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 : //! From the user's perspective, the operations are executed sequentially.
67 : //! Internally, the client knows which operations can be performed in parallel,
68 : //! and which operations act like a "barrier" that require preceding operations
69 : //! to finish. The calling code just needs to call the schedule-functions in the
70 : //! correct order, and the client will parallelize the operations in a way that
71 : //! is safe.
72 : //!
73 : //! The caller should be careful with deletion, though. They should not delete
74 : //! local files that have been scheduled for upload but not yet finished uploading.
75 : //! Otherwise the upload will fail. To wait for an upload to finish, use
76 : //! the 'wait_completion' function (more on that later.)
77 : //!
78 : //! All of this relies on the following invariants:
79 : //!
80 : //! - We rely on read-after write consistency in the remote storage.
81 : //! - Layer files are immutable
82 : //!
83 : //! NB: Pageserver assumes that it has exclusive write access to the tenant in remote
84 : //! storage. Different tenants can be attached to different pageservers, but if the
85 : //! same tenant is attached to two pageservers at the same time, they will overwrite
86 : //! each other's index file updates, and confusion will ensue. There's no interlock or
87 : //! mechanism to detect that in the pageserver, we rely on the control plane to ensure
88 : //! that that doesn't happen.
89 : //!
90 : //! ## Implementation Note
91 : //!
92 : //! The *actual* remote state lags behind the *desired* remote state while
93 : //! there are in-flight operations.
94 : //! We keep track of the desired remote state in [`UploadQueueInitialized::dirty`].
95 : //! It is initialized based on the [`IndexPart`] that was passed during init
96 : //! and updated with every `schedule_*` function call.
97 : //! All this is necessary necessary to compute the future [`IndexPart`]s
98 : //! when scheduling an operation while other operations that also affect the
99 : //! remote [`IndexPart`] are in flight.
100 : //!
101 : //! # Retries & Error Handling
102 : //!
103 : //! The client retries operations indefinitely, using exponential back-off.
104 : //! There is no way to force a retry, i.e., interrupt the back-off.
105 : //! This could be built easily.
106 : //!
107 : //! # Cancellation
108 : //!
109 : //! The operations execute as plain [`task_mgr`] tasks, scoped to
110 : //! the client's tenant and timeline.
111 : //! Dropping the client will drop queued operations but not executing operations.
112 : //! These will complete unless the `task_mgr` tasks are cancelled using `task_mgr`
113 : //! APIs, e.g., during pageserver shutdown, timeline delete, or tenant detach.
114 : //!
115 : //! # Completion
116 : //!
117 : //! Once an operation has completed, we update [`UploadQueueInitialized::clean`] immediately,
118 : //! and submit a request through the DeletionQueue to update
119 : //! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
120 : //! validated that our generation is not stale. It is this visible value
121 : //! that is advertized to safekeepers as a signal that that they can
122 : //! delete the WAL up to that LSN.
123 : //!
124 : //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
125 : //! for all pending operations to complete. It does not prevent more
126 : //! operations from getting scheduled.
127 : //!
128 : //! # Crash Consistency
129 : //!
130 : //! We do not persist the upload queue state.
131 : //! If we drop the client, or crash, all unfinished operations are lost.
132 : //!
133 : //! To recover, the following steps need to be taken:
134 : //! - Retrieve the current remote [`IndexPart`]. This gives us a
135 : //! consistent remote state, assuming the user scheduled the operations in
136 : //! the correct order.
137 : //! - Initiate upload queue with that [`IndexPart`].
138 : //! - Reschedule all lost operations by comparing the local filesystem state
139 : //! and remote state as per [`IndexPart`]. This is done in
140 : //! [`Tenant::timeline_init_and_sync`].
141 : //!
142 : //! Note that if we crash during file deletion between the index update
143 : //! that removes the file from the list of files, and deleting the remote file,
144 : //! the file is leaked in the remote storage. Similarly, if a new file is created
145 : //! and uploaded, but the pageserver dies permanently before updating the
146 : //! remote index file, the new file is leaked in remote storage. We accept and
147 : //! tolerate that for now.
148 : //! Note further that we cannot easily fix this by scheduling deletes for every
149 : //! file that is present only on the remote, because we cannot distinguish the
150 : //! following two cases:
151 : //! - (1) We had the file locally, deleted it locally, scheduled a remote delete,
152 : //! but crashed before it finished remotely.
153 : //! - (2) We never had the file locally because we haven't on-demand downloaded
154 : //! it yet.
155 : //!
156 : //! # Downloads
157 : //!
158 : //! In addition to the upload queue, [`RemoteTimelineClient`] has functions for
159 : //! downloading files from the remote storage. Downloads are performed immediately
160 : //! against the `RemoteStorage`, independently of the upload queue.
161 : //!
162 : //! When we attach a tenant, we perform the following steps:
163 : //! - create `Tenant` object in `TenantState::Attaching` state
164 : //! - List timelines that are present in remote storage, and for each:
165 : //! - download their remote [`IndexPart`]s
166 : //! - create `Timeline` struct and a `RemoteTimelineClient`
167 : //! - initialize the client's upload queue with its `IndexPart`
168 : //! - schedule uploads for layers that are only present locally.
169 : //! - After the above is done for each timeline, open the tenant for business by
170 : //! transitioning it from `TenantState::Attaching` to `TenantState::Active` state.
171 : //! This starts the timelines' WAL-receivers and the tenant's GC & Compaction loops.
172 : //!
173 : //! # Operating Without Remote Storage
174 : //!
175 : //! If no remote storage configuration is provided, the [`RemoteTimelineClient`] is
176 : //! not created and the uploads are skipped.
177 : //!
178 : //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
179 : //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
180 :
181 : pub(crate) mod download;
182 : pub mod index;
183 : pub(crate) mod upload;
184 :
185 : use anyhow::Context;
186 : use camino::Utf8Path;
187 : use chrono::{NaiveDateTime, Utc};
188 :
189 : pub(crate) use download::download_initdb_tar_zst;
190 : use pageserver_api::models::TimelineArchivalState;
191 : use pageserver_api::shard::{ShardIndex, TenantShardId};
192 : use scopeguard::ScopeGuard;
193 : use tokio_util::sync::CancellationToken;
194 : pub(crate) use upload::upload_initdb_dir;
195 : use utils::backoff::{
196 : self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
197 : };
198 : use utils::pausable_failpoint;
199 :
200 : use std::collections::{HashMap, VecDeque};
201 : use std::sync::atomic::{AtomicU32, Ordering};
202 : use std::sync::{Arc, Mutex};
203 : use std::time::Duration;
204 :
205 : use remote_storage::{
206 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
207 : };
208 : use std::ops::DerefMut;
209 : use tracing::{debug, error, info, instrument, warn};
210 : use tracing::{info_span, Instrument};
211 : use utils::lsn::Lsn;
212 :
213 : use crate::context::RequestContext;
214 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
215 : use crate::metrics::{
216 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
217 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
218 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
219 : };
220 : use crate::task_mgr::shutdown_token;
221 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
222 : use crate::tenant::remote_timeline_client::download::download_retry;
223 : use crate::tenant::storage_layer::AsLayerDesc;
224 : use crate::tenant::upload_queue::{Delete, UploadQueueStoppedDeletable};
225 : use crate::tenant::TIMELINES_SEGMENT_NAME;
226 : use crate::{
227 : config::PageServerConf,
228 : task_mgr,
229 : task_mgr::TaskKind,
230 : task_mgr::BACKGROUND_RUNTIME,
231 : tenant::metadata::TimelineMetadata,
232 : tenant::upload_queue::{
233 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
234 : },
235 : TENANT_HEATMAP_BASENAME,
236 : };
237 :
238 : use utils::id::{TenantId, TimelineId};
239 :
240 : use self::index::IndexPart;
241 :
242 : use super::metadata::MetadataUpdate;
243 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
244 : use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
245 : use super::Generation;
246 :
247 : pub(crate) use download::{
248 : download_index_part, is_temp_download_file, list_remote_tenant_shards, list_remote_timelines,
249 : };
250 : pub(crate) use index::LayerFileMetadata;
251 :
252 : // Occasional network issues and such can cause remote operations to fail, and
253 : // that's expected. If a download fails, we log it at info-level, and retry.
254 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
255 : // level instead, as repeated failures can mean a more serious problem. If it
256 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
257 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
258 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
259 :
260 : // Similarly log failed uploads and deletions at WARN level, after this many
261 : // retries. Uploads and deletions are retried forever, though.
262 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
263 :
264 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
265 :
266 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
267 :
268 : /// Default buffer size when interfacing with [`tokio::fs::File`].
269 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
270 :
271 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
272 : /// which we warn and skip.
273 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
274 :
275 : pub enum MaybeDeletedIndexPart {
276 : IndexPart(IndexPart),
277 : Deleted(IndexPart),
278 : }
279 :
280 0 : #[derive(Debug, thiserror::Error)]
281 : pub enum PersistIndexPartWithDeletedFlagError {
282 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
283 : AlreadyInProgress(NaiveDateTime),
284 : #[error("the deleted_flag was already set, value is {0:?}")]
285 : AlreadyDeleted(NaiveDateTime),
286 : #[error(transparent)]
287 : Other(#[from] anyhow::Error),
288 : }
289 :
290 0 : #[derive(Debug, thiserror::Error)]
291 : pub enum WaitCompletionError {
292 : #[error(transparent)]
293 : NotInitialized(NotInitialized),
294 : #[error("wait_completion aborted because upload queue was stopped")]
295 : UploadQueueShutDownOrStopped,
296 : }
297 :
298 : /// A client for accessing a timeline's data in remote storage.
299 : ///
300 : /// This takes care of managing the number of connections, and balancing them
301 : /// across tenants. This also handles retries of failed uploads.
302 : ///
303 : /// Upload and delete requests are ordered so that before a deletion is
304 : /// performed, we wait for all preceding uploads to finish. This ensures sure
305 : /// that if you perform a compaction operation that reshuffles data in layer
306 : /// files, we don't have a transient state where the old files have already been
307 : /// deleted, but new files have not yet been uploaded.
308 : ///
309 : /// Similarly, this enforces an order between index-file uploads, and layer
310 : /// uploads. Before an index-file upload is performed, all preceding layer
311 : /// uploads must be finished.
312 : ///
313 : /// This also maintains a list of remote files, and automatically includes that
314 : /// in the index part file, whenever timeline metadata is uploaded.
315 : ///
316 : /// Downloads are not queued, they are performed immediately.
317 : pub struct RemoteTimelineClient {
318 : conf: &'static PageServerConf,
319 :
320 : runtime: tokio::runtime::Handle,
321 :
322 : tenant_shard_id: TenantShardId,
323 : timeline_id: TimelineId,
324 : generation: Generation,
325 :
326 : upload_queue: Mutex<UploadQueue>,
327 :
328 : pub(crate) metrics: Arc<RemoteTimelineClientMetrics>,
329 :
330 : storage_impl: GenericRemoteStorage,
331 :
332 : deletion_queue_client: DeletionQueueClient,
333 :
334 : cancel: CancellationToken,
335 : }
336 :
337 : impl RemoteTimelineClient {
338 : ///
339 : /// Create a remote storage client for given timeline
340 : ///
341 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
342 : /// by calling init_upload_queue.
343 : ///
344 410 : pub fn new(
345 410 : remote_storage: GenericRemoteStorage,
346 410 : deletion_queue_client: DeletionQueueClient,
347 410 : conf: &'static PageServerConf,
348 410 : tenant_shard_id: TenantShardId,
349 410 : timeline_id: TimelineId,
350 410 : generation: Generation,
351 410 : ) -> RemoteTimelineClient {
352 410 : RemoteTimelineClient {
353 410 : conf,
354 410 : runtime: if cfg!(test) {
355 : // remote_timeline_client.rs tests rely on current-thread runtime
356 410 : tokio::runtime::Handle::current()
357 : } else {
358 0 : BACKGROUND_RUNTIME.handle().clone()
359 : },
360 410 : tenant_shard_id,
361 410 : timeline_id,
362 410 : generation,
363 410 : storage_impl: remote_storage,
364 410 : deletion_queue_client,
365 410 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
366 410 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
367 410 : &tenant_shard_id,
368 410 : &timeline_id,
369 410 : )),
370 410 : cancel: CancellationToken::new(),
371 410 : }
372 410 : }
373 :
374 : /// Initialize the upload queue for a remote storage that already received
375 : /// an index file upload, i.e., it's not empty.
