Line data Source code
1 : //!
2 : //! Management HTTP API
3 : //!
4 : use std::cmp::Reverse;
5 : use std::collections::BinaryHeap;
6 : use std::collections::HashMap;
7 : use std::str::FromStr;
8 : use std::sync::Arc;
9 : use std::time::Duration;
10 :
11 : use anyhow::{anyhow, Context, Result};
12 : use enumset::EnumSet;
13 : use futures::StreamExt;
14 : use futures::TryFutureExt;
15 : use humantime::format_rfc3339;
16 : use hyper::header;
17 : use hyper::StatusCode;
18 : use hyper::{Body, Request, Response, Uri};
19 : use metrics::launch_timestamp::LaunchTimestamp;
20 : use pageserver_api::models::virtual_file::IoMode;
21 : use pageserver_api::models::DownloadRemoteLayersTaskSpawnRequest;
22 : use pageserver_api::models::IngestAuxFilesRequest;
23 : use pageserver_api::models::ListAuxFilesRequest;
24 : use pageserver_api::models::LocationConfig;
25 : use pageserver_api::models::LocationConfigListResponse;
26 : use pageserver_api::models::LocationConfigMode;
27 : use pageserver_api::models::LsnLease;
28 : use pageserver_api::models::LsnLeaseRequest;
29 : use pageserver_api::models::OffloadedTimelineInfo;
30 : use pageserver_api::models::ShardParameters;
31 : use pageserver_api::models::TenantConfigPatchRequest;
32 : use pageserver_api::models::TenantDetails;
33 : use pageserver_api::models::TenantLocationConfigRequest;
34 : use pageserver_api::models::TenantLocationConfigResponse;
35 : use pageserver_api::models::TenantScanRemoteStorageResponse;
36 : use pageserver_api::models::TenantScanRemoteStorageShard;
37 : use pageserver_api::models::TenantShardLocation;
38 : use pageserver_api::models::TenantShardSplitRequest;
39 : use pageserver_api::models::TenantShardSplitResponse;
40 : use pageserver_api::models::TenantSorting;
41 : use pageserver_api::models::TenantState;
42 : use pageserver_api::models::TimelineArchivalConfigRequest;
43 : use pageserver_api::models::TimelineCreateRequestMode;
44 : use pageserver_api::models::TimelineCreateRequestModeImportPgdata;
45 : use pageserver_api::models::TimelinesInfoAndOffloaded;
46 : use pageserver_api::models::TopTenantShardItem;
47 : use pageserver_api::models::TopTenantShardsRequest;
48 : use pageserver_api::models::TopTenantShardsResponse;
49 : use pageserver_api::shard::ShardCount;
50 : use pageserver_api::shard::TenantShardId;
51 : use remote_storage::DownloadError;
52 : use remote_storage::GenericRemoteStorage;
53 : use remote_storage::TimeTravelError;
54 : use tenant_size_model::{svg::SvgBranchKind, SizeResult, StorageModel};
55 : use tokio_util::io::StreamReader;
56 : use tokio_util::sync::CancellationToken;
57 : use tracing::*;
58 : use utils::auth::JwtAuth;
59 : use utils::failpoint_support::failpoints_handler;
60 : use utils::http::endpoint::{
61 : profile_cpu_handler, profile_heap_handler, prometheus_metrics_handler, request_span,
62 : };
63 : use utils::http::request::must_parse_query_param;
64 : use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};
65 :
66 : use crate::config::PageServerConf;
67 : use crate::context::{DownloadBehavior, RequestContext};
68 : use crate::deletion_queue::DeletionQueueClient;
69 : use crate::pgdatadir_mapping::LsnForTimestamp;
70 : use crate::task_mgr::TaskKind;
71 : use crate::tenant::config::{LocationConf, TenantConfOpt};
72 : use crate::tenant::mgr::GetActiveTenantError;
73 : use crate::tenant::mgr::{
74 : GetTenantError, TenantManager, TenantMapError, TenantMapInsertError, TenantSlotError,
75 : TenantSlotUpsertError, TenantStateError,
76 : };
77 : use crate::tenant::mgr::{TenantSlot, UpsertLocationError};
78 : use crate::tenant::remote_timeline_client;
79 : use crate::tenant::remote_timeline_client::download_index_part;
80 : use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
81 : use crate::tenant::remote_timeline_client::list_remote_timelines;
82 : use crate::tenant::secondary::SecondaryController;
83 : use crate::tenant::size::ModelInputs;
84 : use crate::tenant::storage_layer::LayerAccessStatsReset;
85 : use crate::tenant::storage_layer::LayerName;
86 : use crate::tenant::timeline::import_pgdata;
87 : use crate::tenant::timeline::offload::offload_timeline;
88 : use crate::tenant::timeline::offload::OffloadError;
89 : use crate::tenant::timeline::CompactFlags;
90 : use crate::tenant::timeline::CompactOptions;
91 : use crate::tenant::timeline::CompactRequest;
92 : use crate::tenant::timeline::CompactionError;
93 : use crate::tenant::timeline::Timeline;
94 : use crate::tenant::GetTimelineError;
95 : use crate::tenant::OffloadedTimeline;
96 : use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
97 : use crate::DEFAULT_PG_VERSION;
98 : use crate::{disk_usage_eviction_task, tenant};
99 : use pageserver_api::models::{
100 : CompactInfoResponse, StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest,
101 : TimelineGcRequest, TimelineInfo,
102 : };
103 : use utils::{
104 : auth::SwappableJwtAuth,
105 : generation::Generation,
106 : http::{
107 : endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
108 : error::{ApiError, HttpErrorBody},
109 : json::{json_request, json_request_maybe, json_response},
110 : request::parse_request_param,
111 : RequestExt, RouterBuilder,
112 : },
113 : id::{TenantId, TimelineId},
114 : lsn::Lsn,
115 : };
116 :
117 : // For APIs that require an Active tenant, how long should we block waiting for that state?
118 : // This is not functionally necessary (clients will retry), but avoids generating a lot of
119 : // failed API calls while tenants are activating.
120 : #[cfg(not(feature = "testing"))]
121 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(5000);
122 :
123 : // Tests run on slow/oversubscribed nodes, and may need to wait much longer for tenants to
124 : // finish attaching, if calls to remote storage are slow.
125 : #[cfg(feature = "testing")]
126 : pub(crate) const ACTIVE_TENANT_TIMEOUT: Duration = Duration::from_millis(30000);
127 :
128 : pub struct State {
129 : conf: &'static PageServerConf,
130 : tenant_manager: Arc<TenantManager>,
131 : auth: Option<Arc<SwappableJwtAuth>>,
132 : allowlist_routes: &'static [&'static str],
133 : remote_storage: GenericRemoteStorage,
134 : broker_client: storage_broker::BrokerClientChannel,
135 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
136 : deletion_queue_client: DeletionQueueClient,
137 : secondary_controller: SecondaryController,
138 : latest_utilization: tokio::sync::Mutex<Option<(std::time::Instant, bytes::Bytes)>>,
139 : }
140 :
141 : impl State {
142 : #[allow(clippy::too_many_arguments)]
143 0 : pub fn new(
144 0 : conf: &'static PageServerConf,
145 0 : tenant_manager: Arc<TenantManager>,
146 0 : auth: Option<Arc<SwappableJwtAuth>>,
147 0 : remote_storage: GenericRemoteStorage,
148 0 : broker_client: storage_broker::BrokerClientChannel,
149 0 : disk_usage_eviction_state: Arc<disk_usage_eviction_task::State>,
150 0 : deletion_queue_client: DeletionQueueClient,
151 0 : secondary_controller: SecondaryController,
152 0 : ) -> anyhow::Result<Self> {
153 0 : let allowlist_routes = &[
154 0 : "/v1/status",
155 0 : "/v1/doc",
156 0 : "/swagger.yml",
157 0 : "/metrics",
158 0 : "/profile/cpu",
159 0 : "/profile/heap",
160 0 : ];
161 0 : Ok(Self {
162 0 : conf,
163 0 : tenant_manager,
164 0 : auth,
165 0 : allowlist_routes,
166 0 : remote_storage,
167 0 : broker_client,
168 0 : disk_usage_eviction_state,
169 0 : deletion_queue_client,
170 0 : secondary_controller,
171 0 : latest_utilization: Default::default(),
172 0 : })
173 0 : }
174 : }
175 :
176 : #[inline(always)]
177 0 : fn get_state(request: &Request<Body>) -> &State {
178 0 : request
179 0 : .data::<Arc<State>>()
180 0 : .expect("unknown state type")
181 0 : .as_ref()
182 0 : }
183 :
184 : #[inline(always)]
185 0 : fn get_config(request: &Request<Body>) -> &'static PageServerConf {
186 0 : get_state(request).conf
187 0 : }
188 :
189 : /// Check that the requester is authorized to operate on given tenant
190 0 : fn check_permission(request: &Request<Body>, tenant_id: Option<TenantId>) -> Result<(), ApiError> {
191 0 : check_permission_with(request, |claims| {
192 0 : crate::auth::check_permission(claims, tenant_id)
193 0 : })
194 0 : }
195 :
196 : impl From<PageReconstructError> for ApiError {
197 0 : fn from(pre: PageReconstructError) -> ApiError {
198 0 : match pre {
199 0 : PageReconstructError::Other(other) => ApiError::InternalServerError(other),
200 0 : PageReconstructError::MissingKey(e) => ApiError::InternalServerError(e.into()),
201 0 : PageReconstructError::Cancelled => ApiError::Cancelled,
202 0 : PageReconstructError::AncestorLsnTimeout(e) => ApiError::Timeout(format!("{e}").into()),
203 0 : PageReconstructError::WalRedo(pre) => ApiError::InternalServerError(pre),
204 : }
205 0 : }
206 : }
207 :
208 : impl From<TenantMapInsertError> for ApiError {
209 0 : fn from(tmie: TenantMapInsertError) -> ApiError {
210 0 : match tmie {
211 0 : TenantMapInsertError::SlotError(e) => e.into(),
212 0 : TenantMapInsertError::SlotUpsertError(e) => e.into(),
213 0 : TenantMapInsertError::Other(e) => ApiError::InternalServerError(e),
214 : }
215 0 : }
216 : }
217 :
218 : impl From<TenantSlotError> for ApiError {
219 0 : fn from(e: TenantSlotError) -> ApiError {
220 : use TenantSlotError::*;
221 0 : match e {
222 0 : NotFound(tenant_id) => {
223 0 : ApiError::NotFound(anyhow::anyhow!("NotFound: tenant {tenant_id}").into())
224 : }
225 : InProgress => {
226 0 : ApiError::ResourceUnavailable("Tenant is being modified concurrently".into())
227 : }
228 0 : MapState(e) => e.into(),
229 : }
230 0 : }
231 : }
232 :
233 : impl From<TenantSlotUpsertError> for ApiError {
234 0 : fn from(e: TenantSlotUpsertError) -> ApiError {
235 : use TenantSlotUpsertError::*;
236 0 : match e {
237 0 : InternalError(e) => ApiError::InternalServerError(anyhow::anyhow!("{e}")),
238 0 : MapState(e) => e.into(),
239 0 : ShuttingDown(_) => ApiError::ShuttingDown,
240 : }
241 0 : }
242 : }
243 :
244 : impl From<UpsertLocationError> for ApiError {
245 0 : fn from(e: UpsertLocationError) -> ApiError {
246 : use UpsertLocationError::*;
247 0 : match e {
248 0 : BadRequest(e) => ApiError::BadRequest(e),
249 0 : Unavailable(_) => ApiError::ShuttingDown,
250 0 : e @ InProgress => ApiError::Conflict(format!("{e}")),
251 0 : Flush(e) | InternalError(e) => ApiError::InternalServerError(e),
252 : }
253 0 : }
254 : }
255 :
256 : impl From<TenantMapError> for ApiError {
257 0 : fn from(e: TenantMapError) -> ApiError {
258 : use TenantMapError::*;
259 0 : match e {
260 : StillInitializing | ShuttingDown => {
261 0 : ApiError::ResourceUnavailable(format!("{e}").into())
262 0 : }
263 0 : }
264 0 : }
265 : }
266 :
267 : impl From<TenantStateError> for ApiError {
268 0 : fn from(tse: TenantStateError) -> ApiError {
269 0 : match tse {
270 : TenantStateError::IsStopping(_) => {
271 0 : ApiError::ResourceUnavailable("Tenant is stopping".into())
272 : }
273 0 : TenantStateError::SlotError(e) => e.into(),
274 0 : TenantStateError::SlotUpsertError(e) => e.into(),
275 0 : TenantStateError::Other(e) => ApiError::InternalServerError(anyhow!(e)),
276 : }
277 0 : }
278 : }
279 :
280 : impl From<GetTenantError> for ApiError {
281 0 : fn from(tse: GetTenantError) -> ApiError {
282 0 : match tse {
283 0 : GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {tid}").into()),
284 0 : GetTenantError::ShardNotFound(tid) => {
285 0 : ApiError::NotFound(anyhow!("tenant {tid}").into())
286 : }
287 : GetTenantError::NotActive(_) => {
288 : // Why is this not `ApiError::NotFound`?
289 : // Because we must be careful to never return 404 for a tenant if it does
290 : // in fact exist locally. If we did, the caller could draw the conclusion
291 : // that it can attach the tenant to another PS and we'd be in split-brain.
