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 : 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 : 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 : 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 : 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 : 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 : 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 : e @ HasArchivedParent(_) => {
328 0 : ApiError::PreconditionFailed(e.to_string().into_boxed_str())
329 : }
330 0 : HasUnarchivedChildren(children) => ApiError::PreconditionFailed(
331 0 : format!(
332 0 : "Cannot archive timeline which has non-archived child timelines: {children:?}"
333 0 : )
334 0 : .into_boxed_str(),
335 0 : ),
336 0 : a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
337 0 : Other(e) => ApiError::InternalServerError(e),
338 : }
339 0 : }
340 : }
341 :
342 : impl From<crate::tenant::mgr::DeleteTimelineError> for ApiError {
343 0 : fn from(value: crate::tenant::mgr::DeleteTimelineError) -> Self {
344 : use crate::tenant::mgr::DeleteTimelineError::*;
345 0 : match value {
346 : // Report Precondition failed so client can distinguish between
347 : // "tenant is missing" case from "timeline is missing"
348 0 : Tenant(GetTenantError::NotFound(..)) => ApiError::PreconditionFailed(
349 0 : "Requested tenant is missing".to_owned().into_boxed_str(),
350 0 : ),
351 0 : Tenant(t) => ApiError::from(t),
352 0 : Timeline(t) => ApiError::from(t),
353 : }
354 0 : }
355 : }
356 :
357 : impl From<crate::tenant::mgr::DeleteTenantError> for ApiError {
358 0 : fn from(value: crate::tenant::mgr::DeleteTenantError) -> Self {
359 : use crate::tenant::mgr::DeleteTenantError::*;
360 0 : match value {
361 0 : SlotError(e) => e.into(),
362 0 : Other(o) => ApiError::InternalServerError(o),
363 0 : Cancelled => ApiError::ShuttingDown,
364 : }
365 0 : }
366 : }
367 :
368 : // Helper function to construct a TimelineInfo struct for a timeline
369 0 : async fn build_timeline_info(
370 0 : timeline: &Arc<Timeline>,
371 0 : include_non_incremental_logical_size: bool,
372 0 : force_await_initial_logical_size: bool,
373 0 : ctx: &RequestContext,
374 0 : ) -> anyhow::Result<TimelineInfo> {
375 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
376 0 :
377 0 : if force_await_initial_logical_size {
378 0 : timeline.clone().await_initial_logical_size().await
379 0 : }
380 :
381 0 : let mut info = build_timeline_info_common(
382 0 : timeline,
383 0 : ctx,
384 0 : tenant::timeline::GetLogicalSizePriority::Background,
385 0 : )
386 0 : .await?;
387 0 : if include_non_incremental_logical_size {
388 : // XXX we should be using spawn_ondemand_logical_size_calculation here.
389 : // Otherwise, if someone deletes the timeline / detaches the tenant while
390 : // we're executing this function, we will outlive the timeline on-disk state.
391 : info.current_logical_size_non_incremental = Some(
392 0 : timeline
393 0 : .get_current_logical_size_non_incremental(info.last_record_lsn, ctx)
394 0 : .await?,
395 : );
396 0 : }
397 0 : Ok(info)
398 0 : }
399 :
400 0 : async fn build_timeline_info_common(
401 0 : timeline: &Arc<Timeline>,
402 0 : ctx: &RequestContext,
403 0 : logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
404 0 : ) -> anyhow::Result<TimelineInfo> {
405 0 : crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
406 0 : let initdb_lsn = timeline.initdb_lsn;
407 0 : let last_record_lsn = timeline.get_last_record_lsn();
408 0 : let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
409 0 : let guard = timeline.last_received_wal.lock().unwrap();
410 0 : if let Some(info) = guard.as_ref() {
411 0 : (
412 0 : Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
413 0 : Some(info.last_received_msg_lsn),
414 0 : Some(info.last_received_msg_ts),
415 0 : )
416 : } else {
417 0 : (None, None, None)
418 : }
419 : };
420 :
421 0 : let ancestor_timeline_id = timeline.get_ancestor_timeline_id();
422 0 : let ancestor_lsn = match timeline.get_ancestor_lsn() {
423 0 : Lsn(0) => None,
424 0 : lsn @ Lsn(_) => Some(lsn),
425 : };
426 0 : let current_logical_size = timeline.get_current_logical_size(logical_size_task_priority, ctx);
427 0 : let current_physical_size = Some(timeline.layer_size_sum().await);
428 0 : let state = timeline.current_state();
429 0 : // Report is_archived = false if the timeline is still loading
430 0 : let is_archived = timeline.is_archived().unwrap_or(false);
431 0 : let remote_consistent_lsn_projected = timeline
432 0 : .get_remote_consistent_lsn_projected()
433 0 : .unwrap_or(Lsn(0));
434 0 : let remote_consistent_lsn_visible = timeline
435 0 : .get_remote_consistent_lsn_visible()
436 0 : .unwrap_or(Lsn(0));
437 0 :
438 0 : let walreceiver_status = timeline.walreceiver_status();
439 0 :
440 0 : let (pitr_history_size, within_ancestor_pitr) = timeline.get_pitr_history_stats();
441 :
442 0 : let info = TimelineInfo {
443 0 : tenant_id: timeline.tenant_shard_id,
444 0 : timeline_id: timeline.timeline_id,
445 0 : ancestor_timeline_id,
446 0 : ancestor_lsn,
447 0 : disk_consistent_lsn: timeline.get_disk_consistent_lsn(),
448 0 : remote_consistent_lsn: remote_consistent_lsn_projected,
449 0 : remote_consistent_lsn_visible,
450 0 : initdb_lsn,
451 0 : last_record_lsn,
452 0 : prev_record_lsn: Some(timeline.get_prev_record_lsn()),
453 0 : latest_gc_cutoff_lsn: *timeline.get_latest_gc_cutoff_lsn(),
454 0 : current_logical_size: current_logical_size.size_dont_care_about_accuracy(),
455 0 : current_logical_size_is_accurate: match current_logical_size.accuracy() {
456 0 : tenant::timeline::logical_size::Accuracy::Approximate => false,
457 0 : tenant::timeline::logical_size::Accuracy::Exact => true,
458 : },
459 0 : directory_entries_counts: timeline.get_directory_metrics().to_vec(),
460 0 : current_physical_size,
461 0 : current_logical_size_non_incremental: None,
462 0 : pitr_history_size,
463 0 : within_ancestor_pitr,
464 0 : timeline_dir_layer_file_size_sum: None,
465 0 : wal_source_connstr,
466 0 : last_received_msg_lsn,
467 0 : last_received_msg_ts,
468 0 : pg_version: timeline.pg_version,
469 0 :
470 0 : state,
471 0 : is_archived: Some(is_archived),
472 0 :
473 0 : walreceiver_status,
474 0 :
475 0 : last_aux_file_policy: timeline.last_aux_file_policy.load(),
476 0 : };
477 0 : Ok(info)
478 0 : }
479 :
480 : // healthcheck handler
481 0 : async fn status_handler(
482 0 : request: Request<Body>,
483 0 : _cancel: CancellationToken,
484 0 : ) -> Result<Response<Body>, ApiError> {
485 0 : check_permission(&request, None)?;
486 0 : let config = get_config(&request);
487 0 : json_response(StatusCode::OK, StatusResponse { id: config.id })
488 0 : }
489 :
490 0 : async fn reload_auth_validation_keys_handler(
491 0 : request: Request<Body>,
492 0 : _cancel: CancellationToken,
493 0 : ) -> Result<Response<Body>, ApiError> {
494 0 : check_permission(&request, None)?;
495 0 : let config = get_config(&request);
496 0 : let state = get_state(&request);
497 0 : let Some(shared_auth) = &state.auth else {
498 0 : return json_response(StatusCode::BAD_REQUEST, ());
499 : };
500 : // unwrap is ok because check is performed when creating config, so path is set and exists
501 0 : let key_path = config.auth_validation_public_key_path.as_ref().unwrap();
502 0 : info!("Reloading public key(s) for verifying JWT tokens from {key_path:?}");
503 :
504 0 : match JwtAuth::from_key_path(key_path) {
505 0 : Ok(new_auth) => {
506 0 : shared_auth.swap(new_auth);
507 0 : json_response(StatusCode::OK, ())
508 : }
509 0 : Err(e) => {
510 0 : let err_msg = "Error reloading public keys";
511 0 : warn!("Error reloading public keys from {key_path:?}: {e:}");
512 0 : json_response(
513 0 : StatusCode::INTERNAL_SERVER_ERROR,
514 0 : HttpErrorBody::from_msg(err_msg.to_string()),
515 0 : )
516 : }
517 : }
518 0 : }
519 :
520 0 : async fn timeline_create_handler(
521 0 : mut request: Request<Body>,
522 0 : _cancel: CancellationToken,
523 0 : ) -> Result<Response<Body>, ApiError> {
524 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
525 0 : let request_data: TimelineCreateRequest = json_request(&mut request).await?;
526 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
527 :
528 0 : let new_timeline_id = request_data.new_timeline_id;
529 0 :
530 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
531 0 :
532 0 : let state = get_state(&request);
533 :
534 0 : async {
535 0 : let tenant = state
536 0 : .tenant_manager
537 0 : .get_attached_tenant_shard(tenant_shard_id)?;
538 :
539 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
540 :
541 0 : if let Some(ancestor_id) = request_data.ancestor_timeline_id.as_ref() {
542 0 : tracing::info!(%ancestor_id, "starting to branch");
543 : } else {
544 0 : tracing::info!("bootstrapping");
545 : }
546 :
547 0 : match tenant
548 0 : .create_timeline(
549 0 : new_timeline_id,
550 0 : request_data.ancestor_timeline_id,
551 0 : request_data.ancestor_start_lsn,
552 0 : request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
553 0 : request_data.existing_initdb_timeline_id,
554 0 : state.broker_client.clone(),
555 0 : &ctx,
556 0 : )
557 0 : .await
558 : {
559 0 : Ok(new_timeline) => {
560 : // Created. Construct a TimelineInfo for it.
561 0 : let timeline_info = build_timeline_info_common(
562 0 : &new_timeline,
563 0 : &ctx,
564 0 : tenant::timeline::GetLogicalSizePriority::User,
565 0 : )
566 0 : .await
567 0 : .map_err(ApiError::InternalServerError)?;
568 0 : json_response(StatusCode::CREATED, timeline_info)
569 : }
570 0 : Err(_) if tenant.cancel.is_cancelled() => {
571 0 : // In case we get some ugly error type during shutdown, cast it into a clean 503.
572 0 : json_response(
573 0 : StatusCode::SERVICE_UNAVAILABLE,
574 0 : HttpErrorBody::from_msg("Tenant shutting down".to_string()),
575 0 : )
576 : }
577 0 : Err(e @ tenant::CreateTimelineError::Conflict) => {
578 0 : json_response(StatusCode::CONFLICT, HttpErrorBody::from_msg(e.to_string()))
579 : }
580 0 : Err(e @ tenant::CreateTimelineError::AlreadyCreating) => json_response(
581 0 : StatusCode::TOO_MANY_REQUESTS,
582 0 : HttpErrorBody::from_msg(e.to_string()),
583 0 : ),
584 0 : Err(tenant::CreateTimelineError::AncestorLsn(err)) => json_response(
585 0 : StatusCode::NOT_ACCEPTABLE,
586 0 : HttpErrorBody::from_msg(format!("{err:#}")),
587 0 : ),
588 0 : Err(e @ tenant::CreateTimelineError::AncestorNotActive) => json_response(
589 0 : StatusCode::SERVICE_UNAVAILABLE,
590 0 : HttpErrorBody::from_msg(e.to_string()),
591 0 : ),
592 0 : Err(tenant::CreateTimelineError::ShuttingDown) => json_response(
593 0 : StatusCode::SERVICE_UNAVAILABLE,
594 0 : HttpErrorBody::from_msg("tenant shutting down".to_string()),
595 0 : ),
596 0 : Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
597 : }
598 0 : }
599 0 : .instrument(info_span!("timeline_create",
600 : tenant_id = %tenant_shard_id.tenant_id,
601 0 : shard_id = %tenant_shard_id.shard_slug(),
602 : timeline_id = %new_timeline_id,
603 : lsn=?request_data.ancestor_start_lsn,
604 : pg_version=?request_data.pg_version
605 : ))
606 0 : .await
607 0 : }
608 :
609 0 : async fn timeline_list_handler(
610 0 : request: Request<Body>,
611 0 : _cancel: CancellationToken,
612 0 : ) -> Result<Response<Body>, ApiError> {
613 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
614 0 : let include_non_incremental_logical_size: Option<bool> =
615 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
616 0 : let force_await_initial_logical_size: Option<bool> =
617 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
618 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
619 :
620 0 : let state = get_state(&request);
621 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
622 :
623 0 : let response_data = async {
624 0 : let tenant = state
625 0 : .tenant_manager
626 0 : .get_attached_tenant_shard(tenant_shard_id)?;
627 :
628 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
629 :
630 0 : let timelines = tenant.list_timelines();
631 0 :
632 0 : let mut response_data = Vec::with_capacity(timelines.len());
633 0 : for timeline in timelines {
634 0 : let timeline_info = build_timeline_info(
635 0 : &timeline,
636 0 : include_non_incremental_logical_size.unwrap_or(false),
637 0 : force_await_initial_logical_size.unwrap_or(false),
638 0 : &ctx,
639 0 : )
640 0 : .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
641 0 : .await
642 0 : .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
643 0 : .map_err(ApiError::InternalServerError)?;
644 :
645 0 : response_data.push(timeline_info);
646 : }
647 0 : Ok::<Vec<TimelineInfo>, ApiError>(response_data)
648 0 : }
649 0 : .instrument(info_span!("timeline_list",
650 : tenant_id = %tenant_shard_id.tenant_id,
651 0 : shard_id = %tenant_shard_id.shard_slug()))
652 0 : .await?;
653 :
654 0 : json_response(StatusCode::OK, response_data)
655 0 : }
656 :
657 0 : async fn timeline_preserve_initdb_handler(
658 0 : request: Request<Body>,
659 0 : _cancel: CancellationToken,
660 0 : ) -> Result<Response<Body>, ApiError> {
661 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
662 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
663 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
664 0 : let state = get_state(&request);
665 :
666 : // Part of the process for disaster recovery from safekeeper-stored WAL:
667 : // If we don't recover into a new timeline but want to keep the timeline ID,
668 : // then the initdb archive is deleted. This endpoint copies it to a different
669 : // location where timeline recreation cand find it.