376 : /// The given `index_part` must be the one on the remote.
377 6 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
378 6 : let mut upload_queue = self.upload_queue.lock().unwrap();
379 6 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
380 6 : self.update_remote_physical_size_gauge(Some(index_part));
381 6 : info!(
382 0 : "initialized upload queue from remote index with {} layer files",
383 0 : index_part.layer_metadata.len()
384 : );
385 6 : Ok(())
386 6 : }
387 :
388 : /// Initialize the upload queue for the case where the remote storage is empty,
389 : /// i.e., it doesn't have an `IndexPart`.
390 404 : pub fn init_upload_queue_for_empty_remote(
391 404 : &self,
392 404 : local_metadata: &TimelineMetadata,
393 404 : ) -> anyhow::Result<()> {
394 404 : let mut upload_queue = self.upload_queue.lock().unwrap();
395 404 : upload_queue.initialize_empty_remote(local_metadata)?;
396 404 : self.update_remote_physical_size_gauge(None);
397 404 : info!("initialized upload queue as empty");
398 404 : Ok(())
399 404 : }
400 :
401 : /// Initialize the queue in stopped state. Used in startup path
402 : /// to continue deletion operation interrupted by pageserver crash or restart.
403 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
404 0 : &self,
405 0 : index_part: &IndexPart,
406 0 : ) -> anyhow::Result<()> {
407 : // FIXME: consider newtype for DeletedIndexPart.
408 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
409 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
410 0 : ))?;
411 :
412 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
413 0 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
414 0 : self.update_remote_physical_size_gauge(Some(index_part));
415 0 : self.stop_impl(&mut upload_queue);
416 0 :
417 0 : upload_queue
418 0 : .stopped_mut()
419 0 : .expect("stopped above")
420 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
421 0 :
422 0 : Ok(())
423 0 : }
424 :
425 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
426 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
427 0 : match &mut *self.upload_queue.lock().unwrap() {
428 0 : UploadQueue::Uninitialized => None,
429 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
430 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
431 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
432 0 : .upload_queue_for_deletion
433 0 : .get_last_remote_consistent_lsn_projected(),
434 : }
435 0 : }
436 :
437 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
438 0 : match &mut *self.upload_queue.lock().unwrap() {
439 0 : UploadQueue::Uninitialized => None,
440 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
441 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
442 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
443 0 : q.upload_queue_for_deletion
444 0 : .get_last_remote_consistent_lsn_visible(),
445 0 : ),
446 : }
447 0 : }
448 :
449 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
450 : /// client is currently initialized.
451 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
452 0 : self.upload_queue
453 0 : .lock()
454 0 : .unwrap()
455 0 : .initialized_mut()
456 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
457 0 : .unwrap_or(false)
458 0 : }
459 :
460 : /// Returns whether the timeline is archived.
461 : /// Return None if the remote index_part hasn't been downloaded yet.
462 0 : pub(crate) fn is_archived(&self) -> Option<bool> {
463 0 : self.upload_queue
464 0 : .lock()
465 0 : .unwrap()
466 0 : .initialized_mut()
467 0 : .map(|q| q.clean.0.archived_at.is_some())
468 0 : .ok()
469 0 : }
470 :
471 1806 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
472 1806 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
473 1402 : current_remote_index_part
474 1402 : .layer_metadata
475 1402 : .values()
476 17647 : .map(|ilmd| ilmd.file_size)
477 1402 : .sum()
478 : } else {
479 404 : 0
480 : };
481 1806 : self.metrics.remote_physical_size_gauge.set(size);
482 1806 : }
483 :
484 2 : pub fn get_remote_physical_size(&self) -> u64 {
485 2 : self.metrics.remote_physical_size_gauge.get()
486 2 : }
487 :
488 : //
489 : // Download operations.
490 : //
491 : // These don't use the per-timeline queue. They do use the global semaphore in
492 : // S3Bucket, to limit the total number of concurrent operations, though.
493 : //
494 :
495 : /// Download index file
496 20 : pub async fn download_index_file(
497 20 : &self,
498 20 : cancel: &CancellationToken,
499 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
500 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
501 20 : &RemoteOpFileKind::Index,
502 20 : &RemoteOpKind::Download,
503 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
504 20 : reason: "no need for a downloads gauge",
505 20 : },
506 20 : );
507 :
508 20 : let (index_part, index_generation, index_last_modified) = download::download_index_part(
509 20 : &self.storage_impl,
510 20 : &self.tenant_shard_id,
511 20 : &self.timeline_id,
512 20 : self.generation,
513 20 : cancel,
514 20 : )
515 20 : .measure_remote_op(
516 20 : RemoteOpFileKind::Index,
517 20 : RemoteOpKind::Download,
518 20 : Arc::clone(&self.metrics),
519 20 : )
520 73 : .await?;
521 :
522 : // Defense in depth: monotonicity of generation numbers is an important correctness guarantee, so when we see a very
523 : // old index, we do extra checks in case this is the result of backward time-travel of the generation number (e.g.
524 : // in case of a bug in the service that issues generation numbers). Indices are allowed to be old, but we expect that
525 : // when we load an old index we are loading the _latest_ index: if we are asked to load an old index and there is
526 : // also a newer index available, that is surprising.
527 : const INDEX_AGE_CHECKS_THRESHOLD: Duration = Duration::from_secs(14 * 24 * 3600);
528 20 : let index_age = index_last_modified.elapsed().unwrap_or_else(|e| {
529 0 : if e.duration() > Duration::from_secs(5) {
530 : // We only warn if the S3 clock and our local clock are >5s out: because this is a low resolution
531 : // timestamp, it is common to be out by at least 1 second.
532 0 : tracing::warn!("Index has modification time in the future: {e}");
533 0 : }
534 0 : Duration::ZERO
535 20 : });
536 20 : if index_age > INDEX_AGE_CHECKS_THRESHOLD {
537 0 : tracing::info!(
538 : ?index_generation,
539 0 : age = index_age.as_secs_f64(),
540 0 : "Loaded an old index, checking for other indices..."
541 : );
542 :
543 : // Find the highest-generation index
544 0 : let (_latest_index_part, latest_index_generation, latest_index_mtime) =
545 0 : download::download_index_part(
546 0 : &self.storage_impl,
547 0 : &self.tenant_shard_id,
548 0 : &self.timeline_id,
549 0 : Generation::MAX,
550 0 : cancel,
551 0 : )
552 0 : .await?;
553 :
554 0 : if latest_index_generation > index_generation {
555 : // Unexpected! Why are we loading such an old index if a more recent one exists?
556 0 : tracing::warn!(
557 : ?index_generation,
558 : ?latest_index_generation,
559 : ?latest_index_mtime,
560 0 : "Found a newer index while loading an old one"
561 : );
562 0 : }
563 20 : }
564 :
565 20 : if index_part.deleted_at.is_some() {
566 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
567 : } else {
568 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
569 : }
570 20 : }
571 :
572 : /// Download a (layer) file from `path`, into local filesystem.
573 : ///
574 : /// 'layer_metadata' is the metadata from the remote index file.
575 : ///
576 : /// On success, returns the size of the downloaded file.
577 6 : pub async fn download_layer_file(
578 6 : &self,
579 6 : layer_file_name: &LayerName,
580 6 : layer_metadata: &LayerFileMetadata,
581 6 : local_path: &Utf8Path,
582 6 : cancel: &CancellationToken,
583 6 : ctx: &RequestContext,
584 6 : ) -> Result<u64, DownloadError> {
585 6 : let downloaded_size = {
586 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
587 6 : &RemoteOpFileKind::Layer,
588 6 : &RemoteOpKind::Download,
589 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
590 6 : reason: "no need for a downloads gauge",
591 6 : },
592 6 : );
593 6 : download::download_layer_file(
594 6 : self.conf,
595 6 : &self.storage_impl,
596 6 : self.tenant_shard_id,
597 6 : self.timeline_id,
598 6 : layer_file_name,
599 6 : layer_metadata,
600 6 : local_path,
601 6 : cancel,
602 6 : ctx,
603 6 : )
604 6 : .measure_remote_op(
605 6 : RemoteOpFileKind::Layer,
606 6 : RemoteOpKind::Download,
607 6 : Arc::clone(&self.metrics),
608 6 : )
609 72 : .await?
610 : };
611 :
612 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
613 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
614 6 :
615 6 : Ok(downloaded_size)
616 6 : }
617 :
618 : //
619 : // Upload operations.
620 : //
621 :
622 : /// Launch an index-file upload operation in the background, with
623 : /// fully updated metadata.
624 : ///
625 : /// This should only be used to upload initial metadata to remote storage.
626 : ///
627 : /// The upload will be added to the queue immediately, but it
628 : /// won't be performed until all previously scheduled layer file
629 : /// upload operations have completed successfully. This is to
630 : /// ensure that when the index file claims that layers X, Y and Z
631 : /// exist in remote storage, they really do. To wait for the upload
632 : /// to complete, use `wait_completion`.
633 : ///
634 : /// If there were any changes to the list of files, i.e. if any
635 : /// layer file uploads were scheduled, since the last index file
636 : /// upload, those will be included too.
637 228 : pub fn schedule_index_upload_for_full_metadata_update(
638 228 : self: &Arc<Self>,
639 228 : metadata: &TimelineMetadata,
640 228 : ) -> anyhow::Result<()> {
641 228 : let mut guard = self.upload_queue.lock().unwrap();
642 228 : let upload_queue = guard.initialized_mut()?;
643 :
644 : // As documented in the struct definition, it's ok for latest_metadata to be
645 : // ahead of what's _actually_ on the remote during index upload.
646 228 : upload_queue.dirty.metadata = metadata.clone();
647 228 :
648 228 : self.schedule_index_upload(upload_queue)?;
649 :
650 228 : Ok(())
651 228 : }
652 :
653 : /// Launch an index-file upload operation in the background, with only parts of the metadata
654 : /// updated.
655 : ///
656 : /// This is the regular way of updating metadata on layer flushes or Gc.
657 : ///
658 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
659 : /// `index_part.json`, while being more clear on what values update regularly.
660 1140 : pub(crate) fn schedule_index_upload_for_metadata_update(
661 1140 : self: &Arc<Self>,
662 1140 : update: &MetadataUpdate,
663 1140 : ) -> anyhow::Result<()> {
664 1140 : let mut guard = self.upload_queue.lock().unwrap();
665 1140 : let upload_queue = guard.initialized_mut()?;
666 :
667 1140 : upload_queue.dirty.metadata.apply(update);
668 1140 :
669 1140 : self.schedule_index_upload(upload_queue)?;
670 :
671 1140 : Ok(())
672 1140 : }
673 :
674 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
675 : ///
676 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
677 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
678 : /// been in the queue yet.
679 0 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
680 0 : self: &Arc<Self>,
681 0 : state: TimelineArchivalState,
682 0 : ) -> anyhow::Result<bool> {
683 0 : let mut guard = self.upload_queue.lock().unwrap();
684 0 : let upload_queue = guard.initialized_mut()?;
685 :
686 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
687 : /// change needed to set archived_at.