292 0 : ApiError::ResourceUnavailable("Tenant not yet active".into())
293 : }
294 0 : GetTenantError::MapState(e) => ApiError::ResourceUnavailable(format!("{e}").into()),
295 : }
296 0 : }
297 : }
298 :
299 : impl From<GetTimelineError> for ApiError {
300 0 : fn from(gte: GetTimelineError) -> Self {
301 0 : // Rationale: tenant is activated only after eligble timelines activate
302 0 : ApiError::NotFound(gte.into())
303 0 : }
304 : }
305 :
306 : impl From<GetActiveTenantError> for ApiError {
307 0 : fn from(e: GetActiveTenantError) -> ApiError {
308 0 : match e {
309 0 : GetActiveTenantError::Broken(reason) => {
310 0 : ApiError::InternalServerError(anyhow!("tenant is broken: {}", reason))
311 : }
312 : GetActiveTenantError::WillNotBecomeActive(TenantState::Stopping { .. }) => {
313 0 : ApiError::ShuttingDown
314 : }
315 0 : GetActiveTenantError::WillNotBecomeActive(_) => ApiError::Conflict(format!("{}", e)),
316 0 : GetActiveTenantError::Cancelled => ApiError::ShuttingDown,
317 0 : GetActiveTenantError::NotFound(gte) => gte.into(),
318 : GetActiveTenantError::WaitForActiveTimeout { .. } => {
319 0 : ApiError::ResourceUnavailable(format!("{}", e).into())
320 : }
321 : GetActiveTenantError::SwitchedTenant => {
322 : // in our HTTP handlers, this error doesn't happen
323 : // TODO: separate error types
324 0 : ApiError::ResourceUnavailable("switched tenant".into())
325 : }
326 : }
327 0 : }
328 : }
329 :
330 : impl From<crate::tenant::DeleteTimelineError> for ApiError {
331 0 : fn from(value: crate::tenant::DeleteTimelineError) -> Self {
332 : use crate::tenant::DeleteTimelineError::*;
333 0 : match value {
334 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
335 0 : HasChildren(children) => ApiError::PreconditionFailed(
336 0 : format!("Cannot delete timeline which has child timelines: {children:?}")
337 0 : .into_boxed_str(),
338 0 : ),
339 0 : a @ AlreadyInProgress(_) => ApiError::Conflict(a.to_string()),
340 0 : Cancelled => ApiError::ResourceUnavailable("shutting down".into()),
341 0 : Other(e) => ApiError::InternalServerError(e),
342 : }
343 0 : }
344 : }
345 :
346 : impl From<crate::tenant::TimelineArchivalError> for ApiError {
347 0 : fn from(value: crate::tenant::TimelineArchivalError) -> Self {
348 : use crate::tenant::TimelineArchivalError::*;
349 0 : match value {
350 0 : NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
351 0 : Timeout => ApiError::Timeout("hit pageserver internal timeout".into()),
352 0 : Cancelled => ApiError::ShuttingDown,
353 0 : e @ HasArchivedParent(_) => {
354 0 : ApiError::PreconditionFailed(e.to_string().into_boxed_str())
355 : }
356 0 : HasUnarchivedChildren(children) => ApiError::PreconditionFailed(
357 0 : format!(
358 0 : "Cannot archive timeline which has non-archived child timelines: {children:?}"
359 0 : )
360 0 : .into_boxed_str(),
361 0 : ),
362 0 : a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
363 0 : Other(e) => ApiError::InternalServerError(e),
364 : }
365 0 : }
366 : }
367 :
368 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
369 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
370 : use crate::tenant::mgr::DeleteTimelineError::*;
371 0 : match value {
372 : // Report Precondition failed so client can distinguish between
373 : // "tenant is missing" case from "timeline is missing"
374 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
375 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
376 0 : ),
377 0 : Tenant(t) => ApiError::from(t),
378 0 : Timeline(t) => ApiError::from(t),
379 : }
380 0 : }
381 : }
382 :
383 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
384 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
385 : use crate::tenant::mgr::DeleteTenantError::*;
386 0 : match value {
387 0 : SlotError(e) => e.into(),
388 0 : Other(o) => ApiError::InternalServerError(o),
389 0 : Cancelled => ApiError::ShuttingDown,
390 : }
391 0 : }
392 : }
393 :
394 : impl From<crate::tenant::secondary::SecondaryTenantError> for ApiError {
395 0 : fn from(ste: crate::tenant::secondary::SecondaryTenantError) -> ApiError {
396 : use crate::tenant::secondary::SecondaryTenantError;
397 0 : match ste {
398 0 : SecondaryTenantError::GetTenant(gte) => gte.into(),
399 0 : SecondaryTenantError::ShuttingDown => ApiError::ShuttingDown,
400 : }
401 0 : }
402 : }
403 :
404 : // Helper function to construct a TimelineInfo struct for a timeline
405 0 : async fn build_timeline_info(
406 0 : timeline: &Arc<Timeline>,
407 0 : include_non_incremental_logical_size: bool,
408 0 : force_await_initial_logical_size: bool,
409 0 : ctx: &RequestContext,
410 0 : ) -> anyhow::Result<TimelineInfo> {
411 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
412 0 :
413 0 : if force_await_initial_logical_size {
414 0 : timeline.clone().await_initial_logical_size().await
415 0 : }
416 :
417 0 : let mut info = build_timeline_info_common(
418 0 : timeline,
419 0 : ctx,
420 0 : tenant::timeline::GetLogicalSizePriority::Background,
421 0 : )
422 0 : .await?;
423 0 : if include_non_incremental_logical_size {
424 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
425 : // Otherwise, if someone deletes the timeline / detaches the tenant while
426 : // we're executing this function, we will outlive the timeline on-disk state.
427 : info.current_logical_size_non_incremental = Some(
428 0 : timeline
429 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
430 0 : .await?,
431 : );
432 0 : }
433 0 : Ok(info)
434 0 : }
435 :
436 0 : async fn build_timeline_info_common(
437 0 : timeline: &Arc<Timeline>,
438 0 : ctx: &RequestContext,
439 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
440 0 : ) -> anyhow::Result<TimelineInfo> {
441 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
442 0 : let initdb_lsn = timeline.initdb_lsn;
443 0 : let last_record_lsn = timeline.get_last_record_lsn();
444 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
445 0 : let guard = timeline.last_received_wal.lock().unwrap();
446 0 : if let Some(info) = guard.as_ref() {
447 0 : (
448 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
449 0 : Some(info.last_received_msg_lsn),
450 0 : Some(info.last_received_msg_ts),
451 0 : )
452 : } else {
453 0 : (None, None, None)
454 : }
455 : };
456 :
457 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
458 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
459 0 : Lsn(0) => None,
460 0 : lsn @ Lsn(_) => Some(lsn),
461 : };
462 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
463 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
464 0 : let state = timeline.current_state();
465 0 : // Report is_archived = false if the timeline is still loading
466 0 : let is_archived = timeline.is_archived().unwrap_or(false);
467 0 : let remote_consistent_lsn_projected = timeline
468 0 : .get_remote_consistent_lsn_projected()
469 0 : .unwrap_or(Lsn(0));
470 0 : let remote_consistent_lsn_visible = timeline
471 0 : .get_remote_consistent_lsn_visible()
472 0 : .unwrap_or(Lsn(0));
473 0 :
474 0 : let walreceiver_status = timeline.walreceiver_status();
475 0 :
476 0 : let (pitr_history_size, within_ancestor_pitr) = timeline.get_pitr_history_stats();
477 :
478 0 : let info = TimelineInfo {
479 0 : tenant_id: timeline.tenant_shard_id,
480 0 : timeline_id: timeline.timeline_id,
481 0 : ancestor_timeline_id,
482 0 : ancestor_lsn,
483 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
484 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
485 0 : remote_consistent_lsn_visible,
486 0 : initdb_lsn,
487 0 : last_record_lsn,
488 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
489 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
490 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
491 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
492 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
493 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
494 : },
495 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
496 0 : current_physical_size,
497 0 : current_logical_size_non_incremental: None,
498 0 : pitr_history_size,
499 0 : within_ancestor_pitr,
500 0 : timeline_dir_layer_file_size_sum: None,
501 0 : wal_source_connstr,
502 0 : last_received_msg_lsn,
503 0 : last_received_msg_ts,
504 0 : pg_version: timeline.pg_version,
505 0 :
506 0 : state,
507 0 : is_archived: Some(is_archived),
508 0 :
509 0 : walreceiver_status,
510 0 : };
511 0 : Ok(info)
512 0 : }
513 :
514 0 : fn build_timeline_offloaded_info(offloaded: &Arc<OffloadedTimeline>) -> OffloadedTimelineInfo {
515 0 : 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 0 : ..
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 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 0 : StatusCode::INTERNAL_SERVER_ERROR,
566 0 : HttpErrorBody::from_msg(err_msg.to_string()),
567 0 : )
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 0 : pg_version: _,
595 0 : } => tenant::CreateTimelineParams::Branch(tenant::CreateTimelineParamsBranch {
596 0 : new_timeline_id,
597 0 : ancestor_timeline_id,
598 0 : ancestor_start_lsn,
599 0 : }),
600 : TimelineCreateRequestMode::ImportPgdata {
601 : import_pgdata:
602 : TimelineCreateRequestModeImportPgdata {
603 0 : location,
604 0 : idempotency_key,
605 0 : },
606 0 : } => tenant::CreateTimelineParams::ImportPgdata(tenant::CreateTimelineParamsImportPgdata {
607 0 : idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new(
608 0 : idempotency_key.0,
609 0 : ),
610 0 : new_timeline_id,
611 : location: {
612 0 : use import_pgdata::index_part_format::Location;
613 0 : use pageserver_api::models::ImportPgdataLocation;
614 0 : match location {
615 : #[cfg(feature = "testing")]
616 0 : ImportPgdataLocation::LocalFs { path } => Location::LocalFs { path },
617 : ImportPgdataLocation::AwsS3 {
618 0 : region,
619 0 : bucket,
620 0 : key,
621 0 : } => Location::AwsS3 {
622 0 : region,
623 0 : bucket,
624 0 : key,
625 0 : },
626 : }
627 : },
628 : }),
629 : };
630 :
631 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
632 0 :
633 0 : let state = get_state(&request);
634 :
635 0 : async {
636 0 : let tenant = state
637 0 : .tenant_manager
638 0 : .get_attached_tenant_shard(tenant_shard_id)?;
639 :
640 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
641 :
642 : // earlier versions of the code had pg_version and ancestor_lsn in the span
643 : // => continue to provide that information, but, through a log message that doesn't require us to destructure
644 0 : tracing::info!(?params, "creating timeline");
645 :
646 0 : match tenant
647 0 : .create_timeline(params, state.broker_client.clone(), &ctx)
648 0 : .await
649 : {
650 0 : Ok(new_timeline) => {
651 : // Created. Construct a TimelineInfo for it.
652 0 : let timeline_info = build_timeline_info_common(
653 0 : &new_timeline,
654 0 : &ctx,
655 0 : tenant::timeline::GetLogicalSizePriority::User,
656 0 : )
657 0 : .await
658 0 : .map_err(ApiError::InternalServerError)?;
659 0 : json_response(StatusCode::CREATED, timeline_info)
660 : }
661 0 : Err(_) if tenant.cancel.is_cancelled() => {
662 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
663 0 : json_response(
664 0 : StatusCode::SERVICE_UNAVAILABLE,
665 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
666 0 : )
667 : }
668 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
669 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
670 : }
671 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
672 0 : StatusCode::TOO_MANY_REQUESTS,
673 0 : HttpErrorBody::from_msg(e.to_string()),
674 0 : ),
675 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
676 0 : StatusCode::NOT_ACCEPTABLE,
677 0 : HttpErrorBody::from_msg(format!("{err:#}")),
678 0 : ),
679 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
680 0 : StatusCode::SERVICE_UNAVAILABLE,
681 0 : HttpErrorBody::from_msg(e.to_string()),
682 0 : ),
683 0 : Err(e @ tenant::CreateTimelineError::AncestorArchived) => json_response(
684 0 : StatusCode::NOT_ACCEPTABLE,
685 0 : HttpErrorBody::from_msg(e.to_string()),
686 0 : ),
687 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
688 0 : StatusCode::SERVICE_UNAVAILABLE,
689 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
690 0 : ),
691 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
692 : }
693 0 : }
694 0 : .instrument(info_span!("timeline_create",
695 : tenant_id = %tenant_shard_id.tenant_id,
696 0 : shard_id = %tenant_shard_id.shard_slug(),
697 : timeline_id = %new_timeline_id,
698 : ))
699 0 : .await
700 0 : }
701 :
702 0 : async fn timeline_list_handler(
703 0 : request: Request<Body>,
704 0 : _cancel: CancellationToken,
705 0 : ) -> Result<Response<Body>, ApiError> {
706 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
707 0 : let include_non_incremental_logical_size: Option<bool> =
708 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
709 0 : let force_await_initial_logical_size: Option<bool> =
710 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
711 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
712 :
713 0 : let state = get_state(&request);
714 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
715 :
716 0 : let response_data = async {
717 0 : let tenant = state
718 0 : .tenant_manager
719 0 : .get_attached_tenant_shard(tenant_shard_id)?;
720 :
721 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
722 :
723 0 : let timelines = tenant.list_timelines();
724 0 :
725 0 : let mut response_data = Vec::with_capacity(timelines.len());
726 0 : for timeline in timelines {
727 0 : let timeline_info = build_timeline_info(
728 0 : &timeline,
729 0 : include_non_incremental_logical_size.unwrap_or(false),
730 0 : force_await_initial_logical_size.unwrap_or(false),
731 0 : &ctx,
732 0 : )
733 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
734 0 : .await
735 0 : .context("Failed to build timeline info")
736 0 : .map_err(ApiError::InternalServerError)?;
737 :
738 0 : response_data.push(timeline_info);
739 : }
740 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
741 0 : }
742 0 : .instrument(info_span!("timeline_list",
743 : tenant_id = %tenant_shard_id.tenant_id,
744 0 : shard_id = %tenant_shard_id.shard_slug()))
745 0 : .await?;
746 :
747 0 : json_response(StatusCode::OK, response_data)
748 0 : }
749 :
750 0 : async fn timeline_and_offloaded_list_handler(
751 0 : request: Request<Body>,
752 0 : _cancel: CancellationToken,
753 0 : ) -> Result<Response<Body>, ApiError> {
754 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
755 0 : let include_non_incremental_logical_size: Option<bool> =
756 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
757 0 : let force_await_initial_logical_size: Option<bool> =
758 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
759 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
760 :
761 0 : let state = get_state(&request);
762 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
763 :
764 0 : let response_data = async {
765 0 : let tenant = state
766 0 : .tenant_manager
767 0 : .get_attached_tenant_shard(tenant_shard_id)?;
768 :
769 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
770 :
771 0 : let (timelines, offloadeds) = tenant.list_timelines_and_offloaded();
772 0 :
773 0 : let mut timeline_infos = Vec::with_capacity(timelines.len());
774 0 : for timeline in timelines {
775 0 : let timeline_info = build_timeline_info(
776 0 : &timeline,
777 0 : include_non_incremental_logical_size.unwrap_or(false),
778 0 : force_await_initial_logical_size.unwrap_or(false),
779 0 : &ctx,
780 0 : )
781 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
782 0 : .await
783 0 : .context("Failed to build timeline info")
784 0 : .map_err(ApiError::InternalServerError)?;
785 :
786 0 : timeline_infos.push(timeline_info);
787 : }
788 0 : let offloaded_infos = offloadeds
789 0 : .into_iter()
790 0 : .map(|offloaded| build_timeline_offloaded_info(&offloaded))
791 0 : .collect::<Vec<_>>();
792 0 : let res = TimelinesInfoAndOffloaded {
793 0 : timelines: timeline_infos,
794 0 : offloaded: offloaded_infos,
795 0 : };
796 0 : Ok::<TimelinesInfoAndOffloaded, ApiError>(res)
797 0 : }
798 0 : .instrument(info_span!("timeline_and_offloaded_list",
799 : tenant_id = %tenant_shard_id.tenant_id,
800 0 : shard_id = %tenant_shard_id.shard_slug()))
801 0 : .await?;
802 :
803 0 : json_response(StatusCode::OK, response_data)
804 0 : }
805 :
806 0 : async fn timeline_preserve_initdb_handler(
807 0 : request: Request<Body>,
808 0 : _cancel: CancellationToken,
809 0 : ) -> Result<Response<Body>, ApiError> {
810 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
811 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
812 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
813 0 : let state = get_state(&request);
814 :
815 : // Part of the process for disaster recovery from safekeeper-stored WAL:
816 : // If we don't recover into a new timeline but want to keep the timeline ID,
817 : // then the initdb archive is deleted. This endpoint copies it to a different
818 : // location where timeline recreation cand find it.
