0 : let ctx = &ctx;
2608 :
2609 : // Flush the upload queues of all timelines before detaching ancestor. We do the same thing again
2610 : // during shutdown. This early upload ensures the pageserver does not need to upload too many
2611 : // things and creates downtime during timeline reloads.
2612 0 : for timeline in tenant.list_timelines() {
2613 0 : timeline
2614 0 : .remote_client
2615 0 : .wait_completion()
2616 0 : .await
2617 0 : .map_err(|e| {
2618 0 : ApiError::PreconditionFailed(format!("cannot drain upload queue: {e}").into())
2619 0 : })?;
2620 : }
2621 :
2622 0 : tracing::info!("all timeline upload queues are drained");
2623 :
2624 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2625 0 : let ctx = &ctx.with_scope_timeline(&timeline);
2626 :
2627 0 : let progress = timeline
2628 0 : .prepare_to_detach_from_ancestor(&tenant, options, behavior, ctx)
2629 0 : .await?;
2630 :
2631 : // uncomment to allow early as possible Tenant::drop
2632 : // drop(tenant);
2633 :
2634 0 : let resp = match progress {
2635 0 : detach_ancestor::Progress::Prepared(attempt, prepared) => {
2636 : // it would be great to tag the guard on to the tenant activation future
2637 0 : let reparented_timelines = state
2638 0 : .tenant_manager
2639 0 : .complete_detaching_timeline_ancestor(
2640 0 : tenant_shard_id,
2641 0 : timeline_id,
2642 0 : prepared,
2643 0 : behavior,
2644 0 : attempt,
2645 0 : ctx,
2646 0 : )
2647 0 : .await?;
2648 :
2649 0 : AncestorDetached {
2650 0 : reparented_timelines,
2651 0 : }
2652 : }
2653 0 : detach_ancestor::Progress::Done(resp) => resp,
2654 : };
2655 :
2656 0 : json_response(StatusCode::OK, resp)
2657 0 : }
2658 0 : .instrument(span)
2659 0 : .await
2660 0 : }
2661 :
2662 0 : async fn deletion_queue_flush(
2663 0 : r: Request<Body>,
2664 0 : cancel: CancellationToken,
2665 0 : ) -> Result<Response<Body>, ApiError> {
2666 0 : let state = get_state(&r);
2667 :
2668 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
2669 :
2670 0 : let flush = async {
2671 0 : if execute {
2672 0 : state.deletion_queue_client.flush_execute().await
2673 : } else {
2674 0 : state.deletion_queue_client.flush().await
2675 : }
2676 0 : }
2677 : // DeletionQueueError's only case is shutting down.
2678 0 : .map_err(|_| ApiError::ShuttingDown);
2679 :
2680 0 : tokio::select! {
2681 0 : res = flush => {
2682 0 : res.map(|()| json_response(StatusCode::OK, ()))?
2683 : }
2684 0 : _ = cancel.cancelled() => {
2685 0 : Err(ApiError::ShuttingDown)
2686 : }
2687 : }
2688 0 : }
2689 :
2690 0 : async fn getpage_at_lsn_handler(
2691 0 : request: Request<Body>,
2692 0 : cancel: CancellationToken,
2693 0 : ) -> Result<Response<Body>, ApiError> {
2694 0 : getpage_at_lsn_handler_inner(false, request, cancel).await
2695 0 : }
2696 :
2697 0 : async fn touchpage_at_lsn_handler(
2698 0 : request: Request<Body>,
2699 0 : cancel: CancellationToken,
2700 0 : ) -> Result<Response<Body>, ApiError> {
2701 0 : getpage_at_lsn_handler_inner(true, request, cancel).await
2702 0 : }
2703 :
2704 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
2705 0 : async fn getpage_at_lsn_handler_inner(
2706 0 : touch: bool,
2707 0 : request: Request<Body>,
2708 0 : _cancel: CancellationToken,
2709 0 : ) -> Result<Response<Body>, ApiError> {
2710 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2711 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2712 : // Require pageserver admin permission for this API instead of only tenant-level token.
2713 0 : check_permission(&request, None)?;
2714 0 : let state = get_state(&request);
2715 :
2716 : struct Key(pageserver_api::key::Key);
2717 :
2718 : impl std::str::FromStr for Key {
2719 : type Err = anyhow::Error;
2720 :
2721 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
2722 0 : pageserver_api::key::Key::from_hex(s).map(Key)
2723 0 : }
2724 : }
2725 :
2726 0 : let key: Key = parse_query_param(&request, "key")?
2727 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
2728 0 : let lsn: Option<Lsn> = parse_query_param(&request, "lsn")?;
2729 :
2730 0 : async {
2731 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2732 0 : let ctx = RequestContextBuilder::new(TaskKind::MgmtRequest)
2733 0 : .download_behavior(DownloadBehavior::Download)
2734 0 : .scope(context::Scope::new_timeline(&timeline))
2735 0 : .read_path_debug(true)
2736 0 : .root();
2737 :
2738 : // Use last_record_lsn if no lsn is provided
2739 0 : let lsn = lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2740 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
2741 :
2742 0 : if touch {
2743 0 : json_response(StatusCode::OK, ())
2744 : } else {
2745 0 : Result::<_, ApiError>::Ok(
2746 0 : Response::builder()
2747 0 : .status(StatusCode::OK)
2748 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2749 0 : .body(hyper::Body::from(page))
2750 0 : .unwrap(),
2751 0 : )
2752 : }
2753 0 : }
2754 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2755 0 : .await
2756 0 : }
2757 :
2758 0 : async fn timeline_collect_keyspace(
2759 0 : request: Request<Body>,
2760 0 : _cancel: CancellationToken,
2761 0 : ) -> Result<Response<Body>, ApiError> {
2762 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2763 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2764 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2765 0 : let state = get_state(&request);
2766 :
2767 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2768 :
2769 0 : async {
2770 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2771 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download).with_scope_timeline(&timeline);
2772 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2773 0 : let (dense_ks, sparse_ks) = timeline
2774 0 : .collect_keyspace(at_lsn, &ctx)
2775 0 : .await
2776 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2777 :
2778 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2779 : // Therefore, we split dense/sparse keys in this API.
2780 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2781 :
2782 0 : json_response(StatusCode::OK, res)
2783 0 : }
2784 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2785 0 : .await
2786 0 : }
2787 :
2788 0 : async fn active_timeline_of_active_tenant(
2789 0 : tenant_manager: &TenantManager,
2790 0 : tenant_shard_id: TenantShardId,
2791 0 : timeline_id: TimelineId,
2792 0 : ) -> Result<Arc<Timeline>, ApiError> {
2793 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2794 :
2795 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2796 :
2797 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2798 0 : }
2799 :
2800 0 : async fn always_panic_handler(
2801 0 : req: Request<Body>,
2802 0 : _cancel: CancellationToken,
2803 0 : ) -> Result<Response<Body>, ApiError> {
2804 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2805 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2806 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2807 0 : let query = req.uri().query();
2808 0 : let _ = std::panic::catch_unwind(|| {
2809 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2810 : });
2811 0 : json_response(StatusCode::NO_CONTENT, ())
2812 0 : }
2813 :
2814 0 : async fn disk_usage_eviction_run(
2815 0 : mut r: Request<Body>,
2816 0 : cancel: CancellationToken,
2817 0 : ) -> Result<Response<Body>, ApiError> {
2818 0 : check_permission(&r, None)?;
2819 :
2820 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2821 : struct Config {
2822 : /// How many bytes to evict before reporting that pressure is relieved.