670 :
671 0 : async {
672 0 : let tenant = state
673 0 : .tenant_manager
674 0 : .get_attached_tenant_shard(tenant_shard_id)?;
675 :
676 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
677 :
678 0 : timeline
679 0 : .preserve_initdb_archive()
680 0 : .await
681 0 : .context("preserving initdb archive")
682 0 : .map_err(ApiError::InternalServerError)?;
683 :
684 0 : Ok::<_, ApiError>(())
685 0 : }
686 0 : .instrument(info_span!("timeline_preserve_initdb_archive",
687 : tenant_id = %tenant_shard_id.tenant_id,
688 0 : shard_id = %tenant_shard_id.shard_slug(),
689 : %timeline_id))
690 0 : .await?;
691 :
692 0 : json_response(StatusCode::OK, ())
693 0 : }
694 :
695 0 : async fn timeline_archival_config_handler(
696 0 : mut request: Request<Body>,
697 0 : _cancel: CancellationToken,
698 0 : ) -> Result<Response<Body>, ApiError> {
699 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
700 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
701 :
702 0 : let request_data: TimelineArchivalConfigRequest = json_request(&mut request).await?;
703 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
704 0 : let state = get_state(&request);
705 :
706 0 : async {
707 0 : let tenant = state
708 0 : .tenant_manager
709 0 : .get_attached_tenant_shard(tenant_shard_id)?;
710 :
711 0 : tenant
712 0 : .apply_timeline_archival_config(timeline_id, request_data.state)
713 0 : .await?;
714 0 : Ok::<_, ApiError>(())
715 0 : }
716 0 : .instrument(info_span!("timeline_archival_config",
717 : tenant_id = %tenant_shard_id.tenant_id,
718 0 : shard_id = %tenant_shard_id.shard_slug(),
719 : state = ?request_data.state,
720 : %timeline_id))
721 0 : .await?;
722 :
723 0 : json_response(StatusCode::OK, ())
724 0 : }
725 :
726 0 : async fn timeline_detail_handler(
727 0 : request: Request<Body>,
728 0 : _cancel: CancellationToken,
729 0 : ) -> Result<Response<Body>, ApiError> {
730 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
731 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
732 0 : let include_non_incremental_logical_size: Option<bool> =
733 0 : parse_query_param(&request, "include-non-incremental-logical-size")?;
734 0 : let force_await_initial_logical_size: Option<bool> =
735 0 : parse_query_param(&request, "force-await-initial-logical-size")?;
736 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
737 :
738 : // Logical size calculation needs downloading.
739 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
740 0 : let state = get_state(&request);
741 :
742 0 : let timeline_info = async {
743 0 : let tenant = state
744 0 : .tenant_manager
745 0 : .get_attached_tenant_shard(tenant_shard_id)?;
746 :
747 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
748 :
749 0 : let timeline = tenant.get_timeline(timeline_id, false)?;
750 :
751 0 : let timeline_info = build_timeline_info(
752 0 : &timeline,
753 0 : include_non_incremental_logical_size.unwrap_or(false),
754 0 : force_await_initial_logical_size.unwrap_or(false),
755 0 : &ctx,
756 0 : )
757 0 : .await
758 0 : .context("get local timeline info")
759 0 : .map_err(ApiError::InternalServerError)?;
760 :
761 0 : Ok::<_, ApiError>(timeline_info)
762 0 : }
763 0 : .instrument(info_span!("timeline_detail",
764 : tenant_id = %tenant_shard_id.tenant_id,
765 0 : shard_id = %tenant_shard_id.shard_slug(),
766 : %timeline_id))
767 0 : .await?;
768 :
769 0 : json_response(StatusCode::OK, timeline_info)
770 0 : }
771 :
772 0 : async fn get_lsn_by_timestamp_handler(
773 0 : request: Request<Body>,
774 0 : cancel: CancellationToken,
775 0 : ) -> Result<Response<Body>, ApiError> {
776 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
777 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
778 0 : let state = get_state(&request);
779 0 :
780 0 : if !tenant_shard_id.is_shard_zero() {
781 : // Requires SLRU contents, which are only stored on shard zero
782 0 : return Err(ApiError::BadRequest(anyhow!(
783 0 : "Size calculations are only available on shard zero"
784 0 : )));
785 0 : }
786 :
787 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
788 0 : let timestamp_raw = must_get_query_param(&request, "timestamp")?;
789 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
790 0 : .with_context(|| format!("Invalid time: {:?}", timestamp_raw))
791 0 : .map_err(ApiError::BadRequest)?;
792 0 : let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
793 :
794 0 : let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
795 0 :
796 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
797 :
798 0 : let timeline =
799 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
800 0 : .await?;
801 0 : let result = timeline
802 0 : .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
803 0 : .await?;
804 :
805 : #[derive(serde::Serialize, Debug)]
806 : struct Result {
807 : lsn: Lsn,
808 : kind: &'static str,
809 : #[serde(default)]
810 : #[serde(skip_serializing_if = "Option::is_none")]
811 : #[serde(flatten)]
812 : lease: Option<LsnLease>,
813 : }
814 0 : let (lsn, kind) = match result {
815 0 : LsnForTimestamp::Present(lsn) => (lsn, "present"),
816 0 : LsnForTimestamp::Future(lsn) => (lsn, "future"),
817 0 : LsnForTimestamp::Past(lsn) => (lsn, "past"),
818 0 : LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
819 : };
820 :
821 0 : let lease = if with_lease {
822 0 : timeline
823 0 : .make_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
824 0 : .inspect_err(|_| {
825 0 : warn!("fail to grant a lease to {}", lsn);
826 0 : })
827 0 : .ok()
828 : } else {
829 0 : None
830 : };
831 :
832 0 : let result = Result { lsn, kind, lease };
833 0 : let valid_until = result
834 0 : .lease
835 0 : .as_ref()
836 0 : .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
837 0 : tracing::info!(
838 : lsn=?result.lsn,
839 : kind=%result.kind,
840 : timestamp=%timestamp_raw,
841 : valid_until=?valid_until,
842 0 : "lsn_by_timestamp finished"
843 : );
844 0 : json_response(StatusCode::OK, result)
845 0 : }
846 :
847 0 : async fn get_timestamp_of_lsn_handler(
848 0 : request: Request<Body>,
849 0 : _cancel: CancellationToken,
850 0 : ) -> Result<Response<Body>, ApiError> {
851 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
852 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
853 0 : let state = get_state(&request);
854 0 :
855 0 : if !tenant_shard_id.is_shard_zero() {
856 : // Requires SLRU contents, which are only stored on shard zero
857 0 : return Err(ApiError::BadRequest(anyhow!(
858 0 : "Size calculations are only available on shard zero"
859 0 : )));
860 0 : }
861 :
862 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
863 :
864 0 : let lsn_str = must_get_query_param(&request, "lsn")?;
865 0 : let lsn = Lsn::from_str(&lsn_str)
866 0 : .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
867 0 : .map_err(ApiError::BadRequest)?;
868 :
869 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
870 0 : let timeline =
871 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
872 0 : .await?;
873 0 : let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
874 :
875 0 : match result {
876 0 : Some(time) => {
877 0 : let time = format_rfc3339(
878 0 : postgres_ffi::try_from_pg_timestamp(time).map_err(ApiError::InternalServerError)?,
879 : )
880 0 : .to_string();
881 0 : json_response(StatusCode::OK, time)
882 : }
883 0 : None => Err(ApiError::NotFound(
884 0 : anyhow::anyhow!("Timestamp for lsn {} not found", lsn).into(),
885 0 : )),
886 : }
887 0 : }
888 :
889 0 : async fn timeline_delete_handler(
890 0 : request: Request<Body>,
891 0 : _cancel: CancellationToken,
892 0 : ) -> Result<Response<Body>, ApiError> {
893 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
894 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
895 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
896 :
897 0 : let state = get_state(&request);
898 :
899 0 : let tenant = state
900 0 : .tenant_manager
901 0 : .get_attached_tenant_shard(tenant_shard_id)
902 0 : .map_err(|e| {
903 0 : match e {
904 : // GetTenantError has a built-in conversion to ApiError, but in this context we don't
905 : // want to treat missing tenants as 404, to avoid ambiguity with successful deletions.
906 0 : GetTenantError::NotFound(_) => ApiError::PreconditionFailed(
907 0 : "Requested tenant is missing".to_string().into_boxed_str(),
908 0 : ),
909 0 : e => e.into(),
910 : }
911 0 : })?;
912 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
913 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))
914 0 : .await?;
915 :
916 0 : json_response(StatusCode::ACCEPTED, ())
917 0 : }
918 :
919 0 : async fn tenant_reset_handler(
920 0 : request: Request<Body>,
921 0 : _cancel: CancellationToken,
922 0 : ) -> Result<Response<Body>, ApiError> {
923 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
924 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
925 :
926 0 : let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
927 :
928 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
929 0 : let state = get_state(&request);
930 0 : state
931 0 : .tenant_manager
932 0 : .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
933 0 : .await
934 0 : .map_err(ApiError::InternalServerError)?;
935 :
936 0 : json_response(StatusCode::OK, ())
937 0 : }
938 :
939 0 : async fn tenant_list_handler(
940 0 : request: Request<Body>,
941 0 : _cancel: CancellationToken,
942 0 : ) -> Result<Response<Body>, ApiError> {
943 0 : check_permission(&request, None)?;
944 0 : let state = get_state(&request);
945 :
946 0 : let response_data = state
947 0 : .tenant_manager
948 0 : .list_tenants()
949 0 : .map_err(|_| {
950 0 : ApiError::ResourceUnavailable("Tenant map is initializing or shutting down".into())
951 0 : })?
952 0 : .iter()
953 0 : .map(|(id, state, gen)| TenantInfo {
954 0 : id: *id,
955 0 : state: state.clone(),
956 0 : current_physical_size: None,
957 0 : attachment_status: state.attachment_status(),
958 0 : generation: (*gen)
959 0 : .into()
960 0 : .expect("Tenants are always attached with a generation"),
961 0 : gc_blocking: None,
962 0 : })
963 0 : .collect::<Vec<TenantInfo>>();
964 0 :
965 0 : json_response(StatusCode::OK, response_data)
966 0 : }
967 :
968 0 : async fn tenant_status(
969 0 : request: Request<Body>,
970 0 : _cancel: CancellationToken,
971 0 : ) -> Result<Response<Body>, ApiError> {
972 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
973 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
974 0 : let state = get_state(&request);
975 0 :
976 0 : // In tests, sometimes we want to query the state of a tenant without auto-activating it if it's currently waiting.
977 0 : let activate = true;
978 : #[cfg(feature = "testing")]
979 0 : let activate = parse_query_param(&request, "activate")?.unwrap_or(activate);
980 :
981 0 : let tenant_info = async {
982 0 : let tenant = state
983 0 : .tenant_manager
984 0 : .get_attached_tenant_shard(tenant_shard_id)?;
985 :
986 0 : if activate {
987 : // This is advisory: we prefer to let the tenant activate on-demand when this function is
988 : // called, but it is still valid to return 200 and describe the current state of the tenant
989 : // if it doesn't make it into an active state.
990 0 : tenant
991 0 : .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
992 0 : .await
993 0 : .ok();
994 0 : }
995 :
996 : // Calculate total physical size of all timelines
997 0 : let mut current_physical_size = 0;
998 0 : for timeline in tenant.list_timelines().iter() {
999 0 : current_physical_size += timeline.layer_size_sum().await;
1000 : }
1001 :
1002 0 : let state = tenant.current_state();
1003 0 : Result::<_, ApiError>::Ok(TenantDetails {
1004 0 : tenant_info: TenantInfo {
1005 0 : id: tenant_shard_id,
1006 0 : state: state.clone(),
1007 0 : current_physical_size: Some(current_physical_size),
1008 0 : attachment_status: state.attachment_status(),
1009 0 : generation: tenant
1010 0 : .generation()
1011 0 : .into()
1012 0 : .expect("Tenants are always attached with a generation"),
1013 0 : gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
1014 0 : },
1015 0 : walredo: tenant.wal_redo_manager_status(),
1016 0 : timelines: tenant.list_timeline_ids(),
1017 0 : })
1018 0 : }
1019 0 : .instrument(info_span!("tenant_status_handler",
1020 : tenant_id = %tenant_shard_id.tenant_id,
1021 0 : shard_id = %tenant_shard_id.shard_slug()))
1022 0 : .await?;
1023 :
1024 0 : json_response(StatusCode::OK, tenant_info)
1025 0 : }
1026 :
1027 0 : async fn tenant_delete_handler(
1028 0 : request: Request<Body>,
1029 0 : _cancel: CancellationToken,
1030 0 : ) -> Result<Response<Body>, ApiError> {
1031 : // TODO openapi spec
1032 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1033 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1034 :
1035 0 : let state = get_state(&request);
1036 0 :
1037 0 : state
1038 0 : .tenant_manager
1039 0 : .delete_tenant(tenant_shard_id)
1040 0 : .instrument(info_span!("tenant_delete_handler",
1041 : tenant_id = %tenant_shard_id.tenant_id,
1042 0 : shard_id = %tenant_shard_id.shard_slug()
1043 : ))
1044 0 : .await?;
1045 :
1046 0 : json_response(StatusCode::OK, ())
1047 0 : }
1048 :
1049 : /// HTTP endpoint to query the current tenant_size of a tenant.