688 0 : fn need_change(
689 0 : archived_at: &Option<NaiveDateTime>,
690 0 : state: TimelineArchivalState,
691 0 : ) -> Option<bool> {
692 0 : match (archived_at, state) {
693 : (Some(_), TimelineArchivalState::Archived)
694 : | (None, TimelineArchivalState::Unarchived) => {
695 : // Nothing to do
696 0 : tracing::info!("intended state matches present state");
697 0 : None
698 : }
699 0 : (None, TimelineArchivalState::Archived) => Some(true),
700 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
701 : }
702 0 : }
703 0 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
704 :
705 0 : if let Some(archived_at_set) = need_upload_scheduled {
706 0 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
707 0 : upload_queue.dirty.archived_at = intended_archived_at;
708 0 : self.schedule_index_upload(upload_queue)?;
709 0 : }
710 :
711 0 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
712 0 : Ok(need_wait)
713 0 : }
714 :
715 : ///
716 : /// Launch an index-file upload operation in the background, if necessary.
717 : ///
718 : /// Use this function to schedule the update of the index file after
719 : /// scheduling file uploads or deletions. If no file uploads or deletions
720 : /// have been scheduled since the last index file upload, this does
721 : /// nothing.
722 : ///
723 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
724 : /// the upload to the upload queue and returns quickly.
725 370 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
726 370 : let mut guard = self.upload_queue.lock().unwrap();
727 370 : let upload_queue = guard.initialized_mut()?;
728 :
729 370 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
730 14 : self.schedule_index_upload(upload_queue)?;
731 356 : }
732 :
733 370 : Ok(())
734 370 : }
735 :
736 : /// Launch an index-file upload operation in the background (internal function)
737 1428 : fn schedule_index_upload(
738 1428 : self: &Arc<Self>,
739 1428 : upload_queue: &mut UploadQueueInitialized,
740 1428 : ) -> Result<(), NotInitialized> {
741 1428 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
742 1428 : // fix up the duplicated field
743 1428 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
744 1428 :
745 1428 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
746 1428 : // look like a retryable error
747 1428 : let void = std::io::sink();
748 1428 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
749 1428 :
750 1428 : let index_part = &upload_queue.dirty;
751 1428 :
752 1428 : info!(
753 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
754 0 : index_part.layer_metadata.len(),
755 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
756 : );
757 :
758 1428 : let op = UploadOp::UploadMetadata {
759 1428 : uploaded: Box::new(index_part.clone()),
760 1428 : };
761 1428 : self.metric_begin(&op);
762 1428 : upload_queue.queued_operations.push_back(op);
763 1428 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
764 1428 :
765 1428 : // Launch the task immediately, if possible
766 1428 : self.launch_queued_tasks(upload_queue);
767 1428 : Ok(())
768 1428 : }
769 :
770 : /// Reparent this timeline to a new parent.
771 : ///
772 : /// A retryable step of timeline ancestor detach.
773 0 : pub(crate) async fn schedule_reparenting_and_wait(
774 0 : self: &Arc<Self>,
775 0 : new_parent: &TimelineId,
776 0 : ) -> anyhow::Result<()> {
777 0 : let receiver = {
778 0 : let mut guard = self.upload_queue.lock().unwrap();
779 0 : let upload_queue = guard.initialized_mut()?;
780 :
781 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
782 0 : return Err(anyhow::anyhow!(
783 0 : "cannot reparent without a current ancestor"
784 0 : ));
785 : };
786 :
787 0 : let uploaded = &upload_queue.clean.0.metadata;
788 0 :
789 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
790 : // nothing to do
791 0 : None
792 : } else {
793 0 : upload_queue.dirty.metadata.reparent(new_parent);
794 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
795 0 :
796 0 : self.schedule_index_upload(upload_queue)?;
797 :
798 0 : Some(self.schedule_barrier0(upload_queue))
799 : }
800 : };
801 :
802 0 : if let Some(receiver) = receiver {
803 0 : Self::wait_completion0(receiver).await?;
804 0 : }
805 0 : Ok(())
806 0 : }
807 :
808 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
809 : /// detaching from ancestor and waits for it to complete.
810 : ///
811 : /// This is used with `Timeline::detach_ancestor` functionality.
812 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
813 0 : self: &Arc<Self>,
814 0 : layers: &[Layer],
815 0 : adopted: (TimelineId, Lsn),
816 0 : ) -> anyhow::Result<()> {
817 0 : let barrier = {
818 0 : let mut guard = self.upload_queue.lock().unwrap();
819 0 : let upload_queue = guard.initialized_mut()?;
820 :
821 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
822 0 : None
823 : } else {
824 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
825 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
826 :
827 0 : for layer in layers {
828 0 : let prev = upload_queue
829 0 : .dirty
830 0 : .layer_metadata
831 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
832 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
833 : }
834 :
835 0 : self.schedule_index_upload(upload_queue)?;
836 :
837 0 : Some(self.schedule_barrier0(upload_queue))
838 : }
839 : };
840 :
841 0 : if let Some(barrier) = barrier {
842 0 : Self::wait_completion0(barrier).await?;
843 0 : }
844 0 : Ok(())
845 0 : }
846 :
847 : /// Adds a gc blocking reason for this timeline if one does not exist already.
848 : ///
849 : /// A retryable step of timeline detach ancestor.
850 : ///
851 : /// Returns a future which waits until the completion of the upload.
852 0 : pub(crate) fn schedule_insert_gc_block_reason(
853 0 : self: &Arc<Self>,
854 0 : reason: index::GcBlockingReason,
855 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
856 0 : {
857 0 : let maybe_barrier = {
858 0 : let mut guard = self.upload_queue.lock().unwrap();
859 0 : let upload_queue = guard.initialized_mut()?;
860 :
861 0 : if let index::GcBlockingReason::DetachAncestor = reason {
862 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
863 0 : drop(guard);
864 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
865 0 : }
866 0 : }
867 :
868 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
869 :
870 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
871 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
872 0 :
873 0 : match (current, uploaded) {
874 0 : (x, y) if wanted(x) && wanted(y) => None,
875 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
876 : // Usual case: !wanted(x) && !wanted(y)
877 : //
878 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
879 : // turn on and off some reason.
880 0 : (x, y) => {
881 0 : if !wanted(x) && wanted(y) {
882 : // this could be avoided by having external in-memory synchronization, like
883 : // timeline detach ancestor
884 0 : warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
885 0 : }
886 :
887 : // at this point, the metadata must always show that there is a parent
888 0 : upload_queue.dirty.gc_blocking = current
889 0 : .map(|x| x.with_reason(reason))
890 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
891 0 : self.schedule_index_upload(upload_queue)?;
892 0 : Some(self.schedule_barrier0(upload_queue))
893 : }
894 : }
895 : };
896 :
897 0 : Ok(async move {
898 0 : if let Some(barrier) = maybe_barrier {
899 0 : Self::wait_completion0(barrier).await?;
900 0 : }
901 0 : Ok(())
902 0 : })
903 0 : }
904 :
905 : /// Removes a gc blocking reason for this timeline if one exists.
906 : ///
907 : /// A retryable step of timeline detach ancestor.
908 : ///
909 : /// Returns a future which waits until the completion of the upload.
910 0 : pub(crate) fn schedule_remove_gc_block_reason(
911 0 : self: &Arc<Self>,
912 0 : reason: index::GcBlockingReason,
913 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
914 0 : {
915 0 : let maybe_barrier = {
916 0 : let mut guard = self.upload_queue.lock().unwrap();
917 0 : let upload_queue = guard.initialized_mut()?;
918 :
919 0 : if let index::GcBlockingReason::DetachAncestor = reason {
920 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
921 0 : drop(guard);
922 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
923 0 : }
924 0 : }
925 :
926 0 : let wanted = |x: Option<&index::GcBlocking>| {
927 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
928 0 : };
929 :
930 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
931 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
932 0 :
933 0 : match (current, uploaded) {
934 0 : (x, y) if wanted(x) && wanted(y) => None,
935 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
936 0 : (x, y) => {
937 0 : if !wanted(x) && wanted(y) {
938 0 : warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
939 0 : }
940 :
941 0 : upload_queue.dirty.gc_blocking =
942 0 : current.as_ref().and_then(|x| x.without_reason(reason));
943 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
944 : // FIXME: bogus ?
945 0 : self.schedule_index_upload(upload_queue)?;
946 0 : Some(self.schedule_barrier0(upload_queue))
947 : }
948 : }
949 : };
950 :
951 0 : Ok(async move {
952 0 : if let Some(barrier) = maybe_barrier {
953 0 : Self::wait_completion0(barrier).await?;
954 0 : }
955 0 : Ok(())
956 0 : })
957 0 : }
958 :
959 : /// Launch an upload operation in the background; the file is added to be included in next
960 : /// `index_part.json` upload.
961 1184 : pub(crate) fn schedule_layer_file_upload(
962 1184 : self: &Arc<Self>,
963 1184 : layer: ResidentLayer,
964 1184 : ) -> Result<(), NotInitialized> {
965 1184 : let mut guard = self.upload_queue.lock().unwrap();
966 1184 : let upload_queue = guard.initialized_mut()?;
967 :
968 1184 : self.schedule_layer_file_upload0(upload_queue, layer);
969 1184 : self.launch_queued_tasks(upload_queue);
970 1184 : Ok(())
971 1184 : }
972 :
973 1514 : fn schedule_layer_file_upload0(
974 1514 : self: &Arc<Self>,
975 1514 : upload_queue: &mut UploadQueueInitialized,
976 1514 : layer: ResidentLayer,
977 1514 : ) {
978 1514 : let metadata = layer.metadata();
979 1514 :
980 1514 : upload_queue
981 1514 : .dirty
982 1514 : .layer_metadata
983 1514 : .insert(layer.layer_desc().layer_name(), metadata.clone());
984 1514 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
985 1514 :
986 1514 : info!(
987 : gen=?metadata.generation,
988 : shard=?metadata.shard,
989 0 : "scheduled layer file upload {layer}",
990 : );
991 :
992 1514 : let op = UploadOp::UploadLayer(layer, metadata);
993 1514 : self.metric_begin(&op);
994 1514 : upload_queue.queued_operations.push_back(op);
995 1514 : }
996 :
997 : /// Launch a delete operation in the background.
998 : ///
999 : /// The operation does not modify local filesystem state.
1000 : ///
1001 : /// Note: This schedules an index file upload before the deletions. The
1002 : /// deletion won't actually be performed, until all previously scheduled
1003 : /// upload operations, and the index file upload, have completed
1004 : /// successfully.
1005 8 : pub fn schedule_layer_file_deletion(
1006 8 : self: &Arc<Self>,
1007 8 : names: &[LayerName],
1008 8 : ) -> anyhow::Result<()> {
1009 8 : let mut guard = self.upload_queue.lock().unwrap();
1010 8 : let upload_queue = guard.initialized_mut()?;
1011 :
1012 8 : let with_metadata = self
1013 8 : .schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned())?;
1014 :
1015 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1016 8 :
1017 8 : // Launch the tasks immediately, if possible
1018 8 : self.launch_queued_tasks(upload_queue);
1019 8 : Ok(())
1020 8 : }
1021 :
1022 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1023 : /// layer files, leaving them dangling.
1024 : ///
1025 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1026 : /// is invoked on them.
1027 4 : pub(crate) fn schedule_gc_update(
1028 4 : self: &Arc<Self>,
1029 4 : gc_layers: &[Layer],
1030 4 : ) -> Result<(), NotInitialized> {
1031 4 : let mut guard = self.upload_queue.lock().unwrap();
1032 4 : let upload_queue = guard.initialized_mut()?;
1033 :
1034 : // just forget the return value; after uploading the next index_part.json, we can consider
1035 : // the layer files as "dangling". this is fine, at worst case we create work for the
1036 : // scrubber.
1037 :
1038 4 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1039 4 :
1040 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1041 :
1042 4 : self.launch_queued_tasks(upload_queue);
1043 4 :
1044 4 : Ok(())
1045 4 : }
1046 :
1047 : /// Update the remote index file, removing the to-be-deleted files from the index,
1048 : /// allowing scheduling of actual deletions later.
