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 mod manifest;
184 : pub(crate) mod upload;
185 :
186 : use anyhow::Context;
187 : use camino::Utf8Path;
188 : use chrono::{NaiveDateTime, Utc};
189 :
190 : pub(crate) use download::download_initdb_tar_zst;
191 : use pageserver_api::models::TimelineArchivalState;
192 : use pageserver_api::shard::{ShardIndex, TenantShardId};
193 : use regex::Regex;
194 : use scopeguard::ScopeGuard;
195 : use tokio_util::sync::CancellationToken;
196 : use utils::backoff::{
197 : self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
198 : };
199 : use utils::pausable_failpoint;
200 :
201 : use std::collections::{HashMap, VecDeque};
202 : use std::sync::atomic::{AtomicU32, Ordering};
203 : use std::sync::{Arc, Mutex, OnceLock};
204 : use std::time::Duration;
205 :
206 : use remote_storage::{
207 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
208 : };
209 : use std::ops::DerefMut;
210 : use tracing::{debug, error, info, instrument, warn};
211 : use tracing::{info_span, Instrument};
212 : use utils::lsn::Lsn;
213 :
214 : use crate::context::RequestContext;
215 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
216 : use crate::metrics::{
217 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
218 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
219 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
220 : };
221 : use crate::task_mgr::shutdown_token;
222 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
223 : use crate::tenant::remote_timeline_client::download::download_retry;
224 : use crate::tenant::storage_layer::AsLayerDesc;
225 : use crate::tenant::upload_queue::{Delete, UploadQueueStoppedDeletable};
226 : use crate::tenant::TIMELINES_SEGMENT_NAME;
227 : use crate::{
228 : config::PageServerConf,
229 : task_mgr,
230 : task_mgr::TaskKind,
231 : task_mgr::BACKGROUND_RUNTIME,
232 : tenant::metadata::TimelineMetadata,
233 : tenant::upload_queue::{
234 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
235 : },
236 : TENANT_HEATMAP_BASENAME,
237 : };
238 :
239 : use utils::id::{TenantId, TimelineId};
240 :
241 : use self::index::IndexPart;
242 :
243 : use super::metadata::MetadataUpdate;
244 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
245 : use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
246 : use super::Generation;
247 :
248 : pub(crate) use download::{
249 : download_index_part, download_tenant_manifest, is_temp_download_file,
250 : list_remote_tenant_shards, list_remote_timelines,
251 : };
252 : pub(crate) use index::LayerFileMetadata;
253 : pub(crate) use upload::upload_initdb_dir;
254 :
255 : // Occasional network issues and such can cause remote operations to fail, and
256 : // that's expected. If a download fails, we log it at info-level, and retry.
257 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
258 : // level instead, as repeated failures can mean a more serious problem. If it
259 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
260 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
261 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
262 :
263 : // Similarly log failed uploads and deletions at WARN level, after this many
264 : // retries. Uploads and deletions are retried forever, though.
265 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
266 :
267 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
268 :
269 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
270 :
271 : /// Default buffer size when interfacing with [`tokio::fs::File`].
272 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
273 :
274 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
275 : /// which we warn and skip.
276 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
277 :
278 : pub enum MaybeDeletedIndexPart {
279 : IndexPart(IndexPart),
280 : Deleted(IndexPart),
281 : }
282 :
283 0 : #[derive(Debug, thiserror::Error)]
284 : pub enum PersistIndexPartWithDeletedFlagError {
285 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
286 : AlreadyInProgress(NaiveDateTime),
287 : #[error("the deleted_flag was already set, value is {0:?}")]
288 : AlreadyDeleted(NaiveDateTime),
289 : #[error(transparent)]
290 : Other(#[from] anyhow::Error),
291 : }
292 :
293 0 : #[derive(Debug, thiserror::Error)]
294 : pub enum WaitCompletionError {
295 : #[error(transparent)]
296 : NotInitialized(NotInitialized),
297 : #[error("wait_completion aborted because upload queue was stopped")]
298 : UploadQueueShutDownOrStopped,
299 : }
300 :
301 0 : #[derive(Debug, thiserror::Error)]
302 : #[error("Upload queue either in unexpected state or hasn't downloaded manifest yet")]
303 : pub struct UploadQueueNotReadyError;
304 :
305 : /// A client for accessing a timeline's data in remote storage.
