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