1049 62 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1050 62 : self: &Arc<Self>,
1051 62 : upload_queue: &mut UploadQueueInitialized,
1052 62 : names: I,
1053 62 : ) -> Result<Vec<(LayerName, LayerFileMetadata)>, NotInitialized>
1054 62 : where
1055 62 : I: IntoIterator<Item = LayerName>,
1056 62 : {
1057 62 : // Decorate our list of names with each name's metadata, dropping
1058 62 : // names that are unexpectedly missing from our metadata. This metadata
1059 62 : // is later used when physically deleting layers, to construct key paths.
1060 62 : let with_metadata: Vec<_> = names
1061 62 : .into_iter()
1062 456 : .filter_map(|name| {
1063 456 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1064 :
1065 456 : if let Some(meta) = meta {
1066 422 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1067 422 : Some((name, meta))
1068 : } else {
1069 : // This can only happen if we forgot to to schedule the file upload
1070 : // before scheduling the delete. Log it because it is a rare/strange
1071 : // situation, and in case something is misbehaving, we'd like to know which
1072 : // layers experienced this.
1073 34 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1074 34 : None
1075 : }
1076 456 : })
1077 62 : .collect();
1078 :
1079 : #[cfg(feature = "testing")]
1080 484 : for (name, metadata) in &with_metadata {
1081 422 : let gen = metadata.generation;
1082 422 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
1083 0 : if unexpected == gen {
1084 0 : tracing::error!("{name} was unlinked twice with same generation");
1085 : } else {
1086 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
1087 : }
1088 422 : }
1089 : }
1090 :
1091 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1092 : // index_part update, because that needs to be uploaded before we can actually delete the
1093 : // files.
1094 62 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1095 46 : self.schedule_index_upload(upload_queue)?;
1096 16 : }
1097 :
1098 62 : Ok(with_metadata)
1099 62 : }
1100 :
1101 : /// Schedules deletion for layer files which have previously been unlinked from the
1102 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1103 457 : pub(crate) fn schedule_deletion_of_unlinked(
1104 457 : self: &Arc<Self>,
1105 457 : layers: Vec<(LayerName, LayerFileMetadata)>,
1106 457 : ) -> anyhow::Result<()> {
1107 457 : let mut guard = self.upload_queue.lock().unwrap();
1108 457 : let upload_queue = guard.initialized_mut()?;
1109 :
1110 457 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1111 457 : self.launch_queued_tasks(upload_queue);
1112 457 : Ok(())
1113 457 : }
1114 :
1115 465 : fn schedule_deletion_of_unlinked0(
1116 465 : self: &Arc<Self>,
1117 465 : upload_queue: &mut UploadQueueInitialized,
1118 465 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1119 465 : ) {
1120 465 : // Filter out any layers which were not created by this tenant shard. These are
1121 465 : // layers that originate from some ancestor shard after a split, and may still
1122 465 : // be referenced by other shards. We are free to delete them locally and remove
1123 465 : // them from our index (and would have already done so when we reach this point
1124 465 : // in the code), but we may not delete them remotely.
1125 465 : with_metadata.retain(|(name, meta)| {
1126 459 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1127 459 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1128 459 : if !retain {
1129 0 : tracing::debug!(
1130 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1131 : meta.shard
1132 : );
1133 459 : }
1134 459 : retain
1135 465 : });
1136 :
1137 924 : for (name, meta) in &with_metadata {
1138 459 : info!(
1139 0 : "scheduling deletion of layer {}{} (shard {})",
1140 0 : name,
1141 0 : meta.generation.get_suffix(),
1142 : meta.shard
1143 : );
1144 : }
1145 :
1146 : #[cfg(feature = "testing")]
1147 924 : for (name, meta) in &with_metadata {
1148 459 : let gen = meta.generation;
1149 459 : match upload_queue.dangling_files.remove(name) {
1150 420 : Some(same) if same == gen => { /* expected */ }
1151 0 : Some(other) => {
1152 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
1153 : }
1154 : None => {
1155 39 : tracing::error!("{name} was unlinked but was not dangling");
1156 : }
1157 : }
1158 : }
1159 :
1160 : // schedule the actual deletions
1161 465 : if with_metadata.is_empty() {
1162 : // avoid scheduling the op & bumping the metric
1163 6 : return;
1164 459 : }
1165 459 : let op = UploadOp::Delete(Delete {
1166 459 : layers: with_metadata,
1167 459 : });
1168 459 : self.metric_begin(&op);
1169 459 : upload_queue.queued_operations.push_back(op);
1170 465 : }
1171 :
1172 : /// Schedules a compaction update to the remote `index_part.json`.
1173 : ///
1174 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1175 50 : pub(crate) fn schedule_compaction_update(
1176 50 : self: &Arc<Self>,
1177 50 : compacted_from: &[Layer],
1178 50 : compacted_to: &[ResidentLayer],
1179 50 : ) -> Result<(), NotInitialized> {
1180 50 : let mut guard = self.upload_queue.lock().unwrap();
1181 50 : let upload_queue = guard.initialized_mut()?;
1182 :
1183 380 : for layer in compacted_to {
1184 330 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1185 330 : }
1186 :
1187 450 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1188 50 :
1189 50 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1190 50 : self.launch_queued_tasks(upload_queue);
1191 50 :
1192 50 : Ok(())
1193 50 : }
1194 :
1195 : /// Wait for all previously scheduled uploads/deletions to complete
1196 1316 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1197 1316 : let receiver = {
1198 1316 : let mut guard = self.upload_queue.lock().unwrap();
1199 1316 : let upload_queue = guard
1200 1316 : .initialized_mut()
1201 1316 : .map_err(WaitCompletionError::NotInitialized)?;
1202 1316 : self.schedule_barrier0(upload_queue)
1203 1316 : };
1204 1316 :
1205 1316 : Self::wait_completion0(receiver).await
1206 1316 : }
1207 :
1208 1316 : async fn wait_completion0(
1209 1316 : mut receiver: tokio::sync::watch::Receiver<()>,
1210 1316 : ) -> Result<(), WaitCompletionError> {
1211 1316 : if receiver.changed().await.is_err() {
1212 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1213 1316 : }
1214 1316 :
1215 1316 : Ok(())
1216 1316 : }
1217 :
1218 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1219 6 : let mut guard = self.upload_queue.lock().unwrap();
1220 6 : let upload_queue = guard.initialized_mut()?;
1221 6 : self.schedule_barrier0(upload_queue);
1222 6 : Ok(())
1223 6 : }
1224 :
1225 1322 : fn schedule_barrier0(
1226 1322 : self: &Arc<Self>,
1227 1322 : upload_queue: &mut UploadQueueInitialized,
1228 1322 : ) -> tokio::sync::watch::Receiver<()> {
1229 1322 : let (sender, receiver) = tokio::sync::watch::channel(());
1230 1322 : let barrier_op = UploadOp::Barrier(sender);
1231 1322 :
1232 1322 : upload_queue.queued_operations.push_back(barrier_op);
1233 1322 : // Don't count this kind of operation!
1234 1322 :
1235 1322 : // Launch the task immediately, if possible
1236 1322 : self.launch_queued_tasks(upload_queue);
1237 1322 :
1238 1322 : receiver
1239 1322 : }
1240 :
1241 : /// Wait for all previously scheduled operations to complete, and then stop.
1242 : ///
1243 : /// Not cancellation safe
1244 6 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1245 6 : // On cancellation the queue is left in ackward state of refusing new operations but
1246 6 : // proper stop is yet to be called. On cancel the original or some later task must call
1247 6 : // `stop` or `shutdown`.
1248 6 : let sg = scopeguard::guard((), |_| {
1249 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1250 6 : });
1251 :
1252 6 : let fut = {
1253 6 : let mut guard = self.upload_queue.lock().unwrap();
1254 6 : let upload_queue = match &mut *guard {
1255 0 : UploadQueue::Stopped(_) => return,
1256 : UploadQueue::Uninitialized => {
1257 : // transition into Stopped state
1258 0 : self.stop_impl(&mut guard);
1259 0 : return;
1260 : }
1261 6 : UploadQueue::Initialized(ref mut init) => init,
1262 6 : };
1263 6 :
1264 6 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1265 6 : // just don't add more of these as they would never complete.
1266 6 : //
1267 6 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1268 6 : // in every place we would not have to jump through this hoop, and this method could be
1269 6 : // made cancellable.
1270 6 : if !upload_queue.shutting_down {
1271 6 : upload_queue.shutting_down = true;
1272 6 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1273 6 : // this operation is not counted similar to Barrier
1274 6 :
1275 6 : self.launch_queued_tasks(upload_queue);
1276 6 : }
1277 :
1278 6 : upload_queue.shutdown_ready.clone().acquire_owned()
1279 : };
1280 :
1281 6 : let res = fut.await;
1282 :
1283 6 : scopeguard::ScopeGuard::into_inner(sg);
1284 6 :
1285 6 : match res {
1286 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1287 6 : Err(_closed) => {
1288 6 : // expected
1289 6 : }
1290 6 : }
1291 6 :
1292 6 : self.stop();
1293 6 : }
1294 :
1295 : /// Set the deleted_at field in the remote index file.
1296 : ///
1297 : /// This fails if the upload queue has not been `stop()`ed.
1298 : ///
1299 : /// The caller is responsible for calling `stop()` AND for waiting
1300 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1301 : /// Check method [`RemoteTimelineClient::stop`] for details.
1302 0 : #[instrument(skip_all)]
1303 : pub(crate) async fn persist_index_part_with_deleted_flag(
1304 : self: &Arc<Self>,
1305 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1306 : let index_part_with_deleted_at = {
1307 : let mut locked = self.upload_queue.lock().unwrap();
1308 :
1309 : // We must be in stopped state because otherwise
1310 : // we can have inprogress index part upload that can overwrite the file
1311 : // with missing is_deleted flag that we going to set below
1312 : let stopped = locked.stopped_mut()?;
1313 :
1314 : match stopped.deleted_at {
1315 : SetDeletedFlagProgress::NotRunning => (), // proceed
1316 : SetDeletedFlagProgress::InProgress(at) => {
1317 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1318 : }
1319 : SetDeletedFlagProgress::Successful(at) => {
1320 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1321 : }
1322 : };
1323 : let deleted_at = Utc::now().naive_utc();
1324 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1325 :
1326 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1327 : index_part.deleted_at = Some(deleted_at);
1328 : index_part
1329 : };
1330 :
1331 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1332 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1333 0 : let stopped = locked
1334 0 : .stopped_mut()
1335 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1336 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1337 0 : });
1338 :
1339 : pausable_failpoint!("persist_deleted_index_part");
1340 :
1341 : backoff::retry(
1342 0 : || {
1343 0 : upload::upload_index_part(
1344 0 : &self.storage_impl,
1345 0 : &self.tenant_shard_id,
1346 0 : &self.timeline_id,
1347 0 : self.generation,
1348 0 : &index_part_with_deleted_at,
1349 0 : &self.cancel,
1350 0 : )
1351 0 : },
1352 0 : |_e| false,
1353 : 1,
1354 : // have just a couple of attempts
1355 : // when executed as part of timeline deletion this happens in context of api call
1356 : // when executed as part of tenant deletion this happens in the background
1357 : 2,
1358 : "persist_index_part_with_deleted_flag",
1359 : &self.cancel,
1360 : )
1361 : .await
1362 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1363 0 : .and_then(|x| x)?;
1364 :
1365 : // all good, disarm the guard and mark as success
1366 : ScopeGuard::into_inner(undo_deleted_at);
1367 : {
1368 : let mut locked = self.upload_queue.lock().unwrap();
1369 :
1370 : let stopped = locked
1371 : .stopped_mut()
1372 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1373 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1374 : index_part_with_deleted_at
1375 : .deleted_at
1376 : .expect("we set it above"),
1377 : );
1378 : }
1379 :
1380 : Ok(())
1381 : }
1382 :
1383 0 : pub(crate) fn is_deleting(&self) -> bool {
1384 0 : let mut locked = self.upload_queue.lock().unwrap();
1385 0 : locked.stopped_mut().is_ok()
1386 0 : }
1387 :
1388 0 : pub(crate) async fn preserve_initdb_archive(
1389 0 : self: &Arc<Self>,
1390 0 : tenant_id: &TenantId,
1391 0 : timeline_id: &TimelineId,
1392 0 : cancel: &CancellationToken,
1393 0 : ) -> anyhow::Result<()> {
1394 0 : backoff::retry(
1395 0 : || async {
1396 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1397 0 : .await
1398 0 : },
1399 0 : TimeoutOrCancel::caused_by_cancel,
1400 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1401 0 : FAILED_REMOTE_OP_RETRIES,
1402 0 : "preserve_initdb_tar_zst",
1403 0 : &cancel.clone(),
1404 0 : )
1405 0 : .await
1406 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1407 0 : .and_then(|x| x)
1408 0 : .context("backing up initdb archive")?;
1409 0 : Ok(())
1410 0 : }
1411 :
1412 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1413 : ///
1414 : /// This is not normally needed.
