Line data Source code
1 : //! This module manages synchronizing local FS with remote storage.
2 : //!
3 : //! # Overview
4 : //!
5 : //! * [`RemoteTimelineClient`] provides functions related to upload/download of a particular timeline.
6 : //! It contains a queue of pending uploads, and manages the queue, performing uploads in parallel
7 : //! when it's safe to do so.
8 : //!
9 : //! * Stand-alone function, [`list_remote_timelines`], to get list of timelines of a tenant.
10 : //!
11 : //! These functions use the low-level remote storage client, [`remote_storage::RemoteStorage`].
12 : //!
13 : //! # APIs & How To Use Them
14 : //!
15 : //! There is a [RemoteTimelineClient] for each [Timeline][`crate::tenant::Timeline`] in the system,
16 : //! unless the pageserver is configured without remote storage.
17 : //!
18 : //! We allocate the client instance in [Timeline][`crate::tenant::Timeline`], i.e.,
19 : //! either in [`crate::tenant::mgr`] during startup or when creating a new
20 : //! timeline.
21 : //! However, the client does not become ready for use until we've initialized its upload queue:
22 : //!
23 : //! - For timelines that already have some state on the remote storage, we use
24 : //! [`RemoteTimelineClient::init_upload_queue`] .
25 : //! - For newly created timelines, we use
26 : //! [`RemoteTimelineClient::init_upload_queue_for_empty_remote`].
27 : //!
28 : //! The former takes the remote's [`IndexPart`] as an argument, possibly retrieved
29 : //! using [`list_remote_timelines`]. We'll elaborate on [`IndexPart`] in the next section.
30 : //!
31 : //! Whenever we've created/updated/deleted a file in a timeline directory, we schedule
32 : //! the corresponding remote operation with the timeline's [`RemoteTimelineClient`]:
33 : //!
34 : //! - [`RemoteTimelineClient::schedule_layer_file_upload`] when we've created a new layer file.
35 : //! - [`RemoteTimelineClient::schedule_index_upload_for_metadata_update`] when we've updated the timeline metadata file.
36 : //! - [`RemoteTimelineClient::schedule_index_upload_for_file_changes`] to upload an updated index file, after we've scheduled file uploads
37 : //! - [`RemoteTimelineClient::schedule_layer_file_deletion`] when we've deleted one or more layer files.
38 : //!
39 : //! Internally, these functions create [`UploadOp`]s and put them in a queue.
40 : //!
41 : //! There are also APIs for downloading files.
42 : //! These are not part of the aforementioned queuing and will not be discussed
43 : //! further here, except in the section covering tenant attach.
44 : //!
45 : //! # Remote Storage Structure & [`IndexPart`] Index File
46 : //!
47 : //! The "directory structure" in the remote storage mirrors the local directory structure, with paths
48 : //! like `tenants/<tenant_id>/timelines/<timeline_id>/<layer filename>`.
49 : //! Yet instead of keeping the `metadata` file remotely, we wrap it with more
50 : //! data in an "index file" aka [`IndexPart`], containing the list of **all** remote
51 : //! files for a given timeline.
52 : //! If a file is not referenced from [`IndexPart`], it's not part of the remote storage state.
53 : //!
54 : //! Having the `IndexPart` also avoids expensive and slow `S3 list` commands.
55 : //!
56 : //! # Consistency
57 : //!
58 : //! To have a consistent remote structure, it's important that uploads and
59 : //! deletions are performed in the right order. For example, the index file
60 : //! contains a list of layer files, so it must not be uploaded until all the
61 : //! layer files that are in its list have been successfully uploaded.
62 : //!
63 : //! The contract between client and its user is that the user is responsible of
64 : //! scheduling operations in an order that keeps the remote consistent as
65 : //! described above.
66 : //!
67 : //! From the user's perspective, the operations are executed sequentially.
68 : //! Internally, the client knows which operations can be performed in parallel,
69 : //! and which operations act like a "barrier" that require preceding operations
70 : //! to finish. The calling code just needs to call the schedule-functions in the
71 : //! correct order, and the client will parallelize the operations in a way that
72 : //! is safe. For more details, see `UploadOp::can_bypass`.
73 : //!
74 : //! All of this relies on the following invariants:
75 : //!
76 : //! - We rely on read-after write consistency in the remote storage.
77 : //! - Layer files are immutable.
78 : //!
79 : //! NB: Pageserver assumes that it has exclusive write access to the tenant in remote
80 : //! storage. Different tenants can be attached to different pageservers, but if the
81 : //! same tenant is attached to two pageservers at the same time, they will overwrite
82 : //! each other's index file updates, and confusion will ensue. There's no interlock or
83 : //! mechanism to detect that in the pageserver, we rely on the control plane to ensure
84 : //! that that doesn't happen.
85 : //!
86 : //! ## Implementation Note
87 : //!
88 : //! The *actual* remote state lags behind the *desired* remote state while
89 : //! there are in-flight operations.
90 : //! We keep track of the desired remote state in [`UploadQueueInitialized::dirty`].
91 : //! It is initialized based on the [`IndexPart`] that was passed during init
92 : //! and updated with every `schedule_*` function call.
93 : //! All this is necessary necessary to compute the future [`IndexPart`]s
94 : //! when scheduling an operation while other operations that also affect the
95 : //! remote [`IndexPart`] are in flight.
96 : //!
97 : //! # Retries & Error Handling
98 : //!
99 : //! The client retries operations indefinitely, using exponential back-off.
100 : //! There is no way to force a retry, i.e., interrupt the back-off.
101 : //! This could be built easily.
102 : //!
103 : //! # Cancellation
104 : //!
105 : //! The operations execute as plain [`task_mgr`] tasks, scoped to
106 : //! the client's tenant and timeline.
107 : //! Dropping the client will drop queued operations but not executing operations.
108 : //! These will complete unless the `task_mgr` tasks are cancelled using `task_mgr`
109 : //! APIs, e.g., during pageserver shutdown, timeline delete, or tenant detach.
110 : //!
111 : //! # Completion
112 : //!
113 : //! Once an operation has completed, we update [`UploadQueueInitialized::clean`] immediately,
114 : //! and submit a request through the DeletionQueue to update
115 : //! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
116 : //! validated that our generation is not stale. It is this visible value
117 : //! that is advertized to safekeepers as a signal that that they can
118 : //! delete the WAL up to that LSN.
119 : //!
120 : //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
121 : //! for all pending operations to complete. It does not prevent more
122 : //! operations from getting scheduled.
123 : //!
124 : //! # Crash Consistency
125 : //!
126 : //! We do not persist the upload queue state.
127 : //! If we drop the client, or crash, all unfinished operations are lost.
128 : //!
129 : //! To recover, the following steps need to be taken:
130 : //! - Retrieve the current remote [`IndexPart`]. This gives us a
131 : //! consistent remote state, assuming the user scheduled the operations in
132 : //! the correct order.
133 : //! - Initiate upload queue with that [`IndexPart`].
134 : //! - Reschedule all lost operations by comparing the local filesystem state
135 : //! and remote state as per [`IndexPart`]. This is done in
136 : //! [`TenantShard::timeline_init_and_sync`].
137 : //!
138 : //! Note that if we crash during file deletion between the index update
139 : //! that removes the file from the list of files, and deleting the remote file,
140 : //! the file is leaked in the remote storage. Similarly, if a new file is created
141 : //! and uploaded, but the pageserver dies permanently before updating the
142 : //! remote index file, the new file is leaked in remote storage. We accept and
143 : //! tolerate that for now.
144 : //! Note further that we cannot easily fix this by scheduling deletes for every
145 : //! file that is present only on the remote, because we cannot distinguish the
146 : //! following two cases:
147 : //! - (1) We had the file locally, deleted it locally, scheduled a remote delete,
148 : //! but crashed before it finished remotely.
149 : //! - (2) We never had the file locally because we haven't on-demand downloaded
150 : //! it yet.
151 : //!
152 : //! # Downloads
153 : //!
154 : //! In addition to the upload queue, [`RemoteTimelineClient`] has functions for
155 : //! downloading files from the remote storage. Downloads are performed immediately
156 : //! against the `RemoteStorage`, independently of the upload queue.
157 : //!
158 : //! When we attach a tenant, we perform the following steps:
159 : //! - create `Tenant` object in `TenantState::Attaching` state
160 : //! - List timelines that are present in remote storage, and for each:
161 : //! - download their remote [`IndexPart`]s
162 : //! - create `Timeline` struct and a `RemoteTimelineClient`
163 : //! - initialize the client's upload queue with its `IndexPart`
164 : //! - schedule uploads for layers that are only present locally.
165 : //! - After the above is done for each timeline, open the tenant for business by
166 : //! transitioning it from `TenantState::Attaching` to `TenantState::Active` state.
167 : //! This starts the timelines' WAL-receivers and the tenant's GC & Compaction loops.
168 : //!
169 : //! # Operating Without Remote Storage
170 : //!
171 : //! If no remote storage configuration is provided, the [`RemoteTimelineClient`] is
172 : //! not created and the uploads are skipped.
173 : //!
174 : //! [`TenantShard::timeline_init_and_sync`]: super::TenantShard::timeline_init_and_sync
175 : //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
176 :
177 : pub(crate) mod download;
178 : pub mod index;
179 : pub mod manifest;
180 : pub(crate) mod upload;
181 :
182 : use std::collections::{HashMap, HashSet, VecDeque};
183 : use std::ops::DerefMut;
184 : use std::sync::atomic::{AtomicU32, Ordering};
185 : use std::sync::{Arc, Mutex, OnceLock};
186 : use std::time::Duration;
187 :
188 : use anyhow::Context;
189 : use camino::Utf8Path;
190 : use chrono::{NaiveDateTime, Utc};
191 : pub(crate) use download::{
192 : download_index_part, download_initdb_tar_zst, download_tenant_manifest, is_temp_download_file,
193 : list_remote_tenant_shards, list_remote_timelines,
194 : };
195 : use index::GcCompactionState;
196 : pub(crate) use index::LayerFileMetadata;
197 : use pageserver_api::models::{RelSizeMigration, TimelineArchivalState, TimelineVisibilityState};
198 : use pageserver_api::shard::{ShardIndex, TenantShardId};
199 : use regex::Regex;
200 : use remote_storage::{
201 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
202 : };
203 : use scopeguard::ScopeGuard;
204 : use tokio_util::sync::CancellationToken;
205 : use tracing::{Instrument, debug, error, info, info_span, instrument, warn};
206 : pub(crate) use upload::upload_initdb_dir;
207 : use utils::backoff::{
208 : self, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS, exponential_backoff,
209 : };
210 : use utils::id::{TenantId, TimelineId};
211 : use utils::lsn::Lsn;
212 : use utils::pausable_failpoint;
213 : use utils::shard::ShardNumber;
214 :
215 : use self::index::IndexPart;
216 : use super::config::AttachedLocationConfig;
217 : use super::metadata::MetadataUpdate;
218 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
219 : use super::timeline::import_pgdata;
220 : use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
221 : use super::{DeleteTimelineError, Generation};
222 : use crate::config::PageServerConf;
223 : use crate::context::RequestContext;
224 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
225 : use crate::metrics::{
226 : MeasureRemoteOp, REMOTE_ONDEMAND_DOWNLOADED_BYTES, REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
227 : RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
228 : RemoteTimelineClientMetricsCallTrackSize,
229 : };
230 : use crate::task_mgr::{BACKGROUND_RUNTIME, TaskKind, shutdown_token};
231 : use crate::tenant::metadata::TimelineMetadata;
232 : use crate::tenant::remote_timeline_client::download::download_retry;
233 : use crate::tenant::storage_layer::AsLayerDesc;
234 : use crate::tenant::upload_queue::{
235 : Delete, OpType, UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped,
236 : UploadQueueStoppedDeletable, UploadTask,
237 : };
238 : use crate::tenant::{TIMELINES_SEGMENT_NAME, debug_assert_current_span_has_tenant_and_timeline_id};
239 : use crate::{TENANT_HEATMAP_BASENAME, task_mgr};
240 :
241 : // Occasional network issues and such can cause remote operations to fail, and
242 : // that's expected. If a download fails, we log it at info-level, and retry.
243 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
244 : // level instead, as repeated failures can mean a more serious problem. If it
245 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
246 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
247 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
248 :
249 : // Similarly log failed uploads and deletions at WARN level, after this many
250 : // retries. Uploads and deletions are retried forever, though.
251 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
252 :
253 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
254 :
255 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
256 :
257 : /// Default buffer size when interfacing with [`tokio::fs::File`].
258 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
259 :
260 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
261 : /// which we warn and skip.
262 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
263 :
264 : pub enum MaybeDeletedIndexPart {
265 : IndexPart(IndexPart),
266 : Deleted(IndexPart),
267 : }
268 :
269 : #[derive(Debug, thiserror::Error)]
270 : pub enum PersistIndexPartWithDeletedFlagError {
271 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
272 : AlreadyInProgress(NaiveDateTime),
273 : #[error("the deleted_flag was already set, value is {0:?}")]
274 : AlreadyDeleted(NaiveDateTime),
275 : #[error(transparent)]
276 : Other(#[from] anyhow::Error),
277 : }
278 :
279 : #[derive(Debug, thiserror::Error)]
280 : pub enum WaitCompletionError {
281 : #[error(transparent)]
282 : NotInitialized(NotInitialized),
283 : #[error("wait_completion aborted because upload queue was stopped")]
284 : UploadQueueShutDownOrStopped,
285 : }
286 :
287 : #[derive(Debug, thiserror::Error)]
288 : #[error("Upload queue either in unexpected state or hasn't downloaded manifest yet")]
289 : pub struct UploadQueueNotReadyError;
290 :
291 : #[derive(Debug, thiserror::Error)]
292 : pub enum ShutdownIfArchivedError {
293 : #[error(transparent)]
294 : NotInitialized(NotInitialized),
295 : #[error("timeline is not archived")]
296 : NotArchived,
297 : }
298 :
299 : /// Behavioral modes that enable seamless live migration.
300 : ///
301 : /// See docs/rfcs/028-pageserver-migration.md to understand how these fit in.
302 : struct RemoteTimelineClientConfig {
303 : /// If this is false, then update to remote_consistent_lsn are dropped rather
304 : /// than being submitted to DeletionQueue for validation. This behavior is
305 : /// used when a tenant attachment is known to have a stale generation number,
306 : /// such that validation attempts will always fail. This is not necessary
307 : /// for correctness, but avoids spamming error statistics with failed validations
308 : /// when doing migrations of tenants.
309 : process_remote_consistent_lsn_updates: bool,
310 :
311 : /// If this is true, then object deletions are held in a buffer in RemoteTimelineClient
312 : /// rather than being submitted to the DeletionQueue. This behavior is used when a tenant
313 : /// is known to be multi-attached, in order to avoid disrupting other attached tenants
314 : /// whose generations' metadata refers to the deleted objects.
315 : block_deletions: bool,
316 : }
317 :
318 : /// RemoteTimelineClientConfig's state is entirely driven by LocationConf, but we do
319 : /// not carry the entire LocationConf structure: it's much more than we need. The From
320 : /// impl extracts the subset of the LocationConf that is interesting to RemoteTimelineClient.
321 : impl From<&AttachedLocationConfig> for RemoteTimelineClientConfig {
322 240 : fn from(lc: &AttachedLocationConfig) -> Self {
323 240 : Self {
324 240 : block_deletions: !lc.may_delete_layers_hint(),
325 240 : process_remote_consistent_lsn_updates: lc.may_upload_layers_hint(),
326 240 : }
327 240 : }
328 : }
329 :
330 : /// A client for accessing a timeline's data in remote storage.