819 :
820 0 : async {
821 0 : let tenant = state
822 0 : .tenant_manager
823 0 : .get_attached_tenant_shard(tenant_shard_id)?;
824 :
825 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
826 :
827 0 : timeline
828 0 : .preserve_initdb_archive()
829 0 : .await
830 0 : .context("preserving initdb archive")
831 0 : .map_err(ApiError::InternalServerError)?;
832 :
833 0 : Ok::<_, ApiError>(())
834 0 : }
835 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
836 : tenant_id = %tenant_shard_id.tenant_id,
837 0 : shard_id = %tenant_shard_id.shard_slug(),
838 : %timeline_id))
839 0 : .await?;
840 :
841 0 : json_response(StatusCode::OK, ())
842 0 : }
843 :
844 0 : async fn timeline_archival_config_handler(
845 0 : mut request: Request<Body>,
846 0 : _cancel: CancellationToken,
847 0 : ) -> Result<Response<Body>, ApiError> {
848 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
849 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
850 :
851 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
852 :
853 0 : let request_data: TimelineArchivalConfigRequest = json_request(&mut request).await?;
854 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
855 0 : let state = get_state(&request);
856 :
857 0 : async {
858 0 : let tenant = state
859 0 : .tenant_manager
860 0 : .get_attached_tenant_shard(tenant_shard_id)?;
861 :
862 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
863 :
864 0 : tenant
865 0 : .apply_timeline_archival_config(
866 0 : timeline_id,
867 0 : request_data.state,
868 0 : state.broker_client.clone(),
869 0 : ctx,
870 0 : )
871 0 : .await?;
872 0 : Ok::<_, ApiError>(())
873 0 : }
874 0 : .instrument(info_span!("timeline_archival_config",
875 : tenant_id = %tenant_shard_id.tenant_id,
876 0 : shard_id = %tenant_shard_id.shard_slug(),
877 : state = ?request_data.state,
878 : %timeline_id))
879 0 : .await?;
880 :
881 0 : json_response(StatusCode::OK, ())
882 0 : }
883 :
884 0 : async fn timeline_detail_handler(
885 0 : request: Request<Body>,
886 0 : _cancel: CancellationToken,
887 0 : ) -> Result<Response<Body>, ApiError> {
888 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
889 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
890 0 : let include_non_incremental_logical_size: Option<bool> =
891 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
892 0 : let force_await_initial_logical_size: Option<bool> =
893 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
894 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
895 :
896 : // Logical size calculation needs downloading.
897 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
898 0 : let state = get_state(&request);
899 :
900 0 : let timeline_info = async {
901 0 : let tenant = state
902 0 : .tenant_manager
903 0 : .get_attached_tenant_shard(tenant_shard_id)?;
904 :
905 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
906 :
907 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
908 :
909 0 : let timeline_info = build_timeline_info(
910 0 : &timeline,
911 0 : include_non_incremental_logical_size.unwrap_or(false),
912 0 : force_await_initial_logical_size.unwrap_or(false),
913 0 : &ctx,
914 0 : )
915 0 : .await
916 0 : .context("get local timeline info")
917 0 : .map_err(ApiError::InternalServerError)?;
918 :
919 0 : Ok::<_, ApiError>(timeline_info)
920 0 : }
921 0 : .instrument(info_span!("timeline_detail",
922 : tenant_id = %tenant_shard_id.tenant_id,
923 0 : shard_id = %tenant_shard_id.shard_slug(),
924 : %timeline_id))
925 0 : .await?;
926 :
927 0 : json_response(StatusCode::OK, timeline_info)
928 0 : }
929 :
930 0 : async fn get_lsn_by_timestamp_handler(
931 0 : request: Request<Body>,
932 0 : cancel: CancellationToken,
933 0 : ) -> Result<Response<Body>, ApiError> {
934 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
935 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
936 0 : let state = get_state(&request);
937 0 :
938 0 : if !tenant_shard_id.is_shard_zero() {
939 : // Requires SLRU contents, which are only stored on shard zero
940 0 : return Err(ApiError::BadRequest(anyhow!(
941 0 : "Size calculations are only available on shard zero"
942 0 : )));
943 0 : }
944 :
945 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
946 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
947 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
948 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
949 0 : .map_err(ApiError::BadRequest)?;
950 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
951 :
952 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
953 0 :
954 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
955 :
956 0 : let timeline =
957 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
958 0 : .await?;
959 0 : let result = timeline
960 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
961 0 : .await?;
962 :
963 : #[derive(serde::Serialize, Debug)]
964 : struct Result {
965 : lsn: Lsn,
966 : kind: &'static str,
967 : #[serde(default)]
968 : #[serde(skip_serializing_if = "Option::is_none")]
969 : #[serde(flatten)]
970 : lease: Option<LsnLease>,
971 : }
972 0 : let (lsn, kind) = match result {
973 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
974 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
975 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
976 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
977 : };
978 :
979 0 : let lease = if with_lease {
980 0 : timeline
981 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
982 0 : .inspect_err(|_| {
983 0 : warn!("fail to grant a lease to {}", lsn);
984 0 : })
985 0 : .ok()
986 : } else {
987 0 : None
988 : };
989 :
990 0 : let result = Result { lsn, kind, lease };
991 0 : let valid_until = result
992 0 : .lease
993 0 : .as_ref()
994 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
995 0 : tracing::info!(
996 : lsn=?result.lsn,
997 : kind=%result.kind,
998 : timestamp=%timestamp_raw,
999 : valid_until=?valid_until,
1000 0 : "lsn_by_timestamp finished"
1001 : );
1002 0 : json_response(StatusCode::OK, result)
1003 0 : }
1004 :
1005 0 : async fn get_timestamp_of_lsn_handler(
1006 0 : request: Request<Body>,
1007 0 : _cancel: CancellationToken,
1008 0 : ) -> Result<Response<Body>, ApiError> {
1009 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1010 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1011 0 : let state = get_state(&request);
1012 0 :
1013 0 : if !tenant_shard_id.is_shard_zero() {
1014 : // Requires SLRU contents, which are only stored on shard zero
1015 0 : return Err(ApiError::BadRequest(anyhow!(
1016 0 : "Size calculations are only available on shard zero"
1017 0 : )));
1018 0 : }
1019 :
1020 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1021 :
1022 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
1023 0 : let lsn = Lsn::from_str(&lsn_str)
1024 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
1025 0 : .map_err(ApiError::BadRequest)?;
1026 :
1027 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1028 0 : let timeline =
1029 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1030 0 : .await?;
1031 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
1032 :
1033 0 : match result {
1034 0 : Some(time) => {
1035 0 : let time = format_rfc3339(
1036 0 : postgres_ffi::try_from_pg_timestamp(time).map_err(ApiError::InternalServerError)?,
1037 : )
1038 0 : .to_string();
1039 0 : json_response(StatusCode::OK, time)
1040 : }
1041 0 : None => Err(ApiError::NotFound(
1042 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
1043 0 : )),
1044 : }
1045 0 : }
1046 :
1047 0 : async fn timeline_delete_handler(
1048 0 : request: Request<Body>,
1049 0 : _cancel: CancellationToken,
1050 0 : ) -> Result<Response<Body>, ApiError> {
1051 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1052 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1053 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1054 :
1055 0 : let state = get_state(&request);
1056 :
1057 0 : let tenant = state
1058 0 : .tenant_manager
1059 0 : .get_attached_tenant_shard(tenant_shard_id)
1060 0 : .map_err(|e| {
1061 0 : match e {
1062 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
1063 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
1064 : GetTenantError::NotFound(_) | GetTenantError::ShardNotFound(_) => {
1065 0 : ApiError::PreconditionFailed(
1066 0 : "Requested tenant is missing".to_string().into_boxed_str(),
1067 0 : )
1068 : }
1069 0 : e => e.into(),
1070 : }
1071 0 : })?;
1072 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1073 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))
1074 0 : .await?;
1075 :
1076 0 : json_response(StatusCode::ACCEPTED, ())
1077 0 : }
1078 :
1079 0 : async fn tenant_reset_handler(
1080 0 : request: Request<Body>,
1081 0 : _cancel: CancellationToken,
1082 0 : ) -> Result<Response<Body>, ApiError> {
1083 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1084 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1085 :
1086 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
1087 :
1088 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1089 0 : let state = get_state(&request);
1090 0 : state
1091 0 : .tenant_manager
1092 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
1093 0 : .await
1094 0 : .map_err(ApiError::InternalServerError)?;
1095 :
1096 0 : json_response(StatusCode::OK, ())
1097 0 : }
1098 :
1099 0 : async fn tenant_list_handler(
1100 0 : request: Request<Body>,
1101 0 : _cancel: CancellationToken,
1102 0 : ) -> Result<Response<Body>, ApiError> {
1103 0 : check_permission(&request, None)?;
1104 0 : let state = get_state(&request);
1105 :
1106 0 : let response_data = state
1107 0 : .tenant_manager
1108 0 : .list_tenants()
1109 0 : .map_err(|_| {
1110 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
1111 0 : })?
1112 0 : .iter()
1113 0 : .map(|(id, state, gen)| TenantInfo {
1114 0 : id: *id,
1115 0 : state: state.clone(),
1116 0 : current_physical_size: None,
1117 0 : attachment_status: state.attachment_status(),
1118 0 : generation: (*gen)
1119 0 : .into()
1120 0 : .expect("Tenants are always attached with a generation"),
1121 0 : gc_blocking: None,
1122 0 : })
1123 0 : .collect::<Vec<TenantInfo>>();
1124 0 :
1125 0 : json_response(StatusCode::OK, response_data)
1126 0 : }
1127 :
1128 0 : async fn tenant_status(
1129 0 : request: Request<Body>,
1130 0 : _cancel: CancellationToken,
1131 0 : ) -> Result<Response<Body>, ApiError> {
1132 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1133 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1134 0 : let state = get_state(&request);
1135 0 :
1136 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
1137 0 : let activate = true;
1138 : #[cfg(feature = "testing")]
1139 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
1140 :
1141 0 : let tenant_info = async {
1142 0 : let tenant = state
1143 0 : .tenant_manager
1144 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1145 :
1146 0 : if activate {
1147 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
1148 : // called, but it is still valid to return 200 and describe the current state of the tenant
1149 : // if it doesn't make it into an active state.
1150 0 : tenant
1151 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
1152 0 : .await
1153 0 : .ok();
1154 0 : }
1155 :
1156 : // Calculate total physical size of all timelines
1157 0 : let mut current_physical_size = 0;
1158 0 : for timeline in tenant.list_timelines().iter() {
1159 0 : current_physical_size += timeline.layer_size_sum().await;
1160 : }
1161 :
1162 0 : let state = tenant.current_state();
1163 0 : Result::<_, ApiError>::Ok(TenantDetails {
1164 0 : tenant_info: TenantInfo {
1165 0 : id: tenant_shard_id,
1166 0 : state: state.clone(),
1167 0 : current_physical_size: Some(current_physical_size),
1168 0 : attachment_status: state.attachment_status(),
1169 0 : generation: tenant
1170 0 : .generation()
1171 0 : .into()
1172 0 : .expect("Tenants are always attached with a generation"),
1173 0 : gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
1174 0 : },
1175 0 : walredo: tenant.wal_redo_manager_status(),
1176 0 : timelines: tenant.list_timeline_ids(),
1177 0 : })
1178 0 : }
1179 0 : .instrument(info_span!("tenant_status_handler",
1180 : tenant_id = %tenant_shard_id.tenant_id,
1181 0 : shard_id = %tenant_shard_id.shard_slug()))
1182 0 : .await?;
1183 :
1184 0 : json_response(StatusCode::OK, tenant_info)
1185 0 : }
1186 :
1187 0 : async fn tenant_delete_handler(
1188 0 : request: Request<Body>,
1189 0 : _cancel: CancellationToken,
1190 0 : ) -> Result<Response<Body>, ApiError> {
1191 : // TODO openapi spec
1192 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1193 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1194 :
1195 0 : let state = get_state(&request);
1196 0 :
1197 0 : state
1198 0 : .tenant_manager
1199 0 : .delete_tenant(tenant_shard_id)
1200 0 : .instrument(info_span!("tenant_delete_handler",
1201 : tenant_id = %tenant_shard_id.tenant_id,
1202 0 : shard_id = %tenant_shard_id.shard_slug()
1203 : ))
1204 0 : .await?;
1205 :
1206 0 : json_response(StatusCode::OK, ())
1207 0 : }
1208 :
1209 : /// HTTP endpoint to query the current tenant_size of a tenant.
1210 : ///
1211 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1212 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1213 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1214 : /// values.
1215 : ///
1216 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1217 : /// (only if it is shorter than the real cutoff).
1218 : ///
1219 : /// Note: we don't update the cached size and prometheus metric here.
1220 : /// The retention period might be different, and it's nice to have a method to just calculate it
1221 : /// without modifying anything anyway.
1222 0 : async fn tenant_size_handler(
1223 0 : request: Request<Body>,
1224 0 : cancel: CancellationToken,
1225 0 : ) -> Result<Response<Body>, ApiError> {
1226 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1227 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1228 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1229 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1230 0 : let headers = request.headers();
1231 0 : let state = get_state(&request);
1232 0 :
1233 0 : if !tenant_shard_id.is_shard_zero() {
1234 0 : return Err(ApiError::BadRequest(anyhow!(
1235 0 : "Size calculations are only available on shard zero"
1236 0 : )));
1237 0 : }
1238 0 :
1239 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1240 0 : let tenant = state
1241 0 : .tenant_manager
1242 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1243 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1244 :
1245 : // this can be long operation
1246 0 : let inputs = tenant
1247 0 : .gather_size_inputs(
1248 0 : retention_period,
1249 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1250 0 : &cancel,
1251 0 : &ctx,
1252 0 : )
1253 0 : .await
1254 0 : .map_err(|e| match e {
1255 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1256 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1257 0 : })?;
1258 :
1259 0 : let mut sizes = None;
1260 0 : let accepts_html = headers
1261 0 : .get(header::ACCEPT)
1262 0 : .map(|v| v == "text/html")
1263 0 : .unwrap_or_default();
1264 0 : if !inputs_only.unwrap_or(false) {
1265 0 : let storage_model = inputs.calculate_model();
1266 0 : let size = storage_model.calculate();
1267 0 :
1268 0 : // If request header expects html, return html
1269 0 : if accepts_html {
1270 0 : return synthetic_size_html_response(inputs, storage_model, size);
1271 0 : }
1272 0 : sizes = Some(size);
1273 0 : } else if accepts_html {
1274 0 : return Err(ApiError::BadRequest(anyhow!(
1275 0 : "inputs_only parameter is incompatible with html output request"
1276 0 : )));
1277 0 : }
1278 :
1279 : /// The type resides in the pageserver not to expose `ModelInputs`.
1280 : #[derive(serde::Serialize)]
1281 : struct TenantHistorySize {
1282 : id: TenantId,
1283 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1284 : ///
1285 : /// Will be none if `?inputs_only=true` was given.
1286 : size: Option<u64>,
1287 : /// Size of each segment used in the model.
1288 : /// Will be null if `?inputs_only=true` was given.