2823 : evict_bytes: u64,
2824 :
2825 : #[serde(default)]
2826 : eviction_order: pageserver_api::config::EvictionOrder,
2827 : }
2828 :
2829 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2830 : struct Usage {
2831 : // remains unchanged after instantiation of the struct
2832 : evict_bytes: u64,
2833 : // updated by `add_available_bytes`
2834 : freed_bytes: u64,
2835 : }
2836 :
2837 : impl crate::disk_usage_eviction_task::Usage for Usage {
2838 0 : fn has_pressure(&self) -> bool {
2839 0 : self.evict_bytes > self.freed_bytes
2840 0 : }
2841 :
2842 0 : fn add_available_bytes(&mut self, bytes: u64) {
2843 0 : self.freed_bytes += bytes;
2844 0 : }
2845 : }
2846 :
2847 0 : let config = json_request::<Config>(&mut r).await?;
2848 :
2849 0 : let usage = Usage {
2850 0 : evict_bytes: config.evict_bytes,
2851 0 : freed_bytes: 0,
2852 0 : };
2853 :
2854 0 : let state = get_state(&r);
2855 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2856 :
2857 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2858 0 : &eviction_state,
2859 0 : &state.remote_storage,
2860 0 : usage,
2861 0 : &state.tenant_manager,
2862 0 : config.eviction_order.into(),
2863 0 : &cancel,
2864 0 : )
2865 0 : .await;
2866 :
2867 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2868 :
2869 0 : let res = res.map_err(ApiError::InternalServerError)?;
2870 :
2871 0 : json_response(StatusCode::OK, res)
2872 0 : }
2873 :
2874 0 : async fn secondary_upload_handler(
2875 0 : request: Request<Body>,
2876 0 : _cancel: CancellationToken,
2877 0 : ) -> Result<Response<Body>, ApiError> {
2878 0 : let state = get_state(&request);
2879 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2880 0 : state
2881 0 : .secondary_controller
2882 0 : .upload_tenant(tenant_shard_id)
2883 0 : .await?;
2884 :
2885 0 : json_response(StatusCode::OK, ())
2886 0 : }
2887 :
2888 0 : async fn tenant_scan_remote_handler(
2889 0 : request: Request<Body>,
2890 0 : cancel: CancellationToken,
2891 0 : ) -> Result<Response<Body>, ApiError> {
2892 0 : let state = get_state(&request);
2893 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2894 :
2895 0 : let mut response = TenantScanRemoteStorageResponse::default();
2896 :
2897 0 : let (shards, _other_keys) =
2898 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2899 0 : .await
2900 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2901 :
2902 0 : for tenant_shard_id in shards {
2903 0 : let (timeline_ids, _other_keys) =
2904 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2905 0 : .await
2906 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2907 :
2908 0 : let mut generation = Generation::none();
2909 0 : for timeline_id in timeline_ids {
2910 0 : match download_index_part(
2911 0 : &state.remote_storage,
2912 0 : &tenant_shard_id,
2913 0 : &timeline_id,
2914 : Generation::MAX,
2915 0 : &cancel,
2916 : )
2917 0 : .instrument(info_span!("download_index_part",
2918 : tenant_id=%tenant_shard_id.tenant_id,
2919 0 : shard_id=%tenant_shard_id.shard_slug(),
2920 : %timeline_id))
2921 0 : .await
2922 : {
2923 0 : Ok((index_part, index_generation, _index_mtime)) => {
2924 0 : tracing::info!(
2925 0 : "Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2926 0 : index_part.layer_metadata.len(),
2927 0 : index_part.metadata.disk_consistent_lsn()
2928 : );
2929 0 : generation = std::cmp::max(generation, index_generation);
2930 : }
2931 : Err(DownloadError::NotFound) => {
2932 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2933 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2934 0 : tracing::info!(
2935 0 : "Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping"
2936 : );
2937 0 : continue;
2938 : }
2939 0 : Err(e) => {
2940 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2941 : }
2942 : };
2943 : }
2944 :
2945 0 : let result =
2946 0 : download_tenant_manifest(&state.remote_storage, &tenant_shard_id, generation, &cancel)
2947 0 : .instrument(info_span!("download_tenant_manifest",
2948 : tenant_id=%tenant_shard_id.tenant_id,
2949 0 : shard_id=%tenant_shard_id.shard_slug()))
2950 0 : .await;
2951 0 : let stripe_size = match result {
2952 0 : Ok((manifest, _, _)) => manifest.stripe_size,
2953 0 : Err(DownloadError::NotFound) => None,
2954 0 : Err(err) => return Err(ApiError::InternalServerError(anyhow!(err))),
2955 : };
2956 :
2957 0 : response.shards.push(TenantScanRemoteStorageShard {
2958 0 : tenant_shard_id,
2959 0 : generation: generation.into(),
2960 0 : stripe_size,
2961 0 : });
2962 : }
2963 :
2964 0 : if response.shards.is_empty() {
2965 0 : return Err(ApiError::NotFound(
2966 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2967 0 : ));
2968 0 : }
2969 :
2970 0 : json_response(StatusCode::OK, response)
2971 0 : }
2972 :
2973 0 : async fn secondary_download_handler(
2974 0 : request: Request<Body>,
2975 0 : _cancel: CancellationToken,
2976 0 : ) -> Result<Response<Body>, ApiError> {
2977 0 : let state = get_state(&request);
2978 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2979 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2980 :
2981 : // We don't need this to issue the download request, but:
2982 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2983 : // - we will use this to provide status feedback in the response
2984 0 : let Some(secondary_tenant) = state
2985 0 : .tenant_manager
2986 0 : .get_secondary_tenant_shard(tenant_shard_id)
2987 : else {
2988 0 : return Err(ApiError::NotFound(
2989 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2990 0 : ));
2991 : };
2992 :
2993 0 : let timeout = wait.unwrap_or(Duration::MAX);
2994 :
2995 0 : let result = tokio::time::timeout(
2996 0 : timeout,
2997 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2998 0 : )
2999 0 : .await;
3000 :
3001 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
3002 :
3003 0 : let status = match result {
3004 : Ok(Ok(())) => {
3005 0 : if progress.layers_downloaded >= progress.layers_total {
3006 : // Download job ran to completion
3007 0 : StatusCode::OK
3008 : } else {
3009 : // Download dropped out without errors because it ran out of time budget
3010 0 : StatusCode::ACCEPTED
3011 : }
3012 : }
3013 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
3014 : // okay. We could get an error here in the unlikely edge case that the tenant
3015 : // was detached between our check above and executing the download job.