1415 0 : pub(crate) async fn upload_layer_file(
1416 0 : self: &Arc<Self>,
1417 0 : uploaded: &ResidentLayer,
1418 0 : cancel: &CancellationToken,
1419 0 : ) -> anyhow::Result<()> {
1420 0 : let remote_path = remote_layer_path(
1421 0 : &self.tenant_shard_id.tenant_id,
1422 0 : &self.timeline_id,
1423 0 : self.tenant_shard_id.to_index(),
1424 0 : &uploaded.layer_desc().layer_name(),
1425 0 : uploaded.metadata().generation,
1426 0 : );
1427 0 :
1428 0 : backoff::retry(
1429 0 : || async {
1430 0 : upload::upload_timeline_layer(
1431 0 : &self.storage_impl,
1432 0 : uploaded.local_path(),
1433 0 : &remote_path,
1434 0 : uploaded.metadata().file_size,
1435 0 : cancel,
1436 0 : )
1437 0 : .await
1438 0 : },
1439 0 : TimeoutOrCancel::caused_by_cancel,
1440 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1441 0 : FAILED_REMOTE_OP_RETRIES,
1442 0 : "upload a layer without adding it to latest files",
1443 0 : cancel,
1444 0 : )
1445 0 : .await
1446 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1447 0 : .and_then(|x| x)
1448 0 : .context("upload a layer without adding it to latest files")
1449 0 : }
1450 :
1451 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1452 : /// not added to be part of a future `index_part.json` upload.
1453 0 : pub(crate) async fn copy_timeline_layer(
1454 0 : self: &Arc<Self>,
1455 0 : adopted: &Layer,
1456 0 : adopted_as: &Layer,
1457 0 : cancel: &CancellationToken,
1458 0 : ) -> anyhow::Result<()> {
1459 0 : let source_remote_path = remote_layer_path(
1460 0 : &self.tenant_shard_id.tenant_id,
1461 0 : &adopted
1462 0 : .get_timeline_id()
1463 0 : .expect("Source timeline should be alive"),
1464 0 : self.tenant_shard_id.to_index(),
1465 0 : &adopted.layer_desc().layer_name(),
1466 0 : adopted.metadata().generation,
1467 0 : );
1468 0 :
1469 0 : let target_remote_path = remote_layer_path(
1470 0 : &self.tenant_shard_id.tenant_id,
1471 0 : &self.timeline_id,
1472 0 : self.tenant_shard_id.to_index(),
1473 0 : &adopted_as.layer_desc().layer_name(),
1474 0 : adopted_as.metadata().generation,
1475 0 : );
1476 0 :
1477 0 : backoff::retry(
1478 0 : || async {
1479 0 : upload::copy_timeline_layer(
1480 0 : &self.storage_impl,
1481 0 : &source_remote_path,
1482 0 : &target_remote_path,
1483 0 : cancel,
1484 0 : )
1485 0 : .await
1486 0 : },
1487 0 : TimeoutOrCancel::caused_by_cancel,
1488 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1489 0 : FAILED_REMOTE_OP_RETRIES,
1490 0 : "copy timeline layer",
1491 0 : cancel,
1492 0 : )
1493 0 : .await
1494 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1495 0 : .and_then(|x| x)
1496 0 : .context("remote copy timeline layer")
1497 0 : }
1498 :
1499 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1500 0 : match tokio::time::timeout(
1501 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1502 0 : self.deletion_queue_client.flush_immediate(),
1503 0 : )
1504 0 : .await
1505 : {
1506 0 : Ok(result) => result,
1507 0 : Err(_timeout) => {
1508 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1509 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1510 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1511 0 : tracing::warn!(
1512 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1513 : );
1514 0 : Ok(())
1515 : }
1516 : }
1517 0 : }
1518 :
1519 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1520 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1521 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1522 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
1523 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1524 :
1525 0 : let layers: Vec<RemotePath> = {
1526 0 : let mut locked = self.upload_queue.lock().unwrap();
1527 0 : let stopped = locked.stopped_mut()?;
1528 :
1529 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1530 0 : anyhow::bail!("deleted_at is not set")
1531 0 : }
1532 0 :
1533 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1534 :
1535 0 : stopped
1536 0 : .upload_queue_for_deletion
1537 0 : .dirty
1538 0 : .layer_metadata
1539 0 : .drain()
1540 0 : .filter(|(_file_name, meta)| {
1541 0 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1542 0 : // all shards anyway, we _could_ delete these, but
1543 0 : // - it creates a potential race if other shards are still
1544 0 : // using the layers while this shard deletes them.
1545 0 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1546 0 : // these timelines are present but corrupt (their index exists but some layers don't)
1547 0 : //
1548 0 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1549 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1550 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1551 0 : })
1552 0 : .map(|(file_name, meta)| {
1553 0 : remote_layer_path(
1554 0 : &self.tenant_shard_id.tenant_id,
1555 0 : &self.timeline_id,
1556 0 : meta.shard,
1557 0 : &file_name,
1558 0 : meta.generation,
1559 0 : )
1560 0 : })
1561 0 : .collect()
1562 0 : };
1563 0 :
1564 0 : let layer_deletion_count = layers.len();
1565 0 : self.deletion_queue_client.push_immediate(layers).await?;
1566 :
1567 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1568 : // inexistant objects are not considered errors.
1569 0 : let initdb_path =
1570 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1571 0 : self.deletion_queue_client
1572 0 : .push_immediate(vec![initdb_path])
1573 0 : .await?;
1574 :
1575 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1576 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1577 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1578 0 :
1579 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1580 0 : // taking the burden of listing all the layers that we already know we should delete.
1581 0 : self.flush_deletion_queue().await?;
1582 :
1583 0 : let cancel = shutdown_token();
1584 :
1585 0 : let remaining = download_retry(
1586 0 : || async {
1587 0 : self.storage_impl
1588 0 : .list(
1589 0 : Some(&timeline_storage_path),
1590 0 : ListingMode::NoDelimiter,
1591 0 : None,
1592 0 : &cancel,
1593 0 : )
1594 0 : .await
1595 0 : },
1596 0 : "list remaining files",
1597 0 : &cancel,
1598 0 : )
1599 0 : .await
1600 0 : .context("list files remaining files")?
1601 : .keys;
1602 :
1603 : // We will delete the current index_part object last, since it acts as a deletion
1604 : // marker via its deleted_at attribute
1605 0 : let latest_index = remaining
1606 0 : .iter()
1607 0 : .filter(|o| {
1608 0 : o.key
1609 0 : .object_name()
1610 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1611 0 : .unwrap_or(false)
1612 0 : })
1613 0 : .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
1614 0 : .max_by_key(|i| i.1)
1615 0 : .map(|i| i.0.clone())
1616 0 : .unwrap_or(
1617 0 : // No generation-suffixed indices, assume we are dealing with
1618 0 : // a legacy index.
1619 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1620 0 : );
1621 0 :
1622 0 : let remaining_layers: Vec<RemotePath> = remaining
1623 0 : .into_iter()
1624 0 : .filter_map(|o| {
1625 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1626 0 : None
1627 : } else {
1628 0 : Some(o.key)
1629 : }
1630 0 : })
1631 0 : .inspect(|path| {
1632 0 : if let Some(name) = path.object_name() {
1633 0 : info!(%name, "deleting a file not referenced from index_part.json");
1634 : } else {
1635 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1636 : }
1637 0 : })
1638 0 : .collect();
1639 0 :
1640 0 : let not_referenced_count = remaining_layers.len();
1641 0 : if !remaining_layers.is_empty() {
1642 0 : self.deletion_queue_client
1643 0 : .push_immediate(remaining_layers)
1644 0 : .await?;
1645 0 : }
1646 :
1647 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1648 0 : Err(anyhow::anyhow!(
1649 0 : "failpoint: timeline-delete-before-index-delete"
1650 0 : ))?
1651 0 : });
1652 :
1653 0 : debug!("enqueuing index part deletion");
1654 0 : self.deletion_queue_client
1655 0 : .push_immediate([latest_index].to_vec())
1656 0 : .await?;
1657 :
1658 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1659 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1660 0 : self.flush_deletion_queue().await?;
1661 :
1662 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1663 0 : Err(anyhow::anyhow!(
1664 0 : "failpoint: timeline-delete-after-index-delete"
1665 0 : ))?
1666 0 : });
1667 :
1668 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1669 :
1670 0 : Ok(())
1671 0 : }
1672 :
1673 : ///
1674 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1675 : /// the ordering constraints.
1676 : ///
1677 : /// The caller needs to already hold the `upload_queue` lock.
1678 7506 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1679 12050 : while let Some(next_op) = upload_queue.queued_operations.front() {
1680 : // Can we run this task now?
1681 8718 : let can_run_now = match next_op {
1682 : UploadOp::UploadLayer(..) => {
1683 : // Can always be scheduled.
1684 1512 : true
1685 : }
1686 : UploadOp::UploadMetadata { .. } => {
1687 : // These can only be performed after all the preceding operations
1688 : // have finished.
1689 4393 : upload_queue.inprogress_tasks.is_empty()
1690 : }
1691 : UploadOp::Delete(_) => {
1692 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1693 341 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1694 : }
1695 :
1696 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1697 2472 : upload_queue.inprogress_tasks.is_empty()
1698 : }
1699 : };
1700 :
1701 : // If we cannot launch this task, don't look any further.
1702 : //
1703 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1704 : // them now, but we don't try to do that currently. For example, if the frontmost task
1705 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1706 : // could still start layer uploads that were scheduled later.
1707 8718 : if !can_run_now {
1708 4168 : break;
1709 4550 : }
1710 4550 :
1711 4550 : if let UploadOp::Shutdown = next_op {
1712 : // leave the op in the queue but do not start more tasks; it will be dropped when
1713 : // the stop is called.
1714 6 : upload_queue.shutdown_ready.close();
1715 6 : break;
1716 4544 : }
1717 4544 :
1718 4544 : // We can launch this task. Remove it from the queue first.