306 : ///
307 : /// This takes care of managing the number of connections, and balancing them
308 : /// across tenants. This also handles retries of failed uploads.
309 : ///
310 : /// Upload and delete requests are ordered so that before a deletion is
311 : /// performed, we wait for all preceding uploads to finish. This ensures sure
312 : /// that if you perform a compaction operation that reshuffles data in layer
313 : /// files, we don't have a transient state where the old files have already been
314 : /// deleted, but new files have not yet been uploaded.
315 : ///
316 : /// Similarly, this enforces an order between index-file uploads, and layer
317 : /// uploads. Before an index-file upload is performed, all preceding layer
318 : /// uploads must be finished.
319 : ///
320 : /// This also maintains a list of remote files, and automatically includes that
321 : /// in the index part file, whenever timeline metadata is uploaded.
322 : ///
323 : /// Downloads are not queued, they are performed immediately.
324 : pub struct RemoteTimelineClient {
325 : conf: &'static PageServerConf,
326 :
327 : runtime: tokio::runtime::Handle,
328 :
329 : tenant_shard_id: TenantShardId,
330 : timeline_id: TimelineId,
331 : generation: Generation,
332 :
333 : upload_queue: Mutex<UploadQueue>,
334 :
335 : pub(crate) metrics: Arc<RemoteTimelineClientMetrics>,
336 :
337 : storage_impl: GenericRemoteStorage,
338 :
339 : deletion_queue_client: DeletionQueueClient,
340 :
341 : cancel: CancellationToken,
342 : }
343 :
344 : impl RemoteTimelineClient {
345 : ///
346 : /// Create a remote storage client for given timeline
347 : ///
348 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
349 : /// by calling init_upload_queue.
350 : ///
351 414 : pub fn new(
352 414 : remote_storage: GenericRemoteStorage,
353 414 : deletion_queue_client: DeletionQueueClient,
354 414 : conf: &'static PageServerConf,
355 414 : tenant_shard_id: TenantShardId,
356 414 : timeline_id: TimelineId,
357 414 : generation: Generation,
358 414 : ) -> RemoteTimelineClient {
359 414 : RemoteTimelineClient {
360 414 : conf,
361 414 : runtime: if cfg!(test) {
362 : // remote_timeline_client.rs tests rely on current-thread runtime
363 414 : tokio::runtime::Handle::current()
364 : } else {
365 0 : BACKGROUND_RUNTIME.handle().clone()
366 : },
367 414 : tenant_shard_id,
368 414 : timeline_id,
369 414 : generation,
370 414 : storage_impl: remote_storage,
371 414 : deletion_queue_client,
372 414 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
373 414 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
374 414 : &tenant_shard_id,
375 414 : &timeline_id,
376 414 : )),
377 414 : cancel: CancellationToken::new(),
378 414 : }
379 414 : }
380 :
381 : /// Initialize the upload queue for a remote storage that already received
382 : /// an index file upload, i.e., it's not empty.
383 : /// The given `index_part` must be the one on the remote.
384 6 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
385 6 : let mut upload_queue = self.upload_queue.lock().unwrap();
386 6 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
387 6 : self.update_remote_physical_size_gauge(Some(index_part));
388 6 : info!(
389 0 : "initialized upload queue from remote index with {} layer files",
390 0 : index_part.layer_metadata.len()
391 : );
392 6 : Ok(())
393 6 : }
394 :
395 : /// Initialize the upload queue for the case where the remote storage is empty,
396 : /// i.e., it doesn't have an `IndexPart`.
397 408 : pub fn init_upload_queue_for_empty_remote(
398 408 : &self,
399 408 : local_metadata: &TimelineMetadata,
400 408 : ) -> anyhow::Result<()> {
401 408 : let mut upload_queue = self.upload_queue.lock().unwrap();
402 408 : upload_queue.initialize_empty_remote(local_metadata)?;
403 408 : self.update_remote_physical_size_gauge(None);
404 408 : info!("initialized upload queue as empty");
405 408 : Ok(())
406 408 : }
407 :
408 : /// Initialize the queue in stopped state. Used in startup path
409 : /// to continue deletion operation interrupted by pageserver crash or restart.