3016 0 : Ok(Err(e)) => return Err(e.into()),
3017 : // A timeout is not an error: we have started the download, we're just not done
3018 : // yet. The caller will get a response body indicating status.
3019 0 : Err(_) => StatusCode::ACCEPTED,
3020 : };
3021 :
3022 0 : json_response(status, progress)
3023 0 : }
3024 :
3025 0 : async fn wait_lsn_handler(
3026 0 : mut request: Request<Body>,
3027 0 : cancel: CancellationToken,
3028 0 : ) -> Result<Response<Body>, ApiError> {
3029 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3030 0 : let wait_lsn_request: TenantWaitLsnRequest = json_request(&mut request).await?;
3031 :
3032 0 : let state = get_state(&request);
3033 0 : let tenant = state
3034 0 : .tenant_manager
3035 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3036 :
3037 0 : let mut wait_futures = Vec::default();
3038 0 : for timeline in tenant.list_timelines() {
3039 0 : let Some(lsn) = wait_lsn_request.timelines.get(&timeline.timeline_id) else {
3040 0 : continue;
3041 : };
3042 :
3043 0 : let fut = {
3044 0 : let timeline = timeline.clone();
3045 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
3046 0 : async move {
3047 0 : timeline
3048 0 : .wait_lsn(
3049 0 : *lsn,
3050 0 : WaitLsnWaiter::HttpEndpoint,
3051 0 : WaitLsnTimeout::Custom(wait_lsn_request.timeout),
3052 0 : &ctx,
3053 0 : )
3054 0 : .await
3055 0 : }
3056 : };
3057 0 : wait_futures.push(fut);
3058 : }
3059 :
3060 0 : if wait_futures.is_empty() {
3061 0 : return json_response(StatusCode::NOT_FOUND, ());
3062 0 : }
3063 :
3064 0 : let all_done = tokio::select! {
3065 0 : results = join_all(wait_futures) => {
3066 0 : results.iter().all(|res| res.is_ok())
3067 : },
3068 0 : _ = cancel.cancelled() => {
3069 0 : return Err(ApiError::Cancelled);
3070 : }
3071 : };
3072 :
3073 0 : let status = if all_done {
3074 0 : StatusCode::OK
3075 : } else {
3076 0 : StatusCode::ACCEPTED
3077 : };
3078 :
3079 0 : json_response(status, ())
3080 0 : }
3081 :
3082 0 : async fn secondary_status_handler(
3083 0 : request: Request<Body>,
3084 0 : _cancel: CancellationToken,
3085 0 : ) -> Result<Response<Body>, ApiError> {
3086 0 : let state = get_state(&request);
3087 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3088 :
3089 0 : let Some(secondary_tenant) = state
3090 0 : .tenant_manager
3091 0 : .get_secondary_tenant_shard(tenant_shard_id)
3092 : else {
3093 0 : return Err(ApiError::NotFound(
3094 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
3095 0 : ));
3096 : };
3097 :
3098 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
3099 :
3100 0 : json_response(StatusCode::OK, progress)
3101 0 : }
3102 :
3103 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
3104 0 : json_response(
3105 : StatusCode::NOT_FOUND,
3106 0 : HttpErrorBody::from_msg("page not found".to_owned()),
3107 : )
3108 0 : }
3109 :
3110 0 : async fn post_tracing_event_handler(
3111 0 : mut r: Request<Body>,
3112 0 : _cancel: CancellationToken,
3113 0 : ) -> Result<Response<Body>, ApiError> {
3114 0 : #[derive(Debug, serde::Deserialize)]
3115 : #[serde(rename_all = "lowercase")]
3116 : enum Level {
3117 : Error,
3118 : Warn,
3119 : Info,
3120 : Debug,
3121 : Trace,
3122 : }
3123 0 : #[derive(Debug, serde::Deserialize)]
3124 : struct Request {
3125 : level: Level,
3126 : message: String,
3127 : }
3128 0 : let body: Request = json_request(&mut r)
3129 0 : .await
3130 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
3131 :
3132 0 : match body.level {
3133 0 : Level::Error => tracing::error!(?body.message),
3134 0 : Level::Warn => tracing::warn!(?body.message),
3135 0 : Level::Info => tracing::info!(?body.message),
3136 0 : Level::Debug => tracing::debug!(?body.message),
3137 0 : Level::Trace => tracing::trace!(?body.message),
3138 : }
3139 :
3140 0 : json_response(StatusCode::OK, ())
3141 0 : }
3142 :
3143 0 : async fn put_io_engine_handler(
3144 0 : mut r: Request<Body>,
3145 0 : _cancel: CancellationToken,
3146 0 : ) -> Result<Response<Body>, ApiError> {
3147 0 : check_permission(&r, None)?;
3148 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
3149 0 : crate::virtual_file::io_engine::set(kind);
3150 0 : json_response(StatusCode::OK, ())
3151 0 : }
3152 :
3153 0 : async fn put_io_mode_handler(
3154 0 : mut r: Request<Body>,
3155 0 : _cancel: CancellationToken,
3156 0 : ) -> Result<Response<Body>, ApiError> {
3157 0 : check_permission(&r, None)?;
3158 0 : let mode: IoMode = json_request(&mut r).await?;
3159 0 : crate::virtual_file::set_io_mode(mode);
3160 0 : json_response(StatusCode::OK, ())
3161 0 : }
3162 :
3163 : /// Polled by control plane.
3164 : ///
3165 : /// See [`crate::utilization`].
3166 0 : async fn get_utilization(
3167 0 : r: Request<Body>,
3168 0 : _cancel: CancellationToken,
3169 0 : ) -> Result<Response<Body>, ApiError> {
3170 0 : fail::fail_point!("get-utilization-http-handler", |_| {
3171 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
3172 0 : });
3173 :
3174 : // this probably could be completely public, but lets make that change later.
3175 0 : check_permission(&r, None)?;
3176 :
3177 0 : let state = get_state(&r);
3178 0 : let mut g = state.latest_utilization.lock().await;
3179 :
3180 0 : let regenerate_every = Duration::from_secs(1);
3181 0 : let still_valid = g
3182 0 : .as_ref()
3183 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
3184 :
3185 : // avoid needless statvfs calls even though those should be non-blocking fast.
3186 : // regenerate at most 1Hz to allow polling at any rate.