331 : ///
332 : /// This takes care of managing the number of connections, and balancing them
333 : /// across tenants. This also handles retries of failed uploads.
334 : ///
335 : /// Upload and delete requests are ordered so that before a deletion is
336 : /// performed, we wait for all preceding uploads to finish. This ensures sure
337 : /// that if you perform a compaction operation that reshuffles data in layer
338 : /// files, we don't have a transient state where the old files have already been
339 : /// deleted, but new files have not yet been uploaded.
340 : ///
341 : /// Similarly, this enforces an order between index-file uploads, and layer
342 : /// uploads. Before an index-file upload is performed, all preceding layer
343 : /// uploads must be finished.
344 : ///
345 : /// This also maintains a list of remote files, and automatically includes that
346 : /// in the index part file, whenever timeline metadata is uploaded.
347 : ///
348 : /// Downloads are not queued, they are performed immediately.
349 : pub(crate) struct RemoteTimelineClient {
350 : conf: &'static PageServerConf,
351 :
352 : runtime: tokio::runtime::Handle,
353 :
354 : tenant_shard_id: TenantShardId,
355 : timeline_id: TimelineId,
356 : generation: Generation,
357 :
358 : upload_queue: Mutex<UploadQueue>,
359 :
360 : pub(crate) metrics: Arc<RemoteTimelineClientMetrics>,
361 :
362 : storage_impl: GenericRemoteStorage,
363 :
364 : deletion_queue_client: DeletionQueueClient,
365 :
366 : /// Subset of tenant configuration used to control upload behaviors during migrations
367 : config: std::sync::RwLock<RemoteTimelineClientConfig>,
368 :
369 : cancel: CancellationToken,
370 : }
371 :
372 : impl Drop for RemoteTimelineClient {
373 10 : fn drop(&mut self) {
374 10 : debug!("dropping RemoteTimelineClient");
375 10 : }
376 : }
377 :
378 : impl RemoteTimelineClient {
379 : ///
380 : /// Create a remote storage client for given timeline
381 : ///
382 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
383 : /// by calling init_upload_queue.
384 : ///
385 235 : pub(crate) fn new(
386 235 : remote_storage: GenericRemoteStorage,
387 235 : deletion_queue_client: DeletionQueueClient,
388 235 : conf: &'static PageServerConf,
389 235 : tenant_shard_id: TenantShardId,
390 235 : timeline_id: TimelineId,
391 235 : generation: Generation,
392 235 : location_conf: &AttachedLocationConfig,
393 235 : ) -> RemoteTimelineClient {
394 : RemoteTimelineClient {
395 235 : conf,
396 235 : runtime: if cfg!(test) {
397 : // remote_timeline_client.rs tests rely on current-thread runtime
398 235 : tokio::runtime::Handle::current()
399 : } else {
400 0 : BACKGROUND_RUNTIME.handle().clone()
401 : },
402 235 : tenant_shard_id,
403 235 : timeline_id,
404 235 : generation,
405 235 : storage_impl: remote_storage,
406 235 : deletion_queue_client,
407 235 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
408 235 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
409 235 : &tenant_shard_id,
410 235 : &timeline_id,
411 : )),
412 235 : config: std::sync::RwLock::new(RemoteTimelineClientConfig::from(location_conf)),
413 235 : cancel: CancellationToken::new(),
414 : }
415 235 : }
416 :
417 : /// Initialize the upload queue for a remote storage that already received
418 : /// an index file upload, i.e., it's not empty.
419 : /// The given `index_part` must be the one on the remote.
420 3 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
421 : // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
422 : // certainly no point in starting more upload tasks than this.
423 3 : let inprogress_limit = self
424 3 : .conf
425 3 : .remote_storage_config
426 3 : .as_ref()
427 3 : .map_or(0, |r| r.concurrency_limit());
428 3 : let mut upload_queue = self.upload_queue.lock().unwrap();
429 3 : upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
430 3 : self.update_remote_physical_size_gauge(Some(index_part));
431 3 : info!(
432 0 : "initialized upload queue from remote index with {} layer files",
433 0 : index_part.layer_metadata.len()
434 : );
435 3 : Ok(())
436 3 : }
437 :
438 : /// Initialize the upload queue for the case where the remote storage is empty,
439 : /// i.e., it doesn't have an `IndexPart`.
440 : ///
441 : /// `rel_size_v2_status` needs to be carried over during branching, and that's why
442 : /// it's passed in here.
443 232 : pub fn init_upload_queue_for_empty_remote(
444 232 : &self,
445 232 : local_metadata: &TimelineMetadata,
446 232 : rel_size_v2_status: Option<RelSizeMigration>,
447 232 : ) -> anyhow::Result<()> {
448 : // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
449 : // certainly no point in starting more upload tasks than this.
450 232 : let inprogress_limit = self
451 232 : .conf
452 232 : .remote_storage_config
453 232 : .as_ref()
454 232 : .map_or(0, |r| r.concurrency_limit());
455 232 : let mut upload_queue = self.upload_queue.lock().unwrap();
456 232 : let initialized_queue =
457 232 : upload_queue.initialize_empty_remote(local_metadata, inprogress_limit)?;
458 232 : initialized_queue.dirty.rel_size_migration = rel_size_v2_status;
459 232 : self.update_remote_physical_size_gauge(None);
460 232 : info!("initialized upload queue as empty");
461 232 : Ok(())
462 232 : }
463 :
464 : /// Initialize the queue in stopped state. Used in startup path
465 : /// to continue deletion operation interrupted by pageserver crash or restart.
466 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
467 0 : &self,
468 0 : index_part: &IndexPart,
469 0 : ) -> anyhow::Result<()> {
470 : // FIXME: consider newtype for DeletedIndexPart.
471 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
472 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
473 0 : ))?;
474 0 : let inprogress_limit = self
475 0 : .conf
476 0 : .remote_storage_config
477 0 : .as_ref()
478 0 : .map_or(0, |r| r.concurrency_limit());
479 :
480 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
481 0 : upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
482 0 : self.update_remote_physical_size_gauge(Some(index_part));
483 0 : self.stop_impl(&mut upload_queue);
484 :
485 0 : upload_queue
486 0 : .stopped_mut()
487 0 : .expect("stopped above")
488 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
489 :
490 0 : Ok(())
491 0 : }
492 :
493 : /// Notify this client of a change to its parent tenant's config, as this may cause us to
494 : /// take action (unblocking deletions when transitioning from AttachedMulti to AttachedSingle)
495 0 : pub(super) fn update_config(&self, location_conf: &AttachedLocationConfig) {
496 0 : let new_conf = RemoteTimelineClientConfig::from(location_conf);
497 0 : let unblocked = !new_conf.block_deletions;
498 :
499 : // Update config before draining deletions, so that we don't race with more being
500 : // inserted. This can result in deletions happening our of order, but that does not
501 : // violate any invariants: deletions only need to be ordered relative to upload of the index
502 : // that dereferences the deleted objects, and we are not changing that order.
503 0 : *self.config.write().unwrap() = new_conf;
504 :
505 0 : if unblocked {
506 : // If we may now delete layers, drain any that were blocked in our old
507 : // configuration state
508 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
509 :
510 0 : if let Ok(queue) = queue_locked.initialized_mut() {
511 0 : let blocked_deletions = std::mem::take(&mut queue.blocked_deletions);
512 0 : for d in blocked_deletions {
513 0 : if let Err(e) = self.deletion_queue_client.push_layers(
514 0 : self.tenant_shard_id,
515 0 : self.timeline_id,
516 0 : self.generation,
517 0 : d.layers,
518 0 : ) {
519 : // This could happen if the pageserver is shut down while a tenant
520 : // is transitioning from a deletion-blocked state: we will leak some
521 : // S3 objects in this case.
522 0 : warn!("Failed to drain blocked deletions: {}", e);
523 0 : break;
524 0 : }
525 : }
526 0 : }
527 0 : }
528 0 : }
529 :
530 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
531 2 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
532 2 : match &mut *self.upload_queue.lock().unwrap() {
533 0 : UploadQueue::Uninitialized => None,
534 2 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
535 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
536 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
537 0 : .upload_queue_for_deletion
538 0 : .get_last_remote_consistent_lsn_projected(),
539 : }
540 2 : }
541 :
542 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
543 0 : match &mut *self.upload_queue.lock().unwrap() {
544 0 : UploadQueue::Uninitialized => None,
545 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
546 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
547 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
548 0 : q.upload_queue_for_deletion
549 0 : .get_last_remote_consistent_lsn_visible(),
550 0 : ),
551 : }
552 0 : }
553 :
554 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
555 : /// client is currently initialized.
556 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
557 0 : self.upload_queue
558 0 : .lock()
559 0 : .unwrap()
560 0 : .initialized_mut()
561 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
562 0 : .unwrap_or(false)
563 0 : }
564 :
565 : /// Returns whether the timeline is archived.
566 : /// Return None if the remote index_part hasn't been downloaded yet.
567 8 : pub(crate) fn is_archived(&self) -> Option<bool> {
568 8 : self.upload_queue
569 8 : .lock()
570 8 : .unwrap()
571 8 : .initialized_mut()
572 8 : .map(|q| q.clean.0.archived_at.is_some())
573 8 : .ok()
574 8 : }
575 :
576 : /// Returns true if the timeline is invisible in synthetic size calculations.
577 8 : pub(crate) fn is_invisible(&self) -> Option<bool> {
578 8 : self.upload_queue
579 8 : .lock()
580 8 : .unwrap()
581 8 : .initialized_mut()
582 8 : .map(|q| q.clean.0.marked_invisible_at.is_some())
583 8 : .ok()
584 8 : }
585 :
586 : /// Returns `Ok(Some(timestamp))` if the timeline has been archived, `Ok(None)` if the timeline hasn't been archived.
587 : ///
588 : /// Return Err(_) if the remote index_part hasn't been downloaded yet, or the timeline hasn't been stopped yet.
589 1 : pub(crate) fn archived_at_stopped_queue(
590 1 : &self,
591 1 : ) -> Result<Option<NaiveDateTime>, UploadQueueNotReadyError> {
592 1 : self.upload_queue
593 1 : .lock()
594 1 : .unwrap()
595 1 : .stopped_mut()
596 1 : .map(|q| q.upload_queue_for_deletion.clean.0.archived_at)
597 1 : .map_err(|_| UploadQueueNotReadyError)
598 1 : }
599 :
600 1005 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
601 1005 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
602 773 : current_remote_index_part
603 773 : .layer_metadata
604 773 : .values()
605 773 : .map(|ilmd| ilmd.file_size)
606 773 : .sum()
607 : } else {
608 232 : 0
609 : };
610 1005 : self.metrics.remote_physical_size_gauge.set(size);
611 1005 : }
612 :
613 1 : pub fn get_remote_physical_size(&self) -> u64 {
614 1 : self.metrics.remote_physical_size_gauge.get()
615 1 : }
616 :
617 : //
618 : // Download operations.
619 : //
620 : // These don't use the per-timeline queue. They do use the global semaphore in
621 : // S3Bucket, to limit the total number of concurrent operations, though.
622 : //
623 :
624 : /// Download index file
625 10 : pub async fn download_index_file(
626 10 : &self,
627 10 : cancel: &CancellationToken,
628 10 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
629 10 : let _unfinished_gauge_guard = self.metrics.call_begin(
630 10 : &RemoteOpFileKind::Index,
631 10 : &RemoteOpKind::Download,
632 10 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
633 10 : reason: "no need for a downloads gauge",
634 10 : },
635 10 : );
636 :
637 10 : let (index_part, index_generation, index_last_modified) = download::download_index_part(
638 10 : &self.storage_impl,
639 10 : &self.tenant_shard_id,
640 10 : &self.timeline_id,
641 10 : self.generation,
642 10 : cancel,
643 10 : )
644 10 : .measure_remote_op(
645 10 : Option::<TaskKind>::None,
646 10 : RemoteOpFileKind::Index,
647 10 : RemoteOpKind::Download,
648 10 : Arc::clone(&self.metrics),
649 10 : )
650 10 : .await?;
651 :
652 : // Defense in depth: monotonicity of generation numbers is an important correctness guarantee, so when we see a very
653 : // old index, we do extra checks in case this is the result of backward time-travel of the generation number (e.g.
654 : // in case of a bug in the service that issues generation numbers). Indices are allowed to be old, but we expect that
655 : // when we load an old index we are loading the _latest_ index: if we are asked to load an old index and there is
656 : // also a newer index available, that is surprising.
657 : const INDEX_AGE_CHECKS_THRESHOLD: Duration = Duration::from_secs(14 * 24 * 3600);
658 10 : let index_age = index_last_modified.elapsed().unwrap_or_else(|e| {
659 0 : if e.duration() > Duration::from_secs(5) {
660 : // We only warn if the S3 clock and our local clock are >5s out: because this is a low resolution
661 : // timestamp, it is common to be out by at least 1 second.
662 0 : tracing::warn!("Index has modification time in the future: {e}");
663 0 : }
664 0 : Duration::ZERO
665 0 : });
666 10 : if index_age > INDEX_AGE_CHECKS_THRESHOLD {
667 0 : tracing::info!(
668 : ?index_generation,
669 0 : age = index_age.as_secs_f64(),
670 0 : "Loaded an old index, checking for other indices..."
671 : );
672 :
673 : // Find the highest-generation index
674 0 : let (_latest_index_part, latest_index_generation, latest_index_mtime) =
675 0 : download::download_index_part(
676 0 : &self.storage_impl,
677 0 : &self.tenant_shard_id,
678 0 : &self.timeline_id,
679 0 : Generation::MAX,
680 0 : cancel,
681 0 : )
682 0 : .await?;
683 :
684 0 : if latest_index_generation > index_generation {
685 : // Unexpected! Why are we loading such an old index if a more recent one exists?
686 : // We will refuse to proceed, as there is no reasonable scenario where this should happen, but
687 : // there _is_ a clear bug/corruption scenario where it would happen (controller sets the generation
688 : // backwards).
689 0 : tracing::error!(
690 : ?index_generation,
691 : ?latest_index_generation,
692 : ?latest_index_mtime,
693 0 : "Found a newer index while loading an old one"
694 : );
695 0 : return Err(DownloadError::Fatal(
696 0 : "Index age exceeds threshold and a newer index exists".into(),
697 0 : ));
698 0 : }
699 10 : }
700 :
701 10 : if index_part.deleted_at.is_some() {
702 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
703 : } else {
704 10 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
705 : }
706 10 : }
707 :
708 : /// Download a (layer) file from `path`, into local filesystem.
709 : ///
710 : /// 'layer_metadata' is the metadata from the remote index file.
711 : ///
712 : /// On success, returns the size of the downloaded file.