1050 : ///
1051 : /// This is not used by consumption metrics under [`crate::consumption_metrics`], but can be used
1052 : /// to debug any of the calculations. Requires `tenant_id` request parameter, supports
1053 : /// `inputs_only=true|false` (default false) which supports debugging failure to calculate model
1054 : /// values.
1055 : ///
1056 : /// 'retention_period' query parameter overrides the cutoff that is used to calculate the size
1057 : /// (only if it is shorter than the real cutoff).
1058 : ///
1059 : /// Note: we don't update the cached size and prometheus metric here.
1060 : /// The retention period might be different, and it's nice to have a method to just calculate it
1061 : /// without modifying anything anyway.
1062 0 : async fn tenant_size_handler(
1063 0 : request: Request<Body>,
1064 0 : cancel: CancellationToken,
1065 0 : ) -> Result<Response<Body>, ApiError> {
1066 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1067 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1068 0 : let inputs_only: Option<bool> = parse_query_param(&request, "inputs_only")?;
1069 0 : let retention_period: Option<u64> = parse_query_param(&request, "retention_period")?;
1070 0 : let headers = request.headers();
1071 0 : let state = get_state(&request);
1072 0 :
1073 0 : if !tenant_shard_id.is_shard_zero() {
1074 0 : return Err(ApiError::BadRequest(anyhow!(
1075 0 : "Size calculations are only available on shard zero"
1076 0 : )));
1077 0 : }
1078 0 :
1079 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1080 0 : let tenant = state
1081 0 : .tenant_manager
1082 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1083 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1084 :
1085 : // this can be long operation
1086 0 : let inputs = tenant
1087 0 : .gather_size_inputs(
1088 0 : retention_period,
1089 0 : LogicalSizeCalculationCause::TenantSizeHandler,
1090 0 : &cancel,
1091 0 : &ctx,
1092 0 : )
1093 0 : .await
1094 0 : .map_err(|e| match e {
1095 0 : crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
1096 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
1097 0 : })?;
1098 :
1099 0 : let mut sizes = None;
1100 0 : let accepts_html = headers
1101 0 : .get(header::ACCEPT)
1102 0 : .map(|v| v == "text/html")
1103 0 : .unwrap_or_default();
1104 0 : if !inputs_only.unwrap_or(false) {
1105 0 : let storage_model = inputs.calculate_model();
1106 0 : let size = storage_model.calculate();
1107 0 :
1108 0 : // If request header expects html, return html
1109 0 : if accepts_html {
1110 0 : return synthetic_size_html_response(inputs, storage_model, size);
1111 0 : }
1112 0 : sizes = Some(size);
1113 0 : } else if accepts_html {
1114 0 : return Err(ApiError::BadRequest(anyhow!(
1115 0 : "inputs_only parameter is incompatible with html output request"
1116 0 : )));
1117 0 : }
1118 :
1119 : /// The type resides in the pageserver not to expose `ModelInputs`.
1120 : #[derive(serde::Serialize)]
1121 : struct TenantHistorySize {
1122 : id: TenantId,
1123 : /// Size is a mixture of WAL and logical size, so the unit is bytes.
1124 : ///
1125 : /// Will be none if `?inputs_only=true` was given.
1126 : size: Option<u64>,
1127 : /// Size of each segment used in the model.
1128 : /// Will be null if `?inputs_only=true` was given.
1129 : segment_sizes: Option<Vec<tenant_size_model::SegmentSizeResult>>,
1130 : inputs: crate::tenant::size::ModelInputs,
1131 : }
1132 :
1133 0 : json_response(
1134 0 : StatusCode::OK,
1135 0 : TenantHistorySize {
1136 0 : id: tenant_shard_id.tenant_id,
1137 0 : size: sizes.as_ref().map(|x| x.total_size),
1138 0 : segment_sizes: sizes.map(|x| x.segments),
1139 0 : inputs,
1140 0 : },
1141 0 : )
1142 0 : }
1143 :
1144 0 : async fn tenant_shard_split_handler(
1145 0 : mut request: Request<Body>,
1146 0 : _cancel: CancellationToken,
1147 0 : ) -> Result<Response<Body>, ApiError> {
1148 0 : let req: TenantShardSplitRequest = json_request(&mut request).await?;
1149 :
1150 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1151 0 : let state = get_state(&request);
1152 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1153 :
1154 0 : let tenant = state
1155 0 : .tenant_manager
1156 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1157 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1158 :
1159 0 : let new_shards = state
1160 0 : .tenant_manager
1161 0 : .shard_split(
1162 0 : tenant,
1163 0 : ShardCount::new(req.new_shard_count),
1164 0 : req.new_stripe_size,
1165 0 : &ctx,
1166 0 : )
1167 0 : .await
1168 0 : .map_err(ApiError::InternalServerError)?;
1169 :
1170 0 : json_response(StatusCode::OK, TenantShardSplitResponse { new_shards })
1171 0 : }
1172 :
1173 0 : async fn layer_map_info_handler(
1174 0 : request: Request<Body>,
1175 0 : _cancel: CancellationToken,
1176 0 : ) -> Result<Response<Body>, ApiError> {
1177 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1178 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1179 0 : let reset: LayerAccessStatsReset =
1180 0 : parse_query_param(&request, "reset")?.unwrap_or(LayerAccessStatsReset::NoReset);
1181 0 : let state = get_state(&request);
1182 0 :
1183 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1184 :
1185 0 : let timeline =
1186 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1187 0 : .await?;
1188 0 : let layer_map_info = timeline
1189 0 : .layer_map_info(reset)
1190 0 : .await
1191 0 : .map_err(|_shutdown| ApiError::ShuttingDown)?;
1192 :
1193 0 : json_response(StatusCode::OK, layer_map_info)
1194 0 : }
1195 :
1196 0 : async fn layer_download_handler(
1197 0 : request: Request<Body>,
1198 0 : _cancel: CancellationToken,
1199 0 : ) -> Result<Response<Body>, ApiError> {
1200 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1201 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1202 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1203 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1204 0 : let layer_name = LayerName::from_str(layer_file_name)
1205 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1206 0 : let state = get_state(&request);
1207 :
1208 0 : let timeline =
1209 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1210 0 : .await?;
1211 0 : let downloaded = timeline
1212 0 : .download_layer(&layer_name)
1213 0 : .await
1214 0 : .map_err(ApiError::InternalServerError)?;
1215 :
1216 0 : match downloaded {
1217 0 : Some(true) => json_response(StatusCode::OK, ()),
1218 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1219 0 : None => json_response(
1220 0 : StatusCode::BAD_REQUEST,
1221 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1222 0 : ),
1223 : }
1224 0 : }
1225 :
1226 0 : async fn evict_timeline_layer_handler(
1227 0 : request: Request<Body>,
1228 0 : _cancel: CancellationToken,
1229 0 : ) -> Result<Response<Body>, ApiError> {
1230 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1231 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1232 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1233 0 : let layer_file_name = get_request_param(&request, "layer_file_name")?;
1234 0 : let state = get_state(&request);
1235 :
1236 0 : let layer_name = LayerName::from_str(layer_file_name)
1237 0 : .map_err(|s| ApiError::BadRequest(anyhow::anyhow!(s)))?;
1238 :
1239 0 : let timeline =
1240 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1241 0 : .await?;
1242 0 : let evicted = timeline
1243 0 : .evict_layer(&layer_name)
1244 0 : .await
1245 0 : .map_err(ApiError::InternalServerError)?;
1246 :
1247 0 : match evicted {
1248 0 : Some(true) => json_response(StatusCode::OK, ()),
1249 0 : Some(false) => json_response(StatusCode::NOT_MODIFIED, ()),
1250 0 : None => json_response(
1251 0 : StatusCode::BAD_REQUEST,
1252 0 : format!("Layer {tenant_shard_id}/{timeline_id}/{layer_file_name} not found"),
1253 0 : ),
1254 : }
1255 0 : }
1256 :
1257 0 : async fn timeline_gc_blocking_handler(
1258 0 : request: Request<Body>,
1259 0 : _cancel: CancellationToken,
1260 0 : ) -> Result<Response<Body>, ApiError> {
1261 0 : block_or_unblock_gc(request, true).await
1262 0 : }
1263 :
1264 0 : async fn timeline_gc_unblocking_handler(
1265 0 : request: Request<Body>,
1266 0 : _cancel: CancellationToken,
1267 0 : ) -> Result<Response<Body>, ApiError> {
1268 0 : block_or_unblock_gc(request, false).await
1269 0 : }
1270 :
1271 : /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
1272 : ///
1273 : /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
1274 0 : async fn block_or_unblock_gc(
1275 0 : request: Request<Body>,
1276 0 : block: bool,
1277 0 : ) -> Result<Response<Body>, ApiError> {
1278 : use crate::tenant::{
1279 : remote_timeline_client::WaitCompletionError, upload_queue::NotInitialized,
1280 : };
1281 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1282 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1283 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1284 0 : let state = get_state(&request);
1285 :
1286 0 : let tenant = state
1287 0 : .tenant_manager
1288 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1289 :
1290 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1291 :
1292 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1293 :
1294 0 : let fut = async {
1295 0 : if block {
1296 0 : timeline.block_gc(&tenant).await.map(|_| ())
1297 : } else {
1298 0 : timeline.unblock_gc(&tenant).await
1299 : }
1300 0 : };
1301 :
1302 0 : let span = tracing::info_span!(
1303 : "block_or_unblock_gc",
1304 : tenant_id = %tenant_shard_id.tenant_id,
1305 0 : shard_id = %tenant_shard_id.shard_slug(),
1306 : timeline_id = %timeline_id,
1307 : block = block,
1308 : );
1309 :
1310 0 : let res = fut.instrument(span).await;
1311 :
1312 0 : res.map_err(|e| {
1313 0 : if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
1314 0 : ApiError::ShuttingDown
1315 : } else {
1316 0 : ApiError::InternalServerError(e)
1317 : }
1318 0 : })?;
1319 :
1320 0 : json_response(StatusCode::OK, ())
1321 0 : }
1322 :
1323 : /// Get tenant_size SVG graph along with the JSON data.
1324 0 : fn synthetic_size_html_response(
1325 0 : inputs: ModelInputs,
1326 0 : storage_model: StorageModel,
1327 0 : sizes: SizeResult,
1328 0 : ) -> Result<Response<Body>, ApiError> {
1329 0 : let mut timeline_ids: Vec<String> = Vec::new();
1330 0 : let mut timeline_map: HashMap<TimelineId, usize> = HashMap::new();
1331 0 : for (index, ti) in inputs.timeline_inputs.iter().enumerate() {
1332 0 : timeline_map.insert(ti.timeline_id, index);
1333 0 : timeline_ids.push(ti.timeline_id.to_string());
1334 0 : }
1335 0 : let seg_to_branch: Vec<(usize, SvgBranchKind)> = inputs
1336 0 : .segments
1337 0 : .iter()
1338 0 : .map(|seg| {
1339 0 : (
1340 0 : *timeline_map.get(&seg.timeline_id).unwrap(),
1341 0 : seg.kind.into(),
1342 0 : )
1343 0 : })
1344 0 : .collect();
1345 :
1346 0 : let svg =
1347 0 : tenant_size_model::svg::draw_svg(&storage_model, &timeline_ids, &seg_to_branch, &sizes)
1348 0 : .map_err(ApiError::InternalServerError)?;
1349 :
1350 0 : let mut response = String::new();
1351 :
1352 : use std::fmt::Write;
1353 0 : write!(response, "<html>\n<body>\n").unwrap();
1354 0 : write!(response, "<div>\n{svg}\n</div>").unwrap();
1355 0 : writeln!(response, "Project size: {}", sizes.total_size).unwrap();
1356 0 : writeln!(response, "<pre>").unwrap();
1357 0 : writeln!(
1358 0 : response,
1359 0 : "{}",
1360 0 : serde_json::to_string_pretty(&inputs).unwrap()
1361 0 : )
1362 0 : .unwrap();
1363 0 : writeln!(
1364 0 : response,
1365 0 : "{}",
1366 0 : serde_json::to_string_pretty(&sizes.segments).unwrap()
1367 0 : )
1368 0 : .unwrap();
1369 0 : writeln!(response, "</pre>").unwrap();
1370 0 : write!(response, "</body>\n</html>\n").unwrap();
1371 0 :
1372 0 : html_response(StatusCode::OK, response)
1373 0 : }
1374 :
1375 0 : pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>, ApiError> {
1376 0 : let response = Response::builder()
1377 0 : .status(status)
1378 0 : .header(header::CONTENT_TYPE, "text/html")
1379 0 : .body(Body::from(data.as_bytes().to_vec()))
1380 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1381 0 : Ok(response)
1382 0 : }
1383 :
1384 0 : async fn get_tenant_config_handler(
1385 0 : request: Request<Body>,
1386 0 : _cancel: CancellationToken,
1387 0 : ) -> Result<Response<Body>, ApiError> {
1388 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1389 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1390 0 : let state = get_state(&request);
1391 :
1392 0 : let tenant = state
1393 0 : .tenant_manager
1394 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1395 :
1396 0 : let response = HashMap::from([
1397 : (
1398 : "tenant_specific_overrides",
1399 0 : serde_json::to_value(tenant.tenant_specific_overrides())
1400 0 : .context("serializing tenant specific overrides")
1401 0 : .map_err(ApiError::InternalServerError)?,
1402 : ),
1403 : (
1404 0 : "effective_config",
1405 0 : serde_json::to_value(tenant.effective_config())
1406 0 : .context("serializing effective config")
1407 0 : .map_err(ApiError::InternalServerError)?,
1408 : ),
1409 : ]);
1410 :
1411 0 : json_response(StatusCode::OK, response)
1412 0 : }
1413 :
1414 0 : async fn update_tenant_config_handler(
1415 0 : mut request: Request<Body>,
1416 0 : _cancel: CancellationToken,
1417 0 : ) -> Result<Response<Body>, ApiError> {
1418 0 : let request_data: TenantConfigRequest = json_request(&mut request).await?;
1419 0 : let tenant_id = request_data.tenant_id;
1420 0 : check_permission(&request, Some(tenant_id))?;
1421 :
1422 0 : let new_tenant_conf =
1423 0 : TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1424 :
1425 0 : let state = get_state(&request);
1426 0 :
1427 0 : let tenant_shard_id = TenantShardId::unsharded(tenant_id);
1428 :
1429 0 : let tenant = state
1430 0 : .tenant_manager
1431 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1432 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1433 :
1434 : // This is a legacy API that only operates on attached tenants: the preferred
1435 : // API to use is the location_config/ endpoint, which lets the caller provide
1436 : // the full LocationConf.