1719 4544 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1720 4544 :
1721 4544 : debug!("starting op: {}", next_op);
1722 :
1723 : // Update the counters
1724 4544 : match next_op {
1725 1512 : UploadOp::UploadLayer(_, _) => {
1726 1512 : upload_queue.num_inprogress_layer_uploads += 1;
1727 1512 : }
1728 1410 : UploadOp::UploadMetadata { .. } => {
1729 1410 : upload_queue.num_inprogress_metadata_uploads += 1;
1730 1410 : }
1731 300 : UploadOp::Delete(_) => {
1732 300 : upload_queue.num_inprogress_deletions += 1;
1733 300 : }
1734 1322 : UploadOp::Barrier(sender) => {
1735 1322 : sender.send_replace(());
1736 1322 : continue;
1737 : }
1738 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1739 : };
1740 :
1741 : // Assign unique ID to this task
1742 3222 : upload_queue.task_counter += 1;
1743 3222 : let upload_task_id = upload_queue.task_counter;
1744 3222 :
1745 3222 : // Add it to the in-progress map
1746 3222 : let task = Arc::new(UploadTask {
1747 3222 : task_id: upload_task_id,
1748 3222 : op: next_op,
1749 3222 : retries: AtomicU32::new(0),
1750 3222 : });
1751 3222 : upload_queue
1752 3222 : .inprogress_tasks
1753 3222 : .insert(task.task_id, Arc::clone(&task));
1754 3222 :
1755 3222 : // Spawn task to perform the task
1756 3222 : let self_rc = Arc::clone(self);
1757 3222 : let tenant_shard_id = self.tenant_shard_id;
1758 3222 : let timeline_id = self.timeline_id;
1759 3222 : task_mgr::spawn(
1760 3222 : &self.runtime,
1761 3222 : TaskKind::RemoteUploadTask,
1762 3222 : self.tenant_shard_id,
1763 3222 : Some(self.timeline_id),
1764 3222 : "remote upload",
1765 3119 : async move {
1766 41733 : self_rc.perform_upload_task(task).await;
1767 3047 : Ok(())
1768 3047 : }
1769 3222 : .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)),
1770 : );
1771 :
1772 : // Loop back to process next task
1773 : }
1774 7506 : }
1775 :
1776 : ///
1777 : /// Perform an upload task.
1778 : ///
1779 : /// The task is in the `inprogress_tasks` list. This function will try to
1780 : /// execute it, retrying forever. On successful completion, the task is
1781 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1782 : /// queue that were waiting by the completion are launched.
1783 : ///
1784 : /// The task can be shut down, however. That leads to stopping the whole
1785 : /// queue.
1786 : ///
1787 3119 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1788 3119 : let cancel = shutdown_token();
1789 : // Loop to retry until it completes.
1790 : loop {
1791 : // If we're requested to shut down, close up shop and exit.
1792 : //
1793 : // Note: We only check for the shutdown requests between retries, so
1794 : // if a shutdown request arrives while we're busy uploading, in the
1795 : // upload::upload:*() call below, we will wait not exit until it has
1796 : // finished. We probably could cancel the upload by simply dropping
1797 : // the Future, but we're not 100% sure if the remote storage library
1798 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1799 : // upload finishes or times out soon enough.
1800 3119 : if cancel.is_cancelled() {
1801 0 : info!("upload task cancelled by shutdown request");
1802 0 : self.stop();
1803 0 : return;
1804 3119 : }
1805 :
1806 3119 : let upload_result: anyhow::Result<()> = match &task.op {
1807 1419 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1808 1419 : let local_path = layer.local_path();
1809 1419 :
1810 1419 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1811 1419 : // the metadata in the upload should always match our current generation.
1812 1419 : assert_eq!(layer_metadata.generation, self.generation);
1813 :
1814 1419 : let remote_path = remote_layer_path(
1815 1419 : &self.tenant_shard_id.tenant_id,
1816 1419 : &self.timeline_id,
1817 1419 : layer_metadata.shard,
1818 1419 : &layer.layer_desc().layer_name(),
1819 1419 : layer_metadata.generation,
1820 1419 : );
1821 1419 :
1822 1419 : upload::upload_timeline_layer(
1823 1419 : &self.storage_impl,
1824 1419 : local_path,
1825 1419 : &remote_path,
1826 1419 : layer_metadata.file_size,
1827 1419 : &self.cancel,
1828 1419 : )
1829 1419 : .measure_remote_op(
1830 1419 : RemoteOpFileKind::Layer,
1831 1419 : RemoteOpKind::Upload,
1832 1419 : Arc::clone(&self.metrics),
1833 1419 : )
1834 36622 : .await
1835 : }
1836 1400 : UploadOp::UploadMetadata { ref uploaded } => {
1837 1400 : let res = upload::upload_index_part(
1838 1400 : &self.storage_impl,
1839 1400 : &self.tenant_shard_id,
1840 1400 : &self.timeline_id,
1841 1400 : self.generation,
1842 1400 : uploaded,
1843 1400 : &self.cancel,
1844 1400 : )
1845 1400 : .measure_remote_op(
1846 1400 : RemoteOpFileKind::Index,
1847 1400 : RemoteOpKind::Upload,
1848 1400 : Arc::clone(&self.metrics),
1849 1400 : )
1850 4857 : .await;
1851 1396 : if res.is_ok() {
1852 1396 : self.update_remote_physical_size_gauge(Some(uploaded));
1853 1396 : let mention_having_future_layers = if cfg!(feature = "testing") {
1854 1396 : uploaded
1855 1396 : .layer_metadata
1856 1396 : .keys()
1857 17617 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
1858 : } else {
1859 0 : false
1860 : };
1861 1396 : if mention_having_future_layers {
1862 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1863 13 : tracing::info!(
1864 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
1865 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
1866 : );
1867 1383 : }
1868 0 : }
1869 1396 : res
1870 : }
1871 300 : UploadOp::Delete(delete) => {
1872 300 : pausable_failpoint!("before-delete-layer-pausable");
1873 300 : self.deletion_queue_client
1874 300 : .push_layers(
1875 300 : self.tenant_shard_id,
1876 300 : self.timeline_id,
1877 300 : self.generation,
1878 300 : delete.layers.clone(),
1879 300 : )
1880 0 : .await
1881 300 : .map_err(|e| anyhow::anyhow!(e))
1882 : }
1883 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1884 : // unreachable. Barrier operations are handled synchronously in
1885 : // launch_queued_tasks
1886 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1887 0 : break;
1888 : }
1889 : };
1890 :
1891 0 : match upload_result {
1892 : Ok(()) => {
1893 3047 : break;
1894 : }
1895 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1896 0 : // loop around to do the proper stopping
1897 0 : continue;
1898 : }
1899 0 : Err(e) => {
1900 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1901 0 :
1902 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1903 0 : // or other external reasons. Such issues are relatively regular, so log them
1904 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1905 0 : //
1906 0 : // (See similar logic for downloads in `download::download_retry`)
1907 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1908 0 : info!(
1909 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1910 0 : task.op, retries, e
1911 : );
1912 : } else {
1913 0 : warn!(
1914 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1915 0 : task.op, retries, e
1916 : );
1917 : }
1918 :
1919 : // sleep until it's time to retry, or we're cancelled
1920 0 : exponential_backoff(
1921 0 : retries,
1922 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1923 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1924 0 : &cancel,
1925 0 : )
1926 0 : .await;
1927 : }
1928 : }
1929 : }
1930 :
1931 3047 : let retries = task.retries.load(Ordering::SeqCst);
1932 3047 : if retries > 0 {
1933 0 : info!(
1934 0 : "remote task {} completed successfully after {} retries",
1935 0 : task.op, retries
1936 : );
1937 : } else {
1938 3047 : debug!("remote task {} completed successfully", task.op);
1939 : }
1940 :
1941 : // The task has completed successfully. Remove it from the in-progress list.
1942 3047 : let lsn_update = {
1943 3047 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1944 3047 : let upload_queue = match upload_queue_guard.deref_mut() {
1945 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1946 0 : UploadQueue::Stopped(_stopped) => {
1947 0 : None
1948 : },
1949 3047 : UploadQueue::Initialized(qi) => { Some(qi) }
1950 : };
1951 :
1952 3047 : let upload_queue = match upload_queue {
1953 3047 : Some(upload_queue) => upload_queue,
1954 : None => {
1955 0 : info!("another concurrent task already stopped the queue");
1956 0 : return;
1957 : }
1958 : };
1959 :
1960 3047 : upload_queue.inprogress_tasks.remove(&task.task_id);
1961 :
1962 3047 : let lsn_update = match task.op {
1963 : UploadOp::UploadLayer(_, _) => {
1964 1351 : upload_queue.num_inprogress_layer_uploads -= 1;
1965 1351 : None
1966 : }
1967 1396 : UploadOp::UploadMetadata { ref uploaded } => {
1968 1396 : upload_queue.num_inprogress_metadata_uploads -= 1;
1969 1396 :
1970 1396 : // the task id is reused as a monotonicity check for storing the "clean"
1971 1396 : // IndexPart.
1972 1396 : let last_updater = upload_queue.clean.1;
1973 1396 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
1974 1396 : let monotone = is_later || last_updater.is_none();
1975 :
1976 1396 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
1977 :
1978 : // not taking ownership is wasteful
1979 1396 : upload_queue.clean.0.clone_from(uploaded);
1980 1396 : upload_queue.clean.1 = Some(task.task_id);
1981 1396 :
1982 1396 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
1983 1396 :
1984 1396 : if self.generation.is_none() {
1985 : // Legacy mode: skip validating generation
1986 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
1987 0 : None
1988 : } else {
1989 1396 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
1990 : }
1991 : }
1992 : UploadOp::Delete(_) => {
1993 300 : upload_queue.num_inprogress_deletions -= 1;
1994 300 : None
1995 : }
1996 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
1997 : };
1998 :
1999 : // Launch any queued tasks that were unblocked by this one.
2000 3047 : self.launch_queued_tasks(upload_queue);
2001 3047 : lsn_update
2002 : };
2003 :
2004 3047 : if let Some((lsn, slot)) = lsn_update {
2005 : // Updates to the remote_consistent_lsn we advertise to pageservers
2006 : // are all routed through the DeletionQueue, to enforce important
2007 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2008 1396 : self.deletion_queue_client
2009 1396 : .update_remote_consistent_lsn(
2010 1396 : self.tenant_shard_id,
2011 1396 : self.timeline_id,
2012 1396 : self.generation,
2013 1396 : lsn,
2014 1396 : slot,
2015 1396 : )
2016 0 : .await;
2017 1651 : }
2018 :
2019 3047 : self.metric_end(&task.op);
2020 3047 : }
2021 :
2022 6454 : fn metric_impl(
2023 6454 : &self,
2024 6454 : op: &UploadOp,
2025 6454 : ) -> Option<(
2026 6454 : RemoteOpFileKind,
2027 6454 : RemoteOpKind,
2028 6454 : RemoteTimelineClientMetricsCallTrackSize,
2029 6454 : )> {
2030 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2031 6454 : let res = match op {
2032 2865 : UploadOp::UploadLayer(_, m) => (
2033 2865 : RemoteOpFileKind::Layer,
2034 2865 : RemoteOpKind::Upload,
2035 2865 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2036 2865 : ),
2037 2824 : UploadOp::UploadMetadata { .. } => (
2038 2824 : RemoteOpFileKind::Index,
2039 2824 : RemoteOpKind::Upload,
2040 2824 : DontTrackSize {
2041 2824 : reason: "metadata uploads are tiny",
2042 2824 : },
2043 2824 : ),
2044 759 : UploadOp::Delete(_delete) => (
2045 759 : RemoteOpFileKind::Layer,
2046 759 : RemoteOpKind::Delete,
2047 759 : DontTrackSize {
2048 759 : reason: "should we track deletes? positive or negative sign?",
2049 759 : },
2050 759 : ),
2051 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2052 : // we do not account these
2053 6 : return None;
2054 : }
2055 : };
2056 6448 : Some(res)
2057 6454 : }
2058 :
2059 3401 : fn metric_begin(&self, op: &UploadOp) {
2060 3401 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2061 3401 : Some(x) => x,
2062 0 : None => return,
2063 : };
2064 3401 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2065 3401 : guard.will_decrement_manually(); // in metric_end(), see right below
2066 3401 : }
2067 :
2068 3053 : fn metric_end(&self, op: &UploadOp) {
2069 3053 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2070 3047 : Some(x) => x,
2071 6 : None => return,
2072 : };
2073 3047 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2074 3053 : }
2075 :
2076 : /// Close the upload queue for new operations and cancel queued operations.
2077 : ///
2078 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2079 : ///
2080 : /// In-progress operations will still be running after this function returns.
2081 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2082 : /// to wait for them to complete, after calling this function.