410 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
411 0 : &self,
412 0 : index_part: &IndexPart,
413 0 : ) -> anyhow::Result<()> {
414 : // FIXME: consider newtype for DeletedIndexPart.
415 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
416 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
417 0 : ))?;
418 :
419 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
420 0 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
421 0 : self.update_remote_physical_size_gauge(Some(index_part));
422 0 : self.stop_impl(&mut upload_queue);
423 0 :
424 0 : upload_queue
425 0 : .stopped_mut()
426 0 : .expect("stopped above")
427 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
428 0 :
429 0 : Ok(())
430 0 : }
431 :
432 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
433 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
434 0 : match &mut *self.upload_queue.lock().unwrap() {
435 0 : UploadQueue::Uninitialized => None,
436 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
437 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
438 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
439 0 : .upload_queue_for_deletion
440 0 : .get_last_remote_consistent_lsn_projected(),
441 : }
442 0 : }
443 :
444 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
445 0 : match &mut *self.upload_queue.lock().unwrap() {
446 0 : UploadQueue::Uninitialized => None,
447 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
448 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
449 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
450 0 : q.upload_queue_for_deletion
451 0 : .get_last_remote_consistent_lsn_visible(),
452 0 : ),
453 : }
454 0 : }
455 :
456 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
457 : /// client is currently initialized.
458 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
459 0 : self.upload_queue
460 0 : .lock()
461 0 : .unwrap()
462 0 : .initialized_mut()
463 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
464 0 : .unwrap_or(false)
465 0 : }
466 :
467 : /// Returns whether the timeline is archived.
468 : /// Return None if the remote index_part hasn't been downloaded yet.
469 0 : pub(crate) fn is_archived(&self) -> Option<bool> {
470 0 : self.upload_queue
471 0 : .lock()
472 0 : .unwrap()
473 0 : .initialized_mut()
474 0 : .map(|q| q.clean.0.archived_at.is_some())
475 0 : .ok()
476 0 : }
477 :
478 : /// Returns `Ok(Some(timestamp))` if the timeline has been archived, `Ok(None)` if the timeline hasn't been archived.
479 : ///
480 : /// Return Err(_) if the remote index_part hasn't been downloaded yet, or the timeline hasn't been stopped yet.
481 0 : pub(crate) fn archived_at_stopped_queue(
482 0 : &self,
483 0 : ) -> Result<Option<NaiveDateTime>, UploadQueueNotReadyError> {
484 0 : self.upload_queue
485 0 : .lock()
486 0 : .unwrap()
487 0 : .stopped_mut()
488 0 : .map(|q| q.upload_queue_for_deletion.clean.0.archived_at)
489 0 : .map_err(|_| UploadQueueNotReadyError)
490 0 : }
491 :
492 1812 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
493 1812 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
494 1404 : current_remote_index_part
495 1404 : .layer_metadata
496 1404 : .values()
497 17652 : .map(|ilmd| ilmd.file_size)
498 1404 : .sum()
499 : } else {
500 408 : 0
501 : };
502 1812 : self.metrics.remote_physical_size_gauge.set(size);
503 1812 : }
504 :
505 2 : pub fn get_remote_physical_size(&self) -> u64 {
506 2 : self.metrics.remote_physical_size_gauge.get()
507 2 : }
508 :
509 : //
510 : // Download operations.
511 : //
512 : // These don't use the per-timeline queue. They do use the global semaphore in
513 : // S3Bucket, to limit the total number of concurrent operations, though.