1437 0 : let location_conf = LocationConf::attached_single(
1438 0 : new_tenant_conf.clone(),
1439 0 : tenant.get_generation(),
1440 0 : &ShardParameters::default(),
1441 0 : );
1442 0 :
1443 0 : crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
1444 0 : .await
1445 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
1446 0 : tenant.set_new_tenant_config(new_tenant_conf);
1447 0 :
1448 0 : json_response(StatusCode::OK, ())
1449 0 : }
1450 :
1451 0 : async fn put_tenant_location_config_handler(
1452 0 : mut request: Request<Body>,
1453 0 : _cancel: CancellationToken,
1454 0 : ) -> Result<Response<Body>, ApiError> {
1455 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1456 :
1457 0 : let request_data: TenantLocationConfigRequest = json_request(&mut request).await?;
1458 0 : let flush = parse_query_param(&request, "flush_ms")?.map(Duration::from_millis);
1459 0 : let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
1460 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1461 :
1462 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
1463 0 : let state = get_state(&request);
1464 0 : let conf = state.conf;
1465 0 :
1466 0 : // The `Detached` state is special, it doesn't upsert a tenant, it removes
1467 0 : // its local disk content and drops it from memory.
1468 0 : if let LocationConfigMode::Detached = request_data.config.mode {
1469 0 : if let Err(e) = state
1470 0 : .tenant_manager
1471 0 : .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
1472 0 : .instrument(info_span!("tenant_detach",
1473 : tenant_id = %tenant_shard_id.tenant_id,
1474 0 : shard_id = %tenant_shard_id.shard_slug()
1475 : ))
1476 0 : .await
1477 : {
1478 0 : match e {
1479 0 : TenantStateError::SlotError(TenantSlotError::NotFound(_)) => {
1480 0 : // This API is idempotent: a NotFound on a detach is fine.
1481 0 : }
1482 0 : _ => return Err(e.into()),
1483 : }
1484 0 : }
1485 0 : return json_response(StatusCode::OK, ());
1486 0 : }
1487 :
1488 0 : let location_conf =
1489 0 : LocationConf::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
1490 :
1491 : // lazy==true queues up for activation or jumps the queue like normal when a compute connects,
1492 : // similar to at startup ordering.
1493 0 : let spawn_mode = if lazy {
1494 0 : tenant::SpawnMode::Lazy
1495 : } else {
1496 0 : tenant::SpawnMode::Eager
1497 : };
1498 :
1499 0 : let tenant = state
1500 0 : .tenant_manager
1501 0 : .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
1502 0 : .await?;
1503 0 : let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
1504 0 : let attached = tenant.is_some();
1505 :
1506 0 : if let Some(_flush_ms) = flush {
1507 0 : match state
1508 0 : .secondary_controller
1509 0 : .upload_tenant(tenant_shard_id)
1510 0 : .await
1511 : {
1512 : Ok(()) => {
1513 0 : tracing::info!("Uploaded heatmap during flush");
1514 : }
1515 0 : Err(e) => {
1516 0 : tracing::warn!("Failed to flush heatmap: {e}");
1517 : }
1518 : }
1519 : } else {
1520 0 : tracing::info!("No flush requested when configuring");
1521 : }
1522 :
1523 : // This API returns a vector of pageservers where the tenant is attached: this is
1524 : // primarily for use in the sharding service. For compatibilty, we also return this
1525 : // when called directly on a pageserver, but the payload is always zero or one shards.
1526 0 : let mut response = TenantLocationConfigResponse {
1527 0 : shards: Vec::new(),
1528 0 : stripe_size: None,
1529 0 : };
1530 0 : if attached {
1531 0 : response.shards.push(TenantShardLocation {
1532 0 : shard_id: tenant_shard_id,
1533 0 : node_id: state.conf.id,
1534 0 : });
1535 0 : if tenant_shard_id.shard_count.count() > 1 {
1536 : // Stripe size should be set if we are attached
1537 0 : debug_assert!(stripe_size.is_some());
1538 0 : response.stripe_size = stripe_size;
1539 0 : }
1540 0 : }
1541 :
1542 0 : json_response(StatusCode::OK, response)
1543 0 : }
1544 :
1545 0 : async fn list_location_config_handler(
1546 0 : request: Request<Body>,
1547 0 : _cancel: CancellationToken,
1548 0 : ) -> Result<Response<Body>, ApiError> {
1549 0 : let state = get_state(&request);
1550 0 : let slots = state.tenant_manager.list();
1551 0 : let result = LocationConfigListResponse {
1552 0 : tenant_shards: slots
1553 0 : .into_iter()
1554 0 : .map(|(tenant_shard_id, slot)| {
1555 0 : let v = match slot {
1556 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1557 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1558 0 : TenantSlot::InProgress(_) => None,
1559 : };
1560 0 : (tenant_shard_id, v)
1561 0 : })
1562 0 : .collect(),
1563 0 : };
1564 0 : json_response(StatusCode::OK, result)
1565 0 : }
1566 :
1567 0 : async fn get_location_config_handler(
1568 0 : request: Request<Body>,
1569 0 : _cancel: CancellationToken,
1570 0 : ) -> Result<Response<Body>, ApiError> {
1571 0 : let state = get_state(&request);
1572 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1573 0 : let slot = state.tenant_manager.get(tenant_shard_id);
1574 :
1575 0 : let Some(slot) = slot else {
1576 0 : return Err(ApiError::NotFound(
1577 0 : anyhow::anyhow!("Tenant shard not found").into(),
1578 0 : ));
1579 : };
1580 :
1581 0 : let result: Option<LocationConfig> = match slot {
1582 0 : TenantSlot::Attached(t) => Some(t.get_location_conf()),
1583 0 : TenantSlot::Secondary(s) => Some(s.get_location_conf()),
1584 0 : TenantSlot::InProgress(_) => None,
1585 : };
1586 :
1587 0 : json_response(StatusCode::OK, result)
1588 0 : }
1589 :
1590 : // Do a time travel recovery on the given tenant/tenant shard. Tenant needs to be detached
1591 : // (from all pageservers) as it invalidates consistency assumptions.
1592 0 : async fn tenant_time_travel_remote_storage_handler(
1593 0 : request: Request<Body>,
1594 0 : cancel: CancellationToken,
1595 0 : ) -> Result<Response<Body>, ApiError> {
1596 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1597 :
1598 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1599 :
1600 0 : let timestamp_raw = must_get_query_param(&request, "travel_to")?;
1601 0 : let timestamp = humantime::parse_rfc3339(×tamp_raw)
1602 0 : .with_context(|| format!("Invalid time for travel_to: {timestamp_raw:?}"))
1603 0 : .map_err(ApiError::BadRequest)?;
1604 :
1605 0 : let done_if_after_raw = must_get_query_param(&request, "done_if_after")?;
1606 0 : let done_if_after = humantime::parse_rfc3339(&done_if_after_raw)
1607 0 : .with_context(|| format!("Invalid time for done_if_after: {done_if_after_raw:?}"))
1608 0 : .map_err(ApiError::BadRequest)?;
1609 :
1610 : // This is just a sanity check to fend off naive wrong usages of the API:
1611 : // the tenant needs to be detached *everywhere*
1612 0 : let state = get_state(&request);
1613 0 : let we_manage_tenant = state.tenant_manager.manages_tenant_shard(tenant_shard_id);
1614 0 : if we_manage_tenant {
1615 0 : return Err(ApiError::BadRequest(anyhow!(
1616 0 : "Tenant {tenant_shard_id} is already attached at this pageserver"
1617 0 : )));
1618 0 : }
1619 0 :
1620 0 : if timestamp > done_if_after {
1621 0 : return Err(ApiError::BadRequest(anyhow!(
1622 0 : "The done_if_after timestamp comes before the timestamp to recover to"
1623 0 : )));
1624 0 : }
1625 0 :
1626 0 : tracing::info!("Issuing time travel request internally. timestamp={timestamp_raw}, done_if_after={done_if_after_raw}");
1627 :
1628 0 : remote_timeline_client::upload::time_travel_recover_tenant(
1629 0 : &state.remote_storage,
1630 0 : &tenant_shard_id,
1631 0 : timestamp,
1632 0 : done_if_after,
1633 0 : &cancel,
1634 0 : )
1635 0 : .await
1636 0 : .map_err(|e| match e {
1637 0 : TimeTravelError::BadInput(e) => {
1638 0 : warn!("bad input error: {e}");
1639 0 : ApiError::BadRequest(anyhow!("bad input error"))
1640 : }
1641 : TimeTravelError::Unimplemented => {
1642 0 : ApiError::BadRequest(anyhow!("unimplemented for the configured remote storage"))
1643 : }
1644 0 : TimeTravelError::Cancelled => ApiError::InternalServerError(anyhow!("cancelled")),
1645 : TimeTravelError::TooManyVersions => {
1646 0 : ApiError::InternalServerError(anyhow!("too many versions in remote storage"))
1647 : }
1648 0 : TimeTravelError::Other(e) => {
1649 0 : warn!("internal error: {e}");
1650 0 : ApiError::InternalServerError(anyhow!("internal error"))
1651 : }
1652 0 : })?;
1653 :
1654 0 : json_response(StatusCode::OK, ())
1655 0 : }
1656 :
1657 : /// Testing helper to transition a tenant to [`crate::tenant::TenantState::Broken`].
1658 0 : async fn handle_tenant_break(
1659 0 : r: Request<Body>,
1660 0 : _cancel: CancellationToken,
1661 0 : ) -> Result<Response<Body>, ApiError> {
1662 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
1663 :
1664 0 : let state = get_state(&r);
1665 0 : state
1666 0 : .tenant_manager
1667 0 : .get_attached_tenant_shard(tenant_shard_id)?
1668 0 : .set_broken("broken from test".to_owned())
1669 0 : .await;
1670 :
1671 0 : json_response(StatusCode::OK, ())
1672 0 : }
1673 :
1674 : // Obtains an lsn lease on the given timeline.
1675 0 : async fn lsn_lease_handler(
1676 0 : mut request: Request<Body>,
1677 0 : _cancel: CancellationToken,
1678 0 : ) -> Result<Response<Body>, ApiError> {
1679 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1680 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1681 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1682 0 : let lsn = json_request::<LsnLeaseRequest>(&mut request).await?.lsn;
1683 :
1684 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1685 0 :
1686 0 : let state = get_state(&request);
1687 :
1688 0 : let timeline =
1689 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1690 0 : .await?;
1691 0 : let result = timeline
1692 0 : .make_lsn_lease(lsn, timeline.get_lsn_lease_length(), &ctx)
1693 0 : .map_err(|e| ApiError::InternalServerError(e.context("lsn lease http handler")))?;
1694 :
1695 0 : json_response(StatusCode::OK, result)
1696 0 : }
1697 :
1698 : // Run GC immediately on given timeline.
1699 0 : async fn timeline_gc_handler(
1700 0 : mut request: Request<Body>,
1701 0 : cancel: CancellationToken,
1702 0 : ) -> Result<Response<Body>, ApiError> {
1703 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1704 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1705 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1706 :
1707 0 : let gc_req: TimelineGcRequest = json_request(&mut request).await?;
1708 :
1709 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1710 0 : let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
1711 :
1712 0 : json_response(StatusCode::OK, gc_result)
1713 0 : }
1714 :
1715 : // Run compaction immediately on given timeline.
1716 0 : async fn timeline_compact_handler(
1717 0 : request: Request<Body>,
1718 0 : cancel: CancellationToken,
1719 0 : ) -> Result<Response<Body>, ApiError> {
1720 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1721 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1722 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1723 :
1724 0 : let state = get_state(&request);
1725 0 :
1726 0 : let mut flags = EnumSet::empty();
1727 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1728 0 : flags |= CompactFlags::ForceRepartition;
1729 0 : }
1730 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1731 0 : flags |= CompactFlags::ForceImageLayerCreation;
1732 0 : }
1733 0 : if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
1734 0 : flags |= CompactFlags::EnhancedGcBottomMostCompaction;
1735 0 : }
1736 0 : if Some(true) == parse_query_param::<_, bool>(&request, "dry_run")? {
1737 0 : flags |= CompactFlags::DryRun;
1738 0 : }
1739 :
1740 0 : let wait_until_uploaded =
1741 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1742 :
1743 0 : async {
1744 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1745 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1746 0 : timeline
1747 0 : .compact(&cancel, flags, &ctx)
1748 0 : .await
1749 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
1750 0 : if wait_until_uploaded {
1751 0 : timeline.remote_client.wait_completion().await
1752 : // XXX map to correct ApiError for the cases where it's due to shutdown
1753 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
1754 0 : }
1755 0 : json_response(StatusCode::OK, ())
1756 0 : }
1757 0 : .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1758 0 : .await
1759 0 : }
1760 :
1761 : // Run checkpoint immediately on given timeline.