2083 14 : pub(crate) fn stop(&self) {
2084 14 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2085 14 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2086 14 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2087 14 : let mut guard = self.upload_queue.lock().unwrap();
2088 14 : self.stop_impl(&mut guard);
2089 14 : }
2090 :
2091 14 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2092 14 : match &mut **guard {
2093 : UploadQueue::Uninitialized => {
2094 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2095 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2096 : }
2097 : UploadQueue::Stopped(_) => {
2098 : // nothing to do
2099 6 : info!("another concurrent task already shut down the queue");
2100 : }
2101 8 : UploadQueue::Initialized(initialized) => {
2102 8 : info!("shutting down upload queue");
2103 :
2104 : // Replace the queue with the Stopped state, taking ownership of the old
2105 : // Initialized queue. We will do some checks on it, and then drop it.
2106 8 : let qi = {
2107 : // Here we preserve working version of the upload queue for possible use during deletions.
2108 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2109 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2110 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2111 8 : let upload_queue_for_deletion = UploadQueueInitialized {
2112 8 : task_counter: 0,
2113 8 : dirty: initialized.dirty.clone(),
2114 8 : clean: initialized.clean.clone(),
2115 8 : latest_files_changes_since_metadata_upload_scheduled: 0,
2116 8 : visible_remote_consistent_lsn: initialized
2117 8 : .visible_remote_consistent_lsn
2118 8 : .clone(),
2119 8 : num_inprogress_layer_uploads: 0,
2120 8 : num_inprogress_metadata_uploads: 0,
2121 8 : num_inprogress_deletions: 0,
2122 8 : inprogress_tasks: HashMap::default(),
2123 8 : queued_operations: VecDeque::default(),
2124 8 : #[cfg(feature = "testing")]
2125 8 : dangling_files: HashMap::default(),
2126 8 : shutting_down: false,
2127 8 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2128 8 : };
2129 8 :
2130 8 : let upload_queue = std::mem::replace(
2131 8 : &mut **guard,
2132 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2133 8 : UploadQueueStoppedDeletable {
2134 8 : upload_queue_for_deletion,
2135 8 : deleted_at: SetDeletedFlagProgress::NotRunning,
2136 8 : },
2137 8 : )),
2138 8 : );
2139 8 : if let UploadQueue::Initialized(qi) = upload_queue {
2140 8 : qi
2141 : } else {
2142 0 : unreachable!("we checked in the match above that it is Initialized");
2143 : }
2144 : };
2145 :
2146 : // consistency check
2147 8 : assert_eq!(
2148 8 : qi.num_inprogress_layer_uploads
2149 8 : + qi.num_inprogress_metadata_uploads
2150 8 : + qi.num_inprogress_deletions,
2151 8 : qi.inprogress_tasks.len()
2152 8 : );
2153 :
2154 : // We don't need to do anything here for in-progress tasks. They will finish
2155 : // on their own, decrement the unfinished-task counter themselves, and observe
2156 : // that the queue is Stopped.
2157 8 : drop(qi.inprogress_tasks);
2158 :
2159 : // Tear down queued ops
2160 8 : for op in qi.queued_operations.into_iter() {
2161 6 : self.metric_end(&op);
2162 6 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2163 6 : // which is exactly what we want to happen.
2164 6 : drop(op);
2165 6 : }
2166 : }
2167 : }
2168 14 : }
2169 :
2170 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2171 : /// externally to RemoteTimelineClient.
2172 0 : pub(crate) fn initialized_upload_queue(
2173 0 : &self,
2174 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2175 0 : let mut inner = self.upload_queue.lock().unwrap();
2176 0 : inner.initialized_mut()?;
2177 0 : Ok(UploadQueueAccessor { inner })
2178 0 : }
2179 : }
2180 :
2181 : pub(crate) struct UploadQueueAccessor<'a> {
2182 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2183 : }
2184 :
2185 : impl UploadQueueAccessor<'_> {
2186 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2187 0 : match &*self.inner {
2188 0 : UploadQueue::Initialized(x) => &x.clean.0,
2189 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2190 0 : unreachable!("checked before constructing")
2191 : }
2192 : }
2193 0 : }
2194 : }
2195 :
2196 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2197 0 : let path = format!("tenants/{tenant_shard_id}");
2198 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2199 0 : }
2200 :
2201 216 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2202 216 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2203 216 : RemotePath::from_string(&path).expect("Failed to construct path")
2204 216 : }
2205 :
2206 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2207 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2208 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2209 0 : }
2210 :
2211 30 : pub fn remote_timeline_path(
2212 30 : tenant_shard_id: &TenantShardId,
2213 30 : timeline_id: &TimelineId,
2214 30 : ) -> RemotePath {
2215 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2216 30 : }
2217 :
2218 : /// Obtains the path of the given Layer in the remote
2219 : ///
2220 : /// Note that the shard component of a remote layer path is _not_ always the same
2221 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2222 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2223 1437 : pub fn remote_layer_path(
2224 1437 : tenant_id: &TenantId,
2225 1437 : timeline_id: &TimelineId,
2226 1437 : shard: ShardIndex,
2227 1437 : layer_file_name: &LayerName,
2228 1437 : generation: Generation,
2229 1437 : ) -> RemotePath {
2230 1437 : // Generation-aware key format
2231 1437 : let path = format!(
2232 1437 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2233 1437 : shard.get_suffix(),
2234 1437 : layer_file_name,
2235 1437 : generation.get_suffix()
2236 1437 : );
2237 1437 :
2238 1437 : RemotePath::from_string(&path).expect("Failed to construct path")
2239 1437 : }
2240 :
2241 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2242 4 : RemotePath::from_string(&format!(
2243 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2244 4 : ))
2245 4 : .expect("Failed to construct path")
2246 4 : }
2247 :
2248 2 : pub fn remote_initdb_preserved_archive_path(
2249 2 : tenant_id: &TenantId,
2250 2 : timeline_id: &TimelineId,
2251 2 : ) -> RemotePath {
2252 2 : RemotePath::from_string(&format!(
2253 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2254 2 : ))
2255 2 : .expect("Failed to construct path")
2256 2 : }
2257 :
2258 1451 : pub fn remote_index_path(
2259 1451 : tenant_shard_id: &TenantShardId,
2260 1451 : timeline_id: &TimelineId,
2261 1451 : generation: Generation,
2262 1451 : ) -> RemotePath {
2263 1451 : RemotePath::from_string(&format!(
2264 1451 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2265 1451 : IndexPart::FILE_NAME,
2266 1451 : generation.get_suffix()
2267 1451 : ))
2268 1451 : .expect("Failed to construct path")
2269 1451 : }
2270 :
2271 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2272 0 : RemotePath::from_string(&format!(
2273 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2274 0 : ))
2275 0 : .expect("Failed to construct path")
2276 0 : }
2277 :
2278 : /// Given the key of an index, parse out the generation part of the name
2279 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2280 18 : let file_name = match path.get_path().file_name() {
2281 18 : Some(f) => f,
2282 : None => {
2283 : // Unexpected: we should be seeing index_part.json paths only
2284 0 : tracing::warn!("Malformed index key {}", path);
2285 0 : return None;
2286 : }
2287 : };
2288 :
2289 18 : match file_name.split_once('-') {
2290 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2291 6 : None => None,
2292 : }
2293 18 : }
2294 :
2295 : #[cfg(test)]
2296 : mod tests {
2297 : use super::*;
2298 : use crate::{
2299 : context::RequestContext,
2300 : tenant::{
2301 : harness::{TenantHarness, TIMELINE_ID},
2302 : storage_layer::layer::local_layer_path,
2303 : Tenant, Timeline,
2304 : },
2305 : DEFAULT_PG_VERSION,
2306 : };
2307 :
2308 : use std::collections::HashSet;
2309 :
2310 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2311 8 : format!("contents for {name}").into()
2312 8 : }
2313 :
2314 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2315 2 : let metadata = TimelineMetadata::new(
2316 2 : disk_consistent_lsn,
2317 2 : None,
2318 2 : None,
2319 2 : Lsn(0),
2320 2 : Lsn(0),
2321 2 : Lsn(0),
2322 2 : // Any version will do
2323 2 : // but it should be consistent with the one in the tests
2324 2 : crate::DEFAULT_PG_VERSION,
2325 2 : );
2326 2 :
2327 2 : // go through serialize + deserialize to fix the header, including checksum
2328 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2329 2 : }
2330 :
2331 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2332 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2333 2 : avec.sort();
2334 2 :
2335 2 : let mut bvec = b.to_vec();
2336 2 : bvec.sort_unstable();
2337 2 :
2338 2 : assert_eq!(avec, bvec);
2339 2 : }
2340 :
2341 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2342 4 : let mut expected: Vec<String> = expected
2343 4 : .iter()
2344 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2345 4 : .collect();
2346 4 : expected.sort();
2347 4 :
2348 4 : let mut found: Vec<String> = Vec::new();
2349 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2350 16 : let entry_name = entry.file_name();
2351 16 : let fname = entry_name.to_str().unwrap();
2352 16 : found.push(String::from(fname));
2353 16 : }
2354 4 : found.sort();
2355 4 :
2356 4 : assert_eq!(found, expected);
2357 4 : }
2358 :
2359 : struct TestSetup {
2360 : harness: TenantHarness,
2361 : tenant: Arc<Tenant>,
2362 : timeline: Arc<Timeline>,
2363 : tenant_ctx: RequestContext,
2364 : }
2365 :
2366 : impl TestSetup {
2367 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2368 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2369 8 : let harness = TenantHarness::create(test_name).await?;
2370 32 : let (tenant, ctx) = harness.load().await;
2371 :
2372 8 : let timeline = tenant
2373 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2374 16 : .await?;
2375 :
2376 8 : Ok(Self {
2377 8 : harness,
2378 8 : tenant,
2379 8 : timeline,
2380 8 : tenant_ctx: ctx,
2381 8 : })
2382 8 : }
2383 :
2384 : /// Construct a RemoteTimelineClient in an arbitrary generation
2385 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2386 10 : Arc::new(RemoteTimelineClient {
2387 10 : conf: self.harness.conf,
2388 10 : runtime: tokio::runtime::Handle::current(),
2389 10 : tenant_shard_id: self.harness.tenant_shard_id,
2390 10 : timeline_id: TIMELINE_ID,
2391 10 : generation,
2392 10 : storage_impl: self.harness.remote_storage.clone(),
2393 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2394 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2395 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2396 10 : &self.harness.tenant_shard_id,
2397 10 : &TIMELINE_ID,
2398 10 : )),
2399 10 : cancel: CancellationToken::new(),
2400 10 : })
2401 10 : }
2402 :
2403 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2404 : /// and timeline_id are present.
2405 6 : fn span(&self) -> tracing::Span {
2406 6 : tracing::info_span!(
2407 : "test",
2408 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2409 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2410 : timeline_id = %TIMELINE_ID
2411 : )
2412 6 : }
2413 : }
2414 :
2415 : // Test scheduling
2416 : #[tokio::test]
2417 2 : async fn upload_scheduling() {
2418 2 : // Test outline:
2419 2 : //
2420 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2421 2 : // Schedule upload of index. Check that it is queued
2422 2 : // let the layer file uploads finish. Check that the index-upload is now started
2423 2 : // let the index-upload finish.
2424 2 : //
2425 2 : // Download back the index.json. Check that the list of files is correct
2426 2 : //
2427 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2428 2 : // let upload finish. Check that deletion is now started
2429 2 : // Schedule another deletion. Check that it's launched immediately.