713 7 : pub async fn download_layer_file(
714 7 : &self,
715 7 : layer_file_name: &LayerName,
716 7 : layer_metadata: &LayerFileMetadata,
717 7 : local_path: &Utf8Path,
718 7 : gate: &utils::sync::gate::Gate,
719 7 : cancel: &CancellationToken,
720 7 : ctx: &RequestContext,
721 7 : ) -> Result<u64, DownloadError> {
722 7 : let downloaded_size = {
723 7 : let _unfinished_gauge_guard = self.metrics.call_begin(
724 7 : &RemoteOpFileKind::Layer,
725 7 : &RemoteOpKind::Download,
726 7 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
727 7 : reason: "no need for a downloads gauge",
728 7 : },
729 7 : );
730 7 : download::download_layer_file(
731 7 : self.conf,
732 7 : &self.storage_impl,
733 7 : self.tenant_shard_id,
734 7 : self.timeline_id,
735 7 : layer_file_name,
736 7 : layer_metadata,
737 7 : local_path,
738 7 : gate,
739 7 : cancel,
740 7 : ctx,
741 7 : )
742 7 : .measure_remote_op(
743 7 : Some(ctx.task_kind()),
744 7 : RemoteOpFileKind::Layer,
745 7 : RemoteOpKind::Download,
746 7 : Arc::clone(&self.metrics),
747 7 : )
748 7 : .await?
749 : };
750 :
751 7 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
752 7 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
753 :
754 7 : Ok(downloaded_size)
755 7 : }
756 :
757 : //
758 : // Upload operations.
759 : //
760 :
761 : /// Launch an index-file upload operation in the background, with
762 : /// fully updated metadata.
763 : ///
764 : /// This should only be used to upload initial metadata to remote storage.
765 : ///
766 : /// The upload will be added to the queue immediately, but it
767 : /// won't be performed until all previously scheduled layer file
768 : /// upload operations have completed successfully. This is to
769 : /// ensure that when the index file claims that layers X, Y and Z
770 : /// exist in remote storage, they really do. To wait for the upload
771 : /// to complete, use `wait_completion`.
772 : ///
773 : /// If there were any changes to the list of files, i.e. if any
774 : /// layer file uploads were scheduled, since the last index file
775 : /// upload, those will be included too.
776 118 : pub fn schedule_index_upload_for_full_metadata_update(
777 118 : self: &Arc<Self>,
778 118 : metadata: &TimelineMetadata,
779 118 : ) -> anyhow::Result<()> {
780 118 : let mut guard = self.upload_queue.lock().unwrap();
781 118 : let upload_queue = guard.initialized_mut()?;
782 :
783 : // As documented in the struct definition, it's ok for latest_metadata to be
784 : // ahead of what's _actually_ on the remote during index upload.
785 118 : upload_queue.dirty.metadata = metadata.clone();
786 :
787 118 : self.schedule_index_upload(upload_queue);
788 :
789 118 : Ok(())
790 118 : }
791 :
792 : /// Launch an index-file upload operation in the background, with only parts of the metadata
793 : /// updated.
794 : ///
795 : /// This is the regular way of updating metadata on layer flushes or Gc.
796 : ///
797 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
798 : /// `index_part.json`, while being more clear on what values update regularly.
799 623 : pub(crate) fn schedule_index_upload_for_metadata_update(
800 623 : self: &Arc<Self>,
801 623 : update: &MetadataUpdate,
802 623 : ) -> anyhow::Result<()> {
803 623 : let mut guard = self.upload_queue.lock().unwrap();
804 623 : let upload_queue = guard.initialized_mut()?;
805 :
806 623 : upload_queue.dirty.metadata.apply(update);
807 :
808 : // Defense in depth: if we somehow generated invalid metadata, do not persist it.
809 623 : upload_queue
810 623 : .dirty
811 623 : .validate()
812 623 : .map_err(|e| anyhow::anyhow!(e))?;
813 :
814 623 : self.schedule_index_upload(upload_queue);
815 :
816 623 : Ok(())
817 623 : }
818 :
819 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
820 : ///
821 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
822 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
823 : /// been in the queue yet.
824 1 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
825 1 : self: &Arc<Self>,
826 1 : state: TimelineArchivalState,
827 1 : ) -> anyhow::Result<bool> {
828 1 : let mut guard = self.upload_queue.lock().unwrap();
829 1 : let upload_queue = guard.initialized_mut()?;
830 :
831 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
832 : /// change needed to set archived_at.
833 2 : fn need_change(
834 2 : archived_at: &Option<NaiveDateTime>,
835 2 : state: TimelineArchivalState,
836 2 : ) -> Option<bool> {
837 2 : match (archived_at, state) {
838 : (Some(_), TimelineArchivalState::Archived)
839 : | (None, TimelineArchivalState::Unarchived) => {
840 : // Nothing to do
841 0 : tracing::info!("intended state matches present state");
842 0 : None
843 : }
844 2 : (None, TimelineArchivalState::Archived) => Some(true),
845 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
846 : }
847 2 : }
848 1 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
849 :
850 1 : if let Some(archived_at_set) = need_upload_scheduled {
851 1 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
852 1 : upload_queue.dirty.archived_at = intended_archived_at;
853 1 : self.schedule_index_upload(upload_queue);
854 0 : }
855 :
856 1 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
857 1 : Ok(need_wait)
858 1 : }
859 :
860 1 : pub(crate) fn schedule_index_upload_for_timeline_invisible_state(
861 1 : self: &Arc<Self>,
862 1 : state: TimelineVisibilityState,
863 1 : ) -> anyhow::Result<()> {
864 1 : let mut guard = self.upload_queue.lock().unwrap();
865 1 : let upload_queue = guard.initialized_mut()?;
866 :
867 1 : fn need_change(
868 1 : marked_invisible_at: &Option<NaiveDateTime>,
869 1 : state: TimelineVisibilityState,
870 1 : ) -> Option<bool> {
871 1 : match (marked_invisible_at, state) {
872 0 : (Some(_), TimelineVisibilityState::Invisible) => Some(false),
873 1 : (None, TimelineVisibilityState::Invisible) => Some(true),
874 0 : (Some(_), TimelineVisibilityState::Visible) => Some(false),
875 0 : (None, TimelineVisibilityState::Visible) => Some(true),
876 : }
877 1 : }
878 :
879 1 : let need_upload_scheduled = need_change(&upload_queue.dirty.marked_invisible_at, state);
880 :
881 1 : if let Some(marked_invisible_at_set) = need_upload_scheduled {
882 1 : let intended_marked_invisible_at =
883 1 : marked_invisible_at_set.then(|| Utc::now().naive_utc());
884 1 : upload_queue.dirty.marked_invisible_at = intended_marked_invisible_at;
885 1 : self.schedule_index_upload(upload_queue);
886 0 : }
887 :
888 1 : Ok(())
889 1 : }
890 :
891 : /// Shuts the timeline client down, but only if the timeline is archived.
892 : ///
893 : /// This function and [`Self::schedule_index_upload_for_timeline_archival_state`] use the
894 : /// same lock to prevent races between unarchival and offloading: unarchival requires the
895 : /// upload queue to be initialized, and leaves behind an upload queue where either dirty
896 : /// or clean has archived_at of `None`. offloading leaves behind an uninitialized upload
897 : /// queue.
898 1 : pub(crate) async fn shutdown_if_archived(
899 1 : self: &Arc<Self>,
900 1 : ) -> Result<(), ShutdownIfArchivedError> {
901 : {
902 1 : let mut guard = self.upload_queue.lock().unwrap();
903 1 : let upload_queue = guard
904 1 : .initialized_mut()
905 1 : .map_err(ShutdownIfArchivedError::NotInitialized)?;
906 :
907 1 : match (
908 1 : upload_queue.dirty.archived_at.is_none(),
909 1 : upload_queue.clean.0.archived_at.is_none(),
910 1 : ) {
911 : // The expected case: the timeline is archived and we don't want to unarchive
912 1 : (false, false) => {}
913 : (true, false) => {
914 0 : tracing::info!("can't shut down timeline: timeline slated for unarchival");
915 0 : return Err(ShutdownIfArchivedError::NotArchived);
916 : }
917 0 : (dirty_archived, true) => {
918 0 : tracing::info!(%dirty_archived, "can't shut down timeline: timeline not archived in remote storage");
919 0 : return Err(ShutdownIfArchivedError::NotArchived);
920 : }
921 : }
922 :
923 : // Set the shutting_down flag while the guard from the archival check is held.
924 : // This prevents a race with unarchival, as initialized_mut will not return
925 : // an upload queue from this point.
926 : // Also launch the queued tasks like shutdown() does.
927 1 : if !upload_queue.shutting_down {
928 1 : upload_queue.shutting_down = true;
929 1 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
930 1 : // this operation is not counted similar to Barrier
931 1 : self.launch_queued_tasks(upload_queue);
932 1 : }
933 : }
934 :
935 1 : self.shutdown().await;
936 :
937 1 : Ok(())
938 1 : }
939 :
940 : /// Launch an index-file upload operation in the background, setting `import_pgdata` field.
941 0 : pub(crate) fn schedule_index_upload_for_import_pgdata_state_update(
942 0 : self: &Arc<Self>,
943 0 : state: Option<import_pgdata::index_part_format::Root>,
944 0 : ) -> anyhow::Result<()> {
945 0 : let mut guard = self.upload_queue.lock().unwrap();
946 0 : let upload_queue = guard.initialized_mut()?;
947 0 : upload_queue.dirty.import_pgdata = state;
948 0 : self.schedule_index_upload(upload_queue);
949 0 : Ok(())
950 0 : }
951 :
952 : /// If the `import_pgdata` field marks the timeline as having an import in progress,
953 : /// launch an index-file upload operation that transitions it to done in the background
954 0 : pub(crate) fn schedule_index_upload_for_import_pgdata_finalize(
955 0 : self: &Arc<Self>,
956 0 : ) -> anyhow::Result<()> {
957 : use import_pgdata::index_part_format;
958 :
959 0 : let mut guard = self.upload_queue.lock().unwrap();
960 0 : let upload_queue = guard.initialized_mut()?;
961 0 : let to_update = match &upload_queue.dirty.import_pgdata {
962 0 : Some(import) if !import.is_done() => Some(import),
963 0 : Some(_) | None => None,
964 : };
965 :
966 0 : if let Some(old) = to_update {
967 0 : let new =
968 0 : index_part_format::Root::V1(index_part_format::V1::Done(index_part_format::Done {
969 0 : idempotency_key: old.idempotency_key().clone(),
970 0 : started_at: *old.started_at(),
971 0 : finished_at: chrono::Utc::now().naive_utc(),
972 0 : }));
973 0 :
974 0 : upload_queue.dirty.import_pgdata = Some(new);
975 0 : self.schedule_index_upload(upload_queue);
976 0 : }
977 :
978 0 : Ok(())
979 0 : }
980 :
981 : /// Launch an index-file upload operation in the background, setting `gc_compaction_state` field.
982 0 : pub(crate) fn schedule_index_upload_for_gc_compaction_state_update(
983 0 : self: &Arc<Self>,
984 0 : gc_compaction_state: GcCompactionState,
985 0 : ) -> anyhow::Result<()> {
986 0 : let mut guard = self.upload_queue.lock().unwrap();
987 0 : let upload_queue = guard.initialized_mut()?;
988 0 : upload_queue.dirty.gc_compaction = Some(gc_compaction_state);
989 0 : self.schedule_index_upload(upload_queue);
990 0 : Ok(())
991 0 : }
992 :
993 : /// Launch an index-file upload operation in the background, setting `rel_size_v2_status` field.
994 0 : pub(crate) fn schedule_index_upload_for_rel_size_v2_status_update(
995 0 : self: &Arc<Self>,
996 0 : rel_size_v2_status: RelSizeMigration,
997 0 : ) -> anyhow::Result<()> {
998 0 : let mut guard = self.upload_queue.lock().unwrap();
999 0 : let upload_queue = guard.initialized_mut()?;
1000 0 : upload_queue.dirty.rel_size_migration = Some(rel_size_v2_status);
1001 : // TODO: allow this operation to bypass the validation check because we might upload the index part
1002 : // with no layers but the flag updated. For now, we just modify the index part in memory and the next
1003 : // upload will include the flag.
1004 : // self.schedule_index_upload(upload_queue);
1005 0 : Ok(())
1006 0 : }
1007 :
1008 : ///
1009 : /// Launch an index-file upload operation in the background, if necessary.
1010 : ///
1011 : /// Use this function to schedule the update of the index file after
1012 : /// scheduling file uploads or deletions. If no file uploads or deletions
1013 : /// have been scheduled since the last index file upload, this does
1014 : /// nothing.
1015 : ///
1016 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
1017 : /// the upload to the upload queue and returns quickly.
1018 83 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
1019 83 : let mut guard = self.upload_queue.lock().unwrap();
1020 83 : let upload_queue = guard.initialized_mut()?;
1021 :
1022 83 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1023 7 : self.schedule_index_upload(upload_queue);
1024 76 : }
1025 :
1026 83 : Ok(())
1027 83 : }
1028 :
1029 : /// Only used in the `patch_index_part` HTTP API to force trigger an index upload.
1030 0 : pub fn force_schedule_index_upload(self: &Arc<Self>) -> Result<(), NotInitialized> {
1031 0 : let mut guard = self.upload_queue.lock().unwrap();
1032 0 : let upload_queue = guard.initialized_mut()?;
1033 0 : self.schedule_index_upload(upload_queue);
1034 0 : Ok(())
1035 0 : }
1036 :
1037 : /// Launch an index-file upload operation in the background (internal function)
1038 797 : fn schedule_index_upload(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1039 797 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
1040 : // fix up the duplicated field
1041 797 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
1042 :
1043 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
1044 : // look like a retryable error
1045 797 : let void = std::io::sink();
1046 797 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
1047 :
1048 797 : let index_part = &upload_queue.dirty;
1049 :
1050 797 : info!(
1051 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
1052 0 : index_part.layer_metadata.len(),
1053 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
1054 : );
1055 :
1056 797 : let op = UploadOp::UploadMetadata {
1057 797 : uploaded: Box::new(index_part.clone()),
1058 797 : };
1059 797 : self.metric_begin(&op);
1060 797 : upload_queue.queued_operations.push_back(op);
1061 797 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
1062 :
1063 : // Launch the task immediately, if possible
1064 797 : self.launch_queued_tasks(upload_queue);
1065 797 : }
1066 :
1067 : /// Reparent this timeline to a new parent.
1068 : ///
1069 : /// A retryable step of timeline ancestor detach.
1070 0 : pub(crate) async fn schedule_reparenting_and_wait(
1071 0 : self: &Arc<Self>,
1072 0 : new_parent: &TimelineId,
1073 0 : ) -> anyhow::Result<()> {
1074 0 : let receiver = {
1075 0 : let mut guard = self.upload_queue.lock().unwrap();
1076 0 : let upload_queue = guard.initialized_mut()?;
1077 :
1078 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
1079 0 : return Err(anyhow::anyhow!(
1080 0 : "cannot reparent without a current ancestor"
1081 0 : ));
1082 : };
1083 :
1084 0 : let uploaded = &upload_queue.clean.0.metadata;
1085 :
1086 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
1087 : // nothing to do
1088 0 : None
1089 : } else {
1090 0 : upload_queue.dirty.metadata.reparent(new_parent);
1091 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
1092 :
1093 0 : self.schedule_index_upload(upload_queue);
1094 :
1095 0 : Some(self.schedule_barrier0(upload_queue))
1096 : }
1097 : };
1098 :
1099 0 : if let Some(receiver) = receiver {
1100 0 : Self::wait_completion0(receiver).await?;
1101 0 : }
1102 0 : Ok(())
1103 0 : }
1104 :
1105 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
1106 : /// detaching from ancestor and waits for it to complete.
1107 : ///
1108 : /// This is used with `Timeline::detach_ancestor` functionality.