1289 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1290 : inputs: crate::tenant::size::ModelInputs,
1291 : }
1292 :
1293 0 : json_response(
1294 0 : StatusCode::OK,
1295 0 : TenantHistorySize {
1296 0 : id: tenant_shard_id.tenant_id,
1297 0 : size: sizes.as_ref().map(|x| x.total_size),
1298 0 : segment_sizes: sizes.map(|x| x.segments),
1299 0 : inputs,
1300 0 : },
1301 0 : )
1302 0 : }
1303 :
1304 0 : async fn tenant_shard_split_handler(
1305 0 : mut request: Request<Body>,
1306 0 : _cancel: CancellationToken,
1307 0 : ) -> Result<Response<Body>, ApiError> {
1308 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1309 :
1310 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1311 0 : let state = get_state(&request);
1312 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1313 :
1314 0 : let tenant = state
1315 0 : .tenant_manager
1316 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1317 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1318 :
1319 0 : let new_shards = state
1320 0 : .tenant_manager
1321 0 : .shard_split(
1322 0 : tenant,
1323 0 : ShardCount::new(req.new_shard_count),
1324 0 : req.new_stripe_size,
1325 0 : &ctx,
1326 0 : )
1327 0 : .await
1328 0 : .map_err(ApiError::InternalServerError)?;
1329 :
1330 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1331 0 : }
1332 :
1333 0 : async fn layer_map_info_handler(
1334 0 : request: Request<Body>,
1335 0 : _cancel: CancellationToken,
1336 0 : ) -> Result<Response<Body>, ApiError> {
1337 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1338 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1339 0 : let reset: LayerAccessStatsReset =
1340 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1341 0 : let state = get_state(&request);
1342 0 :
1343 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1344 :
1345 0 : let timeline =
1346 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1347 0 : .await?;
1348 0 : let layer_map_info = timeline
1349 0 : .layer_map_info(reset)
1350 0 : .await
1351 0 : .map_err(|_shutdown| ApiError::ShuttingDown)?;
1352 :
1353 0 : json_response(StatusCode::OK, layer_map_info)
1354 0 : }
1355 :
1356 0 : #[instrument(skip_all, fields(tenant_id, shard_id, timeline_id, layer_name))]
1357 : async fn timeline_layer_scan_disposable_keys(
1358 : request: Request<Body>,
1359 : cancel: CancellationToken,
1360 : ) -> Result<Response<Body>, ApiError> {
1361 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1362 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1363 : let layer_name: LayerName = parse_request_param(&request, "layer_name")?;
1364 :
1365 : tracing::Span::current().record(
1366 : "tenant_id",
1367 : tracing::field::display(&tenant_shard_id.tenant_id),
1368 : );
1369 : tracing::Span::current().record(
1370 : "shard_id",
1371 : tracing::field::display(tenant_shard_id.shard_slug()),
1372 : );
1373 : tracing::Span::current().record("timeline_id", tracing::field::display(&timeline_id));
1374 : tracing::Span::current().record("layer_name", tracing::field::display(&layer_name));
1375 :
1376 : let state = get_state(&request);
1377 :
1378 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1379 :
1380 : // technically the timeline need not be active for this scan to complete
1381 : let timeline =
1382 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1383 : .await?;
1384 :
1385 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1386 :
1387 : let guard = timeline.layers.read().await;
1388 : let Some(layer) = guard.try_get_from_key(&layer_name.clone().into()) else {
1389 : return Err(ApiError::NotFound(
1390 : anyhow::anyhow!("Layer {tenant_shard_id}/{timeline_id}/{layer_name} not found").into(),
1391 : ));
1392 : };
1393 :
1394 : let resident_layer = layer
1395 : .download_and_keep_resident()
1396 : .await
1397 0 : .map_err(|err| match err {
1398 : tenant::storage_layer::layer::DownloadError::TimelineShutdown
1399 : | tenant::storage_layer::layer::DownloadError::DownloadCancelled => {
1400 0 : ApiError::ShuttingDown
1401 : }
1402 : tenant::storage_layer::layer::DownloadError::ContextAndConfigReallyDeniesDownloads
1403 : | tenant::storage_layer::layer::DownloadError::DownloadRequired
1404 : | tenant::storage_layer::layer::DownloadError::NotFile(_)
1405 : | tenant::storage_layer::layer::DownloadError::DownloadFailed
1406 : | tenant::storage_layer::layer::DownloadError::PreStatFailed(_) => {
1407 0 : ApiError::InternalServerError(err.into())
1408 : }
1409 : #[cfg(test)]
1410 : tenant::storage_layer::layer::DownloadError::Failpoint(_) => {
1411 0 : ApiError::InternalServerError(err.into())
1412 : }
1413 0 : })?;
1414 :
1415 : let keys = resident_layer
1416 : .load_keys(&ctx)
1417 : .await
1418 : .map_err(ApiError::InternalServerError)?;
1419 :
1420 : let shard_identity = timeline.get_shard_identity();
1421 :
1422 : let mut disposable_count = 0;
1423 : let mut not_disposable_count = 0;
1424 : let cancel = cancel.clone();
1425 : for (i, key) in keys.into_iter().enumerate() {
1426 : if shard_identity.is_key_disposable(&key) {
1427 : disposable_count += 1;
1428 : tracing::debug!(key = %key, key.dbg=?key, "disposable key");
1429 : } else {
1430 : not_disposable_count += 1;
1431 : }
1432 : #[allow(clippy::collapsible_if)]
1433 : if i % 10000 == 0 {
1434 : if cancel.is_cancelled() || timeline.cancel.is_cancelled() || timeline.is_stopping() {
1435 : return Err(ApiError::ShuttingDown);
1436 : }
1437 : }
1438 : }
1439 :
1440 : json_response(
1441 : StatusCode::OK,
1442 : pageserver_api::models::ScanDisposableKeysResponse {
1443 : disposable_count,
1444 : not_disposable_count,
1445 : },
1446 : )
1447 : }
1448 :
1449 0 : async fn layer_download_handler(
1450 0 : request: Request<Body>,
1451 0 : _cancel: CancellationToken,
1452 0 : ) -> Result<Response<Body>, ApiError> {
1453 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1454 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1455 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1456 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1457 0 : let layer_name = LayerName::from_str(layer_file_name)
1458 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1459 0 : let state = get_state(&request);
1460 :
1461 0 : let timeline =
1462 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1463 0 : .await?;
1464 0 : let downloaded = timeline
1465 0 : .download_layer(&layer_name)
1466 0 : .await
1467 0 : .map_err(ApiError::InternalServerError)?;
1468 :
1469 0 : match downloaded {
1470 0 : Some(true) => json_response(StatusCode::OK, ()),
1471 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1472 0 : None => json_response(
1473 0 : StatusCode::BAD_REQUEST,
1474 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1475 0 : ),
1476 : }
1477 0 : }
1478 :
1479 0 : async fn evict_timeline_layer_handler(
1480 0 : request: Request<Body>,
1481 0 : _cancel: CancellationToken,
1482 0 : ) -> Result<Response<Body>, ApiError> {
1483 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1484 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1485 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1486 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1487 0 : let state = get_state(&request);
1488 :
1489 0 : let layer_name = LayerName::from_str(layer_file_name)
1490 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1491 :
1492 0 : let timeline =
1493 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1494 0 : .await?;
1495 0 : let evicted = timeline
1496 0 : .evict_layer(&layer_name)
1497 0 : .await
1498 0 : .map_err(ApiError::InternalServerError)?;
1499 :
1500 0 : match evicted {
1501 0 : Some(true) => json_response(StatusCode::OK, ()),
1502 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1503 0 : None => json_response(
1504 0 : StatusCode::BAD_REQUEST,
1505 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1506 0 : ),
1507 : }
1508 0 : }
1509 :
1510 0 : async fn timeline_gc_blocking_handler(
1511 0 : request: Request<Body>,
1512 0 : _cancel: CancellationToken,
1513 0 : ) -> Result<Response<Body>, ApiError> {
1514 0 : block_or_unblock_gc(request, true).await
1515 0 : }
1516 :
1517 0 : async fn timeline_gc_unblocking_handler(
1518 0 : request: Request<Body>,
1519 0 : _cancel: CancellationToken,
1520 0 : ) -> Result<Response<Body>, ApiError> {
1521 0 : block_or_unblock_gc(request, false).await
1522 0 : }
1523 :
1524 : /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
1525 : ///
1526 : /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
1527 0 : async fn block_or_unblock_gc(
1528 0 : request: Request<Body>,
1529 0 : block: bool,
1530 0 : ) -> Result<Response<Body>, ApiError> {
1531 : use crate::tenant::{
1532 : remote_timeline_client::WaitCompletionError, upload_queue::NotInitialized,
1533 : };
1534 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1535 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1536 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1537 0 : let state = get_state(&request);
1538 :
1539 0 : let tenant = state
1540 0 : .tenant_manager
1541 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1542 :
1543 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1544 :
1545 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1546 :
1547 0 : let fut = async {
1548 0 : if block {
1549 0 : timeline.block_gc(&tenant).await.map(|_| ())
1550 : } else {
1551 0 : timeline.unblock_gc(&tenant).await
1552 : }
1553 0 : };
1554 :
1555 0 : let span = tracing::info_span!(
1556 : "block_or_unblock_gc",
1557 : tenant_id = %tenant_shard_id.tenant_id,
1558 0 : shard_id = %tenant_shard_id.shard_slug(),
1559 : timeline_id = %timeline_id,
1560 : block = block,
1561 : );
1562 :
1563 0 : let res = fut.instrument(span).await;
1564 :
1565 0 : res.map_err(|e| {
1566 0 : if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
1567 0 : ApiError::ShuttingDown
1568 : } else {
1569 0 : ApiError::InternalServerError(e)
1570 : }
1571 0 : })?;
1572 :
1573 0 : json_response(StatusCode::OK, ())
1574 0 : }
1575 :
1576 : /// Get tenant_size SVG graph along with the JSON data.
1577 0 : fn synthetic_size_html_response(
1578 0 : inputs: ModelInputs,
1579 0 : storage_model: StorageModel,
1580 0 : sizes: SizeResult,
1581 0 : ) -> Result<Response<Body>, ApiError> {
1582 0 : let mut timeline_ids: Vec<String> = Vec::new();
1583 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1584 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1585 0 : timeline_map.insert(ti.timeline_id, index);
1586 0 : timeline_ids.push(ti.timeline_id.to_string());
1587 0 : }
1588 0 : let seg_to_branch: Vec<(usize, SvgBranchKind)> = inputs
1589 0 : .segments
1590 0 : .iter()
1591 0 : .map(|seg| {
1592 0 : (
1593 0 : *timeline_map.get(&seg.timeline_id).unwrap(),
1594 0 : seg.kind.into(),
1595 0 : )
1596 0 : })
1597 0 : .collect();
1598 :
1599 0 : let svg =
1600 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1601 0 : .map_err(ApiError::InternalServerError)?;
1602 :
1603 0 : let mut response = String::new();
1604 :
1605 : use std::fmt::Write;
1606 0 : write!(response, "<html>\n<body>\n").unwrap();
1607 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1608 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1609 0 : writeln!(response, "<pre>").unwrap();
1610 0 : writeln!(
1611 0 : response,
1612 0 : "{}",
1613 0 : serde_json::to_string_pretty(&inputs).unwrap()
1614 0 : )
1615 0 : .unwrap();
1616 0 : writeln!(
1617 0 : response,
1618 0 : "{}",
1619 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1620 0 : )
1621 0 : .unwrap();
1622 0 : writeln!(response, "</pre>").unwrap();
1623 0 : write!(response, "</body>\n</html>\n").unwrap();
1624 0 :
1625 0 : html_response(StatusCode::OK, response)
1626 0 : }
1627 :
1628 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1629 0 : let response = Response::builder()
1630 0 : .status(status)
1631 0 : .header(header::CONTENT_TYPE, "text/html")
1632 0 : .body(Body::from(data.as_bytes().to_vec()))
1633 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1634 0 : Ok(response)
1635 0 : }
1636 :
1637 0 : async fn get_tenant_config_handler(
1638 0 : request: Request<Body>,
1639 0 : _cancel: CancellationToken,
1640 0 : ) -> Result<Response<Body>, ApiError> {
1641 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1642 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1643 0 : let state = get_state(&request);
1644 :
1645 0 : let tenant = state
1646 0 : .tenant_manager
1647 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1648 :
1649 0 : let response = HashMap::from([
1650 : (
1651 : "tenant_specific_overrides",
1652 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1653 0 : .context("serializing tenant specific overrides")
1654 0 : .map_err(ApiError::InternalServerError)?,
1655 : ),
1656 : (
1657 0 : "effective_config",
1658 0 : serde_json::to_value(tenant.effective_config())
1659 0 : .context("serializing effective config")
1660 0 : .map_err(ApiError::InternalServerError)?,
1661 : ),
1662 : ]);
1663 :
1664 0 : json_response(StatusCode::OK, response)
1665 0 : }
1666 :
1667 0 : async fn update_tenant_config_handler(
1668 0 : mut request: Request<Body>,
1669 0 : _cancel: CancellationToken,
1670 0 : ) -> Result<Response<Body>, ApiError> {
1671 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1672 0 : let tenant_id = request_data.tenant_id;
1673 0 : check_permission(&request, Some(tenant_id))?;
1674 :
1675 0 : let new_tenant_conf =
1676 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1677 :
1678 0 : let state = get_state(&request);
1679 0 :
1680 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1681 :
1682 0 : let tenant = state
1683 0 : .tenant_manager
1684 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1685 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1686 :
1687 : // This is a legacy API that only operates on attached tenants: the preferred
1688 : // API to use is the location_config/ endpoint, which lets the caller provide
1689 : // the full LocationConf.
1690 0 : let location_conf = LocationConf::attached_single(
1691 0 : new_tenant_conf.clone(),
1692 0 : tenant.get_generation(),
1693 0 : &ShardParameters::default(),
1694 0 : );
1695 0 :
1696 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1697 0 : .await
1698 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1699 :
1700 0 : let _ = tenant
1701 0 : .update_tenant_config(|_crnt| Ok(new_tenant_conf.clone()))
1702 0 : .expect("Closure returns Ok()");
1703 0 :
1704 0 : json_response(StatusCode::OK, ())
1705 0 : }
1706 :
1707 0 : async fn patch_tenant_config_handler(
1708 0 : mut request: Request<Body>,
1709 0 : _cancel: CancellationToken,
1710 0 : ) -> Result<Response<Body>, ApiError> {
1711 0 : let request_data: TenantConfigPatchRequest = json_request(&mut request).await?;
1712 0 : let tenant_id = request_data.tenant_id;
1713 0 : check_permission(&request, Some(tenant_id))?;
1714 :
1715 0 : let state = get_state(&request);
1716 0 :
1717 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1718 :
1719 0 : let tenant = state
1720 0 : .tenant_manager
1721 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1722 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1723 :
1724 0 : let updated = tenant
1725 0 : .update_tenant_config(|crnt| crnt.apply_patch(request_data.config.clone()))
1726 0 : .map_err(ApiError::BadRequest)?;
1727 :
1728 : // This is a legacy API that only operates on attached tenants: the preferred
1729 : // API to use is the location_config/ endpoint, which lets the caller provide
1730 : // the full LocationConf.