3187 0 : if !still_valid {
3188 0 : let path = state.conf.tenants_path();
3189 0 : let doc =
3190 0 : crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
3191 0 : .map_err(ApiError::InternalServerError)?;
3192 :
3193 0 : let mut buf = Vec::new();
3194 0 : serde_json::to_writer(&mut buf, &doc)
3195 0 : .context("serialize")
3196 0 : .map_err(ApiError::InternalServerError)?;
3197 :
3198 0 : let body = bytes::Bytes::from(buf);
3199 :
3200 0 : *g = Some((std::time::Instant::now(), body));
3201 0 : }
3202 :
3203 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
3204 0 : let cached = g.as_ref().expect("just set").1.clone();
3205 :
3206 0 : Response::builder()
3207 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
3208 : // thought of using http date header, but that is second precision which does not give any
3209 : // debugging aid
3210 0 : .status(StatusCode::OK)
3211 0 : .body(hyper::Body::from(cached))
3212 0 : .context("build response")
3213 0 : .map_err(ApiError::InternalServerError)
3214 0 : }
3215 :
3216 0 : async fn list_aux_files(
3217 0 : mut request: Request<Body>,
3218 0 : _cancel: CancellationToken,
3219 0 : ) -> Result<Response<Body>, ApiError> {
3220 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3221 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3222 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
3223 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3224 :
3225 0 : let state = get_state(&request);
3226 :
3227 0 : let timeline =
3228 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3229 0 : .await?;
3230 :
3231 0 : let io_concurrency = IoConcurrency::spawn_from_conf(
3232 0 : state.conf.get_vectored_concurrent_io,
3233 0 : timeline.gate.enter().map_err(|_| ApiError::Cancelled)?,
3234 : );
3235 :
3236 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download)
3237 0 : .with_scope_timeline(&timeline);
3238 0 : let files = timeline
3239 0 : .list_aux_files(body.lsn, &ctx, io_concurrency)
3240 0 : .await?;
3241 0 : json_response(StatusCode::OK, files)
3242 0 : }
3243 :
3244 0 : async fn perf_info(
3245 0 : request: Request<Body>,
3246 0 : _cancel: CancellationToken,
3247 0 : ) -> Result<Response<Body>, ApiError> {
3248 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3249 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3250 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3251 :
3252 0 : let state = get_state(&request);
3253 :
3254 0 : let timeline =
3255 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3256 0 : .await?;
3257 :
3258 0 : let result = timeline.perf_info().await;
3259 :
3260 0 : json_response(StatusCode::OK, result)
3261 0 : }
3262 :
3263 0 : async fn ingest_aux_files(
3264 0 : mut request: Request<Body>,
3265 0 : _cancel: CancellationToken,
3266 0 : ) -> Result<Response<Body>, ApiError> {
3267 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3268 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3269 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
3270 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3271 :
3272 0 : let state = get_state(&request);
3273 :
3274 0 : let timeline =
3275 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
3276 0 : .await?;
3277 :
3278 0 : let mut modification = timeline.begin_modification(
3279 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
3280 0 : );
3281 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
3282 0 : for (fname, content) in body.aux_files {
3283 0 : modification
3284 0 : .put_file(&fname, content.as_bytes(), &ctx)
3285 0 : .await
3286 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
3287 : }
3288 0 : modification
3289 0 : .commit(&ctx)
3290 0 : .await
3291 0 : .map_err(ApiError::InternalServerError)?;
3292 :
3293 0 : json_response(StatusCode::OK, ())
3294 0 : }
3295 :
3296 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
3297 : /// candidates for splitting
3298 0 : async fn post_top_tenants(
3299 0 : mut r: Request<Body>,
3300 0 : _cancel: CancellationToken,
3301 0 : ) -> Result<Response<Body>, ApiError> {
3302 0 : check_permission(&r, None)?;
3303 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
3304 0 : let state = get_state(&r);
3305 :
3306 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
3307 0 : match order_by {
3308 0 : TenantSorting::ResidentSize => sizes.resident_size,
3309 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
3310 0 : TenantSorting::MaxLogicalSizePerShard => sizes.max_logical_size_per_shard,
3311 : }
3312 0 : }
3313 :
3314 : #[derive(Eq, PartialEq)]
3315 : struct HeapItem {
3316 : metric: u64,
3317 : sizes: TopTenantShardItem,
3318 : }
3319 :
3320 : impl PartialOrd for HeapItem {
3321 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
3322 0 : Some(self.cmp(other))
3323 0 : }
3324 : }
3325 :
3326 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
3327 : /// supports popping the greatest item but not the smallest.
3328 : impl Ord for HeapItem {
3329 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
3330 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
3331 0 : }
3332 : }
3333 :
3334 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
3335 :
3336 : // FIXME: this is a lot of clones to take this tenant list
3337 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
3338 0 : if let Some(shards_lt) = request.where_shards_lt {
3339 : // Ignore tenants which already have >= this many shards
3340 0 : if tenant_shard_id.shard_count >= shards_lt {
3341 0 : continue;
3342 0 : }
3343 0 : }
3344 :
3345 0 : let sizes = match tenant_slot {
3346 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
3347 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
3348 0 : continue;
3349 : }
3350 : };
3351 0 : let metric = get_size_metric(&sizes, &request.order_by);
3352 :
3353 0 : if let Some(gt) = request.where_gt {
3354 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
3355 0 : if metric <= gt {
3356 0 : continue;
3357 0 : }
3358 0 : };
3359 :
3360 0 : match top_n.peek() {
3361 0 : None => {
3362 0 : // Top N list is empty: candidate becomes first member
3363 0 : top_n.push(HeapItem { metric, sizes });
3364 0 : }
3365 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
3366 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
3367 0 : top_n.push(HeapItem { metric, sizes });
3368 0 : }
3369 0 : Some(i) if i.metric > metric => {
3370 0 : // List is at limit and lowest value is greater than our candidate, drop it.
3371 0 : }
3372 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
3373 : }
3374 :
3375 0 : while top_n.len() > request.limit {
3376 0 : top_n.pop();
3377 0 : }
3378 : }
3379 :
3380 0 : json_response(
3381 : StatusCode::OK,
3382 : TopTenantShardsResponse {
3383 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
3384 : },
3385 : )
3386 0 : }
3387 :
3388 0 : async fn put_tenant_timeline_import_basebackup(
3389 0 : request: Request<Body>,
3390 0 : _cancel: CancellationToken,
3391 0 : ) -> Result<Response<Body>, ApiError> {
3392 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3393 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3394 0 : let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
3395 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3396 0 : let pg_version: PgMajorVersion = must_parse_query_param(&request, "pg_version")?;
3397 :
3398 0 : check_permission(&request, Some(tenant_id))?;
3399 :
3400 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3401 :
3402 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
3403 :
3404 0 : let span = info_span!("import_basebackup",
3405 0 : tenant_id=%tenant_id, timeline_id=%timeline_id, shard_id=%tenant_shard_id.shard_slug(),
3406 : base_lsn=%base_lsn, end_lsn=%end_lsn, pg_version=%pg_version);
3407 0 : async move {
3408 0 : let state = get_state(&request);
3409 0 : let tenant = state
3410 0 : .tenant_manager
3411 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3412 :
3413 0 : let broker_client = state.broker_client.clone();
3414 :
3415 0 : let mut body = StreamReader::new(
3416 0 : request
3417 0 : .into_body()
3418 0 : .map(|res| res.map_err(|error| std::io::Error::other(anyhow::anyhow!(error)))),
3419 : );
3420 :
3421 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
3422 :
3423 0 : let (timeline, timeline_ctx) = tenant
3424 0 : .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
3425 0 : .map_err(ApiError::InternalServerError)
3426 0 : .await?;
3427 :
3428 : // TODO mark timeline as not ready until it reaches end_lsn.