1109 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
1110 0 : self: &Arc<Self>,
1111 0 : layers: &[Layer],
1112 0 : adopted: (TimelineId, Lsn),
1113 0 : ) -> anyhow::Result<()> {
1114 0 : let barrier = {
1115 0 : let mut guard = self.upload_queue.lock().unwrap();
1116 0 : let upload_queue = guard.initialized_mut()?;
1117 :
1118 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
1119 0 : None
1120 : } else {
1121 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
1122 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
1123 :
1124 0 : for layer in layers {
1125 0 : let prev = upload_queue
1126 0 : .dirty
1127 0 : .layer_metadata
1128 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
1129 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
1130 : }
1131 :
1132 0 : self.schedule_index_upload(upload_queue);
1133 :
1134 0 : Some(self.schedule_barrier0(upload_queue))
1135 : }
1136 : };
1137 :
1138 0 : if let Some(barrier) = barrier {
1139 0 : Self::wait_completion0(barrier).await?;
1140 0 : }
1141 0 : Ok(())
1142 0 : }
1143 :
1144 : /// Adds a gc blocking reason for this timeline if one does not exist already.
1145 : ///
1146 : /// A retryable step of timeline detach ancestor.
1147 : ///
1148 : /// Returns a future which waits until the completion of the upload.
1149 0 : pub(crate) fn schedule_insert_gc_block_reason(
1150 0 : self: &Arc<Self>,
1151 0 : reason: index::GcBlockingReason,
1152 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
1153 : {
1154 0 : let maybe_barrier = {
1155 0 : let mut guard = self.upload_queue.lock().unwrap();
1156 0 : let upload_queue = guard.initialized_mut()?;
1157 :
1158 0 : if let index::GcBlockingReason::DetachAncestor = reason {
1159 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
1160 0 : drop(guard);
1161 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
1162 0 : }
1163 0 : }
1164 :
1165 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
1166 :
1167 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
1168 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
1169 :
1170 0 : match (current, uploaded) {
1171 0 : (x, y) if wanted(x) && wanted(y) => None,
1172 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
1173 : // Usual case: !wanted(x) && !wanted(y)
1174 : //
1175 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
1176 : // turn on and off some reason.
1177 0 : (x, y) => {
1178 0 : if !wanted(x) && wanted(y) {
1179 : // this could be avoided by having external in-memory synchronization, like
1180 : // timeline detach ancestor
1181 0 : warn!(
1182 : ?reason,
1183 : op = "insert",
1184 0 : "unexpected: two racing processes to enable and disable a gc blocking reason"
1185 : );
1186 0 : }
1187 :
1188 : // at this point, the metadata must always show that there is a parent
1189 0 : upload_queue.dirty.gc_blocking = current
1190 0 : .map(|x| x.with_reason(reason))
1191 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
1192 0 : self.schedule_index_upload(upload_queue);
1193 0 : Some(self.schedule_barrier0(upload_queue))
1194 : }
1195 : }
1196 : };
1197 :
1198 0 : Ok(async move {
1199 0 : if let Some(barrier) = maybe_barrier {
1200 0 : Self::wait_completion0(barrier).await?;
1201 0 : }
1202 0 : Ok(())
1203 0 : })
1204 0 : }
1205 :
1206 : /// Removes a gc blocking reason for this timeline if one exists.
1207 : ///
1208 : /// A retryable step of timeline detach ancestor.
1209 : ///
1210 : /// Returns a future which waits until the completion of the upload.
1211 0 : pub(crate) fn schedule_remove_gc_block_reason(
1212 0 : self: &Arc<Self>,
1213 0 : reason: index::GcBlockingReason,
1214 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
1215 : {
1216 0 : let maybe_barrier = {
1217 0 : let mut guard = self.upload_queue.lock().unwrap();
1218 0 : let upload_queue = guard.initialized_mut()?;
1219 :
1220 0 : if let index::GcBlockingReason::DetachAncestor = reason {
1221 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
1222 0 : drop(guard);
1223 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
1224 0 : }
1225 0 : }
1226 :
1227 0 : let wanted = |x: Option<&index::GcBlocking>| {
1228 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
1229 0 : };
1230 :
1231 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
1232 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
1233 :
1234 0 : match (current, uploaded) {
1235 0 : (x, y) if wanted(x) && wanted(y) => None,
1236 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
1237 0 : (x, y) => {
1238 0 : if !wanted(x) && wanted(y) {
1239 0 : warn!(
1240 : ?reason,
1241 : op = "remove",
1242 0 : "unexpected: two racing processes to enable and disable a gc blocking reason (remove)"
1243 : );
1244 0 : }
1245 :
1246 : upload_queue.dirty.gc_blocking =
1247 0 : current.as_ref().and_then(|x| x.without_reason(reason));
1248 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
1249 0 : self.schedule_index_upload(upload_queue);
1250 0 : Some(self.schedule_barrier0(upload_queue))
1251 : }
1252 : }
1253 : };
1254 :
1255 0 : Ok(async move {
1256 0 : if let Some(barrier) = maybe_barrier {
1257 0 : Self::wait_completion0(barrier).await?;
1258 0 : }
1259 0 : Ok(())
1260 0 : })
1261 0 : }
1262 :
1263 : /// Launch an upload operation in the background; the file is added to be included in next
1264 : /// `index_part.json` upload.
1265 705 : pub(crate) fn schedule_layer_file_upload(
1266 705 : self: &Arc<Self>,
1267 705 : layer: ResidentLayer,
1268 705 : ) -> Result<(), NotInitialized> {
1269 705 : let mut guard = self.upload_queue.lock().unwrap();
1270 705 : let upload_queue = guard.initialized_mut()?;
1271 :
1272 705 : self.schedule_layer_file_upload0(upload_queue, layer);
1273 705 : self.launch_queued_tasks(upload_queue);
1274 705 : Ok(())
1275 705 : }
1276 :
1277 895 : fn schedule_layer_file_upload0(
1278 895 : self: &Arc<Self>,
1279 895 : upload_queue: &mut UploadQueueInitialized,
1280 895 : layer: ResidentLayer,
1281 895 : ) {
1282 895 : let metadata = layer.metadata();
1283 :
1284 895 : upload_queue
1285 895 : .dirty
1286 895 : .layer_metadata
1287 895 : .insert(layer.layer_desc().layer_name(), metadata.clone());
1288 895 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1289 :
1290 895 : info!(
1291 : gen=?metadata.generation,
1292 : shard=?metadata.shard,
1293 0 : "scheduled layer file upload {layer}",
1294 : );
1295 :
1296 895 : let op = UploadOp::UploadLayer(layer, metadata, None);
1297 895 : self.metric_begin(&op);
1298 895 : upload_queue.queued_operations.push_back(op);
1299 895 : }
1300 :
1301 : /// Launch a delete operation in the background.
1302 : ///
1303 : /// The operation does not modify local filesystem state.
1304 : ///
1305 : /// Note: This schedules an index file upload before the deletions. The
1306 : /// deletion won't actually be performed, until all previously scheduled
1307 : /// upload operations, and the index file upload, have completed
1308 : /// successfully.
1309 4 : pub fn schedule_layer_file_deletion(
1310 4 : self: &Arc<Self>,
1311 4 : names: &[LayerName],
1312 4 : ) -> anyhow::Result<()> {
1313 4 : let mut guard = self.upload_queue.lock().unwrap();
1314 4 : let upload_queue = guard.initialized_mut()?;
1315 :
1316 4 : let with_metadata =
1317 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned());
1318 :
1319 4 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1320 :
1321 : // Launch the tasks immediately, if possible
1322 4 : self.launch_queued_tasks(upload_queue);
1323 4 : Ok(())
1324 4 : }
1325 :
1326 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1327 : /// layer files, leaving them dangling.
1328 : ///
1329 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1330 : /// is invoked on them.
1331 4 : pub(crate) fn schedule_gc_update(
1332 4 : self: &Arc<Self>,
1333 4 : gc_layers: &[Layer],
1334 4 : ) -> Result<(), NotInitialized> {
1335 4 : let mut guard = self.upload_queue.lock().unwrap();
1336 4 : let upload_queue = guard.initialized_mut()?;
1337 :
1338 : // just forget the return value; after uploading the next index_part.json, we can consider
1339 : // the layer files as "dangling". this is fine, at worst case we create work for the
1340 : // scrubber.
1341 :
1342 5 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1343 :
1344 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
1345 :
1346 4 : self.launch_queued_tasks(upload_queue);
1347 :
1348 4 : Ok(())
1349 4 : }
1350 :
1351 0 : pub(crate) fn schedule_unlinking_of_layers_from_index_part<I>(
1352 0 : self: &Arc<Self>,
1353 0 : names: I,
1354 0 : ) -> Result<(), NotInitialized>
1355 0 : where
1356 0 : I: IntoIterator<Item = LayerName>,
1357 : {
1358 0 : let mut guard = self.upload_queue.lock().unwrap();
1359 0 : let upload_queue = guard.initialized_mut()?;
1360 :
1361 0 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
1362 :
1363 0 : Ok(())
1364 0 : }
1365 :
1366 : /// Update the remote index file, removing the to-be-deleted files from the index,
1367 : /// allowing scheduling of actual deletions later.
1368 55 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1369 55 : self: &Arc<Self>,
1370 55 : upload_queue: &mut UploadQueueInitialized,
1371 55 : names: I,
1372 55 : ) -> Vec<(LayerName, LayerFileMetadata)>
1373 55 : where
1374 55 : I: IntoIterator<Item = LayerName>,
1375 : {
1376 : // Decorate our list of names with each name's metadata, dropping
1377 : // names that are unexpectedly missing from our metadata. This metadata
1378 : // is later used when physically deleting layers, to construct key paths.
1379 55 : let with_metadata: Vec<_> = names
1380 55 : .into_iter()
1381 259 : .filter_map(|name| {
1382 259 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1383 :
1384 259 : if let Some(meta) = meta {
1385 259 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1386 259 : Some((name, meta))
1387 : } else {
1388 : // This can only happen if we forgot to to schedule the file upload
1389 : // before scheduling the delete. Log it because it is a rare/strange
1390 : // situation, and in case something is misbehaving, we'd like to know which
1391 : // layers experienced this.
1392 0 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1393 0 : None
1394 : }
1395 259 : })
1396 55 : .collect();
1397 :
1398 : #[cfg(feature = "testing")]
1399 314 : for (name, metadata) in &with_metadata {
1400 259 : let gen_ = metadata.generation;
1401 259 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen_) {
1402 0 : if unexpected == gen_ {
1403 0 : tracing::error!("{name} was unlinked twice with same generation");
1404 : } else {
1405 0 : tracing::error!(
1406 0 : "{name} was unlinked twice with different generations {gen_:?} and {unexpected:?}"
1407 : );
1408 : }
1409 259 : }
1410 : }
1411 :
1412 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1413 : // index_part update, because that needs to be uploaded before we can actually delete the
1414 : // files.
1415 55 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1416 47 : self.schedule_index_upload(upload_queue);
1417 47 : }
1418 :
1419 55 : with_metadata
1420 55 : }
1421 :
1422 : /// Schedules deletion for layer files which have previously been unlinked from the
1423 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1424 261 : pub(crate) fn schedule_deletion_of_unlinked(
1425 261 : self: &Arc<Self>,
1426 261 : layers: Vec<(LayerName, LayerFileMetadata)>,
1427 261 : ) -> anyhow::Result<()> {
1428 261 : let mut guard = self.upload_queue.lock().unwrap();
1429 261 : let upload_queue = guard.initialized_mut()?;
1430 :
1431 261 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1432 261 : self.launch_queued_tasks(upload_queue);
1433 261 : Ok(())
1434 261 : }
1435 :
1436 264 : fn schedule_deletion_of_unlinked0(
1437 264 : self: &Arc<Self>,
1438 264 : upload_queue: &mut UploadQueueInitialized,
1439 264 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1440 264 : ) {
1441 : // Filter out any layers which were not created by this tenant shard. These are
1442 : // layers that originate from some ancestor shard after a split, and may still
1443 : // be referenced by other shards. We are free to delete them locally and remove
1444 : // them from our index (and would have already done so when we reach this point
1445 : // in the code), but we may not delete them remotely.
1446 264 : with_metadata.retain(|(name, meta)| {
1447 261 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1448 261 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1449 261 : if !retain {
1450 0 : tracing::debug!(
1451 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1452 : meta.shard
1453 : );
1454 261 : }
1455 261 : retain
1456 261 : });
1457 :
1458 525 : for (name, meta) in &with_metadata {
1459 261 : info!(
1460 0 : "scheduling deletion of layer {}{} (shard {})",
1461 : name,
1462 0 : meta.generation.get_suffix(),
1463 : meta.shard
1464 : );
1465 : }
1466 :
1467 : #[cfg(feature = "testing")]
1468 525 : for (name, meta) in &with_metadata {
1469 261 : let gen_ = meta.generation;
1470 261 : match upload_queue.dangling_files.remove(name) {
1471 259 : Some(same) if same == gen_ => { /* expected */ }
1472 0 : Some(other) => {
1473 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen_:?}");
1474 : }
1475 : None => {
1476 2 : tracing::error!("{name} was unlinked but was not dangling");
1477 : }
1478 : }
1479 : }
1480 :
1481 : // schedule the actual deletions
1482 264 : if with_metadata.is_empty() {
1483 : // avoid scheduling the op & bumping the metric
1484 3 : return;
1485 261 : }
1486 261 : let op = UploadOp::Delete(Delete {
1487 261 : layers: with_metadata,
1488 261 : });
1489 261 : self.metric_begin(&op);
1490 261 : upload_queue.queued_operations.push_back(op);
1491 264 : }
1492 :
1493 : /// Schedules a compaction update to the remote `index_part.json`.
1494 : ///
1495 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1496 47 : pub(crate) fn schedule_compaction_update(
1497 47 : self: &Arc<Self>,
1498 47 : compacted_from: &[Layer],
1499 47 : compacted_to: &[ResidentLayer],
1500 47 : ) -> Result<(), NotInitialized> {
1501 47 : let mut guard = self.upload_queue.lock().unwrap();
1502 47 : let upload_queue = guard.initialized_mut()?;
1503 :
1504 237 : for layer in compacted_to {
1505 190 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1506 190 : }
1507 :
1508 253 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1509 :
1510 47 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
1511 47 : self.launch_queued_tasks(upload_queue);
1512 :
1513 47 : Ok(())
1514 47 : }
1515 :
1516 : /// Wait for all previously scheduled uploads/deletions to complete
1517 118 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1518 118 : let receiver = {
1519 118 : let mut guard = self.upload_queue.lock().unwrap();
1520 118 : let upload_queue = guard
1521 118 : .initialized_mut()
1522 118 : .map_err(WaitCompletionError::NotInitialized)?;
1523 118 : self.schedule_barrier0(upload_queue)
1524 : };
1525 :
1526 118 : Self::wait_completion0(receiver).await
1527 118 : }
1528 :
1529 118 : async fn wait_completion0(
1530 118 : mut receiver: tokio::sync::watch::Receiver<()>,
1531 118 : ) -> Result<(), WaitCompletionError> {
1532 118 : if receiver.changed().await.is_err() {
1533 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1534 118 : }
1535 :
1536 118 : Ok(())
1537 118 : }
1538 :
1539 3 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1540 3 : let mut guard = self.upload_queue.lock().unwrap();
1541 3 : let upload_queue = guard.initialized_mut()?;
1542 3 : self.schedule_barrier0(upload_queue);
1543 3 : Ok(())
1544 3 : }
1545 :
1546 121 : fn schedule_barrier0(
1547 121 : self: &Arc<Self>,
1548 121 : upload_queue: &mut UploadQueueInitialized,
1549 121 : ) -> tokio::sync::watch::Receiver<()> {
1550 121 : let (sender, receiver) = tokio::sync::watch::channel(());
1551 121 : let barrier_op = UploadOp::Barrier(sender);
1552 :
1553 121 : upload_queue.queued_operations.push_back(barrier_op);
1554 : // Don't count this kind of operation!