514 : //
515 :
516 : /// Download index file
517 20 : pub async fn download_index_file(
518 20 : &self,
519 20 : cancel: &CancellationToken,
520 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
521 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
522 20 : &RemoteOpFileKind::Index,
523 20 : &RemoteOpKind::Download,
524 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
525 20 : reason: "no need for a downloads gauge",
526 20 : },
527 20 : );
528 :
529 20 : let (index_part, index_generation, index_last_modified) = download::download_index_part(
530 20 : &self.storage_impl,
531 20 : &self.tenant_shard_id,
532 20 : &self.timeline_id,
533 20 : self.generation,
534 20 : cancel,
535 20 : )
536 20 : .measure_remote_op(
537 20 : RemoteOpFileKind::Index,
538 20 : RemoteOpKind::Download,
539 20 : Arc::clone(&self.metrics),
540 20 : )
541 98 : .await?;
542 :
543 : // Defense in depth: monotonicity of generation numbers is an important correctness guarantee, so when we see a very
544 : // old index, we do extra checks in case this is the result of backward time-travel of the generation number (e.g.
545 : // in case of a bug in the service that issues generation numbers). Indices are allowed to be old, but we expect that
546 : // when we load an old index we are loading the _latest_ index: if we are asked to load an old index and there is
547 : // also a newer index available, that is surprising.
548 : const INDEX_AGE_CHECKS_THRESHOLD: Duration = Duration::from_secs(14 * 24 * 3600);
549 20 : let index_age = index_last_modified.elapsed().unwrap_or_else(|e| {
550 0 : if e.duration() > Duration::from_secs(5) {
551 : // We only warn if the S3 clock and our local clock are >5s out: because this is a low resolution
552 : // timestamp, it is common to be out by at least 1 second.
553 0 : tracing::warn!("Index has modification time in the future: {e}");
554 0 : }
555 0 : Duration::ZERO
556 20 : });
557 20 : if index_age > INDEX_AGE_CHECKS_THRESHOLD {
558 0 : tracing::info!(
559 : ?index_generation,
560 0 : age = index_age.as_secs_f64(),
561 0 : "Loaded an old index, checking for other indices..."
562 : );
563 :
564 : // Find the highest-generation index
565 0 : let (_latest_index_part, latest_index_generation, latest_index_mtime) =
566 0 : download::download_index_part(
567 0 : &self.storage_impl,
568 0 : &self.tenant_shard_id,
569 0 : &self.timeline_id,
570 0 : Generation::MAX,
571 0 : cancel,
572 0 : )
573 0 : .await?;
574 :
575 0 : if latest_index_generation > index_generation {
576 : // Unexpected! Why are we loading such an old index if a more recent one exists?
577 0 : tracing::warn!(
578 : ?index_generation,
579 : ?latest_index_generation,
580 : ?latest_index_mtime,
581 0 : "Found a newer index while loading an old one"
582 : );
583 0 : }
584 20 : }
585 :
586 20 : if index_part.deleted_at.is_some() {
587 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
588 : } else {
589 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
590 : }
591 20 : }
592 :
593 : /// Download a (layer) file from `path`, into local filesystem.
594 : ///
595 : /// 'layer_metadata' is the metadata from the remote index file.
596 : ///
597 : /// On success, returns the size of the downloaded file.
598 6 : pub async fn download_layer_file(
599 6 : &self,
600 6 : layer_file_name: &LayerName,
601 6 : layer_metadata: &LayerFileMetadata,
602 6 : local_path: &Utf8Path,
603 6 : cancel: &CancellationToken,
604 6 : ctx: &RequestContext,
605 6 : ) -> Result<u64, DownloadError> {
606 6 : let downloaded_size = {
607 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
608 6 : &RemoteOpFileKind::Layer,
609 6 : &RemoteOpKind::Download,
610 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
611 6 : reason: "no need for a downloads gauge",
612 6 : },
613 6 : );
614 6 : download::download_layer_file(
615 6 : self.conf,
616 6 : &self.storage_impl,
617 6 : self.tenant_shard_id,
618 6 : self.timeline_id,
619 6 : layer_file_name,
620 6 : layer_metadata,
621 6 : local_path,
622 6 : cancel,
623 6 : ctx,
624 6 : )
625 6 : .measure_remote_op(
626 6 : RemoteOpFileKind::Layer,
627 6 : RemoteOpKind::Download,
628 6 : Arc::clone(&self.metrics),
629 6 : )
630 80 : .await?