3429 : // We might have some wal to import as well, and we should prevent compute
3430 : // from connecting before that and writing conflicting wal.
3431 : //
3432 : // This is not relevant for pageserver->pageserver migrations, since there's
3433 : // no wal to import. But should be fixed if we want to import from postgres.
3434 :
3435 : // TODO leave clean state on error. For now you can use detach to clean
3436 : // up broken state from a failed import.
3437 :
3438 : // Import basebackup provided via CopyData
3439 0 : info!("importing basebackup");
3440 :
3441 0 : timeline
3442 0 : .import_basebackup_from_tar(
3443 0 : tenant.clone(),
3444 0 : &mut body,
3445 0 : base_lsn,
3446 0 : broker_client,
3447 0 : &timeline_ctx,
3448 0 : )
3449 0 : .await
3450 0 : .map_err(ApiError::InternalServerError)?;
3451 :
3452 : // Read the end of the tar archive.
3453 0 : read_tar_eof(body)
3454 0 : .await
3455 0 : .map_err(ApiError::InternalServerError)?;
3456 :
3457 : // TODO check checksum
3458 : // Meanwhile you can verify client-side by taking fullbackup
3459 : // and checking that it matches in size with what was imported.
3460 : // It wouldn't work if base came from vanilla postgres though,
3461 : // since we discard some log files.
3462 :
3463 0 : info!("done");
3464 0 : json_response(StatusCode::OK, ())
3465 0 : }
3466 0 : .instrument(span)
3467 0 : .await
3468 0 : }
3469 :
3470 0 : async fn put_tenant_timeline_import_wal(
3471 0 : request: Request<Body>,
3472 0 : _cancel: CancellationToken,
3473 0 : ) -> Result<Response<Body>, ApiError> {
3474 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3475 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3476 0 : let start_lsn: Lsn = must_parse_query_param(&request, "start_lsn")?;
3477 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3478 :
3479 0 : check_permission(&request, Some(tenant_id))?;
3480 :
3481 0 : let span = info_span!("import_wal", tenant_id=%tenant_id, timeline_id=%timeline_id, start_lsn=%start_lsn, end_lsn=%end_lsn);
3482 0 : async move {
3483 0 : let state = get_state(&request);
3484 :
3485 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, TenantShardId::unsharded(tenant_id), timeline_id).await?;
3486 0 : let ctx = RequestContextBuilder::new(TaskKind::MgmtRequest)
3487 0 : .download_behavior(DownloadBehavior::Warn)
3488 0 : .scope(context::Scope::new_timeline(&timeline))
3489 0 : .root();
3490 :
3491 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
3492 0 : res.map_err(|error| {
3493 0 : std::io::Error::other( anyhow::anyhow!(error))
3494 0 : })
3495 0 : }));
3496 :
3497 0 : let last_record_lsn = timeline.get_last_record_lsn();
3498 0 : if last_record_lsn != start_lsn {
3499 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}")));
3500 0 : }
3501 :
3502 : // TODO leave clean state on error. For now you can use detach to clean
3503 : // up broken state from a failed import.
3504 :
3505 : // Import wal provided via CopyData
3506 0 : info!("importing wal");
3507 0 : crate::import_datadir::import_wal_from_tar(&timeline, &mut body, start_lsn, end_lsn, &ctx).await.map_err(ApiError::InternalServerError)?;
3508 0 : info!("wal import complete");
3509 :
3510 : // Read the end of the tar archive.
3511 0 : read_tar_eof(body).await.map_err(ApiError::InternalServerError)?;
3512 :
3513 : // TODO Does it make sense to overshoot?
3514 0 : if timeline.get_last_record_lsn() < end_lsn {
3515 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}")));
3516 0 : }
3517 :
3518 : // Flush data to disk, then upload to s3. No need for a forced checkpoint.
3519 : // We only want to persist the data, and it doesn't matter if it's in the
3520 : // shape of deltas or images.
3521 0 : info!("flushing layers");
3522 0 : timeline.freeze_and_flush().await.map_err(|e| match e {
3523 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
3524 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
3525 0 : })?;
3526 :
3527 0 : info!("done");
3528 :
3529 0 : json_response(StatusCode::OK, ())
3530 0 : }.instrument(span).await
3531 0 : }
3532 :
3533 : /// Activate a timeline after its import has completed
3534 : ///
3535 : /// The endpoint is idempotent and callers are expected to retry all
3536 : /// errors until a successful response.
3537 0 : async fn activate_post_import_handler(
3538 0 : request: Request<Body>,
3539 0 : _cancel: CancellationToken,
3540 0 : ) -> Result<Response<Body>, ApiError> {
3541 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3542 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3543 :
3544 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3545 : const DEFAULT_ACTIVATE_TIMEOUT: Duration = Duration::from_secs(1);
3546 0 : let activate_timeout = parse_query_param(&request, "timeline_activate_timeout_ms")?
3547 0 : .map(Duration::from_millis)
3548 0 : .unwrap_or(DEFAULT_ACTIVATE_TIMEOUT);
3549 :
3550 0 : let span = info_span!(
3551 : "activate_post_import_handler",
3552 : tenant_id=%tenant_shard_id.tenant_id,
3553 : timeline_id=%timeline_id,
3554 0 : shard_id=%tenant_shard_id.shard_slug()
3555 : );
3556 :
3557 0 : async move {
3558 0 : let state = get_state(&request);
3559 0 : let tenant = state
3560 0 : .tenant_manager
3561 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3562 :
3563 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
3564 :
3565 0 : tenant.finalize_importing_timeline(timeline_id).await?;
3566 :
3567 0 : match tenant.get_timeline(timeline_id, false) {
3568 0 : Ok(_timeline) => {
3569 0 : // Timeline is already visible. Reset not required: fall through.
3570 0 : }
3571 : Err(GetTimelineError::NotFound { .. }) => {
3572 : // This is crude: we reset the whole tenant such that the new timeline is detected
3573 : // and activated. We can come up with something more granular in the future.
3574 : //
3575 : // Note that we only reset the tenant if required: when the timeline is
3576 : // not present in [`Tenant::timelines`].