1555 :
1556 : // Launch the task immediately, if possible
1557 121 : self.launch_queued_tasks(upload_queue);
1558 :
1559 121 : receiver
1560 121 : }
1561 :
1562 : /// Wait for all previously scheduled operations to complete, and then stop.
1563 : ///
1564 : /// Not cancellation safe
1565 5 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1566 : // On cancellation the queue is left in ackward state of refusing new operations but
1567 : // proper stop is yet to be called. On cancel the original or some later task must call
1568 : // `stop` or `shutdown`.
1569 5 : let sg = scopeguard::guard((), |_| {
1570 0 : tracing::error!(
1571 0 : "RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error"
1572 : )
1573 0 : });
1574 :
1575 4 : let fut = {
1576 5 : let mut guard = self.upload_queue.lock().unwrap();
1577 5 : let upload_queue = match &mut *guard {
1578 : UploadQueue::Stopped(_) => {
1579 1 : scopeguard::ScopeGuard::into_inner(sg);
1580 1 : return;
1581 : }
1582 : UploadQueue::Uninitialized => {
1583 : // transition into Stopped state
1584 0 : self.stop_impl(&mut guard);
1585 0 : scopeguard::ScopeGuard::into_inner(sg);
1586 0 : return;
1587 : }
1588 4 : UploadQueue::Initialized(init) => init,
1589 : };
1590 :
1591 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1592 : // just don't add more of these as they would never complete.
1593 : //
1594 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1595 : // in every place we would not have to jump through this hoop, and this method could be
1596 : // made cancellable.
1597 4 : if !upload_queue.shutting_down {
1598 3 : upload_queue.shutting_down = true;
1599 3 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1600 3 : // this operation is not counted similar to Barrier
1601 3 :
1602 3 : self.launch_queued_tasks(upload_queue);
1603 3 : }
1604 :
1605 4 : upload_queue.shutdown_ready.clone().acquire_owned()
1606 : };
1607 :
1608 4 : let res = fut.await;
1609 :
1610 4 : scopeguard::ScopeGuard::into_inner(sg);
1611 :
1612 4 : match res {
1613 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1614 4 : Err(_closed) => {
1615 4 : // expected
1616 4 : }
1617 : }
1618 :
1619 4 : self.stop();
1620 5 : }
1621 :
1622 : /// Set the deleted_at field in the remote index file.
1623 : ///
1624 : /// This fails if the upload queue has not been `stop()`ed.
1625 : ///
1626 : /// The caller is responsible for calling `stop()` AND for waiting
1627 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1628 : /// Check method [`RemoteTimelineClient::stop`] for details.
1629 : #[instrument(skip_all)]
1630 : pub(crate) async fn persist_index_part_with_deleted_flag(
1631 : self: &Arc<Self>,
1632 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1633 : let index_part_with_deleted_at = {
1634 : let mut locked = self.upload_queue.lock().unwrap();
1635 :
1636 : // We must be in stopped state because otherwise
1637 : // we can have inprogress index part upload that can overwrite the file
1638 : // with missing is_deleted flag that we going to set below
1639 : let stopped = locked.stopped_mut()?;
1640 :
1641 : match stopped.deleted_at {
1642 : SetDeletedFlagProgress::NotRunning => (), // proceed
1643 : SetDeletedFlagProgress::InProgress(at) => {
1644 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1645 : }
1646 : SetDeletedFlagProgress::Successful(at) => {
1647 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1648 : }
1649 : };
1650 : let deleted_at = Utc::now().naive_utc();
1651 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1652 :
1653 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1654 : index_part.deleted_at = Some(deleted_at);
1655 : index_part
1656 : };
1657 :
1658 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1659 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1660 0 : let stopped = locked
1661 0 : .stopped_mut()
1662 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1663 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1664 0 : });
1665 :
1666 : pausable_failpoint!("persist_deleted_index_part");
1667 :
1668 : backoff::retry(
1669 0 : || {
1670 0 : upload::upload_index_part(
1671 0 : &self.storage_impl,
1672 0 : &self.tenant_shard_id,
1673 0 : &self.timeline_id,
1674 0 : self.generation,
1675 0 : &index_part_with_deleted_at,
1676 0 : &self.cancel,
1677 : )
1678 0 : },
1679 : |_e| false,
1680 : 1,
1681 : // have just a couple of attempts
1682 : // when executed as part of timeline deletion this happens in context of api call
1683 : // when executed as part of tenant deletion this happens in the background
1684 : 2,
1685 : "persist_index_part_with_deleted_flag",
1686 : &self.cancel,
1687 : )
1688 : .await
1689 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1690 : .and_then(|x| x)?;
1691 :
1692 : // all good, disarm the guard and mark as success
1693 : ScopeGuard::into_inner(undo_deleted_at);
1694 : {
1695 : let mut locked = self.upload_queue.lock().unwrap();
1696 :
1697 : let stopped = locked
1698 : .stopped_mut()
1699 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1700 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1701 : index_part_with_deleted_at
1702 : .deleted_at
1703 : .expect("we set it above"),
1704 : );
1705 : }
1706 :
1707 : Ok(())
1708 : }
1709 :
1710 0 : pub(crate) fn is_deleting(&self) -> bool {
1711 0 : let mut locked = self.upload_queue.lock().unwrap();
1712 0 : locked.stopped_mut().is_ok()
1713 0 : }
1714 :
1715 0 : pub(crate) async fn preserve_initdb_archive(
1716 0 : self: &Arc<Self>,
1717 0 : tenant_id: &TenantId,
1718 0 : timeline_id: &TimelineId,
1719 0 : cancel: &CancellationToken,
1720 0 : ) -> anyhow::Result<()> {
1721 0 : backoff::retry(
1722 0 : || async {
1723 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1724 0 : .await
1725 0 : },
1726 : TimeoutOrCancel::caused_by_cancel,
1727 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1728 : FAILED_REMOTE_OP_RETRIES,
1729 0 : "preserve_initdb_tar_zst",
1730 0 : &cancel.clone(),
1731 : )
1732 0 : .await
1733 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1734 0 : .and_then(|x| x)
1735 0 : .context("backing up initdb archive")?;
1736 0 : Ok(())
1737 0 : }
1738 :
1739 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1740 : ///
1741 : /// This is not normally needed.
1742 0 : pub(crate) async fn upload_layer_file(
1743 0 : self: &Arc<Self>,
1744 0 : uploaded: &ResidentLayer,
1745 0 : cancel: &CancellationToken,
1746 0 : ) -> anyhow::Result<()> {
1747 0 : let remote_path = remote_layer_path(
1748 0 : &self.tenant_shard_id.tenant_id,
1749 0 : &self.timeline_id,
1750 0 : uploaded.metadata().shard,
1751 0 : &uploaded.layer_desc().layer_name(),
1752 0 : uploaded.metadata().generation,
1753 : );
1754 :
1755 0 : backoff::retry(
1756 0 : || async {
1757 0 : upload::upload_timeline_layer(
1758 0 : &self.storage_impl,
1759 0 : uploaded.local_path(),
1760 0 : &remote_path,
1761 0 : uploaded.metadata().file_size,
1762 0 : cancel,
1763 0 : )
1764 0 : .await
1765 0 : },
1766 : TimeoutOrCancel::caused_by_cancel,
1767 : FAILED_UPLOAD_WARN_THRESHOLD,
1768 : FAILED_REMOTE_OP_RETRIES,
1769 0 : "upload a layer without adding it to latest files",
1770 0 : cancel,
1771 : )
1772 0 : .await
1773 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1774 0 : .and_then(|x| x)
1775 0 : .context("upload a layer without adding it to latest files")
1776 0 : }
1777 :
1778 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1779 : /// not added to be part of a future `index_part.json` upload.
1780 0 : pub(crate) async fn copy_timeline_layer(
1781 0 : self: &Arc<Self>,
1782 0 : adopted: &Layer,
1783 0 : adopted_as: &Layer,
1784 0 : cancel: &CancellationToken,
1785 0 : ) -> anyhow::Result<()> {
1786 0 : let source_remote_path = remote_layer_path(
1787 0 : &self.tenant_shard_id.tenant_id,
1788 0 : &adopted
1789 0 : .get_timeline_id()
1790 0 : .expect("Source timeline should be alive"),
1791 0 : adopted.metadata().shard,
1792 0 : &adopted.layer_desc().layer_name(),
1793 0 : adopted.metadata().generation,
1794 : );
1795 :
1796 0 : let target_remote_path = remote_layer_path(
1797 0 : &self.tenant_shard_id.tenant_id,
1798 0 : &self.timeline_id,
1799 0 : adopted_as.metadata().shard,
1800 0 : &adopted_as.layer_desc().layer_name(),
1801 0 : adopted_as.metadata().generation,
1802 : );
1803 :
1804 0 : backoff::retry(
1805 0 : || async {
1806 0 : upload::copy_timeline_layer(
1807 0 : &self.storage_impl,
1808 0 : &source_remote_path,
1809 0 : &target_remote_path,
1810 0 : cancel,
1811 0 : )
1812 0 : .await
1813 0 : },
1814 : TimeoutOrCancel::caused_by_cancel,
1815 : FAILED_UPLOAD_WARN_THRESHOLD,
1816 : FAILED_REMOTE_OP_RETRIES,
1817 0 : "copy timeline layer",
1818 0 : cancel,
1819 : )
1820 0 : .await
1821 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1822 0 : .and_then(|x| x)
1823 0 : .context("remote copy timeline layer")
1824 0 : }
1825 :
1826 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1827 0 : match tokio::time::timeout(
1828 : DELETION_QUEUE_FLUSH_TIMEOUT,
1829 0 : self.deletion_queue_client.flush_immediate(),
1830 : )
1831 0 : .await
1832 : {
1833 0 : Ok(result) => result,
1834 0 : Err(_timeout) => {
1835 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1836 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1837 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1838 0 : tracing::warn!(
1839 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1840 : );
1841 0 : Ok(())
1842 : }
1843 : }
1844 0 : }
1845 :
1846 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1847 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1848 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1849 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> Result<(), DeleteTimelineError> {
1850 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1851 :
1852 0 : let layers: Vec<RemotePath> = {
1853 0 : let mut locked = self.upload_queue.lock().unwrap();
1854 0 : let stopped = locked.stopped_mut().map_err(DeleteTimelineError::Other)?;
1855 :
1856 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1857 0 : return Err(DeleteTimelineError::Other(anyhow::anyhow!(
1858 0 : "deleted_at is not set"
1859 0 : )));
1860 0 : }
1861 :
1862 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1863 :
1864 0 : stopped
1865 0 : .upload_queue_for_deletion
1866 0 : .dirty
1867 0 : .layer_metadata
1868 0 : .drain()
1869 0 : .filter(|(_file_name, meta)| {
1870 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1871 : // all shards anyway, we _could_ delete these, but
1872 : // - it creates a potential race if other shards are still
1873 : // using the layers while this shard deletes them.
1874 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1875 : // these timelines are present but corrupt (their index exists but some layers don't)
1876 : //
1877 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1878 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1879 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1880 0 : })
1881 0 : .map(|(file_name, meta)| {
1882 0 : remote_layer_path(
1883 0 : &self.tenant_shard_id.tenant_id,
1884 0 : &self.timeline_id,
1885 0 : meta.shard,
1886 0 : &file_name,
1887 0 : meta.generation,
1888 : )
1889 0 : })
1890 0 : .collect()
1891 : };
1892 :
1893 0 : let layer_deletion_count = layers.len();
1894 0 : self.deletion_queue_client
1895 0 : .push_immediate(layers)
1896 0 : .await
1897 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1898 :
1899 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1900 : // inexistant objects are not considered errors.
1901 0 : let initdb_path =
1902 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1903 0 : self.deletion_queue_client
1904 0 : .push_immediate(vec![initdb_path])
1905 0 : .await
1906 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1907 :
1908 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1909 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1910 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1911 :
1912 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1913 : // taking the burden of listing all the layers that we already know we should delete.
1914 0 : self.flush_deletion_queue()
1915 0 : .await
1916 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1917 :
1918 0 : let cancel = shutdown_token();
1919 :
1920 0 : let remaining = download_retry(
1921 0 : || async {
1922 0 : self.storage_impl
1923 0 : .list(
1924 0 : Some(&timeline_storage_path),
1925 0 : ListingMode::NoDelimiter,
1926 0 : None,
1927 0 : &cancel,
1928 0 : )
1929 0 : .await
1930 0 : },
1931 0 : "list remaining files",
1932 0 : &cancel,
1933 : )
1934 0 : .await
1935 0 : .context("list files remaining files")?
1936 : .keys;
1937 :
1938 : // We will delete the current index_part object last, since it acts as a deletion
1939 : // marker via its deleted_at attribute
1940 0 : let latest_index = remaining
1941 0 : .iter()
1942 0 : .filter(|o| {
1943 0 : o.key
1944 0 : .object_name()
1945 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1946 0 : .unwrap_or(false)
1947 0 : })
1948 0 : .filter_map(|o| {
1949 0 : parse_remote_index_path(o.key.clone()).map(|gen_| (o.key.clone(), gen_))
1950 0 : })
1951 0 : .max_by_key(|i| i.1)
1952 0 : .map(|i| i.0.clone())
1953 0 : .unwrap_or(
1954 : // No generation-suffixed indices, assume we are dealing with
1955 : // a legacy index.
1956 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1957 : );
1958 :
1959 0 : let remaining_layers: Vec<RemotePath> = remaining
1960 0 : .into_iter()
1961 0 : .filter_map(|o| {
1962 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1963 0 : None
1964 : } else {
1965 0 : Some(o.key)
1966 : }
1967 0 : })
1968 0 : .inspect(|path| {
1969 0 : if let Some(name) = path.object_name() {
1970 0 : info!(%name, "deleting a file not referenced from index_part.json");
1971 : } else {
1972 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1973 : }
1974 0 : })
1975 0 : .collect();
1976 :
1977 0 : let not_referenced_count = remaining_layers.len();
1978 0 : if !remaining_layers.is_empty() {
1979 0 : self.deletion_queue_client
1980 0 : .push_immediate(remaining_layers)
1981 0 : .await
1982 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1983 0 : }
1984 :
1985 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1986 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1987 0 : "failpoint: timeline-delete-before-index-delete"
1988 0 : )))?
1989 0 : });
1990 :
1991 0 : debug!("enqueuing index part deletion");
1992 0 : self.deletion_queue_client
1993 0 : .push_immediate([latest_index].to_vec())
1994 0 : .await
1995 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1996 :
1997 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1998 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1999 0 : self.flush_deletion_queue()
2000 0 : .await
2001 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
2002 :
2003 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
2004 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
2005 0 : "failpoint: timeline-delete-after-index-delete"
2006 0 : )))?
2007 0 : });
2008 :
2009 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
2010 :
2011 0 : Ok(())
2012 0 : }
2013 :
2014 : /// Pick next tasks from the queue, and start as many of them as possible without violating
2015 : /// the ordering constraints.
2016 : ///
2017 : /// The number of inprogress tasks is limited by `Self::inprogress_tasks`, see `next_ready`.