1731 0 : let location_conf = LocationConf::attached_single(
1732 0 : updated,
1733 0 : tenant.get_generation(),
1734 0 : &ShardParameters::default(),
1735 0 : );
1736 0 :
1737 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1738 0 : .await
1739 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1740 :
1741 0 : json_response(StatusCode::OK, ())
1742 0 : }
1743 :
1744 0 : async fn put_tenant_location_config_handler(
1745 0 : mut request: Request<Body>,
1746 0 : _cancel: CancellationToken,
1747 0 : ) -> Result<Response<Body>, ApiError> {
1748 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1749 :
1750 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1751 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1752 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1753 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1754 :
1755 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1756 0 : let state = get_state(&request);
1757 0 : let conf = state.conf;
1758 0 :
1759 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1760 0 : // its local disk content and drops it from memory.
1761 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1762 0 : if let Err(e) = state
1763 0 : .tenant_manager
1764 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
1765 0 : .instrument(info_span!("tenant_detach",
1766 : tenant_id = %tenant_shard_id.tenant_id,
1767 0 : shard_id = %tenant_shard_id.shard_slug()
1768 : ))
1769 0 : .await
1770 : {
1771 0 : match e {
1772 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1773 0 : // This API is idempotent: a NotFound on a detach is fine.
1774 0 : }
1775 0 : _ => return Err(e.into()),
1776 : }
1777 0 : }
1778 0 : return json_response(StatusCode::OK, ());
1779 0 : }
1780 :
1781 0 : let location_conf =
1782 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1783 :
1784 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1785 : // similar to at startup ordering.
1786 0 : let spawn_mode = if lazy {
1787 0 : tenant::SpawnMode::Lazy
1788 : } else {
1789 0 : tenant::SpawnMode::Eager
1790 : };
1791 :
1792 0 : let tenant = state
1793 0 : .tenant_manager
1794 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1795 0 : .await?;
1796 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1797 0 : let attached = tenant.is_some();
1798 :
1799 0 : if let Some(_flush_ms) = flush {
1800 0 : match state
1801 0 : .secondary_controller
1802 0 : .upload_tenant(tenant_shard_id)
1803 0 : .await
1804 : {
1805 : Ok(()) => {
1806 0 : tracing::info!("Uploaded heatmap during flush");
1807 : }
1808 0 : Err(e) => {
1809 0 : tracing::warn!("Failed to flush heatmap: {e}");
1810 : }
1811 : }
1812 : } else {
1813 0 : tracing::info!("No flush requested when configuring");
1814 : }
1815 :
1816 : // This API returns a vector of pageservers where the tenant is attached: this is
1817 : // primarily for use in the sharding service. For compatibilty, we also return this
1818 : // when called directly on a pageserver, but the payload is always zero or one shards.
1819 0 : let mut response = TenantLocationConfigResponse {
1820 0 : shards: Vec::new(),
1821 0 : stripe_size: None,
1822 0 : };
1823 0 : if attached {
1824 0 : response.shards.push(TenantShardLocation {
1825 0 : shard_id: tenant_shard_id,
1826 0 : node_id: state.conf.id,
1827 0 : });
1828 0 : if tenant_shard_id.shard_count.count() > 1 {
1829 : // Stripe size should be set if we are attached
1830 0 : debug_assert!(stripe_size.is_some());
1831 0 : response.stripe_size = stripe_size;
1832 0 : }
1833 0 : }
1834 :
1835 0 : json_response(StatusCode::OK, response)
1836 0 : }
1837 :
1838 0 : async fn list_location_config_handler(
1839 0 : request: Request<Body>,
1840 0 : _cancel: CancellationToken,
1841 0 : ) -> Result<Response<Body>, ApiError> {
1842 0 : let state = get_state(&request);
1843 0 : let slots = state.tenant_manager.list();
1844 0 : let result = LocationConfigListResponse {
1845 0 : tenant_shards: slots
1846 0 : .into_iter()
1847 0 : .map(|(tenant_shard_id, slot)| {
1848 0 : let v = match slot {
1849 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1850 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1851 0 : TenantSlot::InProgress(_) => None,
1852 : };
1853 0 : (tenant_shard_id, v)
1854 0 : })
1855 0 : .collect(),
1856 0 : };
1857 0 : json_response(StatusCode::OK, result)
1858 0 : }
1859 :
1860 0 : async fn get_location_config_handler(
1861 0 : request: Request<Body>,
1862 0 : _cancel: CancellationToken,
1863 0 : ) -> Result<Response<Body>, ApiError> {
1864 0 : let state = get_state(&request);
1865 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1866 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1867 :
1868 0 : let Some(slot) = slot else {
1869 0 : return Err(ApiError::NotFound(
1870 0 : anyhow::anyhow!("Tenant shard not found").into(),
1871 0 : ));
1872 : };
1873 :
1874 0 : let result: Option<LocationConfig> = match slot {
1875 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1876 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1877 0 : TenantSlot::InProgress(_) => None,
1878 : };
1879 :
1880 0 : json_response(StatusCode::OK, result)
1881 0 : }
1882 :
1883 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1884 : // (from all pageservers) as it invalidates consistency assumptions.
1885 0 : async fn tenant_time_travel_remote_storage_handler(
1886 0 : request: Request<Body>,
1887 0 : cancel: CancellationToken,
1888 0 : ) -> Result<Response<Body>, ApiError> {
1889 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1890 :
1891 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1892 :
1893 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1894 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1895 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1896 0 : .map_err(ApiError::BadRequest)?;
1897 :
1898 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1899 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1900 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1901 0 : .map_err(ApiError::BadRequest)?;
1902 :
1903 : // This is just a sanity check to fend off naive wrong usages of the API:
1904 : // the tenant needs to be detached *everywhere*
1905 0 : let state = get_state(&request);
1906 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1907 0 : if we_manage_tenant {
1908 0 : return Err(ApiError::BadRequest(anyhow!(
1909 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1910 0 : )));
1911 0 : }
1912 0 :
1913 0 : if timestamp > done_if_after {
1914 0 : return Err(ApiError::BadRequest(anyhow!(
1915 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1916 0 : )));
1917 0 : }
1918 0 :
1919 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1920 :
1921 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1922 0 : &state.remote_storage,
1923 0 : &tenant_shard_id,
1924 0 : timestamp,
1925 0 : done_if_after,
1926 0 : &cancel,
1927 0 : )
1928 0 : .await
1929 0 : .map_err(|e| match e {
1930 0 : TimeTravelError::BadInput(e) => {
1931 0 : warn!("bad input error: {e}");
1932 0 : ApiError::BadRequest(anyhow!("bad input error"))
1933 : }
1934 : TimeTravelError::Unimplemented => {
1935 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1936 : }
1937 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1938 : TimeTravelError::TooManyVersions => {
1939 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1940 : }
1941 0 : TimeTravelError::Other(e) => {
1942 0 : warn!("internal error: {e}");
1943 0 : ApiError::InternalServerError(anyhow!("internal error"))
1944 : }
1945 0 : })?;
1946 :
1947 0 : json_response(StatusCode::OK, ())
1948 0 : }
1949 :
1950 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1951 0 : async fn handle_tenant_break(
1952 0 : r: Request<Body>,
1953 0 : _cancel: CancellationToken,
1954 0 : ) -> Result<Response<Body>, ApiError> {
1955 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1956 :
1957 0 : let state = get_state(&r);
1958 0 : state
1959 0 : .tenant_manager
1960 0 : .get_attached_tenant_shard(tenant_shard_id)?
1961 0 : .set_broken("broken from test".to_owned())
1962 0 : .await;
1963 :
1964 0 : json_response(StatusCode::OK, ())
1965 0 : }
1966 :
1967 : // Obtains an lsn lease on the given timeline.
1968 0 : async fn lsn_lease_handler(
1969 0 : mut request: Request<Body>,
1970 0 : _cancel: CancellationToken,
1971 0 : ) -> Result<Response<Body>, ApiError> {
1972 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1973 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1974 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1975 0 : let lsn = json_request::<LsnLeaseRequest>(&mut request).await?.lsn;
1976 :
1977 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1978 0 :
1979 0 : let state = get_state(&request);
1980 :
1981 0 : let timeline =
1982 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1983 0 : .await?;
1984 :
1985 0 : let result = async {
1986 0 : timeline
1987 0 : .init_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
1988 0 : .map_err(|e| {
1989 0 : ApiError::InternalServerError(
1990 0 : e.context(format!("invalid lsn lease request at {lsn}")),
1991 0 : )
1992 0 : })
1993 0 : }
1994 0 : .instrument(info_span!("init_lsn_lease", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1995 0 : .await?;
1996 :
1997 0 : json_response(StatusCode::OK, result)
1998 0 : }
1999 :
2000 : // Run GC immediately on given timeline.
2001 0 : async fn timeline_gc_handler(
2002 0 : mut request: Request<Body>,
2003 0 : cancel: CancellationToken,
2004 0 : ) -> Result<Response<Body>, ApiError> {
2005 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2006 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2007 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2008 :
2009 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
2010 :
2011 0 : let state = get_state(&request);
2012 0 :
2013 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2014 0 : let gc_result = state
2015 0 : .tenant_manager
2016 0 : .immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx)
2017 0 : .await?;
2018 :
2019 0 : json_response(StatusCode::OK, gc_result)
2020 0 : }
2021 :
2022 : // Cancel scheduled compaction tasks
2023 0 : async fn timeline_cancel_compact_handler(
2024 0 : request: Request<Body>,
2025 0 : _cancel: CancellationToken,
2026 0 : ) -> Result<Response<Body>, ApiError> {
2027 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2028 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2029 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2030 0 : let state = get_state(&request);
2031 0 : async {
2032 0 : let tenant = state
2033 0 : .tenant_manager
2034 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2035 0 : tenant.cancel_scheduled_compaction(timeline_id);
2036 0 : json_response(StatusCode::OK, ())
2037 0 : }
2038 0 : .instrument(info_span!("timeline_cancel_compact", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2039 0 : .await
2040 0 : }
2041 :
2042 : // Get compact info of a timeline
2043 0 : async fn timeline_compact_info_handler(
2044 0 : request: Request<Body>,
2045 0 : _cancel: CancellationToken,
2046 0 : ) -> Result<Response<Body>, ApiError> {
2047 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2048 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2049 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2050 0 : let state = get_state(&request);
2051 0 : async {
2052 0 : let tenant = state
2053 0 : .tenant_manager
2054 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2055 0 : let res = tenant.get_scheduled_compaction_tasks(timeline_id);
2056 0 : let mut resp = Vec::new();
2057 0 : for item in res {
2058 0 : resp.push(CompactInfoResponse {
2059 0 : compact_key_range: item.compact_key_range,
2060 0 : compact_lsn_range: item.compact_lsn_range,
2061 0 : sub_compaction: item.sub_compaction,
2062 0 : });
2063 0 : }
2064 0 : json_response(StatusCode::OK, resp)
2065 0 : }
2066 0 : .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2067 0 : .await
2068 0 : }
2069 :
2070 : // Run compaction immediately on given timeline.
2071 0 : async fn timeline_compact_handler(
2072 0 : mut request: Request<Body>,
2073 0 : cancel: CancellationToken,
2074 0 : ) -> Result<Response<Body>, ApiError> {
2075 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2076 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2077 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2078 :
2079 0 : let compact_request = json_request_maybe::<Option<CompactRequest>>(&mut request).await?;
2080 :
2081 0 : let state = get_state(&request);
2082 0 :
2083 0 : let mut flags = EnumSet::empty();
2084 0 :
2085 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2086 0 : flags |= CompactFlags::ForceL0Compaction;
2087 0 : }
2088 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2089 0 : flags |= CompactFlags::ForceRepartition;
2090 0 : }
2091 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2092 0 : flags |= CompactFlags::ForceImageLayerCreation;
2093 0 : }
2094 0 : if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
2095 0 : flags |= CompactFlags::EnhancedGcBottomMostCompaction;
2096 0 : }
2097 0 : if Some(true) == parse_query_param::<_, bool>(&request, "dry_run")? {
2098 0 : flags |= CompactFlags::DryRun;
2099 0 : }
2100 :
2101 0 : let wait_until_uploaded =
2102 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2103 :
2104 0 : let wait_until_scheduled_compaction_done =
2105 0 : parse_query_param::<_, bool>(&request, "wait_until_scheduled_compaction_done")?
2106 0 : .unwrap_or(false);
2107 0 :
2108 0 : let sub_compaction = compact_request
2109 0 : .as_ref()
2110 0 : .map(|r| r.sub_compaction)
2111 0 : .unwrap_or(false);
2112 0 : let sub_compaction_max_job_size_mb = compact_request
2113 0 : .as_ref()
2114 0 : .and_then(|r| r.sub_compaction_max_job_size_mb);
2115 0 :
2116 0 : let options = CompactOptions {
2117 0 : compact_key_range: compact_request
2118 0 : .as_ref()
2119 0 : .and_then(|r| r.compact_key_range.clone()),
2120 0 : compact_lsn_range: compact_request
2121 0 : .as_ref()
2122 0 : .and_then(|r| r.compact_lsn_range.clone()),
2123 0 : flags,
2124 0 : sub_compaction,
2125 0 : sub_compaction_max_job_size_mb,
2126 0 : };
2127 0 :
2128 0 : let scheduled = compact_request
2129 0 : .as_ref()
2130 0 : .map(|r| r.scheduled)
2131 0 : .unwrap_or(false);
2132 :
2133 0 : async {
2134 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2135 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2136 0 : if scheduled {
2137 0 : let tenant = state
2138 0 : .tenant_manager
2139 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2140 0 : let rx = tenant.schedule_compaction(timeline_id, options).await.map_err(ApiError::InternalServerError)?;
2141 0 : if wait_until_scheduled_compaction_done {
2142 : // It is possible that this will take a long time, dropping the HTTP request will not cancel the compaction.
2143 0 : rx.await.ok();
2144 0 : }
2145 : } else {
2146 0 : timeline
2147 0 : .compact_with_options(&cancel, options, &ctx)
2148 0 : .await
2149 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2150 0 : if wait_until_uploaded {
2151 0 : timeline.remote_client.wait_completion().await
2152 : // XXX map to correct ApiError for the cases where it's due to shutdown
2153 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2154 0 : }
2155 : }
2156 0 : json_response(StatusCode::OK, ())
2157 0 : }
2158 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2159 0 : .await
2160 0 : }
2161 :
2162 : // Run offload immediately on given timeline.