631 : };
632 :
633 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
634 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
635 6 :
636 6 : Ok(downloaded_size)
637 6 : }
638 :
639 : //
640 : // Upload operations.
641 : //
642 :
643 : /// Launch an index-file upload operation in the background, with
644 : /// fully updated metadata.
645 : ///
646 : /// This should only be used to upload initial metadata to remote storage.
647 : ///
648 : /// The upload will be added to the queue immediately, but it
649 : /// won't be performed until all previously scheduled layer file
650 : /// upload operations have completed successfully. This is to
651 : /// ensure that when the index file claims that layers X, Y and Z
652 : /// exist in remote storage, they really do. To wait for the upload
653 : /// to complete, use `wait_completion`.
654 : ///
655 : /// If there were any changes to the list of files, i.e. if any
656 : /// layer file uploads were scheduled, since the last index file
657 : /// upload, those will be included too.
658 228 : pub fn schedule_index_upload_for_full_metadata_update(
659 228 : self: &Arc<Self>,
660 228 : metadata: &TimelineMetadata,
661 228 : ) -> anyhow::Result<()> {
662 228 : let mut guard = self.upload_queue.lock().unwrap();
663 228 : let upload_queue = guard.initialized_mut()?;
664 :
665 : // As documented in the struct definition, it's ok for latest_metadata to be
666 : // ahead of what's _actually_ on the remote during index upload.
667 228 : upload_queue.dirty.metadata = metadata.clone();
668 228 :
669 228 : self.schedule_index_upload(upload_queue)?;
670 :
671 228 : Ok(())
672 228 : }
673 :
674 : /// Launch an index-file upload operation in the background, with only parts of the metadata
675 : /// updated.
676 : ///
677 : /// This is the regular way of updating metadata on layer flushes or Gc.
678 : ///
679 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
680 : /// `index_part.json`, while being more clear on what values update regularly.
681 1144 : pub(crate) fn schedule_index_upload_for_metadata_update(
682 1144 : self: &Arc<Self>,
683 1144 : update: &MetadataUpdate,
684 1144 : ) -> anyhow::Result<()> {
685 1144 : let mut guard = self.upload_queue.lock().unwrap();
686 1144 : let upload_queue = guard.initialized_mut()?;
687 :
688 1144 : upload_queue.dirty.metadata.apply(update);
689 1144 :
690 1144 : self.schedule_index_upload(upload_queue)?;
691 :
692 1144 : Ok(())
693 1144 : }
694 :
695 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
696 : ///
697 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
698 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
699 : /// been in the queue yet.
700 0 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
701 0 : self: &Arc<Self>,
702 0 : state: TimelineArchivalState,
703 0 : ) -> anyhow::Result<bool> {
704 0 : let mut guard = self.upload_queue.lock().unwrap();
705 0 : let upload_queue = guard.initialized_mut()?;
706 :
707 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
708 : /// change needed to set archived_at.
709 0 : fn need_change(
710 0 : archived_at: &Option<NaiveDateTime>,
711 0 : state: TimelineArchivalState,
712 0 : ) -> Option<bool> {
713 0 : match (archived_at, state) {
714 : (Some(_), TimelineArchivalState::Archived)
715 : | (None, TimelineArchivalState::Unarchived) => {
716 : // Nothing to do
717 0 : tracing::info!("intended state matches present state");
718 0 : None
719 : }
720 0 : (None, TimelineArchivalState::Archived) => Some(true),
721 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
722 : }
723 0 : }
724 0 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
725 :
726 0 : if let Some(archived_at_set) = need_upload_scheduled {
727 0 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
728 0 : upload_queue.dirty.archived_at = intended_archived_at;
729 0 : self.schedule_index_upload(upload_queue)?;
730 0 : }
731 :
732 0 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
733 0 : Ok(need_wait)
734 0 : }
735 :
736 : ///
737 : /// Launch an index-file upload operation in the background, if necessary.
738 : ///
739 : /// Use this function to schedule the update of the index file after
740 : /// scheduling file uploads or deletions. If no file uploads or deletions
741 : /// have been scheduled since the last index file upload, this does
742 : /// nothing.
743 : ///
744 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
745 : /// the upload to the upload queue and returns quickly.