1762 0 : async fn timeline_checkpoint_handler(
1763 0 : request: Request<Body>,
1764 0 : cancel: CancellationToken,
1765 0 : ) -> Result<Response<Body>, ApiError> {
1766 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1767 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1768 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1769 :
1770 0 : let state = get_state(&request);
1771 0 :
1772 0 : let mut flags = EnumSet::empty();
1773 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_repartition")? {
1774 0 : flags |= CompactFlags::ForceRepartition;
1775 0 : }
1776 0 : if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
1777 0 : flags |= CompactFlags::ForceImageLayerCreation;
1778 0 : }
1779 :
1780 : // By default, checkpoints come with a compaction, but this may be optionally disabled by tests that just want to flush + upload.
1781 0 : let compact = parse_query_param::<_, bool>(&request, "compact")?.unwrap_or(true);
1782 :
1783 0 : let wait_until_uploaded =
1784 0 : parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
1785 :
1786 0 : async {
1787 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1788 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1789 0 : timeline
1790 0 : .freeze_and_flush()
1791 0 : .await
1792 0 : .map_err(|e| {
1793 0 : match e {
1794 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
1795 0 : other => ApiError::InternalServerError(other.into()),
1796 :
1797 : }
1798 0 : })?;
1799 0 : if compact {
1800 0 : timeline
1801 0 : .compact(&cancel, flags, &ctx)
1802 0 : .await
1803 0 : .map_err(|e|
1804 0 : match e {
1805 0 : CompactionError::ShuttingDown => ApiError::ShuttingDown,
1806 0 : CompactionError::Other(e) => ApiError::InternalServerError(e)
1807 0 : }
1808 0 : )?;
1809 0 : }
1810 :
1811 0 : if wait_until_uploaded {
1812 0 : tracing::info!("Waiting for uploads to complete...");
1813 0 : timeline.remote_client.wait_completion().await
1814 : // XXX map to correct ApiError for the cases where it's due to shutdown
1815 0 : .context("wait completion").map_err(ApiError::InternalServerError)?;
1816 0 : tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
1817 0 : }
1818 :
1819 0 : json_response(StatusCode::OK, ())
1820 0 : }
1821 0 : .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
1822 0 : .await
1823 0 : }
1824 :
1825 0 : async fn timeline_download_remote_layers_handler_post(
1826 0 : mut request: Request<Body>,
1827 0 : _cancel: CancellationToken,
1828 0 : ) -> Result<Response<Body>, ApiError> {
1829 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1830 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1831 0 : let body: DownloadRemoteLayersTaskSpawnRequest = json_request(&mut request).await?;
1832 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1833 :
1834 0 : let state = get_state(&request);
1835 :
1836 0 : let timeline =
1837 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1838 0 : .await?;
1839 0 : match timeline.spawn_download_all_remote_layers(body).await {
1840 0 : Ok(st) => json_response(StatusCode::ACCEPTED, st),
1841 0 : Err(st) => json_response(StatusCode::CONFLICT, st),
1842 : }
1843 0 : }
1844 :
1845 0 : async fn timeline_download_remote_layers_handler_get(
1846 0 : request: Request<Body>,
1847 0 : _cancel: CancellationToken,
1848 0 : ) -> Result<Response<Body>, ApiError> {
1849 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1850 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1851 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1852 0 : let state = get_state(&request);
1853 :
1854 0 : let timeline =
1855 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
1856 0 : .await?;
1857 0 : let info = timeline
1858 0 : .get_download_all_remote_layers_task_info()
1859 0 : .context("task never started since last pageserver process start")
1860 0 : .map_err(|e| ApiError::NotFound(e.into()))?;
1861 0 : json_response(StatusCode::OK, info)
1862 0 : }
1863 :
1864 0 : async fn timeline_detach_ancestor_handler(
1865 0 : request: Request<Body>,
1866 0 : _cancel: CancellationToken,
1867 0 : ) -> Result<Response<Body>, ApiError> {
1868 : use crate::tenant::timeline::detach_ancestor;
1869 : use pageserver_api::models::detach_ancestor::AncestorDetached;
1870 :
1871 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1872 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1873 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1874 :
1875 0 : let span = tracing::info_span!("detach_ancestor", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id);
1876 :
1877 0 : async move {
1878 0 : let mut options = detach_ancestor::Options::default();
1879 :
1880 0 : let rewrite_concurrency =
1881 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "rewrite_concurrency")?;
1882 0 : let copy_concurrency =
1883 0 : parse_query_param::<_, std::num::NonZeroUsize>(&request, "copy_concurrency")?;
1884 :
1885 0 : [
1886 0 : (&mut options.rewrite_concurrency, rewrite_concurrency),
1887 0 : (&mut options.copy_concurrency, copy_concurrency),
1888 0 : ]
1889 0 : .into_iter()
1890 0 : .filter_map(|(target, val)| val.map(|val| (target, val)))
1891 0 : .for_each(|(target, val)| *target = val);
1892 0 :
1893 0 : let state = get_state(&request);
1894 :
1895 0 : let tenant = state
1896 0 : .tenant_manager
1897 0 : .get_attached_tenant_shard(tenant_shard_id)?;
1898 :
1899 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
1900 :
1901 0 : let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
1902 0 : let ctx = &ctx;
1903 :
1904 0 : let timeline = tenant.get_timeline(timeline_id, true)?;
1905 :
1906 0 : let progress = timeline
1907 0 : .prepare_to_detach_from_ancestor(&tenant, options, ctx)
1908 0 : .await?;
1909 :
1910 : // uncomment to allow early as possible Tenant::drop
1911 : // drop(tenant);
1912 :
1913 0 : let resp = match progress {
1914 0 : detach_ancestor::Progress::Prepared(attempt, prepared) => {
1915 : // it would be great to tag the guard on to the tenant activation future
1916 0 : let reparented_timelines = state
1917 0 : .tenant_manager
1918 0 : .complete_detaching_timeline_ancestor(
1919 0 : tenant_shard_id,
1920 0 : timeline_id,
1921 0 : prepared,
1922 0 : attempt,
1923 0 : ctx,
1924 0 : )
1925 0 : .await?;
1926 :
1927 0 : AncestorDetached {
1928 0 : reparented_timelines,
1929 0 : }
1930 : }
1931 0 : detach_ancestor::Progress::Done(resp) => resp,
1932 : };
1933 :
1934 0 : json_response(StatusCode::OK, resp)
1935 0 : }
1936 0 : .instrument(span)
1937 0 : .await
1938 0 : }
1939 :
1940 0 : async fn deletion_queue_flush(
1941 0 : r: Request<Body>,
1942 0 : cancel: CancellationToken,
1943 0 : ) -> Result<Response<Body>, ApiError> {
1944 0 : let state = get_state(&r);
1945 :
1946 0 : let execute = parse_query_param(&r, "execute")?.unwrap_or(false);
1947 0 :
1948 0 : let flush = async {
1949 0 : if execute {
1950 0 : state.deletion_queue_client.flush_execute().await
1951 : } else {
1952 0 : state.deletion_queue_client.flush().await
1953 : }
1954 0 : }
1955 : // DeletionQueueError's only case is shutting down.
1956 0 : .map_err(|_| ApiError::ShuttingDown);
1957 0 :
1958 0 : tokio::select! {
1959 0 : res = flush => {
1960 0 : res.map(|()| json_response(StatusCode::OK, ()))?
1961 : }
1962 0 : _ = cancel.cancelled() => {
1963 0 : Err(ApiError::ShuttingDown)
1964 : }
1965 : }
1966 0 : }
1967 :
1968 : /// Try if `GetPage@Lsn` is successful, useful for manual debugging.
1969 0 : async fn getpage_at_lsn_handler(
1970 0 : request: Request<Body>,
1971 0 : _cancel: CancellationToken,
1972 0 : ) -> Result<Response<Body>, ApiError> {
1973 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
1974 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
1975 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
1976 0 : let state = get_state(&request);
1977 :
1978 : struct Key(crate::repository::Key);
1979 :
1980 : impl std::str::FromStr for Key {
1981 : type Err = anyhow::Error;
1982 :
1983 0 : fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
1984 0 : crate::repository::Key::from_hex(s).map(Key)
1985 0 : }
1986 : }
1987 :
1988 0 : let key: Key = parse_query_param(&request, "key")?
1989 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'key' query parameter")))?;
1990 0 : let lsn: Lsn = parse_query_param(&request, "lsn")?
1991 0 : .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
1992 :
1993 0 : async {
1994 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
1995 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
1996 :
1997 0 : let page = timeline.get(key.0, lsn, &ctx).await?;
1998 :
1999 0 : Result::<_, ApiError>::Ok(
2000 0 : Response::builder()
2001 0 : .status(StatusCode::OK)
2002 0 : .header(header::CONTENT_TYPE, "application/octet-stream")
2003 0 : .body(hyper::Body::from(page))
2004 0 : .unwrap(),
2005 0 : )
2006 0 : }
2007 0 : .instrument(info_span!("timeline_get", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2008 0 : .await
2009 0 : }
2010 :
2011 0 : async fn timeline_collect_keyspace(
2012 0 : request: Request<Body>,
2013 0 : _cancel: CancellationToken,
2014 0 : ) -> Result<Response<Body>, ApiError> {
2015 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2016 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2017 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2018 0 : let state = get_state(&request);
2019 :
2020 0 : let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
2021 :
2022 0 : async {
2023 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2024 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
2025 0 : let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
2026 0 : let (dense_ks, sparse_ks) = timeline
2027 0 : .collect_keyspace(at_lsn, &ctx)
2028 0 : .await
2029 0 : .map_err(|e| ApiError::InternalServerError(e.into()))?;
2030 :
2031 : // This API is currently used by pagebench. Pagebench will iterate all keys within the keyspace.
2032 : // Therefore, we split dense/sparse keys in this API.
2033 0 : let res = pageserver_api::models::partitioning::Partitioning { keys: dense_ks, sparse_keys: sparse_ks, at_lsn };
2034 0 :
2035 0 : json_response(StatusCode::OK, res)
2036 0 : }
2037 0 : .instrument(info_span!("timeline_collect_keyspace", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
2038 0 : .await
2039 0 : }
2040 :
2041 0 : async fn active_timeline_of_active_tenant(
2042 0 : tenant_manager: &TenantManager,
2043 0 : tenant_shard_id: TenantShardId,
2044 0 : timeline_id: TimelineId,
2045 0 : ) -> Result<Arc<Timeline>, ApiError> {
2046 0 : let tenant = tenant_manager.get_attached_tenant_shard(tenant_shard_id)?;
2047 :
2048 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2049 :
2050 0 : Ok(tenant.get_timeline(timeline_id, true)?)
2051 0 : }
2052 :
2053 0 : async fn always_panic_handler(
2054 0 : req: Request<Body>,
2055 0 : _cancel: CancellationToken,
2056 0 : ) -> Result<Response<Body>, ApiError> {
2057 0 : // Deliberately cause a panic to exercise the panic hook registered via std::panic::set_hook().
2058 0 : // For pageserver, the relevant panic hook is `tracing_panic_hook` , and the `sentry` crate's wrapper around it.
2059 0 : // Use catch_unwind to ensure that tokio nor hyper are distracted by our panic.
2060 0 : let query = req.uri().query();
2061 0 : let _ = std::panic::catch_unwind(|| {
2062 0 : panic!("unconditional panic for testing panic hook integration; request query: {query:?}")
2063 0 : });
2064 0 : json_response(StatusCode::NO_CONTENT, ())
2065 0 : }
2066 :
2067 0 : async fn disk_usage_eviction_run(
2068 0 : mut r: Request<Body>,
2069 0 : cancel: CancellationToken,
2070 0 : ) -> Result<Response<Body>, ApiError> {
2071 0 : check_permission(&r, None)?;
2072 :
2073 0 : #[derive(Debug, Clone, Copy, serde::Serialize, serde::Deserialize)]
2074 : struct Config {
2075 : /// How many bytes to evict before reporting that pressure is relieved.