2163 0 : async fn timeline_offload_handler(
2164 0 : request: Request<Body>,
2165 0 : _cancel: CancellationToken,
2166 0 : ) -> Result<Response<Body>, ApiError> {
2167 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2168 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2169 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2170 :
2171 0 : let state = get_state(&request);
2172 :
2173 0 : async {
2174 0 : let tenant = state
2175 0 : .tenant_manager
2176 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2177 :
2178 0 : if tenant.get_offloaded_timeline(timeline_id).is_ok() {
2179 0 : return json_response(StatusCode::OK, ());
2180 0 : }
2181 0 : let timeline =
2182 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2183 0 : .await?;
2184 :
2185 0 : if !tenant.timeline_has_no_attached_children(timeline_id) {
2186 0 : return Err(ApiError::PreconditionFailed(
2187 0 : "timeline has attached children".into(),
2188 0 : ));
2189 0 : }
2190 0 : if let (false, reason) = timeline.can_offload() {
2191 0 : return Err(ApiError::PreconditionFailed(
2192 0 : format!("Timeline::can_offload() check failed: {}", reason) .into(),
2193 0 : ));
2194 0 : }
2195 0 : offload_timeline(&tenant, &timeline)
2196 0 : .await
2197 0 : .map_err(|e| {
2198 0 : match e {
2199 0 : OffloadError::Cancelled => ApiError::ResourceUnavailable("Timeline shutting down".into()),
2200 0 : _ => ApiError::InternalServerError(anyhow!(e))
2201 : }
2202 0 : })?;
2203 :
2204 0 : json_response(StatusCode::OK, ())
2205 0 : }
2206 0 : .instrument(info_span!("manual_timeline_offload", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2207 0 : .await
2208 0 : }
2209 :
2210 : // Run checkpoint immediately on given timeline.
2211 0 : async fn timeline_checkpoint_handler(
2212 0 : request: Request<Body>,
2213 0 : cancel: CancellationToken,
2214 0 : ) -> Result<Response<Body>, ApiError> {
2215 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2216 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2217 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2218 :
2219 0 : let state = get_state(&request);
2220 0 :
2221 0 : let mut flags = EnumSet::empty();
2222 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_l0_compaction")? {
2223 0 : flags |= CompactFlags::ForceL0Compaction;
2224 0 : }
2225 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
2226 0 : flags |= CompactFlags::ForceRepartition;
2227 0 : }
2228 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
2229 0 : flags |= CompactFlags::ForceImageLayerCreation;
2230 0 : }
2231 :
2232 : // By default, checkpoints come with a compaction, but this may be optionally disabled by tests that just want to flush + upload.
2233 0 : let compact = parse_query_param::<_, bool>(&request, "compact")?.unwrap_or(true);
2234 :
2235 0 : let wait_until_flushed: bool =
2236 0 : parse_query_param(&request, "wait_until_flushed")?.unwrap_or(true);
2237 :
2238 0 : let wait_until_uploaded =
2239 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
2240 :
2241 0 : async {
2242 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2243 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2244 0 : if wait_until_flushed {
2245 0 : timeline.freeze_and_flush().await
2246 : } else {
2247 0 : timeline.freeze().await.and(Ok(()))
2248 0 : }.map_err(|e| {
2249 0 : match e {
2250 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
2251 0 : other => ApiError::InternalServerError(other.into()),
2252 :
2253 : }
2254 0 : })?;
2255 0 : if compact {
2256 0 : timeline
2257 0 : .compact(&cancel, flags, &ctx)
2258 0 : .await
2259 0 : .map_err(|e|
2260 0 : match e {
2261 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
2262 0 : CompactionError::Offload(e) => ApiError::InternalServerError(anyhow::anyhow!(e)),
2263 0 : CompactionError::Other(e) => ApiError::InternalServerError(e)
2264 0 : }
2265 0 : )?;
2266 0 : }
2267 :
2268 0 : if wait_until_uploaded {
2269 0 : tracing::info!("Waiting for uploads to complete...");
2270 0 : timeline.remote_client.wait_completion().await
2271 : // XXX map to correct ApiError for the cases where it's due to shutdown
2272 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
2273 0 : tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
2274 0 : }
2275 :
2276 0 : json_response(StatusCode::OK, ())
2277 0 : }
2278 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2279 0 : .await
2280 0 : }
2281 :
2282 0 : async fn timeline_download_remote_layers_handler_post(
2283 0 : mut request: Request<Body>,
2284 0 : _cancel: CancellationToken,
2285 0 : ) -> Result<Response<Body>, ApiError> {
2286 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2287 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2288 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
2289 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2290 :
2291 0 : let state = get_state(&request);
2292 :
2293 0 : let timeline =
2294 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2295 0 : .await?;
2296 0 : match timeline.spawn_download_all_remote_layers(body).await {
2297 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
2298 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
2299 : }
2300 0 : }
2301 :
2302 0 : async fn timeline_download_remote_layers_handler_get(
2303 0 : request: Request<Body>,
2304 0 : _cancel: CancellationToken,
2305 0 : ) -> Result<Response<Body>, ApiError> {
2306 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2307 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2308 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2309 0 : let state = get_state(&request);
2310 :
2311 0 : let timeline =
2312 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2313 0 : .await?;
2314 0 : let info = timeline
2315 0 : .get_download_all_remote_layers_task_info()
2316 0 : .context("task never started since last pageserver process start")
2317 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
2318 0 : json_response(StatusCode::OK, info)
2319 0 : }
2320 :
2321 0 : async fn timeline_detach_ancestor_handler(
2322 0 : request: Request<Body>,
2323 0 : _cancel: CancellationToken,
2324 0 : ) -> Result<Response<Body>, ApiError> {
2325 : use crate::tenant::timeline::detach_ancestor;
2326 : use pageserver_api::models::detach_ancestor::AncestorDetached;
2327 :
2328 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2329 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2330 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2331 :
2332 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
2333 :
2334 0 : async move {
2335 0 : let mut options = detach_ancestor::Options::default();
2336 :
2337 0 : let rewrite_concurrency =
2338 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
2339 0 : let copy_concurrency =
2340 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
2341 :
2342 0 : [
2343 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
2344 0 : (&mut options.copy_concurrency, copy_concurrency),
2345 0 : ]
2346 0 : .into_iter()
2347 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
2348 0 : .for_each(|(target, val)| *target = val);
2349 0 :
2350 0 : let state = get_state(&request);
2351 :
2352 0 : let tenant = state
2353 0 : .tenant_manager
2354 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2355 :
2356 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2357 :
2358 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
2359 0 : let ctx = &ctx;
2360 :
2361 : // Flush the upload queues of all timelines before detaching ancestor. We do the same thing again
2362 : // during shutdown. This early upload ensures the pageserver does not need to upload too many
2363 : // things and creates downtime during timeline reloads.
2364 0 : for timeline in tenant.list_timelines() {
2365 0 : timeline
2366 0 : .remote_client
2367 0 : .wait_completion()
2368 0 : .await
2369 0 : .map_err(|e| {
2370 0 : ApiError::PreconditionFailed(format!("cannot drain upload queue: {e}").into())
2371 0 : })?;
2372 : }
2373 :
2374 0 : tracing::info!("all timeline upload queues are drained");
2375 :
2376 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
2377 :
2378 0 : let progress = timeline
2379 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
2380 0 : .await?;
2381 :
2382 : // uncomment to allow early as possible Tenant::drop
2383 : // drop(tenant);
2384 :
2385 0 : let resp = match progress {
2386 0 : detach_ancestor::Progress::Prepared(attempt, prepared) => {
2387 : // it would be great to tag the guard on to the tenant activation future
2388 0 : let reparented_timelines = state
2389 0 : .tenant_manager
2390 0 : .complete_detaching_timeline_ancestor(
2391 0 : tenant_shard_id,
2392 0 : timeline_id,
2393 0 : prepared,
2394 0 : attempt,
2395 0 : ctx,
2396 0 : )
2397 0 : .await?;
2398 :
2399 0 : AncestorDetached {
2400 0 : reparented_timelines,
2401 0 : }
2402 : }
2403 0 : detach_ancestor::Progress::Done(resp) => resp,
2404 : };
2405 :
2406 0 : json_response(StatusCode::OK, resp)
2407 0 : }
2408 0 : .instrument(span)
2409 0 : .await
2410 0 : }
2411 :
2412 0 : async fn deletion_queue_flush(
2413 0 : r: Request<Body>,
2414 0 : cancel: CancellationToken,
2415 0 : ) -> Result<Response<Body>, ApiError> {
2416 0 : let state = get_state(&r);
2417 :
2418 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
2419 0 :
2420 0 : let flush = async {
2421 0 : if execute {
2422 0 : state.deletion_queue_client.flush_execute().await
2423 : } else {
2424 0 : state.deletion_queue_client.flush().await
2425 : }
2426 0 : }
2427 : // DeletionQueueError's only case is shutting down.
2428 0 : .map_err(|_| ApiError::ShuttingDown);
2429 0 :
2430 0 : tokio::select! {
2431 0 : res = flush => {
2432 0 : res.map(|()| json_response(StatusCode::OK, ()))?
2433 : }
2434 0 : _ = cancel.cancelled() => {
2435 0 : Err(ApiError::ShuttingDown)
2436 : }
2437 : }
2438 0 : }
2439 :
2440 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
2441 0 : async fn getpage_at_lsn_handler(
2442 0 : request: Request<Body>,
2443 0 : _cancel: CancellationToken,
2444 0 : ) -> Result<Response<Body>, ApiError> {
2445 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2446 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2447 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2448 0 : let state = get_state(&request);
2449 :
2450 : struct Key(pageserver_api::key::Key);
2451 :
2452 : impl std::str::FromStr for Key {
2453 : type Err = anyhow::Error;
2454 :
2455 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
2456 0 : pageserver_api::key::Key::from_hex(s).map(Key)
2457 0 : }
2458 : }
2459 :
2460 0 : let key: Key = parse_query_param(&request, "key")?
2461 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
2462 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
2463 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
2464 :
2465 0 : async {
2466 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2467 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2468 :
2469 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
2470 :
2471 0 : Result::<_, ApiError>::Ok(
2472 0 : Response::builder()
2473 0 : .status(StatusCode::OK)
2474 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2475 0 : .body(hyper::Body::from(page))
2476 0 : .unwrap(),
2477 0 : )
2478 0 : }
2479 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2480 0 : .await
2481 0 : }
2482 :
2483 0 : async fn timeline_collect_keyspace(
2484 0 : request: Request<Body>,
2485 0 : _cancel: CancellationToken,
2486 0 : ) -> Result<Response<Body>, ApiError> {
2487 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2488 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2489 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2490 0 : let state = get_state(&request);
2491 :
2492 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2493 :
2494 0 : async {
2495 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2496 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2497 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2498 0 : let (dense_ks, sparse_ks) = timeline
2499 0 : .collect_keyspace(at_lsn, &ctx)
2500 0 : .await
2501 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2502 :
2503 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2504 : // Therefore, we split dense/sparse keys in this API.
2505 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2506 0 :
2507 0 : json_response(StatusCode::OK, res)
2508 0 : }
2509 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2510 0 : .await
2511 0 : }
2512 :
2513 0 : async fn active_timeline_of_active_tenant(
2514 0 : tenant_manager: &TenantManager,
2515 0 : tenant_shard_id: TenantShardId,
2516 0 : timeline_id: TimelineId,
2517 0 : ) -> Result<Arc<Timeline>, ApiError> {
2518 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2519 :
2520 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2521 :
2522 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2523 0 : }
2524 :
2525 0 : async fn always_panic_handler(
2526 0 : req: Request<Body>,
2527 0 : _cancel: CancellationToken,
2528 0 : ) -> Result<Response<Body>, ApiError> {
2529 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2530 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2531 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2532 0 : let query = req.uri().query();
2533 0 : let _ = std::panic::catch_unwind(|| {
2534 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2535 0 : });
2536 0 : json_response(StatusCode::NO_CONTENT, ())
2537 0 : }
2538 :
2539 0 : async fn disk_usage_eviction_run(
2540 0 : mut r: Request<Body>,
2541 0 : cancel: CancellationToken,
2542 0 : ) -> Result<Response<Body>, ApiError> {
2543 0 : check_permission(&r, None)?;
2544 :
2545 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2546 : struct Config {
2547 : /// How many bytes to evict before reporting that pressure is relieved.
2548 : evict_bytes: u64,
2549 :
2550 : #[serde(default)]
2551 : eviction_order: pageserver_api::config::EvictionOrder,
2552 : }
2553 :
2554 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2555 : struct Usage {
2556 : // remains unchanged after instantiation of the struct
2557 : evict_bytes: u64,
2558 : // updated by `add_available_bytes`
2559 : freed_bytes: u64,
2560 : }
2561 :
2562 : impl crate::disk_usage_eviction_task::Usage for Usage {
2563 0 : fn has_pressure(&self) -> bool {
2564 0 : self.evict_bytes > self.freed_bytes
2565 0 : }
2566 :
2567 0 : fn add_available_bytes(&mut self, bytes: u64) {
2568 0 : self.freed_bytes += bytes;
2569 0 : }
2570 : }
2571 :
2572 0 : let config = json_request::<Config>(&mut r).await?;
2573 :
2574 0 : let usage = Usage {
2575 0 : evict_bytes: config.evict_bytes,
2576 0 : freed_bytes: 0,
2577 0 : };
2578 0 :
2579 0 : let state = get_state(&r);
2580 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2581 :
2582 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2583 0 : &eviction_state,
2584 0 : &state.remote_storage,
2585 0 : usage,
2586 0 : &state.tenant_manager,
2587 0 : config.eviction_order.into(),
2588 0 : &cancel,
2589 0 : )
2590 0 : .await;
2591 :
2592 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2593 :
2594 0 : let res = res.map_err(ApiError::InternalServerError)?;
2595 :
2596 0 : json_response(StatusCode::OK, res)
2597 0 : }
2598 :
2599 0 : async fn secondary_upload_handler(
2600 0 : request: Request<Body>,
2601 0 : _cancel: CancellationToken,
2602 0 : ) -> Result<Response<Body>, ApiError> {
2603 0 : let state = get_state(&request);
2604 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2605 0 : state
2606 0 : .secondary_controller
2607 0 : .upload_tenant(tenant_shard_id)
2608 0 : .await?;
2609 :
2610 0 : json_response(StatusCode::OK, ())
2611 0 : }
2612 :
2613 0 : async fn tenant_scan_remote_handler(
2614 0 : request: Request<Body>,
2615 0 : cancel: CancellationToken,
2616 0 : ) -> Result<Response<Body>, ApiError> {
2617 0 : let state = get_state(&request);
2618 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2619 :
2620 0 : let mut response = TenantScanRemoteStorageResponse::default();
2621 :
2622 0 : let (shards, _other_keys) =
2623 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2624 0 : .await
2625 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2626 :
2627 0 : for tenant_shard_id in shards {
2628 0 : let (timeline_ids, _other_keys) =
2629 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2630 0 : .await
2631 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2632 :
2633 0 : let mut generation = Generation::none();
2634 0 : for timeline_id in timeline_ids {
2635 0 : match download_index_part(
2636 0 : &state.remote_storage,
2637 0 : &tenant_shard_id,
2638 0 : &timeline_id,
2639 0 : Generation::MAX,
2640 0 : &cancel,
2641 0 : )
2642 0 : .instrument(info_span!("download_index_part",
2643 : tenant_id=%tenant_shard_id.tenant_id,
2644 0 : shard_id=%tenant_shard_id.shard_slug(),
2645 : %timeline_id))
2646 0 : .await
2647 : {
2648 0 : Ok((index_part, index_generation, _index_mtime)) => {
2649 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2650 0 : index_part.layer_metadata.len(), index_part.metadata.disk_consistent_lsn());
2651 0 : generation = std::cmp::max(generation, index_generation);
2652 : }
2653 : Err(DownloadError::NotFound) => {
2654 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2655 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2656 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2657 0 : continue;
2658 : }
2659 0 : Err(e) => {
2660 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2661 : }
2662 : };
2663 : }
2664 :
2665 0 : response.shards.push(TenantScanRemoteStorageShard {
2666 0 : tenant_shard_id,
2667 0 : generation: generation.into(),
2668 0 : });
2669 : }
2670 :
2671 0 : if response.shards.is_empty() {
2672 0 : return Err(ApiError::NotFound(
2673 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2674 0 : ));
2675 0 : }
2676 0 :
2677 0 : json_response(StatusCode::OK, response)
2678 0 : }
2679 :
2680 0 : async fn secondary_download_handler(
2681 0 : request: Request<Body>,
2682 0 : _cancel: CancellationToken,
2683 0 : ) -> Result<Response<Body>, ApiError> {
2684 0 : let state = get_state(&request);
2685 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2686 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2687 :
2688 : // We don't need this to issue the download request, but:
2689 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2690 : // - we will use this to provide status feedback in the response
2691 0 : let Some(secondary_tenant) = state
2692 0 : .tenant_manager
2693 0 : .get_secondary_tenant_shard(tenant_shard_id)
2694 : else {
2695 0 : return Err(ApiError::NotFound(
2696 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2697 0 : ));
2698 : };
2699 :
2700 0 : let timeout = wait.unwrap_or(Duration::MAX);
2701 :
2702 0 : let result = tokio::time::timeout(
2703 0 : timeout,
2704 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2705 0 : )
2706 0 : .await;
2707 :
2708 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2709 :
2710 0 : let status = match result {
2711 : Ok(Ok(())) => {
2712 0 : if progress.layers_downloaded >= progress.layers_total {
2713 : // Download job ran to completion
2714 0 : StatusCode::OK
2715 : } else {
2716 : // Download dropped out without errors because it ran out of time budget
2717 0 : StatusCode::ACCEPTED
2718 : }
2719 : }
2720 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2721 : // okay. We could get an error here in the unlikely edge case that the tenant
2722 : // was detached between our check above and executing the download job.