2018 3799 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
2019 5804 : while let Some((mut next_op, coalesced_ops)) = upload_queue.next_ready() {
2020 2005 : debug!("starting op: {next_op}");
2021 :
2022 : // Prepare upload.
2023 2005 : match &mut next_op {
2024 895 : UploadOp::UploadLayer(layer, meta, mode) => {
2025 895 : if upload_queue
2026 895 : .recently_deleted
2027 895 : .remove(&(layer.layer_desc().layer_name().clone(), meta.generation))
2028 0 : {
2029 0 : *mode = Some(OpType::FlushDeletion);
2030 0 : } else {
2031 895 : *mode = Some(OpType::MayReorder)
2032 : }
2033 : }
2034 780 : UploadOp::UploadMetadata { .. } => {}
2035 209 : UploadOp::Delete(Delete { layers }) => {
2036 418 : for (name, meta) in layers {
2037 209 : upload_queue
2038 209 : .recently_deleted
2039 209 : .insert((name.clone(), meta.generation));
2040 209 : }
2041 : }
2042 121 : UploadOp::Barrier(sender) => {
2043 121 : sender.send_replace(());
2044 121 : continue;
2045 : }
2046 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
2047 : };
2048 :
2049 : // Assign unique ID to this task
2050 1884 : upload_queue.task_counter += 1;
2051 1884 : let upload_task_id = upload_queue.task_counter;
2052 :
2053 : // Add it to the in-progress map
2054 1884 : let task = Arc::new(UploadTask {
2055 1884 : task_id: upload_task_id,
2056 1884 : op: next_op,
2057 1884 : coalesced_ops,
2058 1884 : retries: AtomicU32::new(0),
2059 1884 : });
2060 1884 : upload_queue
2061 1884 : .inprogress_tasks
2062 1884 : .insert(task.task_id, Arc::clone(&task));
2063 :
2064 : // Spawn task to perform the task
2065 1884 : let self_rc = Arc::clone(self);
2066 1884 : let tenant_shard_id = self.tenant_shard_id;
2067 1884 : let timeline_id = self.timeline_id;
2068 1884 : task_mgr::spawn(
2069 1884 : &self.runtime,
2070 1884 : TaskKind::RemoteUploadTask,
2071 1884 : self.tenant_shard_id,
2072 1884 : Some(self.timeline_id),
2073 1884 : "remote upload",
2074 1873 : async move {
2075 1873 : self_rc.perform_upload_task(task).await;
2076 1859 : Ok(())
2077 1859 : }
2078 1884 : .instrument(info_span!(parent: None, "remote_upload", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id, %upload_task_id)),
2079 : );
2080 :
2081 : // Loop back to process next task
2082 : }
2083 3799 : }
2084 :
2085 : ///
2086 : /// Perform an upload task.
2087 : ///
2088 : /// The task is in the `inprogress_tasks` list. This function will try to
2089 : /// execute it, retrying forever. On successful completion, the task is
2090 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
2091 : /// queue that were waiting by the completion are launched.
2092 : ///
2093 : /// The task can be shut down, however. That leads to stopping the whole
2094 : /// queue.
2095 : ///
2096 1873 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
2097 1873 : let cancel = shutdown_token();
2098 : // Loop to retry until it completes.
2099 : loop {
2100 : // If we're requested to shut down, close up shop and exit.
2101 : //
2102 : // Note: We only check for the shutdown requests between retries, so
2103 : // if a shutdown request arrives while we're busy uploading, in the
2104 : // upload::upload:*() call below, we will wait not exit until it has
2105 : // finished. We probably could cancel the upload by simply dropping
2106 : // the Future, but we're not 100% sure if the remote storage library
2107 : // is cancellation safe, so we don't dare to do that. Hopefully, the
2108 : // upload finishes or times out soon enough.
2109 1873 : if cancel.is_cancelled() {
2110 0 : info!("upload task cancelled by shutdown request");
2111 0 : self.stop();
2112 0 : return;
2113 1873 : }
2114 :
2115 : // Assert that we don't modify a layer that's referenced by the current index.
2116 1873 : if cfg!(debug_assertions) {
2117 1873 : let modified = match &task.op {
2118 890 : UploadOp::UploadLayer(layer, layer_metadata, _) => {
2119 890 : vec![(layer.layer_desc().layer_name(), layer_metadata)]
2120 : }
2121 209 : UploadOp::Delete(delete) => {
2122 209 : delete.layers.iter().map(|(n, m)| (n.clone(), m)).collect()
2123 : }
2124 : // These don't modify layers.
2125 774 : UploadOp::UploadMetadata { .. } => Vec::new(),
2126 0 : UploadOp::Barrier(_) => Vec::new(),
2127 0 : UploadOp::Shutdown => Vec::new(),
2128 : };
2129 1873 : if let Ok(queue) = self.upload_queue.lock().unwrap().initialized_mut() {
2130 2968 : for (ref name, metadata) in modified {
2131 1099 : debug_assert!(
2132 1099 : !queue.clean.0.references(name, metadata),
2133 2 : "layer {name} modified while referenced by index",
2134 : );
2135 : }
2136 2 : }
2137 0 : }
2138 :
2139 1871 : let upload_result: anyhow::Result<()> = match &task.op {
2140 890 : UploadOp::UploadLayer(layer, layer_metadata, mode) => {
2141 : // TODO: check if this mechanism can be removed now that can_bypass() performs
2142 : // conflict checks during scheduling.
2143 890 : if let Some(OpType::FlushDeletion) = mode {
2144 0 : if self.config.read().unwrap().block_deletions {
2145 : // Of course, this is not efficient... but usually the queue should be empty.
2146 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2147 0 : let mut detected = false;
2148 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2149 0 : for list in queue.blocked_deletions.iter_mut() {
2150 0 : list.layers.retain(|(name, meta)| {
2151 0 : if name == &layer.layer_desc().layer_name()
2152 0 : && meta.generation == layer_metadata.generation
2153 : {
2154 0 : detected = true;
2155 : // remove the layer from deletion queue
2156 0 : false
2157 : } else {
2158 : // keep the layer
2159 0 : true
2160 : }
2161 0 : });
2162 : }
2163 0 : }
2164 0 : if detected {
2165 0 : info!(
2166 0 : "cancelled blocked deletion of layer {} at gen {:?}",
2167 0 : layer.layer_desc().layer_name(),
2168 : layer_metadata.generation
2169 : );
2170 0 : }
2171 : } else {
2172 : // TODO: we did not guarantee that upload task starts after deletion task, so there could be possibly race conditions
2173 : // that we still get the layer deleted. But this only happens if someone creates a layer immediately after it's deleted,
2174 : // which is not possible in the current system.
2175 0 : info!(
2176 0 : "waiting for deletion queue flush to complete before uploading layer {} at gen {:?}",
2177 0 : layer.layer_desc().layer_name(),
2178 : layer_metadata.generation
2179 : );
2180 : {
2181 : // We are going to flush, we can clean up the recently deleted list.
2182 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2183 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2184 0 : queue.recently_deleted.clear();
2185 0 : }
2186 : }
2187 0 : if let Err(e) = self.deletion_queue_client.flush_execute().await {
2188 0 : warn!(
2189 0 : "failed to flush the deletion queue before uploading layer {} at gen {:?}, still proceeding to upload: {e:#} ",
2190 0 : layer.layer_desc().layer_name(),
2191 : layer_metadata.generation
2192 : );
2193 : } else {
2194 0 : info!(
2195 0 : "done flushing deletion queue before uploading layer {} at gen {:?}",
2196 0 : layer.layer_desc().layer_name(),
2197 : layer_metadata.generation
2198 : );
2199 : }
2200 : }
2201 890 : }
2202 890 : let local_path = layer.local_path();
2203 :
2204 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
2205 : // the metadata in the upload should always match our current generation.
2206 890 : assert_eq!(layer_metadata.generation, self.generation);
2207 :
2208 890 : let remote_path = remote_layer_path(
2209 890 : &self.tenant_shard_id.tenant_id,
2210 890 : &self.timeline_id,
2211 890 : layer_metadata.shard,
2212 890 : &layer.layer_desc().layer_name(),
2213 890 : layer_metadata.generation,
2214 : );
2215 :
2216 890 : upload::upload_timeline_layer(
2217 890 : &self.storage_impl,
2218 890 : local_path,
2219 890 : &remote_path,
2220 890 : layer_metadata.file_size,
2221 890 : &self.cancel,
2222 890 : )
2223 890 : .measure_remote_op(
2224 890 : Some(TaskKind::RemoteUploadTask),
2225 890 : RemoteOpFileKind::Layer,
2226 890 : RemoteOpKind::Upload,
2227 890 : Arc::clone(&self.metrics),
2228 890 : )
2229 890 : .await
2230 : }
2231 774 : UploadOp::UploadMetadata { uploaded } => {
2232 774 : let res = upload::upload_index_part(
2233 774 : &self.storage_impl,
2234 774 : &self.tenant_shard_id,
2235 774 : &self.timeline_id,
2236 774 : self.generation,
2237 774 : uploaded,
2238 774 : &self.cancel,
2239 774 : )
2240 774 : .measure_remote_op(
2241 774 : Some(TaskKind::RemoteUploadTask),
2242 774 : RemoteOpFileKind::Index,
2243 774 : RemoteOpKind::Upload,
2244 774 : Arc::clone(&self.metrics),
2245 774 : )
2246 774 : .await;
2247 770 : if res.is_ok() {
2248 770 : self.update_remote_physical_size_gauge(Some(uploaded));
2249 770 : let mention_having_future_layers = if cfg!(feature = "testing") {
2250 770 : uploaded
2251 770 : .layer_metadata
2252 770 : .keys()
2253 8931 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
2254 : } else {
2255 0 : false
2256 : };
2257 770 : if mention_having_future_layers {
2258 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
2259 42 : tracing::info!(
2260 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
2261 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
2262 : );
2263 728 : }
2264 0 : }
2265 770 : res
2266 : }
2267 : // TODO: this should wait for the deletion to be executed by the deletion queue.
2268 : // Otherwise, the deletion may race with an upload and wrongfully delete a newer
2269 : // file. Some of the above logic attempts to work around this, it should be replaced
2270 : // by the upload queue ordering guarantees (see `can_bypass`). See:
2271 : // <https://github.com/neondatabase/neon/issues/10283>.
2272 207 : UploadOp::Delete(delete) => {
2273 207 : if self.config.read().unwrap().block_deletions {
2274 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2275 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2276 0 : queue.blocked_deletions.push(delete.clone());
2277 0 : }
2278 0 : Ok(())
2279 : } else {
2280 207 : pausable_failpoint!("before-delete-layer-pausable");
2281 207 : self.deletion_queue_client
2282 207 : .push_layers(
2283 207 : self.tenant_shard_id,
2284 207 : self.timeline_id,
2285 207 : self.generation,
2286 207 : delete.layers.clone(),
2287 : )
2288 207 : .map_err(|e| anyhow::anyhow!(e))
2289 : }
2290 : }
2291 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
2292 : // unreachable. Barrier operations are handled synchronously in
2293 : // launch_queued_tasks
2294 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
2295 0 : break;
2296 : }
2297 : };
2298 :
2299 0 : match upload_result {
2300 : Ok(()) => {
2301 1857 : break;
2302 : }
2303 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
2304 : // loop around to do the proper stopping
2305 0 : continue;
2306 : }
2307 0 : Err(e) => {
2308 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
2309 :
2310 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
2311 : // or other external reasons. Such issues are relatively regular, so log them
2312 : // at info level at first, and only WARN if the operation fails repeatedly.
2313 : //
2314 : // (See similar logic for downloads in `download::download_retry`)
2315 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
2316 0 : info!(
2317 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
2318 0 : task.op, retries, e
2319 : );
2320 : } else {
2321 0 : warn!(
2322 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
2323 0 : task.op, retries, e
2324 : );
2325 : }
2326 :
2327 : // sleep until it's time to retry, or we're cancelled
2328 0 : exponential_backoff(
2329 0 : retries,
2330 0 : DEFAULT_BASE_BACKOFF_SECONDS,
2331 0 : DEFAULT_MAX_BACKOFF_SECONDS,
2332 0 : &cancel,
2333 0 : )
2334 0 : .await;
2335 : }
2336 : }
2337 : }
2338 :
2339 1857 : let retries = task.retries.load(Ordering::SeqCst);
2340 1857 : if retries > 0 {
2341 0 : info!(
2342 0 : "remote task {} completed successfully after {} retries",
2343 0 : task.op, retries
2344 : );
2345 : } else {
2346 1857 : debug!("remote task {} completed successfully", task.op);
2347 : }
2348 :
2349 : // The task has completed successfully. Remove it from the in-progress list.
2350 1857 : let lsn_update = {
2351 1857 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
2352 1857 : let upload_queue = match upload_queue_guard.deref_mut() {
2353 0 : UploadQueue::Uninitialized => panic!(
2354 0 : "callers are responsible for ensuring this is only called on an initialized queue"
2355 : ),
2356 0 : UploadQueue::Stopped(_stopped) => None,
2357 1857 : UploadQueue::Initialized(qi) => Some(qi),
2358 : };
2359 :
2360 1857 : let upload_queue = match upload_queue {
2361 1857 : Some(upload_queue) => upload_queue,
2362 : None => {
2363 0 : info!("another concurrent task already stopped the queue");
2364 0 : return;
2365 : }
2366 : };
2367 :
2368 1857 : upload_queue.inprogress_tasks.remove(&task.task_id);
2369 :
2370 1857 : let lsn_update = match task.op {
2371 880 : UploadOp::UploadLayer(_, _, _) => None,
2372 770 : UploadOp::UploadMetadata { ref uploaded } => {
2373 : // the task id is reused as a monotonicity check for storing the "clean"
2374 : // IndexPart.
2375 770 : let last_updater = upload_queue.clean.1;
2376 770 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
2377 770 : let monotone = is_later || last_updater.is_none();
2378 :
2379 770 : assert!(
2380 770 : monotone,
2381 0 : "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}",
2382 0 : task.task_id
2383 : );
2384 :
2385 : // not taking ownership is wasteful
2386 770 : upload_queue.clean.0.clone_from(uploaded);
2387 770 : upload_queue.clean.1 = Some(task.task_id);
2388 :
2389 770 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
2390 770 : self.metrics
2391 770 : .projected_remote_consistent_lsn_gauge
2392 770 : .set(lsn.0);
2393 :
2394 770 : if self.generation.is_none() {
2395 : // Legacy mode: skip validating generation
2396 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
2397 0 : None
2398 770 : } else if self
2399 770 : .config
2400 770 : .read()
2401 770 : .unwrap()
2402 770 : .process_remote_consistent_lsn_updates
2403 : {
2404 770 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
2405 : } else {
2406 : // Our config disables remote_consistent_lsn updates: drop it.
2407 0 : None
2408 : }
2409 : }
2410 207 : UploadOp::Delete(_) => None,
2411 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
2412 : };
2413 :
2414 : // Launch any queued tasks that were unblocked by this one.