746 370 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
747 370 : let mut guard = self.upload_queue.lock().unwrap();
748 370 : let upload_queue = guard.initialized_mut()?;
749 :
750 370 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
751 14 : self.schedule_index_upload(upload_queue)?;
752 356 : }
753 :
754 370 : Ok(())
755 370 : }
756 :
757 : /// Launch an index-file upload operation in the background (internal function)
758 1446 : fn schedule_index_upload(
759 1446 : self: &Arc<Self>,
760 1446 : upload_queue: &mut UploadQueueInitialized,
761 1446 : ) -> Result<(), NotInitialized> {
762 1446 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
763 1446 : // fix up the duplicated field
764 1446 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
765 1446 :
766 1446 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
767 1446 : // look like a retryable error
768 1446 : let void = std::io::sink();
769 1446 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
770 1446 :
771 1446 : let index_part = &upload_queue.dirty;
772 1446 :
773 1446 : info!(
774 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
775 0 : index_part.layer_metadata.len(),
776 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
777 : );
778 :
779 1446 : let op = UploadOp::UploadMetadata {
780 1446 : uploaded: Box::new(index_part.clone()),
781 1446 : };
782 1446 : self.metric_begin(&op);
783 1446 : upload_queue.queued_operations.push_back(op);
784 1446 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
785 1446 :
786 1446 : // Launch the task immediately, if possible
787 1446 : self.launch_queued_tasks(upload_queue);
788 1446 : Ok(())
789 1446 : }
790 :
791 : /// Reparent this timeline to a new parent.
792 : ///
793 : /// A retryable step of timeline ancestor detach.
794 0 : pub(crate) async fn schedule_reparenting_and_wait(
795 0 : self: &Arc<Self>,
796 0 : new_parent: &TimelineId,
797 0 : ) -> anyhow::Result<()> {
798 0 : let receiver = {
799 0 : let mut guard = self.upload_queue.lock().unwrap();
800 0 : let upload_queue = guard.initialized_mut()?;
801 :
802 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
803 0 : return Err(anyhow::anyhow!(
804 0 : "cannot reparent without a current ancestor"
805 0 : ));
806 : };
807 :
808 0 : let uploaded = &upload_queue.clean.0.metadata;
809 0 :
810 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
811 : // nothing to do
812 0 : None
813 : } else {
814 0 : upload_queue.dirty.metadata.reparent(new_parent);
815 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
816 0 :
817 0 : self.schedule_index_upload(upload_queue)?;
818 :
819 0 : Some(self.schedule_barrier0(upload_queue))
820 : }
821 : };
822 :
823 0 : if let Some(receiver) = receiver {
824 0 : Self::wait_completion0(receiver).await?;
825 0 : }
826 0 : Ok(())
827 0 : }
828 :
829 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
830 : /// detaching from ancestor and waits for it to complete.
831 : ///
832 : /// This is used with `Timeline::detach_ancestor` functionality.
833 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
834 0 : self: &Arc<Self>,
835 0 : layers: &[Layer],
836 0 : adopted: (TimelineId, Lsn),
837 0 : ) -> anyhow::Result<()> {
838 0 : let barrier = {
839 0 : let mut guard = self.upload_queue.lock().unwrap();
840 0 : let upload_queue = guard.initialized_mut()?;
841 :
842 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
843 0 : None
844 : } else {
845 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
846 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
847 :
848 0 : for layer in layers {
849 0 : let prev = upload_queue
850 0 : .dirty
851 0 : .layer_metadata
852 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
853 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
854 : }
855 :
856 0 : self.schedule_index_upload(upload_queue)?;
857 :
858 0 : Some(self.schedule_barrier0(upload_queue))
859 : }
860 : };
861 :
862 0 : if let Some(barrier) = barrier {
863 0 : Self::wait_completion0(barrier).await?;
864 0 : }
865 0 : Ok(())
866 0 : }
867 :
868 : /// Adds a gc blocking reason for this timeline if one does not exist already.
869 : ///
870 : /// A retryable step of timeline detach ancestor.
871 : ///
872 : /// Returns a future which waits until the completion of the upload.