3577 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3578 0 : state
3579 0 : .tenant_manager
3580 0 : .reset_tenant(tenant_shard_id, false, &ctx)
3581 0 : .await
3582 0 : .map_err(ApiError::InternalServerError)?;
3583 : }
3584 : Err(GetTimelineError::ShuttingDown) => {
3585 0 : return Err(ApiError::ShuttingDown);
3586 : }
3587 : Err(GetTimelineError::NotActive { .. }) => {
3588 0 : unreachable!("Called get_timeline with active_only=false");
3589 : }
3590 : }
3591 :
3592 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
3593 :
3594 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn)
3595 0 : .with_scope_timeline(&timeline);
3596 :
3597 0 : let result =
3598 0 : tokio::time::timeout(activate_timeout, timeline.wait_to_become_active(&ctx)).await;
3599 0 : match result {
3600 0 : Ok(Ok(())) => {
3601 0 : // fallthrough
3602 0 : }
3603 : // Timeline reached some other state that's not active
3604 : // TODO(vlad): if the tenant is broken, return a permananet error
3605 0 : Ok(Err(_timeline_state)) => {
3606 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(
3607 0 : "Timeline activation failed"
3608 0 : )));
3609 : }
3610 : // Activation timed out
3611 : Err(_) => {
3612 0 : return Err(ApiError::Timeout("Timeline activation timed out".into()));
3613 : }
3614 : }
3615 :
3616 0 : let timeline_info = build_timeline_info(
3617 0 : &timeline, false, // include_non_incremental_logical_size,
3618 0 : false, // force_await_initial_logical_size
3619 0 : &ctx,
3620 0 : )
3621 0 : .await
3622 0 : .context("get local timeline info")
3623 0 : .map_err(ApiError::InternalServerError)?;
3624 :
3625 0 : json_response(StatusCode::OK, timeline_info)
3626 0 : }
3627 0 : .instrument(span)
3628 0 : .await
3629 0 : }
3630 :
3631 : /// Read the end of a tar archive.
3632 : ///
3633 : /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
3634 : /// `tokio_tar` already read the first such block. Read the second all-zeros block,
3635 : /// and check that there is no more data after the EOF marker.
3636 : ///
3637 : /// 'tar' command can also write extra blocks of zeros, up to a record
3638 : /// size, controlled by the --record-size argument. Ignore them too.
3639 0 : async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow::Result<()> {
3640 : use tokio::io::AsyncReadExt;
3641 0 : let mut buf = [0u8; 512];
3642 :
3643 : // Read the all-zeros block, and verify it
3644 0 : let mut total_bytes = 0;
3645 0 : while total_bytes < 512 {
3646 0 : let nbytes = reader.read(&mut buf[total_bytes..]).await?;
3647 0 : total_bytes += nbytes;
3648 0 : if nbytes == 0 {
3649 0 : break;
3650 0 : }
3651 : }
3652 0 : if total_bytes < 512 {
3653 0 : anyhow::bail!("incomplete or invalid tar EOF marker");
3654 0 : }
3655 0 : if !buf.iter().all(|&x| x == 0) {
3656 0 : anyhow::bail!("invalid tar EOF marker");
3657 0 : }
3658 :
3659 : // Drain any extra zero-blocks after the EOF marker
3660 0 : let mut trailing_bytes = 0;
3661 0 : let mut seen_nonzero_bytes = false;
3662 : loop {
3663 0 : let nbytes = reader.read(&mut buf).await?;
3664 0 : trailing_bytes += nbytes;
3665 0 : if !buf.iter().all(|&x| x == 0) {
3666 0 : seen_nonzero_bytes = true;
3667 0 : }
3668 0 : if nbytes == 0 {
3669 0 : break;
3670 0 : }
3671 : }
3672 0 : if seen_nonzero_bytes {
3673 0 : anyhow::bail!("unexpected non-zero bytes after the tar archive");
3674 0 : }
3675 0 : if trailing_bytes % 512 != 0 {
3676 0 : anyhow::bail!(
3677 0 : "unexpected number of zeros ({trailing_bytes}), not divisible by tar block size (512 bytes), after the tar archive"
3678 : );
3679 0 : }
3680 0 : Ok(())
3681 0 : }
3682 :
3683 0 : async fn tenant_evaluate_feature_flag(
3684 0 : request: Request<Body>,
3685 0 : _cancel: CancellationToken,
3686 0 : ) -> Result<Response<Body>, ApiError> {
3687 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
3688 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
3689 :
3690 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3691 0 : let as_type: Option<String> = parse_query_param(&request, "as")?;
3692 :
3693 0 : let state = get_state(&request);
3694 :
3695 0 : async {
3696 0 : let tenant = state
3697 0 : .tenant_manager
3698 0 : .get_attached_tenant_shard(tenant_shard_id)?;
3699 : // TODO: the properties we get here might be stale right after it is collected. But such races are rare (updated every 10s)
3700 : // and we don't need to worry about it for now.
3701 0 : let properties = tenant.feature_resolver.collect_properties();
3702 0 : if as_type.as_deref() == Some("boolean") {
3703 0 : let result = tenant.feature_resolver.evaluate_boolean(&flag);
3704 0 : let result = result.map(|_| true).map_err(|e| e.to_string());
3705 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3706 0 : } else if as_type.as_deref() == Some("multivariate") {
3707 0 : let result = tenant.feature_resolver.evaluate_multivariate(&flag).map_err(|e| e.to_string());
3708 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3709 : } else {
3710 : // Auto infer the type of the feature flag.
3711 0 : let is_boolean = tenant.feature_resolver.is_feature_flag_boolean(&flag).map_err(|e| ApiError::InternalServerError(anyhow::anyhow!("{e}")))?;
3712 0 : if is_boolean {
3713 0 : let result = tenant.feature_resolver.evaluate_boolean(&flag);
3714 0 : let result = result.map(|_| true).map_err(|e| e.to_string());
3715 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3716 : } else {
3717 0 : let result = tenant.feature_resolver.evaluate_multivariate(&flag).map_err(|e| e.to_string());
3718 0 : json_response(StatusCode::OK, json!({ "result": result, "properties": properties }))
3719 : }
3720 : }
3721 0 : }
3722 0 : .instrument(info_span!("tenant_evaluate_feature_flag", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
3723 0 : .await
3724 0 : }
3725 :
3726 0 : async fn force_override_feature_flag_for_testing_put(
3727 0 : request: Request<Body>,
3728 0 : _cancel: CancellationToken,
3729 0 : ) -> Result<Response<Body>, ApiError> {
3730 0 : check_permission(&request, None)?;
3731 :
3732 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3733 0 : let value: String = must_parse_query_param(&request, "value")?;
3734 0 : let state = get_state(&request);
3735 0 : state
3736 0 : .feature_resolver
3737 0 : .force_override_for_testing(&flag, Some(&value));
3738 0 : json_response(StatusCode::OK, ())
3739 0 : }
3740 :
3741 0 : async fn force_override_feature_flag_for_testing_delete(
3742 0 : request: Request<Body>,
3743 0 : _cancel: CancellationToken,
3744 0 : ) -> Result<Response<Body>, ApiError> {
3745 0 : check_permission(&request, None)?;
3746 :
3747 0 : let flag: String = parse_request_param(&request, "flag_key")?;
3748 0 : let state = get_state(&request);
3749 0 : state
3750 0 : .feature_resolver
3751 0 : .force_override_for_testing(&flag, None);
3752 0 : json_response(StatusCode::OK, ())
3753 0 : }
3754 :
3755 0 : async fn update_feature_flag_spec(
3756 0 : mut request: Request<Body>,
3757 0 : _cancel: CancellationToken,
3758 0 : ) -> Result<Response<Body>, ApiError> {
3759 0 : check_permission(&request, None)?;
3760 0 : let body = json_request(&mut request).await?;
3761 0 : let state = get_state(&request);
3762 0 : state
3763 0 : .feature_resolver
3764 0 : .update(body)
3765 0 : .map_err(ApiError::InternalServerError)?;
3766 0 : json_response(StatusCode::OK, ())
3767 0 : }
3768 :
3769 : /// Common functionality of all the HTTP API handlers.