2415 1857 : self.launch_queued_tasks(upload_queue);
2416 1857 : lsn_update
2417 : };
2418 :
2419 1857 : if let Some((lsn, slot)) = lsn_update {
2420 : // Updates to the remote_consistent_lsn we advertise to pageservers
2421 : // are all routed through the DeletionQueue, to enforce important
2422 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2423 770 : self.deletion_queue_client
2424 770 : .update_remote_consistent_lsn(
2425 770 : self.tenant_shard_id,
2426 770 : self.timeline_id,
2427 770 : self.generation,
2428 770 : lsn,
2429 770 : slot,
2430 770 : )
2431 770 : .await;
2432 1087 : }
2433 :
2434 1857 : self.metric_end(&task.op);
2435 1857 : for coalesced_op in &task.coalesced_ops {
2436 4 : self.metric_end(coalesced_op);
2437 4 : }
2438 1857 : }
2439 :
2440 3818 : fn metric_impl(
2441 3818 : &self,
2442 3818 : op: &UploadOp,
2443 3818 : ) -> Option<(
2444 3818 : RemoteOpFileKind,
2445 3818 : RemoteOpKind,
2446 3818 : RemoteTimelineClientMetricsCallTrackSize,
2447 3818 : )> {
2448 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2449 3818 : let res = match op {
2450 1775 : UploadOp::UploadLayer(_, m, _) => (
2451 1775 : RemoteOpFileKind::Layer,
2452 1775 : RemoteOpKind::Upload,
2453 1775 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2454 1775 : ),
2455 1571 : UploadOp::UploadMetadata { .. } => (
2456 1571 : RemoteOpFileKind::Index,
2457 1571 : RemoteOpKind::Upload,
2458 1571 : DontTrackSize {
2459 1571 : reason: "metadata uploads are tiny",
2460 1571 : },
2461 1571 : ),
2462 468 : UploadOp::Delete(_delete) => (
2463 468 : RemoteOpFileKind::Layer,
2464 468 : RemoteOpKind::Delete,
2465 468 : DontTrackSize {
2466 468 : reason: "should we track deletes? positive or negative sign?",
2467 468 : },
2468 468 : ),
2469 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2470 : // we do not account these
2471 4 : return None;
2472 : }
2473 : };
2474 3814 : Some(res)
2475 3818 : }
2476 :
2477 1953 : fn metric_begin(&self, op: &UploadOp) {
2478 1953 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2479 1953 : Some(x) => x,
2480 0 : None => return,
2481 : };
2482 1953 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2483 1953 : guard.will_decrement_manually(); // in metric_end(), see right below
2484 1953 : }
2485 :
2486 1865 : fn metric_end(&self, op: &UploadOp) {
2487 1865 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2488 1861 : Some(x) => x,
2489 4 : None => return,
2490 : };
2491 1861 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2492 1865 : }
2493 :
2494 : /// Close the upload queue for new operations and cancel queued operations.
2495 : ///
2496 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2497 : ///
2498 : /// In-progress operations will still be running after this function returns.
2499 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2500 : /// to wait for them to complete, after calling this function.
2501 9 : pub(crate) fn stop(&self) {
2502 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2503 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2504 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2505 9 : let mut guard = self.upload_queue.lock().unwrap();
2506 9 : self.stop_impl(&mut guard);
2507 9 : }
2508 :
2509 9 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2510 9 : match &mut **guard {
2511 : UploadQueue::Uninitialized => {
2512 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2513 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2514 : }
2515 : UploadQueue::Stopped(_) => {
2516 : // nothing to do
2517 4 : info!("another concurrent task already shut down the queue");
2518 : }
2519 5 : UploadQueue::Initialized(initialized) => {
2520 5 : info!("shutting down upload queue");
2521 :
2522 : // Replace the queue with the Stopped state, taking ownership of the old
2523 : // Initialized queue. We will do some checks on it, and then drop it.
2524 5 : let qi = {
2525 : // Here we preserve working version of the upload queue for possible use during deletions.
2526 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2527 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2528 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2529 5 : let upload_queue_for_deletion = UploadQueueInitialized {
2530 5 : inprogress_limit: initialized.inprogress_limit,
2531 5 : task_counter: 0,
2532 5 : dirty: initialized.dirty.clone(),
2533 5 : clean: initialized.clean.clone(),
2534 5 : latest_files_changes_since_metadata_upload_scheduled: 0,
2535 5 : visible_remote_consistent_lsn: initialized
2536 5 : .visible_remote_consistent_lsn
2537 5 : .clone(),
2538 5 : inprogress_tasks: HashMap::default(),
2539 5 : queued_operations: VecDeque::default(),
2540 5 : #[cfg(feature = "testing")]
2541 5 : dangling_files: HashMap::default(),
2542 5 : blocked_deletions: Vec::new(),
2543 5 : shutting_down: false,
2544 5 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2545 5 : recently_deleted: HashSet::new(),
2546 5 : };
2547 :
2548 5 : let upload_queue = std::mem::replace(
2549 5 : &mut **guard,
2550 5 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2551 5 : UploadQueueStoppedDeletable {
2552 5 : upload_queue_for_deletion,
2553 5 : deleted_at: SetDeletedFlagProgress::NotRunning,
2554 5 : },
2555 5 : )),
2556 : );
2557 5 : if let UploadQueue::Initialized(qi) = upload_queue {
2558 5 : qi
2559 : } else {
2560 0 : unreachable!("we checked in the match above that it is Initialized");
2561 : }
2562 : };
2563 :
2564 : // We don't need to do anything here for in-progress tasks. They will finish
2565 : // on their own, decrement the unfinished-task counter themselves, and observe
2566 : // that the queue is Stopped.
2567 5 : drop(qi.inprogress_tasks);
2568 :
2569 : // Tear down queued ops
2570 5 : for op in qi.queued_operations.into_iter() {
2571 4 : self.metric_end(&op);
2572 4 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2573 4 : // which is exactly what we want to happen.
2574 4 : drop(op);
2575 4 : }
2576 : }
2577 : }
2578 9 : }
2579 :
2580 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2581 : /// externally to RemoteTimelineClient.
2582 0 : pub(crate) fn initialized_upload_queue(
2583 0 : &self,
2584 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2585 0 : let mut inner = self.upload_queue.lock().unwrap();
2586 0 : inner.initialized_mut()?;
2587 0 : Ok(UploadQueueAccessor { inner })
2588 0 : }
2589 :
2590 4 : pub(crate) fn no_pending_work(&self) -> bool {
2591 4 : let inner = self.upload_queue.lock().unwrap();
2592 4 : match &*inner {
2593 : UploadQueue::Uninitialized
2594 0 : | UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => true,
2595 4 : UploadQueue::Stopped(UploadQueueStopped::Deletable(x)) => {
2596 4 : x.upload_queue_for_deletion.no_pending_work()
2597 : }
2598 0 : UploadQueue::Initialized(x) => x.no_pending_work(),
2599 : }
2600 4 : }
2601 :
2602 : /// 'foreign' in the sense that it does not belong to this tenant shard. This method
2603 : /// is used during GC for other shards to get the index of shard zero.
2604 0 : pub(crate) async fn download_foreign_index(
2605 0 : &self,
2606 0 : shard_number: ShardNumber,
2607 0 : cancel: &CancellationToken,
2608 0 : ) -> Result<(IndexPart, Generation, std::time::SystemTime), DownloadError> {
2609 0 : let foreign_shard_id = TenantShardId {
2610 0 : shard_number,
2611 0 : shard_count: self.tenant_shard_id.shard_count,
2612 0 : tenant_id: self.tenant_shard_id.tenant_id,
2613 0 : };
2614 0 : download_index_part(
2615 0 : &self.storage_impl,
2616 0 : &foreign_shard_id,
2617 0 : &self.timeline_id,
2618 0 : Generation::MAX,
2619 0 : cancel,
2620 0 : )
2621 0 : .await
2622 0 : }
2623 : }
2624 :
2625 : pub(crate) struct UploadQueueAccessor<'a> {
2626 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2627 : }
2628 :
2629 : impl UploadQueueAccessor<'_> {
2630 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2631 0 : match &*self.inner {
2632 0 : UploadQueue::Initialized(x) => &x.clean.0,
2633 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2634 0 : unreachable!("checked before constructing")
2635 : }
2636 : }
2637 0 : }
2638 : }
2639 :
2640 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2641 0 : let path = format!("tenants/{tenant_shard_id}");
2642 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2643 0 : }
2644 :
2645 468 : pub fn remote_tenant_manifest_path(
2646 468 : tenant_shard_id: &TenantShardId,
2647 468 : generation: Generation,
2648 468 : ) -> RemotePath {
2649 468 : let path = format!(
2650 468 : "tenants/{tenant_shard_id}/tenant-manifest{}.json",
2651 468 : generation.get_suffix()
2652 : );
2653 468 : RemotePath::from_string(&path).expect("Failed to construct path")
2654 468 : }
2655 :
2656 : /// Prefix to all generations' manifest objects in a tenant shard
2657 119 : pub fn remote_tenant_manifest_prefix(tenant_shard_id: &TenantShardId) -> RemotePath {
2658 119 : let path = format!("tenants/{tenant_shard_id}/tenant-manifest",);
2659 119 : RemotePath::from_string(&path).expect("Failed to construct path")
2660 119 : }
2661 :
2662 134 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2663 134 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2664 134 : RemotePath::from_string(&path).expect("Failed to construct path")
2665 134 : }
2666 :
2667 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2668 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2669 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2670 0 : }
2671 :
2672 15 : pub fn remote_timeline_path(
2673 15 : tenant_shard_id: &TenantShardId,
2674 15 : timeline_id: &TimelineId,
2675 15 : ) -> RemotePath {
2676 15 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2677 15 : }
2678 :
2679 : /// Obtains the path of the given Layer in the remote
2680 : ///
2681 : /// Note that the shard component of a remote layer path is _not_ always the same
2682 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2683 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2684 1076 : pub fn remote_layer_path(
2685 1076 : tenant_id: &TenantId,
2686 1076 : timeline_id: &TimelineId,
2687 1076 : shard: ShardIndex,
2688 1076 : layer_file_name: &LayerName,
2689 1076 : generation: Generation,
2690 1076 : ) -> RemotePath {
2691 : // Generation-aware key format
2692 1076 : let path = format!(
2693 1076 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2694 1076 : shard.get_suffix(),
2695 : layer_file_name,
2696 1076 : generation.get_suffix()
2697 : );
2698 :
2699 1076 : RemotePath::from_string(&path).expect("Failed to construct path")
2700 1076 : }
2701 :
2702 : /// Returns true if a and b have the same layer path within a tenant/timeline. This is essentially
2703 : /// remote_layer_path(a) == remote_layer_path(b) without the string allocations.
2704 : ///
2705 : /// TODO: there should be a variant of LayerName for the physical path that contains information
2706 : /// about the shard and generation, such that this could be replaced by a simple comparison.
2707 892591 : pub fn is_same_remote_layer_path(
2708 892591 : aname: &LayerName,
2709 892591 : ameta: &LayerFileMetadata,
2710 892591 : bname: &LayerName,
2711 892591 : bmeta: &LayerFileMetadata,
2712 892591 : ) -> bool {
2713 : // NB: don't assert remote_layer_path(a) == remote_layer_path(b); too expensive even for debug.
2714 892591 : aname == bname && ameta.shard == bmeta.shard && ameta.generation == bmeta.generation
2715 892591 : }
2716 :
2717 2 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2718 2 : RemotePath::from_string(&format!(
2719 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2720 2 : ))
2721 2 : .expect("Failed to construct path")
2722 2 : }
2723 :
2724 1 : pub fn remote_initdb_preserved_archive_path(
2725 1 : tenant_id: &TenantId,
2726 1 : timeline_id: &TimelineId,
2727 1 : ) -> RemotePath {
2728 1 : RemotePath::from_string(&format!(
2729 1 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2730 1 : ))
2731 1 : .expect("Failed to construct path")
2732 1 : }
2733 :
2734 807 : pub fn remote_index_path(
2735 807 : tenant_shard_id: &TenantShardId,
2736 807 : timeline_id: &TimelineId,
2737 807 : generation: Generation,
2738 807 : ) -> RemotePath {
2739 807 : RemotePath::from_string(&format!(
2740 807 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2741 807 : IndexPart::FILE_NAME,
2742 807 : generation.get_suffix()
2743 807 : ))
2744 807 : .expect("Failed to construct path")
2745 807 : }
2746 :
2747 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2748 0 : RemotePath::from_string(&format!(
2749 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2750 0 : ))
2751 0 : .expect("Failed to construct path")
2752 0 : }
2753 :
2754 : /// Given the key of an index, parse out the generation part of the name
2755 9 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2756 9 : let file_name = match path.get_path().file_name() {
2757 9 : Some(f) => f,
2758 : None => {
2759 : // Unexpected: we should be seeing index_part.json paths only
2760 0 : tracing::warn!("Malformed index key {}", path);
2761 0 : return None;
2762 : }
2763 : };
2764 :
2765 9 : match file_name.split_once('-') {
2766 6 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2767 3 : None => None,
2768 : }
2769 9 : }
2770 :
2771 : /// Given the key of a tenant manifest, parse out the generation number
2772 0 : pub fn parse_remote_tenant_manifest_path(path: RemotePath) -> Option<Generation> {
2773 : static RE: OnceLock<Regex> = OnceLock::new();
2774 0 : let re = RE.get_or_init(|| Regex::new(r".*tenant-manifest-([0-9a-f]{8}).json").unwrap());
2775 0 : re.captures(path.get_path().as_str())
2776 0 : .and_then(|c| c.get(1))
2777 0 : .and_then(|m| Generation::parse_suffix(m.as_str()))
2778 0 : }
2779 :
2780 : #[cfg(test)]
2781 : mod tests {
2782 : use std::collections::HashSet;
2783 :
2784 : use super::*;
2785 : use crate::DEFAULT_PG_VERSION;
2786 : use crate::context::RequestContext;
2787 : use crate::tenant::config::AttachmentMode;
2788 : use crate::tenant::harness::{TIMELINE_ID, TenantHarness};
2789 : use crate::tenant::storage_layer::layer::local_layer_path;
2790 : use crate::tenant::{TenantShard, Timeline};
2791 :
2792 4 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2793 4 : format!("contents for {name}").into()
2794 4 : }
2795 :
2796 1 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2797 1 : let metadata = TimelineMetadata::new(
2798 1 : disk_consistent_lsn,
2799 1 : None,
2800 1 : None,
2801 1 : Lsn(0),
2802 1 : Lsn(0),
2803 1 : Lsn(0),
2804 : // Any version will do
2805 : // but it should be consistent with the one in the tests
2806 : crate::DEFAULT_PG_VERSION,
2807 : );
2808 :
2809 : // go through serialize + deserialize to fix the header, including checksum
2810 1 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2811 1 : }
2812 :
2813 1 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2814 3 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2815 1 : avec.sort();
2816 :
2817 1 : let mut bvec = b.to_vec();
2818 1 : bvec.sort_unstable();
2819 :
2820 1 : assert_eq!(avec, bvec);
2821 1 : }
2822 :
2823 2 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2824 2 : let mut expected: Vec<String> = expected
2825 2 : .iter()
2826 8 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2827 2 : .collect();
2828 2 : expected.sort();
2829 :
2830 2 : let mut found: Vec<String> = Vec::new();
2831 8 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2832 8 : let entry_name = entry.file_name();
2833 8 : let fname = entry_name.to_str().unwrap();
2834 8 : found.push(String::from(fname));
2835 8 : }
2836 2 : found.sort();
2837 :
2838 2 : assert_eq!(found, expected);
2839 2 : }
2840 :
2841 : struct TestSetup {
2842 : harness: TenantHarness,
2843 : tenant: Arc<TenantShard>,
2844 : timeline: Arc<Timeline>,
2845 : tenant_ctx: RequestContext,
2846 : }
2847 :
2848 : impl TestSetup {
2849 4 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2850 4 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2851 4 : let harness = TenantHarness::create(test_name).await?;
2852 4 : let (tenant, ctx) = harness.load().await;
2853 :
2854 4 : let timeline = tenant
2855 4 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2856 4 : .await?;
2857 :
2858 4 : Ok(Self {
2859 4 : harness,
2860 4 : tenant,
2861 4 : timeline,
2862 4 : tenant_ctx: ctx,
2863 4 : })
2864 4 : }
2865 :
2866 : /// Construct a RemoteTimelineClient in an arbitrary generation
2867 5 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2868 5 : let location_conf = AttachedLocationConfig {
2869 5 : generation,
2870 5 : attach_mode: AttachmentMode::Single,
2871 5 : };
2872 5 : Arc::new(RemoteTimelineClient {
2873 5 : conf: self.harness.conf,
2874 5 : runtime: tokio::runtime::Handle::current(),
2875 5 : tenant_shard_id: self.harness.tenant_shard_id,
2876 5 : timeline_id: TIMELINE_ID,
2877 5 : generation,
2878 5 : storage_impl: self.harness.remote_storage.clone(),
2879 5 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2880 5 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2881 5 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2882 5 : &self.harness.tenant_shard_id,
2883 5 : &TIMELINE_ID,
2884 5 : )),
2885 5 : config: std::sync::RwLock::new(RemoteTimelineClientConfig::from(&location_conf)),
2886 5 : cancel: CancellationToken::new(),
2887 5 : })
2888 5 : }
2889 :
2890 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2891 : /// and timeline_id are present.