2076 : evict_bytes: u64,
2077 :
2078 : #[serde(default)]
2079 : eviction_order: pageserver_api::config::EvictionOrder,
2080 : }
2081 :
2082 : #[derive(Debug, Clone, Copy, serde::Serialize)]
2083 : struct Usage {
2084 : // remains unchanged after instantiation of the struct
2085 : evict_bytes: u64,
2086 : // updated by `add_available_bytes`
2087 : freed_bytes: u64,
2088 : }
2089 :
2090 : impl crate::disk_usage_eviction_task::Usage for Usage {
2091 0 : fn has_pressure(&self) -> bool {
2092 0 : self.evict_bytes > self.freed_bytes
2093 0 : }
2094 :
2095 0 : fn add_available_bytes(&mut self, bytes: u64) {
2096 0 : self.freed_bytes += bytes;
2097 0 : }
2098 : }
2099 :
2100 0 : let config = json_request::<Config>(&mut r).await?;
2101 :
2102 0 : let usage = Usage {
2103 0 : evict_bytes: config.evict_bytes,
2104 0 : freed_bytes: 0,
2105 0 : };
2106 0 :
2107 0 : let state = get_state(&r);
2108 0 : let eviction_state = state.disk_usage_eviction_state.clone();
2109 :
2110 0 : let res = crate::disk_usage_eviction_task::disk_usage_eviction_task_iteration_impl(
2111 0 : &eviction_state,
2112 0 : &state.remote_storage,
2113 0 : usage,
2114 0 : &state.tenant_manager,
2115 0 : config.eviction_order.into(),
2116 0 : &cancel,
2117 0 : )
2118 0 : .await;
2119 :
2120 0 : info!(?res, "disk_usage_eviction_task_iteration_impl finished");
2121 :
2122 0 : let res = res.map_err(ApiError::InternalServerError)?;
2123 :
2124 0 : json_response(StatusCode::OK, res)
2125 0 : }
2126 :
2127 0 : async fn secondary_upload_handler(
2128 0 : request: Request<Body>,
2129 0 : _cancel: CancellationToken,
2130 0 : ) -> Result<Response<Body>, ApiError> {
2131 0 : let state = get_state(&request);
2132 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2133 0 : state
2134 0 : .secondary_controller
2135 0 : .upload_tenant(tenant_shard_id)
2136 0 : .await
2137 0 : .map_err(ApiError::InternalServerError)?;
2138 :
2139 0 : json_response(StatusCode::OK, ())
2140 0 : }
2141 :
2142 0 : async fn tenant_scan_remote_handler(
2143 0 : request: Request<Body>,
2144 0 : cancel: CancellationToken,
2145 0 : ) -> Result<Response<Body>, ApiError> {
2146 0 : let state = get_state(&request);
2147 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2148 :
2149 0 : let mut response = TenantScanRemoteStorageResponse::default();
2150 :
2151 0 : let (shards, _other_keys) =
2152 0 : list_remote_tenant_shards(&state.remote_storage, tenant_id, cancel.clone())
2153 0 : .await
2154 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2155 :
2156 0 : for tenant_shard_id in shards {
2157 0 : let (timeline_ids, _other_keys) =
2158 0 : list_remote_timelines(&state.remote_storage, tenant_shard_id, cancel.clone())
2159 0 : .await
2160 0 : .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
2161 :
2162 0 : let mut generation = Generation::none();
2163 0 : for timeline_id in timeline_ids {
2164 0 : match download_index_part(
2165 0 : &state.remote_storage,
2166 0 : &tenant_shard_id,
2167 0 : &timeline_id,
2168 0 : Generation::MAX,
2169 0 : &cancel,
2170 0 : )
2171 0 : .instrument(info_span!("download_index_part",
2172 : tenant_id=%tenant_shard_id.tenant_id,
2173 0 : shard_id=%tenant_shard_id.shard_slug(),
2174 : %timeline_id))
2175 0 : .await
2176 : {
2177 0 : Ok((index_part, index_generation)) => {
2178 0 : tracing::info!("Found timeline {tenant_shard_id}/{timeline_id} metadata (gen {index_generation:?}, {} layers, {} consistent LSN)",
2179 0 : index_part.layer_metadata.len(), index_part.metadata.disk_consistent_lsn());
2180 0 : generation = std::cmp::max(generation, index_generation);
2181 : }
2182 : Err(DownloadError::NotFound) => {
2183 : // This is normal for tenants that were created with multiple shards: they have an unsharded path
2184 : // containing the timeline's initdb tarball but no index. Otherwise it is a bit strange.
2185 0 : tracing::info!("Timeline path {tenant_shard_id}/{timeline_id} exists in remote storage but has no index, skipping");
2186 0 : continue;
2187 : }
2188 0 : Err(e) => {
2189 0 : return Err(ApiError::InternalServerError(anyhow::anyhow!(e)));
2190 : }
2191 : };
2192 : }
2193 :
2194 0 : response.shards.push(TenantScanRemoteStorageShard {
2195 0 : tenant_shard_id,
2196 0 : generation: generation.into(),
2197 0 : });
2198 : }
2199 :
2200 0 : if response.shards.is_empty() {
2201 0 : return Err(ApiError::NotFound(
2202 0 : anyhow::anyhow!("No shards found for tenant ID {tenant_id}").into(),
2203 0 : ));
2204 0 : }
2205 0 :
2206 0 : json_response(StatusCode::OK, response)
2207 0 : }
2208 :
2209 0 : async fn secondary_download_handler(
2210 0 : request: Request<Body>,
2211 0 : _cancel: CancellationToken,
2212 0 : ) -> Result<Response<Body>, ApiError> {
2213 0 : let state = get_state(&request);
2214 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2215 0 : let wait = parse_query_param(&request, "wait_ms")?.map(Duration::from_millis);
2216 :
2217 : // We don't need this to issue the download request, but:
2218 : // - it enables us to cleanly return 404 if we get a request for an absent shard
2219 : // - we will use this to provide status feedback in the response
2220 0 : let Some(secondary_tenant) = state
2221 0 : .tenant_manager
2222 0 : .get_secondary_tenant_shard(tenant_shard_id)
2223 : else {
2224 0 : return Err(ApiError::NotFound(
2225 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2226 0 : ));
2227 : };
2228 :
2229 0 : let timeout = wait.unwrap_or(Duration::MAX);
2230 :
2231 0 : let result = tokio::time::timeout(
2232 0 : timeout,
2233 0 : state.secondary_controller.download_tenant(tenant_shard_id),
2234 0 : )
2235 0 : .await;
2236 :
2237 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2238 :
2239 0 : let status = match result {
2240 : Ok(Ok(())) => {
2241 0 : if progress.layers_downloaded >= progress.layers_total {
2242 : // Download job ran to completion
2243 0 : StatusCode::OK
2244 : } else {
2245 : // Download dropped out without errors because it ran out of time budget
2246 0 : StatusCode::ACCEPTED
2247 : }
2248 : }
2249 : // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
2250 : // okay. We could get an error here in the unlikely edge case that the tenant
2251 : // was detached between our check above and executing the download job.
2252 0 : Ok(Err(e)) => return Err(ApiError::InternalServerError(e)),
2253 : // A timeout is not an error: we have started the download, we're just not done
2254 : // yet. The caller will get a response body indicating status.
2255 0 : Err(_) => StatusCode::ACCEPTED,
2256 : };
2257 :
2258 0 : json_response(status, progress)
2259 0 : }
2260 :
2261 0 : async fn secondary_status_handler(
2262 0 : request: Request<Body>,
2263 0 : _cancel: CancellationToken,
2264 0 : ) -> Result<Response<Body>, ApiError> {
2265 0 : let state = get_state(&request);
2266 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2267 :
2268 0 : let Some(secondary_tenant) = state
2269 0 : .tenant_manager
2270 0 : .get_secondary_tenant_shard(tenant_shard_id)
2271 : else {
2272 0 : return Err(ApiError::NotFound(
2273 0 : anyhow::anyhow!("Shard {} not found", tenant_shard_id).into(),
2274 0 : ));
2275 : };
2276 :
2277 0 : let progress = secondary_tenant.progress.lock().unwrap().clone();
2278 0 :
2279 0 : json_response(StatusCode::OK, progress)
2280 0 : }
2281 :
2282 0 : async fn handler_404(_: Request<Body>) -> Result<Response<Body>, ApiError> {
2283 0 : json_response(
2284 0 : StatusCode::NOT_FOUND,
2285 0 : HttpErrorBody::from_msg("page not found".to_owned()),
2286 0 : )
2287 0 : }
2288 :
2289 0 : async fn post_tracing_event_handler(
2290 0 : mut r: Request<Body>,
2291 0 : _cancel: CancellationToken,
2292 0 : ) -> Result<Response<Body>, ApiError> {
2293 0 : #[derive(Debug, serde::Deserialize)]
2294 : #[serde(rename_all = "lowercase")]
2295 : enum Level {
2296 : Error,
2297 : Warn,
2298 : Info,
2299 : Debug,
2300 : Trace,
2301 : }
2302 0 : #[derive(Debug, serde::Deserialize)]
2303 : struct Request {
2304 : level: Level,
2305 : message: String,
2306 : }
2307 0 : let body: Request = json_request(&mut r)
2308 0 : .await
2309 0 : .map_err(|_| ApiError::BadRequest(anyhow::anyhow!("invalid JSON body")))?;
2310 :
2311 0 : match body.level {
2312 0 : Level::Error => tracing::error!(?body.message),
2313 0 : Level::Warn => tracing::warn!(?body.message),
2314 0 : Level::Info => tracing::info!(?body.message),
2315 0 : Level::Debug => tracing::debug!(?body.message),
2316 0 : Level::Trace => tracing::trace!(?body.message),
2317 : }
2318 :
2319 0 : json_response(StatusCode::OK, ())
2320 0 : }
2321 :
2322 0 : async fn force_aux_policy_switch_handler(
2323 0 : mut r: Request<Body>,
2324 0 : _cancel: CancellationToken,
2325 0 : ) -> Result<Response<Body>, ApiError> {
2326 0 : check_permission(&r, None)?;
2327 0 : let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
2328 0 : let timeline_id: TimelineId = parse_request_param(&r, "timeline_id")?;
2329 0 : let policy: AuxFilePolicy = json_request(&mut r).await?;
2330 :
2331 0 : let state = get_state(&r);
2332 :
2333 0 : let tenant = state
2334 0 : .tenant_manager
2335 0 : .get_attached_tenant_shard(tenant_shard_id)?;
2336 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2337 0 : let timeline =
2338 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2339 0 : .await?;
2340 0 : timeline
2341 0 : .do_switch_aux_policy(policy)
2342 0 : .map_err(ApiError::InternalServerError)?;
2343 :
2344 0 : json_response(StatusCode::OK, ())
2345 0 : }
2346 :
2347 0 : async fn put_io_engine_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 kind: crate::virtual_file::IoEngineKind = json_request(&mut r).await?;
2353 0 : crate::virtual_file::io_engine::set(kind);
2354 0 : json_response(StatusCode::OK, ())
2355 0 : }
2356 :
2357 0 : async fn put_io_alignment_handler(
2358 0 : mut r: Request<Body>,
2359 0 : _cancel: CancellationToken,
2360 0 : ) -> Result<Response<Body>, ApiError> {
2361 0 : check_permission(&r, None)?;
2362 0 : let align: usize = json_request(&mut r).await?;
2363 0 : crate::virtual_file::set_io_buffer_alignment(align).map_err(|align| {
2364 0 : ApiError::PreconditionFailed(
2365 0 : format!("Requested io alignment ({align}) is not a power of two").into(),
2366 0 : )
2367 0 : })?;
2368 0 : json_response(StatusCode::OK, ())
2369 0 : }
2370 :
2371 : /// Polled by control plane.
2372 : ///
2373 : /// See [`crate::utilization`].
2374 0 : async fn get_utilization(
2375 0 : r: Request<Body>,
2376 0 : _cancel: CancellationToken,
2377 0 : ) -> Result<Response<Body>, ApiError> {
2378 0 : fail::fail_point!("get-utilization-http-handler", |_| {
2379 0 : Err(ApiError::ResourceUnavailable("failpoint".into()))
2380 0 : });
2381 :
2382 : // this probably could be completely public, but lets make that change later.
2383 0 : check_permission(&r, None)?;
2384 :
2385 0 : let state = get_state(&r);
2386 0 : let mut g = state.latest_utilization.lock().await;
2387 :
2388 0 : let regenerate_every = Duration::from_secs(1);
2389 0 : let still_valid = g
2390 0 : .as_ref()
2391 0 : .is_some_and(|(captured_at, _)| captured_at.elapsed() < regenerate_every);
2392 0 :
2393 0 : // avoid needless statvfs calls even though those should be non-blocking fast.
2394 0 : // regenerate at most 1Hz to allow polling at any rate.