873 0 : pub(crate) fn schedule_insert_gc_block_reason(
874 0 : self: &Arc<Self>,
875 0 : reason: index::GcBlockingReason,
876 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
877 0 : {
878 0 : let maybe_barrier = {
879 0 : let mut guard = self.upload_queue.lock().unwrap();
880 0 : let upload_queue = guard.initialized_mut()?;
881 :
882 0 : if let index::GcBlockingReason::DetachAncestor = reason {
883 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
884 0 : drop(guard);
885 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
886 0 : }
887 0 : }
888 :
889 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
890 :
891 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
892 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
893 0 :
894 0 : match (current, uploaded) {
895 0 : (x, y) if wanted(x) && wanted(y) => None,
896 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
897 : // Usual case: !wanted(x) && !wanted(y)
898 : //
899 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
900 : // turn on and off some reason.
901 0 : (x, y) => {
902 0 : if !wanted(x) && wanted(y) {
903 : // this could be avoided by having external in-memory synchronization, like
904 : // timeline detach ancestor
905 0 : warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
906 0 : }
907 :
908 : // at this point, the metadata must always show that there is a parent
909 0 : upload_queue.dirty.gc_blocking = current
910 0 : .map(|x| x.with_reason(reason))
911 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
912 0 : self.schedule_index_upload(upload_queue)?;
913 0 : Some(self.schedule_barrier0(upload_queue))
914 : }
915 : }
916 : };
917 :
918 0 : Ok(async move {
919 0 : if let Some(barrier) = maybe_barrier {
920 0 : Self::wait_completion0(barrier).await?;
921 0 : }
922 0 : Ok(())
923 0 : })
924 0 : }
925 :
926 : /// Removes a gc blocking reason for this timeline if one exists.
927 : ///
928 : /// A retryable step of timeline detach ancestor.
929 : ///
930 : /// Returns a future which waits until the completion of the upload.
931 0 : pub(crate) fn schedule_remove_gc_block_reason(
932 0 : self: &Arc<Self>,
933 0 : reason: index::GcBlockingReason,
934 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
935 0 : {
936 0 : let maybe_barrier = {
937 0 : let mut guard = self.upload_queue.lock().unwrap();
938 0 : let upload_queue = guard.initialized_mut()?;
939 :
940 0 : if let index::GcBlockingReason::DetachAncestor = reason {
941 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
942 0 : drop(guard);
943 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
944 0 : }
945 0 : }
946 :
947 0 : let wanted = |x: Option<&index::GcBlocking>| {
948 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
949 0 : };
950 :
951 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
952 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
953 0 :
954 0 : match (current, uploaded) {
955 0 : (x, y) if wanted(x) && wanted(y) => None,
956 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
957 0 : (x, y) => {
958 0 : if !wanted(x) && wanted(y) {
959 0 : warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
960 0 : }
961 :
962 0 : upload_queue.dirty.gc_blocking =
963 0 : current.as_ref().and_then(|x| x.without_reason(reason));
964 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
965 : // FIXME: bogus ?
966 0 : self.schedule_index_upload(upload_queue)?;
967 0 : Some(self.schedule_barrier0(upload_queue))
968 : }
969 : }
970 : };
971 :
972 0 : Ok(async move {
973 0 : if let Some(barrier) = maybe_barrier {
974 0 : Self::wait_completion0(barrier).await?;
975 0 : }
976 0 : Ok(())
977 0 : })
978 0 : }
979 :
980 : /// Launch an upload operation in the background; the file is added to be included in next
981 : /// `index_part.json` upload.