2723 0 : Ok(Err(e)) => return Err(e.into()),
2724 : // A timeout is not an error: we have started the download, we're just not done
2725 : // yet. The caller will get a response body indicating status.
2726 0 : Err(_) => StatusCode::ACCEPTED,
2727 : };
2728 :
2729 0 : json_response(status, progress)
2730 0 : }
2731 :
2732 0 : async fn secondary_status_handler(
2733 0 : request: Request<Body>,
2734 0 : _cancel: CancellationToken,
2735 0 : ) -> Result<Response<Body>, ApiError> {
2736 0 : let state = get_state(&request);
2737 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2738 :
2739 0 : let Some(secondary_tenant) = state
2740 0 : .tenant_manager
2741 0 : .get_secondary_tenant_shard(tenant_shard_id)
2742 : else {
2743 0 : return Err(ApiError::NotFound(
2744 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2745 0 : ));
2746 : };
2747 :
2748 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2749 0 :
2750 0 : json_response(StatusCode::OK, progress)
2751 0 : }
2752 :
2753 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2754 0 : json_response(
2755 0 : StatusCode::NOT_FOUND,
2756 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2757 0 : )
2758 0 : }
2759 :
2760 0 : async fn post_tracing_event_handler(
2761 0 : mut r: Request<Body>,
2762 0 : _cancel: CancellationToken,
2763 0 : ) -> Result<Response<Body>, ApiError> {
2764 0 : #[derive(Debug, serde::Deserialize)]
2765 : #[serde(rename_all = "lowercase")]
2766 : enum Level {
2767 : Error,
2768 : Warn,
2769 : Info,
2770 : Debug,
2771 : Trace,
2772 : }
2773 0 : #[derive(Debug, serde::Deserialize)]
2774 : struct Request {
2775 : level: Level,
2776 : message: String,
2777 : }
2778 0 : let body: Request = json_request(&mut r)
2779 0 : .await
2780 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2781 :
2782 0 : match body.level {
2783 0 : Level::Error => tracing::error!(?body.message),
2784 0 : Level::Warn => tracing::warn!(?body.message),
2785 0 : Level::Info => tracing::info!(?body.message),
2786 0 : Level::Debug => tracing::debug!(?body.message),
2787 0 : Level::Trace => tracing::trace!(?body.message),
2788 : }
2789 :
2790 0 : json_response(StatusCode::OK, ())
2791 0 : }
2792 :
2793 0 : async fn put_io_engine_handler(
2794 0 : mut r: Request<Body>,
2795 0 : _cancel: CancellationToken,
2796 0 : ) -> Result<Response<Body>, ApiError> {
2797 0 : check_permission(&r, None)?;
2798 0 : let kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2799 0 : crate::virtual_file::io_engine::set(kind);
2800 0 : json_response(StatusCode::OK, ())
2801 0 : }
2802 :
2803 0 : async fn put_io_mode_handler(
2804 0 : mut r: Request<Body>,
2805 0 : _cancel: CancellationToken,
2806 0 : ) -> Result<Response<Body>, ApiError> {
2807 0 : check_permission(&r, None)?;
2808 0 : let mode: IoMode = json_request(&mut r).await?;
2809 0 : crate::virtual_file::set_io_mode(mode);
2810 0 : json_response(StatusCode::OK, ())
2811 0 : }
2812 :
2813 : /// Polled by control plane.
2814 : ///
2815 : /// See [`crate::utilization`].
2816 0 : async fn get_utilization(
2817 0 : r: Request<Body>,
2818 0 : _cancel: CancellationToken,
2819 0 : ) -> Result<Response<Body>, ApiError> {
2820 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2821 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2822 0 : });
2823 :
2824 : // this probably could be completely public, but lets make that change later.
2825 0 : check_permission(&r, None)?;
2826 :
2827 0 : let state = get_state(&r);
2828 0 : let mut g = state.latest_utilization.lock().await;
2829 :
2830 0 : let regenerate_every = Duration::from_secs(1);
2831 0 : let still_valid = g
2832 0 : .as_ref()
2833 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2834 0 :
2835 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2836 0 : // regenerate at most 1Hz to allow polling at any rate.
2837 0 : if !still_valid {
2838 0 : let path = state.conf.tenants_path();
2839 0 : let doc =
2840 0 : crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
2841 0 : .map_err(ApiError::InternalServerError)?;
2842 :
2843 0 : let mut buf = Vec::new();
2844 0 : serde_json::to_writer(&mut buf, &doc)
2845 0 : .context("serialize")
2846 0 : .map_err(ApiError::InternalServerError)?;
2847 :
2848 0 : let body = bytes::Bytes::from(buf);
2849 0 :
2850 0 : *g = Some((std::time::Instant::now(), body));
2851 0 : }
2852 :
2853 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2854 0 : let cached = g.as_ref().expect("just set").1.clone();
2855 0 :
2856 0 : Response::builder()
2857 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2858 0 : // thought of using http date header, but that is second precision which does not give any
2859 0 : // debugging aid
2860 0 : .status(StatusCode::OK)
2861 0 : .body(hyper::Body::from(cached))
2862 0 : .context("build response")
2863 0 : .map_err(ApiError::InternalServerError)
2864 0 : }
2865 :
2866 0 : async fn list_aux_files(
2867 0 : mut request: Request<Body>,
2868 0 : _cancel: CancellationToken,
2869 0 : ) -> Result<Response<Body>, ApiError> {
2870 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2871 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2872 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2873 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2874 :
2875 0 : let state = get_state(&request);
2876 :
2877 0 : let timeline =
2878 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2879 0 : .await?;
2880 :
2881 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2882 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2883 0 : json_response(StatusCode::OK, files)
2884 0 : }
2885 :
2886 0 : async fn perf_info(
2887 0 : request: Request<Body>,
2888 0 : _cancel: CancellationToken,
2889 0 : ) -> Result<Response<Body>, ApiError> {
2890 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2891 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2892 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2893 :
2894 0 : let state = get_state(&request);
2895 :
2896 0 : let timeline =
2897 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2898 0 : .await?;
2899 :
2900 0 : let result = timeline.perf_info().await;
2901 :
2902 0 : json_response(StatusCode::OK, result)
2903 0 : }
2904 :
2905 0 : async fn ingest_aux_files(
2906 0 : mut request: Request<Body>,
2907 0 : _cancel: CancellationToken,
2908 0 : ) -> Result<Response<Body>, ApiError> {
2909 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2910 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2911 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2912 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2913 :
2914 0 : let state = get_state(&request);
2915 :
2916 0 : let timeline =
2917 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2918 0 : .await?;
2919 :
2920 0 : let mut modification = timeline.begin_modification(
2921 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
2922 0 : );
2923 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2924 0 : for (fname, content) in body.aux_files {
2925 0 : modification
2926 0 : .put_file(&fname, content.as_bytes(), &ctx)
2927 0 : .await
2928 0 : .map_err(ApiError::InternalServerError)?;
2929 : }
2930 0 : modification
2931 0 : .commit(&ctx)
2932 0 : .await
2933 0 : .map_err(ApiError::InternalServerError)?;
2934 :
2935 0 : json_response(StatusCode::OK, ())
2936 0 : }
2937 :
2938 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2939 : /// candidates for splitting
2940 0 : async fn post_top_tenants(
2941 0 : mut r: Request<Body>,
2942 0 : _cancel: CancellationToken,
2943 0 : ) -> Result<Response<Body>, ApiError> {
2944 0 : check_permission(&r, None)?;
2945 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2946 0 : let state = get_state(&r);
2947 :
2948 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2949 0 : match order_by {
2950 0 : TenantSorting::ResidentSize => sizes.resident_size,
2951 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2952 : }
2953 0 : }
2954 :
2955 : #[derive(Eq, PartialEq)]
2956 : struct HeapItem {
2957 : metric: u64,
2958 : sizes: TopTenantShardItem,
2959 : }
2960 :
2961 : impl PartialOrd for HeapItem {
2962 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2963 0 : Some(self.cmp(other))
2964 0 : }
2965 : }
2966 :
2967 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2968 : /// supports popping the greatest item but not the smallest.
2969 : impl Ord for HeapItem {
2970 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2971 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2972 0 : }
2973 : }
2974 :
2975 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2976 :
2977 : // FIXME: this is a lot of clones to take this tenant list
2978 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2979 0 : if let Some(shards_lt) = request.where_shards_lt {
2980 : // Ignore tenants which already have >= this many shards
2981 0 : if tenant_shard_id.shard_count >= shards_lt {
2982 0 : continue;
2983 0 : }
2984 0 : }
2985 :
2986 0 : let sizes = match tenant_slot {
2987 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2988 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2989 0 : continue;
2990 : }
2991 : };
2992 0 : let metric = get_size_metric(&sizes, &request.order_by);
2993 :
2994 0 : if let Some(gt) = request.where_gt {
2995 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2996 0 : if metric <= gt {
2997 0 : continue;
2998 0 : }
2999 0 : };
3000 :
3001 0 : match top_n.peek() {
3002 0 : None => {
3003 0 : // Top N list is empty: candidate becomes first member
3004 0 : top_n.push(HeapItem { metric, sizes });
3005 0 : }
3006 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
3007 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
3008 0 : top_n.push(HeapItem { metric, sizes });
3009 0 : }
3010 0 : Some(i) if i.metric > metric => {
3011 0 : // List is at limit and lowest value is greater than our candidate, drop it.
3012 0 : }
3013 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
3014 : }
3015 :
3016 0 : while top_n.len() > request.limit {
3017 0 : top_n.pop();
3018 0 : }
3019 : }
3020 :
3021 0 : json_response(
3022 0 : StatusCode::OK,
3023 0 : TopTenantShardsResponse {
3024 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
3025 0 : },
3026 0 : )
3027 0 : }
3028 :
3029 0 : async fn put_tenant_timeline_import_basebackup(
3030 0 : request: Request<Body>,
3031 0 : _cancel: CancellationToken,
3032 0 : ) -> Result<Response<Body>, ApiError> {
3033 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3034 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3035 0 : let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
3036 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3037 0 : let pg_version: u32 = must_parse_query_param(&request, "pg_version")?;
3038 :
3039 0 : check_permission(&request, Some(tenant_id))?;
3040 :
3041 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3042 :
3043 0 : let span = info_span!("import_basebackup", tenant_id=%tenant_id, timeline_id=%timeline_id, base_lsn=%base_lsn, end_lsn=%end_lsn, pg_version=%pg_version);
3044 0 : async move {
3045 0 : let state = get_state(&request);
3046 0 : let tenant = state
3047 0 : .tenant_manager
3048 0 : .get_attached_tenant_shard(TenantShardId::unsharded(tenant_id))?;
3049 :
3050 0 : let broker_client = state.broker_client.clone();
3051 0 :
3052 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
3053 0 : res.map_err(|error| {
3054 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
3055 0 : })
3056 0 : }));
3057 0 :
3058 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
3059 :
3060 0 : let timeline = tenant
3061 0 : .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
3062 0 : .map_err(ApiError::InternalServerError)
3063 0 : .await?;
3064 :
3065 : // TODO mark timeline as not ready until it reaches end_lsn.
3066 : // We might have some wal to import as well, and we should prevent compute
3067 : // from connecting before that and writing conflicting wal.
3068 : //
3069 : // This is not relevant for pageserver->pageserver migrations, since there's
3070 : // no wal to import. But should be fixed if we want to import from postgres.
3071 :
3072 : // TODO leave clean state on error. For now you can use detach to clean
3073 : // up broken state from a failed import.
3074 :
3075 : // Import basebackup provided via CopyData
3076 0 : info!("importing basebackup");
3077 :
3078 0 : timeline
3079 0 : .import_basebackup_from_tar(tenant.clone(), &mut body, base_lsn, broker_client, &ctx)
3080 0 : .await
3081 0 : .map_err(ApiError::InternalServerError)?;
3082 :
3083 : // Read the end of the tar archive.
3084 0 : read_tar_eof(body)
3085 0 : .await
3086 0 : .map_err(ApiError::InternalServerError)?;
3087 :
3088 : // TODO check checksum
3089 : // Meanwhile you can verify client-side by taking fullbackup
3090 : // and checking that it matches in size with what was imported.
3091 : // It wouldn't work if base came from vanilla postgres though,
3092 : // since we discard some log files.