2395 0 : if !still_valid {
2396 0 : let path = state.conf.tenants_path();
2397 0 : let doc =
2398 0 : crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
2399 0 : .map_err(ApiError::InternalServerError)?;
2400 :
2401 0 : let mut buf = Vec::new();
2402 0 : serde_json::to_writer(&mut buf, &doc)
2403 0 : .context("serialize")
2404 0 : .map_err(ApiError::InternalServerError)?;
2405 :
2406 0 : let body = bytes::Bytes::from(buf);
2407 0 :
2408 0 : *g = Some((std::time::Instant::now(), body));
2409 0 : }
2410 :
2411 : // hyper 0.14 doesn't yet have Response::clone so this is a bit of extra legwork
2412 0 : let cached = g.as_ref().expect("just set").1.clone();
2413 0 :
2414 0 : Response::builder()
2415 0 : .header(hyper::http::header::CONTENT_TYPE, "application/json")
2416 0 : // thought of using http date header, but that is second precision which does not give any
2417 0 : // debugging aid
2418 0 : .status(StatusCode::OK)
2419 0 : .body(hyper::Body::from(cached))
2420 0 : .context("build response")
2421 0 : .map_err(ApiError::InternalServerError)
2422 0 : }
2423 :
2424 0 : async fn list_aux_files(
2425 0 : mut request: Request<Body>,
2426 0 : _cancel: CancellationToken,
2427 0 : ) -> Result<Response<Body>, ApiError> {
2428 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2429 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2430 0 : let body: ListAuxFilesRequest = json_request(&mut request).await?;
2431 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2432 :
2433 0 : let state = get_state(&request);
2434 :
2435 0 : let timeline =
2436 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2437 0 : .await?;
2438 :
2439 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2440 0 : let files = timeline.list_aux_files(body.lsn, &ctx).await?;
2441 0 : json_response(StatusCode::OK, files)
2442 0 : }
2443 :
2444 0 : async fn perf_info(
2445 0 : request: Request<Body>,
2446 0 : _cancel: CancellationToken,
2447 0 : ) -> Result<Response<Body>, ApiError> {
2448 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2449 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2450 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2451 :
2452 0 : let state = get_state(&request);
2453 :
2454 0 : let timeline =
2455 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2456 0 : .await?;
2457 :
2458 0 : let result = timeline.perf_info().await;
2459 :
2460 0 : json_response(StatusCode::OK, result)
2461 0 : }
2462 :
2463 0 : async fn ingest_aux_files(
2464 0 : mut request: Request<Body>,
2465 0 : _cancel: CancellationToken,
2466 0 : ) -> Result<Response<Body>, ApiError> {
2467 0 : let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
2468 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2469 0 : let body: IngestAuxFilesRequest = json_request(&mut request).await?;
2470 0 : check_permission(&request, Some(tenant_shard_id.tenant_id))?;
2471 :
2472 0 : let state = get_state(&request);
2473 :
2474 0 : let timeline =
2475 0 : active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
2476 0 : .await?;
2477 :
2478 0 : let mut modification = timeline.begin_modification(
2479 0 : Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
2480 0 : );
2481 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
2482 0 : for (fname, content) in body.aux_files {
2483 0 : modification
2484 0 : .put_file(&fname, content.as_bytes(), &ctx)
2485 0 : .await
2486 0 : .map_err(ApiError::InternalServerError)?;
2487 : }
2488 0 : modification
2489 0 : .commit(&ctx)
2490 0 : .await
2491 0 : .map_err(ApiError::InternalServerError)?;
2492 :
2493 0 : json_response(StatusCode::OK, ())
2494 0 : }
2495 :
2496 : /// Report on the largest tenants on this pageserver, for the storage controller to identify
2497 : /// candidates for splitting
2498 0 : async fn post_top_tenants(
2499 0 : mut r: Request<Body>,
2500 0 : _cancel: CancellationToken,
2501 0 : ) -> Result<Response<Body>, ApiError> {
2502 0 : check_permission(&r, None)?;
2503 0 : let request: TopTenantShardsRequest = json_request(&mut r).await?;
2504 0 : let state = get_state(&r);
2505 :
2506 0 : fn get_size_metric(sizes: &TopTenantShardItem, order_by: &TenantSorting) -> u64 {
2507 0 : match order_by {
2508 0 : TenantSorting::ResidentSize => sizes.resident_size,
2509 0 : TenantSorting::MaxLogicalSize => sizes.max_logical_size,
2510 : }
2511 0 : }
2512 :
2513 : #[derive(Eq, PartialEq)]
2514 : struct HeapItem {
2515 : metric: u64,
2516 : sizes: TopTenantShardItem,
2517 : }
2518 :
2519 : impl PartialOrd for HeapItem {
2520 0 : fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
2521 0 : Some(self.cmp(other))
2522 0 : }
2523 : }
2524 :
2525 : /// Heap items have reverse ordering on their metric: this enables using BinaryHeap, which
2526 : /// supports popping the greatest item but not the smallest.
2527 : impl Ord for HeapItem {
2528 0 : fn cmp(&self, other: &Self) -> std::cmp::Ordering {
2529 0 : Reverse(self.metric).cmp(&Reverse(other.metric))
2530 0 : }
2531 : }
2532 :
2533 0 : let mut top_n: BinaryHeap<HeapItem> = BinaryHeap::with_capacity(request.limit);
2534 :
2535 : // FIXME: this is a lot of clones to take this tenant list
2536 0 : for (tenant_shard_id, tenant_slot) in state.tenant_manager.list() {
2537 0 : if let Some(shards_lt) = request.where_shards_lt {
2538 : // Ignore tenants which already have >= this many shards
2539 0 : if tenant_shard_id.shard_count >= shards_lt {
2540 0 : continue;
2541 0 : }
2542 0 : }
2543 :
2544 0 : let sizes = match tenant_slot {
2545 0 : TenantSlot::Attached(tenant) => tenant.get_sizes(),
2546 : TenantSlot::Secondary(_) | TenantSlot::InProgress(_) => {
2547 0 : continue;
2548 : }
2549 : };
2550 0 : let metric = get_size_metric(&sizes, &request.order_by);
2551 :
2552 0 : if let Some(gt) = request.where_gt {
2553 : // Ignore tenants whose metric is <= the lower size threshold, to do less sorting work
2554 0 : if metric <= gt {
2555 0 : continue;
2556 0 : }
2557 0 : };
2558 :
2559 0 : match top_n.peek() {
2560 0 : None => {
2561 0 : // Top N list is empty: candidate becomes first member
2562 0 : top_n.push(HeapItem { metric, sizes });
2563 0 : }
2564 0 : Some(i) if i.metric > metric && top_n.len() < request.limit => {
2565 0 : // Lowest item in list is greater than our candidate, but we aren't at limit yet: push to end
2566 0 : top_n.push(HeapItem { metric, sizes });
2567 0 : }
2568 0 : Some(i) if i.metric > metric => {
2569 0 : // List is at limit and lowest value is greater than our candidate, drop it.
2570 0 : }
2571 0 : Some(_) => top_n.push(HeapItem { metric, sizes }),
2572 : }
2573 :
2574 0 : while top_n.len() > request.limit {
2575 0 : top_n.pop();
2576 0 : }
2577 : }
2578 :
2579 0 : json_response(
2580 0 : StatusCode::OK,
2581 0 : TopTenantShardsResponse {
2582 0 : shards: top_n.into_iter().map(|i| i.sizes).collect(),
2583 0 : },
2584 0 : )
2585 0 : }
2586 :
2587 0 : async fn put_tenant_timeline_import_basebackup(
2588 0 : request: Request<Body>,
2589 0 : _cancel: CancellationToken,
2590 0 : ) -> Result<Response<Body>, ApiError> {
2591 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2592 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2593 0 : let base_lsn: Lsn = must_parse_query_param(&request, "base_lsn")?;
2594 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
2595 0 : let pg_version: u32 = must_parse_query_param(&request, "pg_version")?;
2596 :
2597 0 : check_permission(&request, Some(tenant_id))?;
2598 :
2599 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
2600 :
2601 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);
2602 0 : async move {
2603 0 : let state = get_state(&request);
2604 0 : let tenant = state
2605 0 : .tenant_manager
2606 0 : .get_attached_tenant_shard(TenantShardId::unsharded(tenant_id))?;
2607 :
2608 0 : let broker_client = state.broker_client.clone();
2609 0 :
2610 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
2611 0 : res.map_err(|error| {
2612 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
2613 0 : })
2614 0 : }));
2615 0 :
2616 0 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2617 :
2618 0 : let timeline = tenant
2619 0 : .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
2620 0 : .map_err(ApiError::InternalServerError)
2621 0 : .await?;
2622 :
2623 : // TODO mark timeline as not ready until it reaches end_lsn.
2624 : // We might have some wal to import as well, and we should prevent compute
2625 : // from connecting before that and writing conflicting wal.
2626 : //
2627 : // This is not relevant for pageserver->pageserver migrations, since there's
2628 : // no wal to import. But should be fixed if we want to import from postgres.
2629 :
2630 : // TODO leave clean state on error. For now you can use detach to clean
2631 : // up broken state from a failed import.
2632 :
2633 : // Import basebackup provided via CopyData
2634 0 : info!("importing basebackup");
2635 :
2636 0 : timeline
2637 0 : .import_basebackup_from_tar(tenant.clone(), &mut body, base_lsn, broker_client, &ctx)
2638 0 : .await
2639 0 : .map_err(ApiError::InternalServerError)?;
2640 :
2641 : // Read the end of the tar archive.
2642 0 : read_tar_eof(body)
2643 0 : .await
2644 0 : .map_err(ApiError::InternalServerError)?;
2645 :
2646 : // TODO check checksum
2647 : // Meanwhile you can verify client-side by taking fullbackup
2648 : // and checking that it matches in size with what was imported.
2649 : // It wouldn't work if base came from vanilla postgres though,
2650 : // since we discard some log files.
2651 :
2652 0 : info!("done");
2653 0 : json_response(StatusCode::OK, ())
2654 0 : }
2655 0 : .instrument(span)
2656 0 : .await
2657 0 : }
2658 :
2659 0 : async fn put_tenant_timeline_import_wal(
2660 0 : request: Request<Body>,
2661 0 : _cancel: CancellationToken,
2662 0 : ) -> Result<Response<Body>, ApiError> {
2663 0 : let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
2664 0 : let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
2665 0 : let start_lsn: Lsn = must_parse_query_param(&request, "start_lsn")?;
2666 0 : let end_lsn: Lsn = must_parse_query_param(&request, "end_lsn")?;
2667 :
2668 0 : check_permission(&request, Some(tenant_id))?;
2669 :
2670 0 : let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
2671 :
2672 0 : let span = info_span!("import_wal", tenant_id=%tenant_id, timeline_id=%timeline_id, start_lsn=%start_lsn, end_lsn=%end_lsn);
2673 0 : async move {
2674 0 : let state = get_state(&request);
2675 :
2676 0 : let timeline = active_timeline_of_active_tenant(&state.tenant_manager, TenantShardId::unsharded(tenant_id), timeline_id).await?;
2677 :
2678 0 : let mut body = StreamReader::new(request.into_body().map(|res| {
2679 0 : res.map_err(|error| {
2680 0 : std::io::Error::new(std::io::ErrorKind::Other, anyhow::anyhow!(error))
2681 0 : })
2682 0 : }));
2683 0 :
2684 0 : let last_record_lsn = timeline.get_last_record_lsn();
2685 0 : if last_record_lsn != start_lsn {
2686 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}")));
2687 0 : }
2688 0 :
2689 0 : // TODO leave clean state on error. For now you can use detach to clean
2690 0 : // up broken state from a failed import.
2691 0 :
2692 0 : // Import wal provided via CopyData
2693 0 : info!("importing wal");
2694 0 : crate::import_datadir::import_wal_from_tar(&timeline, &mut body, start_lsn, end_lsn, &ctx).await.map_err(ApiError::InternalServerError)?;
2695 0 : info!("wal import complete");
2696 :
2697 : // Read the end of the tar archive.
2698 0 : read_tar_eof(body).await.map_err(ApiError::InternalServerError)?;
2699 :
2700 : // TODO Does it make sense to overshoot?
2701 0 : if timeline.get_last_record_lsn() < end_lsn {
2702 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}")));
2703 0 : }
2704 0 :
2705 0 : // Flush data to disk, then upload to s3. No need for a forced checkpoint.
2706 0 : // We only want to persist the data, and it doesn't matter if it's in the
2707 0 : // shape of deltas or images.
2708 0 : info!("flushing layers");
2709 0 : timeline.freeze_and_flush().await.map_err(|e| match e {
2710 0 : tenant::timeline::FlushLayerError::Cancelled => ApiError::ShuttingDown,
2711 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
2712 0 : })?;
2713 :
2714 0 : info!("done");
2715 :
2716 0 : json_response(StatusCode::OK, ())
2717 0 : }.instrument(span).await
2718 0 : }
2719 :
2720 : /// Read the end of a tar archive.
2721 : ///
2722 : /// A tar archive normally ends with two consecutive blocks of zeros, 512 bytes each.
2723 : /// `tokio_tar` already read the first such block. Read the second all-zeros block,
2724 : /// and check that there is no more data after the EOF marker.
2725 : ///
2726 : /// 'tar' command can also write extra blocks of zeros, up to a record
2727 : /// size, controlled by the --record-size argument. Ignore them too.
2728 0 : async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow::Result<()> {
2729 : use tokio::io::AsyncReadExt;
2730 0 : let mut buf = [0u8; 512];
2731 0 :
2732 0 : // Read the all-zeros block, and verify it
2733 0 : let mut total_bytes = 0;
2734 0 : while total_bytes < 512 {
2735 0 : let nbytes = reader.read(&mut buf[total_bytes..]).await?;
2736 0 : total_bytes += nbytes;
2737 0 : if nbytes == 0 {
2738 0 : break;
2739 0 : }
2740 : }
2741 0 : if total_bytes < 512 {
2742 0 : anyhow::bail!("incomplete or invalid tar EOF marker");
2743 0 : }
2744 0 : if !buf.iter().all(|&x| x == 0) {
2745 0 : anyhow::bail!("invalid tar EOF marker");
2746 0 : }
2747 0 :
2748 0 : // Drain any extra zero-blocks after the EOF marker
2749 0 : let mut trailing_bytes = 0;
2750 0 : let mut seen_nonzero_bytes = false;
2751 : loop {
2752 0 : let nbytes = reader.read(&mut buf).await?;
2753 0 : trailing_bytes += nbytes;
2754 0 : if !buf.iter().all(|&x| x == 0) {
2755 0 : seen_nonzero_bytes = true;
2756 0 : }
2757 0 : if nbytes == 0 {
2758 0 : break;
2759 0 : }
2760 : }
2761 0 : if seen_nonzero_bytes {
2762 0 : anyhow::bail!("unexpected non-zero bytes after the tar archive");
2763 0 : }
2764 0 : if trailing_bytes % 512 != 0 {
2765 0 : anyhow::bail!("unexpected number of zeros ({trailing_bytes}), not divisible by tar block size (512 bytes), after the tar archive");
2766 0 : }
2767 0 : Ok(())
2768 0 : }
2769 :
2770 : /// Common functionality of all the HTTP API handlers.