3770 : ///
3771 : /// - Adds a tracing span to each request (by `request_span`)
3772 : /// - Logs the request depending on the request method (by `request_span`)
3773 : /// - Logs the response if it was not successful (by `request_span`
3774 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
3775 : /// Future if the connection to the client is lost, but most of the pageserver code is
3776 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
3777 : /// request with a CancellationToken.
3778 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
3779 0 : where
3780 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3781 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3782 0 : {
3783 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
3784 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
3785 0 : "failpoint".into()
3786 0 : )));
3787 :
3788 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
3789 0 : anyhow::anyhow!("failpoint")
3790 0 : )));
3791 0 : }
3792 :
3793 : // Spawn a new task to handle the request, to protect the handler from unexpected
3794 : // async cancellations. Most pageserver functions are not async cancellation safe.
3795 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
3796 : // with the cancellation token.
3797 0 : let token = CancellationToken::new();
3798 0 : let cancel_guard = token.clone().drop_guard();
3799 0 : let result = request_span(request, move |r| async {
3800 0 : let handle = tokio::spawn(
3801 0 : async {
3802 0 : let token_cloned = token.clone();
3803 0 : let result = handler(r, token).await;
3804 0 : if token_cloned.is_cancelled() {
3805 : // dropguard has executed: we will never turn this result into response.
3806 : //
3807 : // at least temporarily do {:?} logging; these failures are rare enough but
3808 : // could hide difficult errors.
3809 0 : match &result {
3810 0 : Ok(response) => {
3811 0 : let status = response.status();
3812 0 : info!(%status, "Cancelled request finished successfully")
3813 : }
3814 0 : Err(e) => match e {
3815 : ApiError::ShuttingDown | ApiError::ResourceUnavailable(_) => {
3816 : // Don't log this at error severity: they are normal during lifecycle of tenants/process
3817 0 : info!("Cancelled request aborted for shutdown")
3818 : }
3819 : _ => {
3820 : // Log these in a highly visible way, because we have no client to send the response to, but
3821 : // would like to know that something went wrong.
3822 0 : error!("Cancelled request finished with an error: {e:?}")
3823 : }
3824 : },
3825 : }
3826 0 : }
3827 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
3828 : // which should suffice.
3829 : //
3830 : // there is still a chance to lose the result due to race between
3831 : // returning from here and the actual connection closing happening
3832 : // before outer task gets to execute. leaving that up for #5815.
3833 0 : result
3834 0 : }
3835 0 : .in_current_span(),
3836 : );
3837 :
3838 0 : match handle.await {
3839 : // TODO: never actually return Err from here, always Ok(...) so that we can log
3840 : // spanned errors. Call api_error_handler instead and return appropriate Body.
3841 0 : Ok(result) => result,
3842 0 : Err(e) => {
3843 : // The handler task panicked. We have a global panic handler that logs the
3844 : // panic with its backtrace, so no need to log that here. Only log a brief
3845 : // message to make it clear that we returned the error to the client.
3846 0 : error!("HTTP request handler task panicked: {e:#}");
3847 :
3848 : // Don't return an Error here, because then fallback error handler that was
3849 : // installed in make_router() will print the error. Instead, construct the
3850 : // HTTP error response and return that.
3851 0 : Ok(
3852 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
3853 0 : .into_response(),
3854 0 : )
3855 : }
3856 : }
3857 0 : })
3858 0 : .await;
3859 :
3860 0 : cancel_guard.disarm();
3861 :
3862 0 : result
3863 0 : }
3864 :
3865 : /// Like api_handler, but returns an error response if the server is built without
3866 : /// the 'testing' feature.
3867 0 : async fn testing_api_handler<R, H>(
3868 0 : desc: &str,
3869 0 : request: Request<Body>,
3870 0 : handler: H,
3871 0 : ) -> Result<Response<Body>, ApiError>
3872 0 : where
3873 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3874 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3875 0 : {
3876 0 : if cfg!(feature = "testing") {
3877 0 : api_handler(request, handler).await
3878 : } else {
3879 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
3880 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
3881 0 : ))))
3882 0 : .await
3883 : }
3884 0 : }
3885 :
3886 0 : pub fn make_router(
3887 0 : state: Arc<State>,
3888 0 : launch_ts: &'static LaunchTimestamp,
3889 0 : auth: Option<Arc<SwappableJwtAuth>>,
3890 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
3891 0 : let spec = include_bytes!("openapi_spec.yml");
3892 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
3893 0 : if auth.is_some() {
3894 0 : router = router.middleware(auth_middleware(|request| {
3895 0 : let state = get_state(request);
3896 0 : if state.allowlist_routes.contains(&request.uri().path()) {
3897 0 : None
3898 : } else {
3899 0 : state.auth.as_deref()
3900 : }
3901 0 : }))
3902 0 : }
3903 :
3904 0 : router = router.middleware(
3905 0 : endpoint::add_response_header_middleware(
3906 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
3907 0 : &launch_ts.to_string(),
3908 : )
3909 0 : .expect("construct launch timestamp header middleware"),
3910 : );
3911 :
3912 0 : Ok(router
3913 0 : .data(state)
3914 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
3915 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
3916 0 : .get("/profile/heap", |r| request_span(r, profile_heap_handler))
3917 0 : .get("/v1/status", |r| api_handler(r, status_handler))
3918 0 : .put("/v1/failpoints", |r| {
3919 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
3920 0 : })
3921 0 : .post("/v1/reload_auth_validation_keys", |r| {
3922 0 : api_handler(r, reload_auth_validation_keys_handler)
3923 0 : })
3924 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
3925 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
3926 0 : api_handler(r, tenant_status)
3927 0 : })
3928 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
3929 0 : api_handler(r, tenant_delete_handler)
3930 0 : })
3931 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
3932 0 : api_handler(r, tenant_size_handler)
3933 0 : })
3934 0 : .patch("/v1/tenant/config", |r| {
3935 0 : api_handler(r, patch_tenant_config_handler)
3936 0 : })
3937 0 : .put("/v1/tenant/config", |r| {
3938 0 : api_handler(r, update_tenant_config_handler)
3939 0 : })
3940 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
3941 0 : api_handler(r, tenant_shard_split_handler)
3942 0 : })
3943 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
3944 0 : api_handler(r, get_tenant_config_handler)
3945 0 : })
3946 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
3947 0 : api_handler(r, put_tenant_location_config_handler)
3948 0 : })
3949 0 : .get("/v1/location_config", |r| {
3950 0 : api_handler(r, list_location_config_handler)
3951 0 : })
3952 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
3953 0 : api_handler(r, get_location_config_handler)
3954 0 : })
3955 0 : .put(
3956 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
3957 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
3958 : )
3959 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
3960 0 : api_handler(r, timeline_list_handler)
3961 0 : })
3962 0 : .get("/v1/tenant/:tenant_shard_id/timeline_and_offloaded", |r| {