3093 :
3094 0 : info!("done");
3095 0 : json_response(StatusCode::OK, ())
3096 0 : }
3097 0 : .instrument(span)
3098 0 : .await
3099 0 : }
3100 :
3101 0 : async fn put_tenant_timeline_import_wal(
3102 0 : request: Request<Body>,
3103 0 : _cancel: CancellationToken,
3104 0 : ) -> Result<Response<Body>, ApiError> {
3105 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
3106 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
3107 0 : let start_lsn: Lsn = must_parse_query_param(&request, "start_lsn")?;
3108 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
3109 :
3110 0 : check_permission(&request, Some(tenant_id))?;
3111 :
3112 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
3113 :
3114 0 : let span = info_span!("import_wal", tenant_id=%tenant_id, timeline_id=%timeline_id, start_lsn=%start_lsn, end_lsn=%end_lsn);
3115 0 : async move {
3116 0 : let state = get_state(&request);
3117 :
3118 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, TenantShardId::unsharded(tenant_id), timeline_id).await?;
3119 :
3120 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
3121 0 : res.map_err(|error| {
3122 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
3123 0 : })
3124 0 : }));
3125 0 :
3126 0 : let last_record_lsn = timeline.get_last_record_lsn();
3127 0 : if last_record_lsn != start_lsn {
3128 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}")));
3129 0 : }
3130 0 :
3131 0 : // TODO leave clean state on error. For now you can use detach to clean
3132 0 : // up broken state from a failed import.
3133 0 :
3134 0 : // Import wal provided via CopyData
3135 0 : info!("importing wal");
3136 0 : crate::import_datadir::import_wal_from_tar(&timeline, &mut body, start_lsn, end_lsn, &ctx).await.map_err(ApiError::InternalServerError)?;
3137 0 : info!("wal import complete");
3138 :
3139 : // Read the end of the tar archive.
3140 0 : read_tar_eof(body).await.map_err(ApiError::InternalServerError)?;
3141 :
3142 : // TODO Does it make sense to overshoot?
3143 0 : if timeline.get_last_record_lsn() < end_lsn {
3144 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}")));
3145 0 : }
3146 0 :
3147 0 : // Flush data to disk, then upload to s3. No need for a forced checkpoint.
3148 0 : // We only want to persist the data, and it doesn't matter if it's in the
3149 0 : // shape of deltas or images.
3150 0 : info!("flushing layers");
3151 0 : timeline.freeze_and_flush().await.map_err(|e| match e {
3152 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
3153 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
3154 0 : })?;
3155 :
3156 0 : info!("done");
3157 :
3158 0 : json_response(StatusCode::OK, ())
3159 0 : }.instrument(span).await
3160 0 : }
3161 :
3162 : /// Read the end of a tar archive.
3163 : ///
3164 : /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
3165 : /// `tokio_tar` already read the first such block. Read the second all-zeros block,
3166 : /// and check that there is no more data after the EOF marker.
3167 : ///
3168 : /// 'tar' command can also write extra blocks of zeros, up to a record
3169 : /// size, controlled by the --record-size argument. Ignore them too.
3170 0 : async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow::Result<()> {
3171 : use tokio::io::AsyncReadExt;
3172 0 : let mut buf = [0u8; 512];
3173 0 :
3174 0 : // Read the all-zeros block, and verify it
3175 0 : let mut total_bytes = 0;
3176 0 : while total_bytes < 512 {
3177 0 : let nbytes = reader.read(&mut buf[total_bytes..]).await?;
3178 0 : total_bytes += nbytes;
3179 0 : if nbytes == 0 {
3180 0 : break;
3181 0 : }
3182 : }
3183 0 : if total_bytes < 512 {
3184 0 : anyhow::bail!("incomplete or invalid tar EOF marker");
3185 0 : }
3186 0 : if !buf.iter().all(|&x| x == 0) {
3187 0 : anyhow::bail!("invalid tar EOF marker");
3188 0 : }
3189 0 :
3190 0 : // Drain any extra zero-blocks after the EOF marker
3191 0 : let mut trailing_bytes = 0;
3192 0 : let mut seen_nonzero_bytes = false;
3193 : loop {
3194 0 : let nbytes = reader.read(&mut buf).await?;
3195 0 : trailing_bytes += nbytes;
3196 0 : if !buf.iter().all(|&x| x == 0) {
3197 0 : seen_nonzero_bytes = true;
3198 0 : }
3199 0 : if nbytes == 0 {
3200 0 : break;
3201 0 : }
3202 : }
3203 0 : if seen_nonzero_bytes {
3204 0 : anyhow::bail!("unexpected non-zero bytes after the tar archive");
3205 0 : }
3206 0 : if trailing_bytes % 512 != 0 {
3207 0 : anyhow::bail!("unexpected number of zeros ({trailing_bytes}), not divisible by tar block size (512 bytes), after the tar archive");
3208 0 : }
3209 0 : Ok(())
3210 0 : }
3211 :
3212 : /// Common functionality of all the HTTP API handlers.
3213 : ///
3214 : /// - Adds a tracing span to each request (by `request_span`)
3215 : /// - Logs the request depending on the request method (by `request_span`)
3216 : /// - Logs the response if it was not successful (by `request_span`
3217 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
3218 : /// Future if the connection to the client is lost, but most of the pageserver code is
3219 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
3220 : /// request with a CancellationToken.
3221 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
3222 0 : where
3223 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3224 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3225 0 : {
3226 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
3227 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
3228 0 : "failpoint".into()
3229 0 : )));
3230 :
3231 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
3232 0 : anyhow::anyhow!("failpoint")
3233 0 : )));
3234 0 : }
3235 :
3236 : // Spawn a new task to handle the request, to protect the handler from unexpected
3237 : // async cancellations. Most pageserver functions are not async cancellation safe.
3238 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
3239 : // with the cancellation token.
3240 0 : let token = CancellationToken::new();
3241 0 : let cancel_guard = token.clone().drop_guard();
3242 0 : let result = request_span(request, move |r| async {
3243 0 : let handle = tokio::spawn(
3244 0 : async {
3245 0 : let token_cloned = token.clone();
3246 0 : let result = handler(r, token).await;
3247 0 : if token_cloned.is_cancelled() {
3248 : // dropguard has executed: we will never turn this result into response.
3249 : //
3250 : // at least temporarily do {:?} logging; these failures are rare enough but
3251 : // could hide difficult errors.
3252 0 : match &result {
3253 0 : Ok(response) => {
3254 0 : let status = response.status();
3255 0 : info!(%status, "Cancelled request finished successfully")
3256 : }
3257 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
3258 : }
3259 0 : }
3260 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
3261 : // which should suffice.
3262 : //
3263 : // there is still a chance to lose the result due to race between
3264 : // returning from here and the actual connection closing happening
3265 : // before outer task gets to execute. leaving that up for #5815.
3266 0 : result
3267 0 : }
3268 0 : .in_current_span(),
3269 0 : );
3270 0 :
3271 0 : match handle.await {
3272 : // TODO: never actually return Err from here, always Ok(...) so that we can log
3273 : // spanned errors. Call api_error_handler instead and return appropriate Body.
3274 0 : Ok(result) => result,
3275 0 : Err(e) => {
3276 0 : // The handler task panicked. We have a global panic handler that logs the
3277 0 : // panic with its backtrace, so no need to log that here. Only log a brief
3278 0 : // message to make it clear that we returned the error to the client.
3279 0 : error!("HTTP request handler task panicked: {e:#}");
3280 :
3281 : // Don't return an Error here, because then fallback error handler that was
3282 : // installed in make_router() will print the error. Instead, construct the
3283 : // HTTP error response and return that.
3284 0 : Ok(
3285 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
3286 0 : .into_response(),
3287 0 : )
3288 : }
3289 : }
3290 0 : })
3291 0 : .await;
3292 :
3293 0 : cancel_guard.disarm();
3294 0 :
3295 0 : result
3296 0 : }
3297 :
3298 : /// Like api_handler, but returns an error response if the server is built without
3299 : /// the 'testing' feature.
3300 0 : async fn testing_api_handler<R, H>(
3301 0 : desc: &str,
3302 0 : request: Request<Body>,
3303 0 : handler: H,
3304 0 : ) -> Result<Response<Body>, ApiError>
3305 0 : where
3306 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
3307 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
3308 0 : {
3309 0 : if cfg!(feature = "testing") {
3310 0 : api_handler(request, handler).await
3311 : } else {
3312 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
3313 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
3314 0 : ))))
3315 0 : .await
3316 : }
3317 0 : }
3318 :
3319 0 : pub fn make_router(
3320 0 : state: Arc<State>,
3321 0 : launch_ts: &'static LaunchTimestamp,
3322 0 : auth: Option<Arc<SwappableJwtAuth>>,
3323 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
3324 0 : let spec = include_bytes!("openapi_spec.yml");
3325 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
3326 0 : if auth.is_some() {
3327 0 : router = router.middleware(auth_middleware(|request| {
3328 0 : let state = get_state(request);
3329 0 : if state.allowlist_routes.contains(&request.uri().path()) {
3330 0 : None
3331 : } else {
3332 0 : state.auth.as_deref()
3333 : }
3334 0 : }))
3335 0 : }
3336 :
3337 0 : router = router.middleware(
3338 0 : endpoint::add_response_header_middleware(
3339 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
3340 0 : &launch_ts.to_string(),
3341 0 : )
3342 0 : .expect("construct launch timestamp header middleware"),
3343 0 : );
3344 0 :
3345 0 : Ok(router
3346 0 : .data(state)
3347 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
3348 0 : .get("/profile/cpu", |r| request_span(r, profile_cpu_handler))
3349 0 : .get("/profile/heap", |r| request_span(r, profile_heap_handler))
3350 0 : .get("/v1/status", |r| api_handler(r, status_handler))
3351 0 : .put("/v1/failpoints", |r| {
3352 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
3353 0 : })
3354 0 : .post("/v1/reload_auth_validation_keys", |r| {
3355 0 : api_handler(r, reload_auth_validation_keys_handler)
3356 0 : })
3357 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
3358 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
3359 0 : api_handler(r, tenant_status)
3360 0 : })
3361 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
3362 0 : api_handler(r, tenant_delete_handler)
3363 0 : })
3364 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
3365 0 : api_handler(r, tenant_size_handler)
3366 0 : })
3367 0 : .patch("/v1/tenant/config", |r| {
3368 0 : api_handler(r, patch_tenant_config_handler)
3369 0 : })
3370 0 : .put("/v1/tenant/config", |r| {
3371 0 : api_handler(r, update_tenant_config_handler)
3372 0 : })
3373 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
3374 0 : api_handler(r, tenant_shard_split_handler)
3375 0 : })
3376 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
3377 0 : api_handler(r, get_tenant_config_handler)
3378 0 : })
3379 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
3380 0 : api_handler(r, put_tenant_location_config_handler)
3381 0 : })
3382 0 : .get("/v1/location_config", |r| {
3383 0 : api_handler(r, list_location_config_handler)
3384 0 : })
3385 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
3386 0 : api_handler(r, get_location_config_handler)
3387 0 : })
3388 0 : .put(
3389 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
3390 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
3391 0 : )
3392 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
3393 0 : api_handler(r, timeline_list_handler)
3394 0 : })
3395 0 : .get("/v1/tenant/:tenant_shard_id/timeline_and_offloaded", |r| {
3396 0 : api_handler(r, timeline_and_offloaded_list_handler)
3397 0 : })
3398 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
3399 0 : api_handler(r, timeline_create_handler)
3400 0 : })
3401 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
3402 0 : api_handler(r, tenant_reset_handler)
3403 0 : })
3404 0 : .post(
3405 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
3406 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
3407 0 : )
3408 0 : .put(
3409 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/archival_config",
3410 0 : |r| api_handler(r, timeline_archival_config_handler),
3411 0 : )
3412 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3413 0 : api_handler(r, timeline_detail_handler)
3414 0 : })
3415 0 : .get(
3416 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
3417 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
3418 0 : )
3419 0 : .get(
3420 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
3421 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
3422 0 : )
3423 0 : .post(
3424 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
3425 0 : |r| api_handler(r, lsn_lease_handler),
3426 0 : )
3427 0 : .put(
3428 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
3429 0 : |r| api_handler(r, timeline_gc_handler),
3430 0 : )
3431 0 : .get(
3432 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
3433 0 : |r| api_handler(r, timeline_compact_info_handler),
3434 0 : )
3435 0 : .put(
3436 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
3437 0 : |r| api_handler(r, timeline_compact_handler),
3438 0 : )
3439 0 : .delete(
3440 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
3441 0 : |r| api_handler(r, timeline_cancel_compact_handler),
3442 0 : )
3443 0 : .put(
3444 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/offload",
3445 0 : |r| testing_api_handler("attempt timeline offload", r, timeline_offload_handler),
3446 0 : )
3447 0 : .put(
3448 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
3449 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
3450 0 : )
3451 0 : .post(
3452 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
3453 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
3454 0 : )
3455 0 : .get(
3456 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
3457 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
3458 0 : )
3459 0 : .put(
3460 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
3461 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
3462 0 : )
3463 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3464 0 : api_handler(r, timeline_delete_handler)
3465 0 : })
3466 0 : .get(
3467 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
3468 0 : |r| api_handler(r, layer_map_info_handler),
3469 0 : )
3470 0 : .get(
3471 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3472 0 : |r| api_handler(r, layer_download_handler),
3473 0 : )
3474 0 : .delete(
3475 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3476 0 : |r| api_handler(r, evict_timeline_layer_handler),
3477 0 : )
3478 0 : .post(
3479 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_name/scan_disposable_keys",
3480 0 : |r| testing_api_handler("timeline_layer_scan_disposable_keys", r, timeline_layer_scan_disposable_keys),
3481 0 : )
3482 0 : .post(
3483 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
3484 0 : |r| api_handler(r, timeline_gc_blocking_handler),
3485 0 : )
3486 0 : .post(
3487 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
3488 0 : |r| api_handler(r, timeline_gc_unblocking_handler),
3489 0 : )
3490 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
3491 0 : api_handler(r, secondary_upload_handler)
3492 0 : })
3493 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
3494 0 : api_handler(r, tenant_scan_remote_handler)
3495 0 : })
3496 0 : .put("/v1/disk_usage_eviction/run", |r| {
3497 0 : api_handler(r, disk_usage_eviction_run)
3498 0 : })
3499 0 : .put("/v1/deletion_queue/flush", |r| {
3500 0 : api_handler(r, deletion_queue_flush)
3501 0 : })
3502 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
3503 0 : api_handler(r, secondary_status_handler)
3504 0 : })
3505 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
3506 0 : api_handler(r, secondary_download_handler)
3507 0 : })
3508 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
3509 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
3510 0 : })
3511 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
3512 0 : .post("/v1/tracing/event", |r| {
3513 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
3514 0 : })
3515 0 : .get(
3516 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
3517 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
3518 0 : )
3519 0 : .get(
3520 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
3521 0 : |r| api_handler(r, timeline_collect_keyspace),
3522 0 : )
3523 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
3524 0 : .put("/v1/io_mode", |r| api_handler(r, put_io_mode_handler))
3525 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
3526 0 : .post(
3527 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
3528 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
3529 0 : )
3530 0 : .post(
3531 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
3532 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
3533 0 : )
3534 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
3535 0 : .post(
3536 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
3537 0 : |r| testing_api_handler("perf_info", r, perf_info),
3538 0 : )
3539 0 : .put(
3540 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_basebackup",
3541 0 : |r| api_handler(r, put_tenant_timeline_import_basebackup),
3542 0 : )
3543 0 : .put(
3544 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_wal",
3545 0 : |r| api_handler(r, put_tenant_timeline_import_wal),
3546 0 : )
3547 0 : .any(handler_404))
3548 0 : }
|