2771 : ///
2772 : /// - Adds a tracing span to each request (by `request_span`)
2773 : /// - Logs the request depending on the request method (by `request_span`)
2774 : /// - Logs the response if it was not successful (by `request_span`
2775 : /// - Shields the handler function from async cancellations. Hyper can drop the handler
2776 : /// Future if the connection to the client is lost, but most of the pageserver code is
2777 : /// not async cancellation safe. This converts the dropped future into a graceful cancellation
2778 : /// request with a CancellationToken.
2779 0 : async fn api_handler<R, H>(request: Request<Body>, handler: H) -> Result<Response<Body>, ApiError>
2780 0 : where
2781 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2782 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2783 0 : {
2784 0 : if request.uri() != &"/v1/failpoints".parse::<Uri>().unwrap() {
2785 0 : fail::fail_point!("api-503", |_| Err(ApiError::ResourceUnavailable(
2786 0 : "failpoint".into()
2787 0 : )));
2788 :
2789 0 : fail::fail_point!("api-500", |_| Err(ApiError::InternalServerError(
2790 0 : anyhow::anyhow!("failpoint")
2791 0 : )));
2792 0 : }
2793 :
2794 : // Spawn a new task to handle the request, to protect the handler from unexpected
2795 : // async cancellations. Most pageserver functions are not async cancellation safe.
2796 : // We arm a drop-guard, so that if Hyper drops the Future, we signal the task
2797 : // with the cancellation token.
2798 0 : let token = CancellationToken::new();
2799 0 : let cancel_guard = token.clone().drop_guard();
2800 0 : let result = request_span(request, move |r| async {
2801 0 : let handle = tokio::spawn(
2802 0 : async {
2803 0 : let token_cloned = token.clone();
2804 0 : let result = handler(r, token).await;
2805 0 : if token_cloned.is_cancelled() {
2806 : // dropguard has executed: we will never turn this result into response.
2807 : //
2808 : // at least temporarily do {:?} logging; these failures are rare enough but
2809 : // could hide difficult errors.
2810 0 : match &result {
2811 0 : Ok(response) => {
2812 0 : let status = response.status();
2813 0 : info!(%status, "Cancelled request finished successfully")
2814 : }
2815 0 : Err(e) => error!("Cancelled request finished with an error: {e:?}"),
2816 : }
2817 0 : }
2818 : // only logging for cancelled panicked request handlers is the tracing_panic_hook,
2819 : // which should suffice.
2820 : //
2821 : // there is still a chance to lose the result due to race between
2822 : // returning from here and the actual connection closing happening
2823 : // before outer task gets to execute. leaving that up for #5815.
2824 0 : result
2825 0 : }
2826 0 : .in_current_span(),
2827 0 : );
2828 0 :
2829 0 : match handle.await {
2830 : // TODO: never actually return Err from here, always Ok(...) so that we can log
2831 : // spanned errors. Call api_error_handler instead and return appropriate Body.
2832 0 : Ok(result) => result,
2833 0 : Err(e) => {
2834 0 : // The handler task panicked. We have a global panic handler that logs the
2835 0 : // panic with its backtrace, so no need to log that here. Only log a brief
2836 0 : // message to make it clear that we returned the error to the client.
2837 0 : error!("HTTP request handler task panicked: {e:#}");
2838 :
2839 : // Don't return an Error here, because then fallback error handler that was
2840 : // installed in make_router() will print the error. Instead, construct the
2841 : // HTTP error response and return that.
2842 0 : Ok(
2843 0 : ApiError::InternalServerError(anyhow!("HTTP request handler task panicked"))
2844 0 : .into_response(),
2845 0 : )
2846 : }
2847 : }
2848 0 : })
2849 0 : .await;
2850 :
2851 0 : cancel_guard.disarm();
2852 0 :
2853 0 : result
2854 0 : }
2855 :
2856 : /// Like api_handler, but returns an error response if the server is built without
2857 : /// the 'testing' feature.
2858 0 : async fn testing_api_handler<R, H>(
2859 0 : desc: &str,
2860 0 : request: Request<Body>,
2861 0 : handler: H,
2862 0 : ) -> Result<Response<Body>, ApiError>
2863 0 : where
2864 0 : R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
2865 0 : H: FnOnce(Request<Body>, CancellationToken) -> R + Send + Sync + 'static,
2866 0 : {
2867 0 : if cfg!(feature = "testing") {
2868 0 : api_handler(request, handler).await
2869 : } else {
2870 0 : std::future::ready(Err(ApiError::BadRequest(anyhow!(
2871 0 : "Cannot {desc} because pageserver was compiled without testing APIs",
2872 0 : ))))
2873 0 : .await
2874 : }
2875 0 : }
2876 :
2877 0 : pub fn make_router(
2878 0 : state: Arc<State>,
2879 0 : launch_ts: &'static LaunchTimestamp,
2880 0 : auth: Option<Arc<SwappableJwtAuth>>,
2881 0 : ) -> anyhow::Result<RouterBuilder<hyper::Body, ApiError>> {
2882 0 : let spec = include_bytes!("openapi_spec.yml");
2883 0 : let mut router = attach_openapi_ui(endpoint::make_router(), spec, "/swagger.yml", "/v1/doc");
2884 0 : if auth.is_some() {
2885 0 : router = router.middleware(auth_middleware(|request| {
2886 0 : let state = get_state(request);
2887 0 : if state.allowlist_routes.contains(request.uri()) {
2888 0 : None
2889 : } else {
2890 0 : state.auth.as_deref()
2891 : }
2892 0 : }))
2893 0 : }
2894 :
2895 0 : router = router.middleware(
2896 0 : endpoint::add_response_header_middleware(
2897 0 : "PAGESERVER_LAUNCH_TIMESTAMP",
2898 0 : &launch_ts.to_string(),
2899 0 : )
2900 0 : .expect("construct launch timestamp header middleware"),
2901 0 : );
2902 0 :
2903 0 : Ok(router
2904 0 : .data(state)
2905 0 : .get("/metrics", |r| request_span(r, prometheus_metrics_handler))
2906 0 : .get("/v1/status", |r| api_handler(r, status_handler))
2907 0 : .put("/v1/failpoints", |r| {
2908 0 : testing_api_handler("manage failpoints", r, failpoints_handler)
2909 0 : })
2910 0 : .post("/v1/reload_auth_validation_keys", |r| {
2911 0 : api_handler(r, reload_auth_validation_keys_handler)
2912 0 : })
2913 0 : .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
2914 0 : .get("/v1/tenant/:tenant_shard_id", |r| {
2915 0 : api_handler(r, tenant_status)
2916 0 : })
2917 0 : .delete("/v1/tenant/:tenant_shard_id", |r| {
2918 0 : api_handler(r, tenant_delete_handler)
2919 0 : })
2920 0 : .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
2921 0 : api_handler(r, tenant_size_handler)
2922 0 : })
2923 0 : .put("/v1/tenant/config", |r| {
2924 0 : api_handler(r, update_tenant_config_handler)
2925 0 : })
2926 0 : .put("/v1/tenant/:tenant_shard_id/shard_split", |r| {
2927 0 : api_handler(r, tenant_shard_split_handler)
2928 0 : })
2929 0 : .get("/v1/tenant/:tenant_shard_id/config", |r| {
2930 0 : api_handler(r, get_tenant_config_handler)
2931 0 : })
2932 0 : .put("/v1/tenant/:tenant_shard_id/location_config", |r| {
2933 0 : api_handler(r, put_tenant_location_config_handler)
2934 0 : })
2935 0 : .get("/v1/location_config", |r| {
2936 0 : api_handler(r, list_location_config_handler)
2937 0 : })
2938 0 : .get("/v1/location_config/:tenant_shard_id", |r| {
2939 0 : api_handler(r, get_location_config_handler)
2940 0 : })
2941 0 : .put(
2942 0 : "/v1/tenant/:tenant_shard_id/time_travel_remote_storage",
2943 0 : |r| api_handler(r, tenant_time_travel_remote_storage_handler),
2944 0 : )
2945 0 : .get("/v1/tenant/:tenant_shard_id/timeline", |r| {
2946 0 : api_handler(r, timeline_list_handler)
2947 0 : })
2948 0 : .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
2949 0 : api_handler(r, timeline_create_handler)
2950 0 : })
2951 0 : .post("/v1/tenant/:tenant_shard_id/reset", |r| {
2952 0 : api_handler(r, tenant_reset_handler)
2953 0 : })
2954 0 : .post(
2955 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/preserve_initdb_archive",
2956 0 : |r| api_handler(r, timeline_preserve_initdb_handler),
2957 0 : )
2958 0 : .post(
2959 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/archival_config",
2960 0 : |r| api_handler(r, timeline_archival_config_handler),
2961 0 : )
2962 0 : .get("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
2963 0 : api_handler(r, timeline_detail_handler)
2964 0 : })
2965 0 : .get(
2966 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_lsn_by_timestamp",
2967 0 : |r| api_handler(r, get_lsn_by_timestamp_handler),
2968 0 : )
2969 0 : .get(
2970 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
2971 0 : |r| api_handler(r, get_timestamp_of_lsn_handler),
2972 0 : )
2973 0 : .post(
2974 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
2975 0 : |r| api_handler(r, lsn_lease_handler),
2976 0 : )
2977 0 : .put(
2978 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
2979 0 : |r| api_handler(r, timeline_gc_handler),
2980 0 : )
2981 0 : .put(
2982 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
2983 0 : |r| api_handler(r, timeline_compact_handler),
2984 0 : )
2985 0 : .put(
2986 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/checkpoint",
2987 0 : |r| testing_api_handler("run timeline checkpoint", r, timeline_checkpoint_handler),
2988 0 : )
2989 0 : .post(
2990 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2991 0 : |r| api_handler(r, timeline_download_remote_layers_handler_post),
2992 0 : )
2993 0 : .get(
2994 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/download_remote_layers",
2995 0 : |r| api_handler(r, timeline_download_remote_layers_handler_get),
2996 0 : )
2997 0 : .put(
2998 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/detach_ancestor",
2999 0 : |r| api_handler(r, timeline_detach_ancestor_handler),
3000 0 : )
3001 0 : .delete("/v1/tenant/:tenant_shard_id/timeline/:timeline_id", |r| {
3002 0 : api_handler(r, timeline_delete_handler)
3003 0 : })
3004 0 : .get(
3005 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer",
3006 0 : |r| api_handler(r, layer_map_info_handler),
3007 0 : )
3008 0 : .get(
3009 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3010 0 : |r| api_handler(r, layer_download_handler),
3011 0 : )
3012 0 : .delete(
3013 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
3014 0 : |r| api_handler(r, evict_timeline_layer_handler),
3015 0 : )
3016 0 : .post(
3017 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
3018 0 : |r| api_handler(r, timeline_gc_blocking_handler),
3019 0 : )
3020 0 : .post(
3021 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
3022 0 : |r| api_handler(r, timeline_gc_unblocking_handler),
3023 0 : )
3024 0 : .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
3025 0 : api_handler(r, secondary_upload_handler)
3026 0 : })
3027 0 : .get("/v1/tenant/:tenant_id/scan_remote_storage", |r| {
3028 0 : api_handler(r, tenant_scan_remote_handler)
3029 0 : })
3030 0 : .put("/v1/disk_usage_eviction/run", |r| {
3031 0 : api_handler(r, disk_usage_eviction_run)
3032 0 : })
3033 0 : .put("/v1/deletion_queue/flush", |r| {
3034 0 : api_handler(r, deletion_queue_flush)
3035 0 : })
3036 0 : .get("/v1/tenant/:tenant_shard_id/secondary/status", |r| {
3037 0 : api_handler(r, secondary_status_handler)
3038 0 : })
3039 0 : .post("/v1/tenant/:tenant_shard_id/secondary/download", |r| {
3040 0 : api_handler(r, secondary_download_handler)
3041 0 : })
3042 0 : .put("/v1/tenant/:tenant_shard_id/break", |r| {
3043 0 : testing_api_handler("set tenant state to broken", r, handle_tenant_break)
3044 0 : })
3045 0 : .get("/v1/panic", |r| api_handler(r, always_panic_handler))
3046 0 : .post("/v1/tracing/event", |r| {
3047 0 : testing_api_handler("emit a tracing event", r, post_tracing_event_handler)
3048 0 : })
3049 0 : .get(
3050 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/getpage",
3051 0 : |r| testing_api_handler("getpage@lsn", r, getpage_at_lsn_handler),
3052 0 : )
3053 0 : .get(
3054 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/keyspace",
3055 0 : |r| api_handler(r, timeline_collect_keyspace),
3056 0 : )
3057 0 : .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
3058 0 : .put("/v1/io_alignment", |r| {
3059 0 : api_handler(r, put_io_alignment_handler)
3060 0 : })
3061 0 : .put(
3062 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/force_aux_policy_switch",
3063 0 : |r| api_handler(r, force_aux_policy_switch_handler),
3064 0 : )
3065 0 : .get("/v1/utilization", |r| api_handler(r, get_utilization))
3066 0 : .post(
3067 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
3068 0 : |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
3069 0 : )
3070 0 : .post(
3071 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
3072 0 : |r| testing_api_handler("list_aux_files", r, list_aux_files),
3073 0 : )
3074 0 : .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
3075 0 : .post(
3076 0 : "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
3077 0 : |r| testing_api_handler("perf_info", r, perf_info),
3078 0 : )
3079 0 : .put(
3080 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_basebackup",
3081 0 : |r| api_handler(r, put_tenant_timeline_import_basebackup),
3082 0 : )
3083 0 : .put(
3084 0 : "/v1/tenant/:tenant_id/timeline/:timeline_id/import_wal",
3085 0 : |r| api_handler(r, put_tenant_timeline_import_wal),
3086 0 : )
3087 0 : .any(handler_404))
3088 0 : }
|
