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