2430 2 : // Schedule index upload. Check that it's queued
2431 2 :
2432 12 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2433 2 : let span = test_setup.span();
2434 2 : let _guard = span.enter();
2435 2 :
2436 2 : let TestSetup {
2437 2 : harness,
2438 2 : tenant: _tenant,
2439 2 : timeline,
2440 2 : tenant_ctx: _tenant_ctx,
2441 2 : } = test_setup;
2442 2 :
2443 2 : let client = &timeline.remote_client;
2444 2 :
2445 2 : // Download back the index.json, and check that the list of files is correct
2446 2 : let initial_index_part = match client
2447 2 : .download_index_file(&CancellationToken::new())
2448 6 : .await
2449 2 : .unwrap()
2450 2 : {
2451 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2452 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2453 2 : };
2454 2 : let initial_layers = initial_index_part
2455 2 : .layer_metadata
2456 2 : .keys()
2457 2 : .map(|f| f.to_owned())
2458 2 : .collect::<HashSet<LayerName>>();
2459 2 : let initial_layer = {
2460 2 : assert!(initial_layers.len() == 1);
2461 2 : initial_layers.into_iter().next().unwrap()
2462 2 : };
2463 2 :
2464 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2465 2 :
2466 2 : println!("workdir: {}", harness.conf.workdir);
2467 2 :
2468 2 : let remote_timeline_dir = harness
2469 2 : .remote_fs_dir
2470 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2471 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2472 2 :
2473 2 : let generation = harness.generation;
2474 2 : let shard = harness.shard;
2475 2 :
2476 2 : // Create a couple of dummy files, schedule upload for them
2477 2 :
2478 2 : let layers = [
2479 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2480 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2481 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2482 2 : ]
2483 2 : .into_iter()
2484 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2485 6 :
2486 6 : let local_path = local_layer_path(
2487 6 : harness.conf,
2488 6 : &timeline.tenant_shard_id,
2489 6 : &timeline.timeline_id,
2490 6 : &name,
2491 6 : &generation,
2492 6 : );
2493 6 : std::fs::write(&local_path, &contents).unwrap();
2494 6 :
2495 6 : Layer::for_resident(
2496 6 : harness.conf,
2497 6 : &timeline,
2498 6 : local_path,
2499 6 : name,
2500 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2501 6 : )
2502 6 : }).collect::<Vec<_>>();
2503 2 :
2504 2 : client
2505 2 : .schedule_layer_file_upload(layers[0].clone())
2506 2 : .unwrap();
2507 2 : client
2508 2 : .schedule_layer_file_upload(layers[1].clone())
2509 2 : .unwrap();
2510 2 :
2511 2 : // Check that they are started immediately, not queued
2512 2 : //
2513 2 : // this works because we running within block_on, so any futures are now queued up until
2514 2 : // our next await point.
2515 2 : {
2516 2 : let mut guard = client.upload_queue.lock().unwrap();
2517 2 : let upload_queue = guard.initialized_mut().unwrap();
2518 2 : assert!(upload_queue.queued_operations.is_empty());
2519 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2520 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2521 2 :
2522 2 : // also check that `latest_file_changes` was updated
2523 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2524 2 : }
2525 2 :
2526 2 : // Schedule upload of index. Check that it is queued
2527 2 : let metadata = dummy_metadata(Lsn(0x20));
2528 2 : client
2529 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2530 2 : .unwrap();
2531 2 : {
2532 2 : let mut guard = client.upload_queue.lock().unwrap();
2533 2 : let upload_queue = guard.initialized_mut().unwrap();
2534 2 : assert!(upload_queue.queued_operations.len() == 1);
2535 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2536 2 : }
2537 2 :
2538 2 : // Wait for the uploads to finish
2539 2 : client.wait_completion().await.unwrap();
2540 2 : {
2541 2 : let mut guard = client.upload_queue.lock().unwrap();
2542 2 : let upload_queue = guard.initialized_mut().unwrap();
2543 2 :
2544 2 : assert!(upload_queue.queued_operations.is_empty());
2545 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2546 2 : }
2547 2 :
2548 2 : // Download back the index.json, and check that the list of files is correct
2549 2 : let index_part = match client
2550 2 : .download_index_file(&CancellationToken::new())
2551 6 : .await
2552 2 : .unwrap()
2553 2 : {
2554 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2555 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2556 2 : };
2557 2 :
2558 2 : assert_file_list(
2559 2 : &index_part
2560 2 : .layer_metadata
2561 2 : .keys()
2562 6 : .map(|f| f.to_owned())
2563 2 : .collect(),
2564 2 : &[
2565 2 : &initial_layer.to_string(),
2566 2 : &layers[0].layer_desc().layer_name().to_string(),
2567 2 : &layers[1].layer_desc().layer_name().to_string(),
2568 2 : ],
2569 2 : );
2570 2 : assert_eq!(index_part.metadata, metadata);
2571 2 :
2572 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2573 2 : client
2574 2 : .schedule_layer_file_upload(layers[2].clone())
2575 2 : .unwrap();
2576 2 :
2577 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2578 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2579 2 : // spawn_blocking started by the drop.
2580 2 : client
2581 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2582 2 : .unwrap();
2583 2 : {
2584 2 : let mut guard = client.upload_queue.lock().unwrap();
2585 2 : let upload_queue = guard.initialized_mut().unwrap();
2586 2 :
2587 2 : // Deletion schedules upload of the index file, and the file deletion itself
2588 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2589 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2590 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2591 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2592 2 : assert_eq!(
2593 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2594 2 : 0
2595 2 : );
2596 2 : }
2597 2 : assert_remote_files(
2598 2 : &[
2599 2 : &initial_layer.to_string(),
2600 2 : &layers[0].layer_desc().layer_name().to_string(),
2601 2 : &layers[1].layer_desc().layer_name().to_string(),
2602 2 : "index_part.json",
2603 2 : ],
2604 2 : &remote_timeline_dir,
2605 2 : generation,
2606 2 : );
2607 2 :
2608 2 : // Finish them
2609 2 : client.wait_completion().await.unwrap();
2610 2 : harness.deletion_queue.pump().await;
2611 2 :
2612 2 : assert_remote_files(
2613 2 : &[
2614 2 : &initial_layer.to_string(),
2615 2 : &layers[1].layer_desc().layer_name().to_string(),
2616 2 : &layers[2].layer_desc().layer_name().to_string(),
2617 2 : "index_part.json",
2618 2 : ],
2619 2 : &remote_timeline_dir,
2620 2 : generation,
2621 2 : );
2622 2 : }
2623 :
2624 : #[tokio::test]
2625 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2626 2 : // Setup
2627 2 :
2628 2 : let TestSetup {
2629 2 : harness,
2630 2 : tenant: _tenant,
2631 2 : timeline,
2632 2 : ..
2633 12 : } = TestSetup::new("metrics").await.unwrap();
2634 2 : let client = &timeline.remote_client;
2635 2 :
2636 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2637 2 : let local_path = local_layer_path(
2638 2 : harness.conf,
2639 2 : &timeline.tenant_shard_id,
2640 2 : &timeline.timeline_id,
2641 2 : &layer_file_name_1,
2642 2 : &harness.generation,
2643 2 : );
2644 2 : let content_1 = dummy_contents("foo");
2645 2 : std::fs::write(&local_path, &content_1).unwrap();
2646 2 :
2647 2 : let layer_file_1 = Layer::for_resident(
2648 2 : harness.conf,
2649 2 : &timeline,
2650 2 : local_path,
2651 2 : layer_file_name_1.clone(),
2652 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2653 2 : );
2654 2 :
2655 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2656 2 : struct BytesStartedFinished {
2657 2 : started: Option<usize>,
2658 2 : finished: Option<usize>,
2659 2 : }
2660 2 : impl std::ops::Add for BytesStartedFinished {
2661 2 : type Output = Self;
2662 4 : fn add(self, rhs: Self) -> Self::Output {
2663 4 : Self {
2664 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2665 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2666 4 : }
2667 4 : }
2668 2 : }
2669 6 : let get_bytes_started_stopped = || {
2670 6 : let started = client
2671 6 : .metrics
2672 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2673 6 : .map(|v| v.try_into().unwrap());
2674 6 : let stopped = client
2675 6 : .metrics
2676 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2677 6 : .map(|v| v.try_into().unwrap());
2678 6 : BytesStartedFinished {
2679 6 : started,
2680 6 : finished: stopped,
2681 6 : }
2682 6 : };
2683 2 :
2684 2 : // Test
2685 2 : tracing::info!("now doing actual test");
2686 2 :
2687 2 : let actual_a = get_bytes_started_stopped();
2688 2 :
2689 2 : client
2690 2 : .schedule_layer_file_upload(layer_file_1.clone())
2691 2 : .unwrap();
2692 2 :
2693 2 : let actual_b = get_bytes_started_stopped();
2694 2 :
2695 2 : client.wait_completion().await.unwrap();
2696 2 :
2697 2 : let actual_c = get_bytes_started_stopped();
2698 2 :
2699 2 : // Validate
2700 2 :
2701 2 : let expected_b = actual_a
2702 2 : + BytesStartedFinished {
2703 2 : started: Some(content_1.len()),
2704 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2705 2 : finished: Some(0),
2706 2 : };
2707 2 : assert_eq!(actual_b, expected_b);
2708 2 :
2709 2 : let expected_c = actual_a
2710 2 : + BytesStartedFinished {
2711 2 : started: Some(content_1.len()),
2712 2 : finished: Some(content_1.len()),
2713 2 : };
2714 2 : assert_eq!(actual_c, expected_c);
2715 2 : }
2716 :
2717 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2718 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2719 12 : let example_index_part = IndexPart::example();
2720 12 :
2721 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2722 12 :
2723 12 : let index_path = test_state.harness.remote_fs_dir.join(
2724 12 : remote_index_path(
2725 12 : &test_state.harness.tenant_shard_id,
2726 12 : &TIMELINE_ID,
2727 12 : generation,
2728 12 : )
2729 12 : .get_path(),
2730 12 : );
2731 12 :
2732 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2733 12 : .expect("creating test dir should work");
2734 12 :
2735 12 : eprintln!("Writing {index_path}");
2736 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2737 12 : example_index_part
2738 12 : }
2739 :
2740 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2741 : /// index, the IndexPart returned is equal to `expected`
2742 10 : async fn assert_got_index_part(
2743 10 : test_state: &TestSetup,
2744 10 : get_generation: Generation,
2745 10 : expected: &IndexPart,
2746 10 : ) {
2747 10 : let client = test_state.build_client(get_generation);
2748 :
2749 10 : let download_r = client
2750 10 : .download_index_file(&CancellationToken::new())
2751 50 : .await
2752 10 : .expect("download should always succeed");
2753 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2754 10 : match download_r {
2755 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2756 10 : assert_eq!(&index_part, expected);
2757 : }
2758 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2759 : }
2760 10 : }
2761 :
2762 : #[tokio::test]
2763 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2764 12 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2765 2 : let span = test_state.span();
2766 2 : let _guard = span.enter();
2767 2 :
2768 2 : // Simple case: we are in generation N, load the index from generation N - 1
2769 2 : let generation_n = 5;
2770 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2771 2 :
2772 8 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2773 2 :
2774 2 : Ok(())
2775 2 : }
2776 :
2777 : #[tokio::test]
2778 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2779 2 : let test_state = TestSetup::new("index_part_download_ordering")
2780 12 : .await
2781 2 : .unwrap();
2782 2 :
2783 2 : let span = test_state.span();
2784 2 : let _guard = span.enter();
2785 2 :
2786 2 : // A generation-less IndexPart exists in the bucket, we should find it
2787 2 : let generation_n = 5;
2788 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2789 12 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2790 2 :
2791 2 : // If a more recent-than-none generation exists, we should prefer to load that
2792 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2793 12 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2794 2 :
2795 2 : // If a more-recent-than-me generation exists, we should ignore it.
2796 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2797 12 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2798 2 :
2799 2 : // If a directly previous generation exists, _and_ an index exists in my own
2800 2 : // generation, I should prefer my own generation.
2801 2 : let _injected_prev =
2802 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2803 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2804 2 : assert_got_index_part(
2805 2 : &test_state,
2806 2 : Generation::new(generation_n),
2807 2 : &injected_current,
2808 2 : )
2809 6 : .await;
2810 2 :
2811 2 : Ok(())
2812 2 : }
2813 : }
|