3963 0 : api_handler(r, timeline_and_offloaded_list_handler)
3964 0 : })
3965 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
3966 0 : api_handler(r, timeline_create_handler)
3967 0 : })
3968 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
3969 0 : api_handler(r, tenant_reset_handler)
3970 0 : })
3971 0 : .post(
3972 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
3973 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
3974 : )
3975 0 : .put(
3976 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/archival_config",
3977 0 : |r| api_handler(r, timeline_archival_config_handler),
3978 : )
3979 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3980 0 : api_handler(r, timeline_detail_handler)
3981 0 : })
3982 0 : .get(
3983 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
3984 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
3985 : )
3986 0 : .get(
3987 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
3988 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
3989 : )
3990 0 : .post(
3991 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/patch_index_part",
3992 0 : |r| api_handler(r, timeline_patch_index_part_handler),
3993 : )
3994 0 : .post(
3995 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
3996 0 : |r| api_handler(r, lsn_lease_handler),
3997 : )
3998 0 : .put(
3999 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
4000 0 : |r| api_handler(r, timeline_gc_handler),
4001 : )
4002 0 : .get(
4003 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4004 0 : |r| api_handler(r, timeline_compact_info_handler),
4005 : )
4006 0 : .put(
4007 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4008 0 : |r| api_handler(r, timeline_compact_handler),
4009 : )
4010 0 : .delete(
4011 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
4012 0 : |r| api_handler(r, timeline_cancel_compact_handler),
4013 : )
4014 0 : .put(
4015 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/offload",
4016 0 : |r| testing_api_handler("attempt timeline offload", r, timeline_offload_handler),
4017 : )
4018 0 : .put(
4019 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/mark_invisible",
4020 0 : |r| api_handler( r, timeline_mark_invisible_handler),
4021 : )
4022 0 : .put(
4023 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
4024 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
4025 : )
4026 0 : .post(
4027 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
4028 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
4029 : )
4030 0 : .get(
4031 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
4032 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
4033 : )
4034 0 : .put(
4035 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
4036 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
4037 : )
4038 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
4039 0 : api_handler(r, timeline_delete_handler)
4040 0 : })
4041 0 : .get(
4042 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
4043 0 : |r| api_handler(r, layer_map_info_handler),
4044 : )
4045 0 : .post(
4046 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
4047 0 : |r| api_handler(r, timeline_download_heatmap_layers_handler),
4048 : )
4049 0 : .delete(
4050 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_heatmap_layers",
4051 0 : |r| api_handler(r, timeline_shutdown_download_heatmap_layers_handler),
4052 : )
4053 0 : .get(
4054 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
4055 0 : |r| api_handler(r, layer_download_handler),
4056 : )
4057 0 : .delete(
4058 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
4059 0 : |r| api_handler(r, evict_timeline_layer_handler),
4060 : )
4061 0 : .post(
4062 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_name/scan_disposable_keys",
4063 0 : |r| testing_api_handler("timeline_layer_scan_disposable_keys", r, timeline_layer_scan_disposable_keys),
4064 : )
4065 0 : .post(
4066 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
4067 0 : |r| api_handler(r, timeline_gc_blocking_handler),
4068 : )
4069 0 : .post(
4070 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
4071 0 : |r| api_handler(r, timeline_gc_unblocking_handler),
4072 : )
4073 0 : .get(
4074 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/page_trace",
4075 0 : |r| api_handler(r, timeline_page_trace_handler),
4076 : )
4077 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
4078 0 : api_handler(r, secondary_upload_handler)
4079 0 : })
4080 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
4081 0 : api_handler(r, tenant_scan_remote_handler)
4082 0 : })
4083 0 : .put("/v1/disk_usage_eviction/run", |r| {
4084 0 : api_handler(r, disk_usage_eviction_run)
4085 0 : })
4086 0 : .put("/v1/deletion_queue/flush", |r| {
4087 0 : api_handler(r, deletion_queue_flush)
4088 0 : })
4089 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
4090 0 : api_handler(r, secondary_status_handler)
4091 0 : })
4092 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
4093 0 : api_handler(r, secondary_download_handler)
4094 0 : })
4095 0 : .post("/v1/tenant/:tenant_shard_id/wait_lsn", |r| {
4096 0 : api_handler(r, wait_lsn_handler)
4097 0 : })
4098 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
4099 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
4100 0 : })
4101 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
4102 0 : .post("/v1/tracing/event", |r| {
4103 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
4104 0 : })
4105 0 : .get(
4106 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
4107 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
4108 : )
4109 0 : .get(
4110 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/touchpage",
4111 0 : |r| api_handler(r, touchpage_at_lsn_handler),
4112 : )
4113 0 : .get(
4114 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
4115 0 : |r| api_handler(r, timeline_collect_keyspace),
4116 : )
4117 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
4118 0 : .put("/v1/io_mode", |r| api_handler(r, put_io_mode_handler))
4119 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
4120 0 : .post(
4121 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
4122 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
4123 : )
4124 0 : .post(
4125 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
4126 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
4127 : )
4128 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
4129 0 : .post(
4130 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
4131 0 : |r| testing_api_handler("perf_info", r, perf_info),
4132 : )
4133 0 : .put(
4134 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_basebackup",
4135 0 : |r| api_handler(r, put_tenant_timeline_import_basebackup),
4136 : )
4137 0 : .put(
4138 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_wal",
4139 0 : |r| api_handler(r, put_tenant_timeline_import_wal),
4140 : )
4141 0 : .put(
4142 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/activate_post_import",
4143 0 : |r| api_handler(r, activate_post_import_handler),
4144 : )
4145 0 : .get("/v1/tenant/:tenant_shard_id/feature_flag/:flag_key", |r| {
4146 0 : api_handler(r, tenant_evaluate_feature_flag)
4147 0 : })
4148 0 : .put("/v1/feature_flag/:flag_key", |r| {
4149 0 : testing_api_handler("force override feature flag - put", r, force_override_feature_flag_for_testing_put)
4150 0 : })
4151 0 : .delete("/v1/feature_flag/:flag_key", |r| {
4152 0 : testing_api_handler("force override feature flag - delete", r, force_override_feature_flag_for_testing_delete)
4153 0 : })
4154 0 : .post("/v1/feature_flag_spec", |r| {
4155 0 : api_handler(r, update_feature_flag_spec)
4156 0 : })
4157 0 : .any(handler_404))
4158 0 : }
|