2892 3 : fn span(&self) -> tracing::Span {
2893 3 : tracing::info_span!(
2894 : "test",
2895 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2896 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2897 : timeline_id = %TIMELINE_ID
2898 : )
2899 3 : }
2900 : }
2901 :
2902 : // Test scheduling
2903 : #[tokio::test]
2904 1 : async fn upload_scheduling() {
2905 : // Test outline:
2906 : //
2907 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2908 : // Schedule upload of index. Check that it is queued
2909 : // let the layer file uploads finish. Check that the index-upload is now started
2910 : // let the index-upload finish.
2911 : //
2912 : // Download back the index.json. Check that the list of files is correct
2913 : //
2914 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2915 : // let upload finish. Check that deletion is now started
2916 : // Schedule another deletion. Check that it's launched immediately.
2917 : // Schedule index upload. Check that it's queued
2918 :
2919 1 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2920 1 : let span = test_setup.span();
2921 1 : let _guard = span.enter();
2922 :
2923 : let TestSetup {
2924 1 : harness,
2925 1 : tenant: _tenant,
2926 1 : timeline,
2927 1 : tenant_ctx: _tenant_ctx,
2928 1 : } = test_setup;
2929 :
2930 1 : let client = &timeline.remote_client;
2931 :
2932 : // Download back the index.json, and check that the list of files is correct
2933 1 : let initial_index_part = match client
2934 1 : .download_index_file(&CancellationToken::new())
2935 1 : .await
2936 1 : .unwrap()
2937 : {
2938 1 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2939 0 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2940 : };
2941 1 : let initial_layers = initial_index_part
2942 1 : .layer_metadata
2943 1 : .keys()
2944 1 : .map(|f| f.to_owned())
2945 1 : .collect::<HashSet<LayerName>>();
2946 1 : let initial_layer = {
2947 1 : assert!(initial_layers.len() == 1);
2948 1 : initial_layers.into_iter().next().unwrap()
2949 : };
2950 :
2951 1 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2952 :
2953 1 : println!("workdir: {}", harness.conf.workdir);
2954 :
2955 1 : let remote_timeline_dir = harness
2956 1 : .remote_fs_dir
2957 1 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2958 1 : println!("remote_timeline_dir: {remote_timeline_dir}");
2959 :
2960 1 : let generation = harness.generation;
2961 1 : let shard = harness.shard;
2962 :
2963 : // Create a couple of dummy files, schedule upload for them
2964 :
2965 1 : let layers = [
2966 1 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2967 1 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2968 1 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2969 1 : ]
2970 1 : .into_iter()
2971 3 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2972 :
2973 3 : let local_path = local_layer_path(
2974 3 : harness.conf,
2975 3 : &timeline.tenant_shard_id,
2976 3 : &timeline.timeline_id,
2977 3 : &name,
2978 3 : &generation,
2979 : );
2980 3 : std::fs::write(&local_path, &contents).unwrap();
2981 :
2982 3 : Layer::for_resident(
2983 3 : harness.conf,
2984 3 : &timeline,
2985 3 : local_path,
2986 3 : name,
2987 3 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2988 : )
2989 3 : }).collect::<Vec<_>>();
2990 :
2991 1 : client
2992 1 : .schedule_layer_file_upload(layers[0].clone())
2993 1 : .unwrap();
2994 1 : client
2995 1 : .schedule_layer_file_upload(layers[1].clone())
2996 1 : .unwrap();
2997 :
2998 : // Check that they are started immediately, not queued
2999 : //
3000 : // this works because we running within block_on, so any futures are now queued up until
3001 : // our next await point.
3002 : {
3003 1 : let mut guard = client.upload_queue.lock().unwrap();
3004 1 : let upload_queue = guard.initialized_mut().unwrap();
3005 1 : assert!(upload_queue.queued_operations.is_empty());
3006 1 : assert_eq!(upload_queue.inprogress_tasks.len(), 2);
3007 1 : assert_eq!(upload_queue.num_inprogress_layer_uploads(), 2);
3008 :
3009 : // also check that `latest_file_changes` was updated
3010 1 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
3011 : }
3012 :
3013 : // Schedule upload of index. Check that it is queued
3014 1 : let metadata = dummy_metadata(Lsn(0x20));
3015 1 : client
3016 1 : .schedule_index_upload_for_full_metadata_update(&metadata)
3017 1 : .unwrap();
3018 : {
3019 1 : let mut guard = client.upload_queue.lock().unwrap();
3020 1 : let upload_queue = guard.initialized_mut().unwrap();
3021 1 : assert!(upload_queue.queued_operations.len() == 1);
3022 1 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
3023 : }
3024 :
3025 : // Wait for the uploads to finish
3026 1 : client.wait_completion().await.unwrap();
3027 : {
3028 1 : let mut guard = client.upload_queue.lock().unwrap();
3029 1 : let upload_queue = guard.initialized_mut().unwrap();
3030 :
3031 1 : assert!(upload_queue.queued_operations.is_empty());
3032 1 : assert!(upload_queue.inprogress_tasks.is_empty());
3033 : }
3034 :
3035 : // Download back the index.json, and check that the list of files is correct
3036 1 : let index_part = match client
3037 1 : .download_index_file(&CancellationToken::new())
3038 1 : .await
3039 1 : .unwrap()
3040 : {
3041 1 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
3042 0 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
3043 : };
3044 :
3045 1 : assert_file_list(
3046 1 : &index_part
3047 1 : .layer_metadata
3048 1 : .keys()
3049 3 : .map(|f| f.to_owned())
3050 1 : .collect(),
3051 1 : &[
3052 1 : &initial_layer.to_string(),
3053 1 : &layers[0].layer_desc().layer_name().to_string(),
3054 1 : &layers[1].layer_desc().layer_name().to_string(),
3055 1 : ],
3056 : );
3057 1 : assert_eq!(index_part.metadata, metadata);
3058 :
3059 : // Schedule upload and then a deletion. Check that the deletion is queued
3060 1 : client
3061 1 : .schedule_layer_file_upload(layers[2].clone())
3062 1 : .unwrap();
3063 :
3064 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
3065 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
3066 : // spawn_blocking started by the drop.
3067 1 : client
3068 1 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
3069 1 : .unwrap();
3070 : {
3071 1 : let mut guard = client.upload_queue.lock().unwrap();
3072 1 : let upload_queue = guard.initialized_mut().unwrap();
3073 :
3074 : // Deletion schedules upload of the index file, and the file deletion itself
3075 1 : assert_eq!(upload_queue.queued_operations.len(), 2);
3076 1 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
3077 1 : assert_eq!(upload_queue.num_inprogress_layer_uploads(), 1);
3078 1 : assert_eq!(upload_queue.num_inprogress_deletions(), 0);
3079 1 : assert_eq!(
3080 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
3081 : 0
3082 : );
3083 : }
3084 1 : assert_remote_files(
3085 1 : &[
3086 1 : &initial_layer.to_string(),
3087 1 : &layers[0].layer_desc().layer_name().to_string(),
3088 1 : &layers[1].layer_desc().layer_name().to_string(),
3089 1 : "index_part.json",
3090 1 : ],
3091 1 : &remote_timeline_dir,
3092 1 : generation,
3093 : );
3094 :
3095 : // Finish them
3096 1 : client.wait_completion().await.unwrap();
3097 1 : harness.deletion_queue.pump().await;
3098 :
3099 1 : assert_remote_files(
3100 1 : &[
3101 1 : &initial_layer.to_string(),
3102 1 : &layers[1].layer_desc().layer_name().to_string(),
3103 1 : &layers[2].layer_desc().layer_name().to_string(),
3104 1 : "index_part.json",
3105 1 : ],
3106 1 : &remote_timeline_dir,
3107 1 : generation,
3108 1 : );
3109 1 : }
3110 :
3111 : #[tokio::test]
3112 1 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
3113 : // Setup
3114 :
3115 : let TestSetup {
3116 1 : harness,
3117 1 : tenant: _tenant,
3118 1 : timeline,
3119 : ..
3120 1 : } = TestSetup::new("metrics").await.unwrap();
3121 1 : let client = &timeline.remote_client;
3122 :
3123 1 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
3124 1 : let local_path = local_layer_path(
3125 1 : harness.conf,
3126 1 : &timeline.tenant_shard_id,
3127 1 : &timeline.timeline_id,
3128 1 : &layer_file_name_1,
3129 1 : &harness.generation,
3130 : );
3131 1 : let content_1 = dummy_contents("foo");
3132 1 : std::fs::write(&local_path, &content_1).unwrap();
3133 :
3134 1 : let layer_file_1 = Layer::for_resident(
3135 1 : harness.conf,
3136 1 : &timeline,
3137 1 : local_path,
3138 1 : layer_file_name_1.clone(),
3139 1 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
3140 : );
3141 :
3142 : #[derive(Debug, PartialEq, Clone, Copy)]
3143 : struct BytesStartedFinished {
3144 : started: Option<usize>,
3145 : finished: Option<usize>,
3146 : }
3147 : impl std::ops::Add for BytesStartedFinished {
3148 : type Output = Self;
3149 2 : fn add(self, rhs: Self) -> Self::Output {
3150 : Self {
3151 2 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
3152 2 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
3153 : }
3154 2 : }
3155 : }
3156 3 : let get_bytes_started_stopped = || {
3157 3 : let started = client
3158 3 : .metrics
3159 3 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
3160 3 : .map(|v| v.try_into().unwrap());
3161 3 : let stopped = client
3162 3 : .metrics
3163 3 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
3164 3 : .map(|v| v.try_into().unwrap());
3165 3 : BytesStartedFinished {
3166 3 : started,
3167 3 : finished: stopped,
3168 3 : }
3169 3 : };
3170 :
3171 : // Test
3172 1 : tracing::info!("now doing actual test");
3173 :
3174 1 : let actual_a = get_bytes_started_stopped();
3175 :
3176 1 : client
3177 1 : .schedule_layer_file_upload(layer_file_1.clone())
3178 1 : .unwrap();
3179 :
3180 1 : let actual_b = get_bytes_started_stopped();
3181 :
3182 1 : client.wait_completion().await.unwrap();
3183 :
3184 1 : let actual_c = get_bytes_started_stopped();
3185 :
3186 : // Validate
3187 :
3188 1 : let expected_b = actual_a
3189 1 : + BytesStartedFinished {
3190 1 : started: Some(content_1.len()),
3191 1 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
3192 1 : finished: Some(0),
3193 1 : };
3194 1 : assert_eq!(actual_b, expected_b);
3195 :
3196 1 : let expected_c = actual_a
3197 1 : + BytesStartedFinished {
3198 1 : started: Some(content_1.len()),
3199 1 : finished: Some(content_1.len()),
3200 1 : };
3201 1 : assert_eq!(actual_c, expected_c);
3202 1 : }
3203 :
3204 6 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
3205 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
3206 6 : let example_index_part = IndexPart::example();
3207 :
3208 6 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
3209 :
3210 6 : let index_path = test_state.harness.remote_fs_dir.join(
3211 6 : remote_index_path(
3212 6 : &test_state.harness.tenant_shard_id,
3213 6 : &TIMELINE_ID,
3214 6 : generation,
3215 6 : )
3216 6 : .get_path(),
3217 6 : );
3218 :
3219 6 : std::fs::create_dir_all(index_path.parent().unwrap())
3220 6 : .expect("creating test dir should work");
3221 :
3222 6 : eprintln!("Writing {index_path}");
3223 6 : std::fs::write(&index_path, index_part_bytes).unwrap();
3224 6 : example_index_part
3225 6 : }
3226 :
3227 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
3228 : /// index, the IndexPart returned is equal to `expected`
3229 5 : async fn assert_got_index_part(
3230 5 : test_state: &TestSetup,
3231 5 : get_generation: Generation,
3232 5 : expected: &IndexPart,
3233 5 : ) {
3234 5 : let client = test_state.build_client(get_generation);
3235 :
3236 5 : let download_r = client
3237 5 : .download_index_file(&CancellationToken::new())
3238 5 : .await
3239 5 : .expect("download should always succeed");
3240 5 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
3241 5 : match download_r {
3242 5 : MaybeDeletedIndexPart::IndexPart(index_part) => {
3243 5 : assert_eq!(&index_part, expected);
3244 : }
3245 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
3246 : }
3247 5 : }
3248 :
3249 : #[tokio::test]
3250 1 : async fn index_part_download_simple() -> anyhow::Result<()> {
3251 1 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
3252 1 : let span = test_state.span();
3253 1 : let _guard = span.enter();
3254 :
3255 : // Simple case: we are in generation N, load the index from generation N - 1
3256 1 : let generation_n = 5;
3257 1 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
3258 :
3259 1 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
3260 :
3261 2 : Ok(())
3262 1 : }
3263 :
3264 : #[tokio::test]
3265 1 : async fn index_part_download_ordering() -> anyhow::Result<()> {
3266 1 : let test_state = TestSetup::new("index_part_download_ordering")
3267 1 : .await
3268 1 : .unwrap();
3269 :
3270 1 : let span = test_state.span();
3271 1 : let _guard = span.enter();
3272 :
3273 : // A generation-less IndexPart exists in the bucket, we should find it
3274 1 : let generation_n = 5;
3275 1 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
3276 1 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
3277 :
3278 : // If a more recent-than-none generation exists, we should prefer to load that
3279 1 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
3280 1 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
3281 :
3282 : // If a more-recent-than-me generation exists, we should ignore it.
3283 1 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
3284 1 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
3285 :
3286 : // If a directly previous generation exists, _and_ an index exists in my own
3287 : // generation, I should prefer my own generation.
3288 1 : let _injected_prev =
3289 1 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
3290 1 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
3291 1 : assert_got_index_part(
3292 1 : &test_state,
3293 1 : Generation::new(generation_n),
3294 1 : &injected_current,
3295 1 : )
3296 1 : .await;
3297 :
3298 2 : Ok(())
3299 1 : }
3300 : }
|