982 1188 : pub(crate) fn schedule_layer_file_upload(
983 1188 : self: &Arc<Self>,
984 1188 : layer: ResidentLayer,
985 1188 : ) -> Result<(), NotInitialized> {
986 1188 : let mut guard = self.upload_queue.lock().unwrap();
987 1188 : let upload_queue = guard.initialized_mut()?;
988 :
989 1188 : self.schedule_layer_file_upload0(upload_queue, layer);
990 1188 : self.launch_queued_tasks(upload_queue);
991 1188 : Ok(())
992 1188 : }
993 :
994 1536 : fn schedule_layer_file_upload0(
995 1536 : self: &Arc<Self>,
996 1536 : upload_queue: &mut UploadQueueInitialized,
997 1536 : layer: ResidentLayer,
998 1536 : ) {
999 1536 : let metadata = layer.metadata();
1000 1536 :
1001 1536 : upload_queue
1002 1536 : .dirty
1003 1536 : .layer_metadata
1004 1536 : .insert(layer.layer_desc().layer_name(), metadata.clone());
1005 1536 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1006 1536 :
1007 1536 : info!(
1008 : gen=?metadata.generation,
1009 : shard=?metadata.shard,
1010 0 : "scheduled layer file upload {layer}",
1011 : );
1012 :
1013 1536 : let op = UploadOp::UploadLayer(layer, metadata);
1014 1536 : self.metric_begin(&op);
1015 1536 : upload_queue.queued_operations.push_back(op);
1016 1536 : }
1017 :
1018 : /// Launch a delete operation in the background.
1019 : ///
1020 : /// The operation does not modify local filesystem state.
1021 : ///
1022 : /// Note: This schedules an index file upload before the deletions. The
1023 : /// deletion won't actually be performed, until all previously scheduled
1024 : /// upload operations, and the index file upload, have completed
1025 : /// successfully.
1026 8 : pub fn schedule_layer_file_deletion(
1027 8 : self: &Arc<Self>,
1028 8 : names: &[LayerName],
1029 8 : ) -> anyhow::Result<()> {
1030 8 : let mut guard = self.upload_queue.lock().unwrap();
1031 8 : let upload_queue = guard.initialized_mut()?;
1032 :
1033 8 : let with_metadata = self
1034 8 : .schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned())?;
1035 :
1036 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1037 8 :
1038 8 : // Launch the tasks immediately, if possible
1039 8 : self.launch_queued_tasks(upload_queue);
1040 8 : Ok(())
1041 8 : }
1042 :
1043 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1044 : /// layer files, leaving them dangling.
1045 : ///
1046 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1047 : /// is invoked on them.
1048 4 : pub(crate) fn schedule_gc_update(
1049 4 : self: &Arc<Self>,
1050 4 : gc_layers: &[Layer],
1051 4 : ) -> Result<(), NotInitialized> {
1052 4 : let mut guard = self.upload_queue.lock().unwrap();
1053 4 : let upload_queue = guard.initialized_mut()?;
1054 :
1055 : // just forget the return value; after uploading the next index_part.json, we can consider
1056 : // the layer files as "dangling". this is fine, at worst case we create work for the
1057 : // scrubber.
1058 :
1059 4 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1060 4 :
1061 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1062 :
1063 4 : self.launch_queued_tasks(upload_queue);
1064 4 :
1065 4 : Ok(())
1066 4 : }
1067 :
1068 : /// Update the remote index file, removing the to-be-deleted files from the index,
1069 : /// allowing scheduling of actual deletions later.
1070 76 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1071 76 : self: &Arc<Self>,
1072 76 : upload_queue: &mut UploadQueueInitialized,
1073 76 : names: I,
1074 76 : ) -> Result<Vec<(LayerName, LayerFileMetadata)>, NotInitialized>
1075 76 : where
1076 76 : I: IntoIterator<Item = LayerName>,
1077 76 : {
1078 76 : // Decorate our list of names with each name's metadata, dropping
1079 76 : // names that are unexpectedly missing from our metadata. This metadata
1080 76 : // is later used when physically deleting layers, to construct key paths.
1081 76 : let with_metadata: Vec<_> = names
1082 76 : .into_iter()
1083 482 : .filter_map(|name| {
1084 482 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1085 :
1086 482 : if let Some(meta) = meta {
1087 436 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1088 436 : Some((name, meta))
1089 : } else {
1090 : // This can only happen if we forgot to to schedule the file upload
1091 : // before scheduling the delete. Log it because it is a rare/strange
1092 : // situation, and in case something is misbehaving, we'd like to know which
1093 : // layers experienced this.
1094 46 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1095 46 : None
1096 : }
1097 482 : })
1098 76 : .collect();
1099 :
1100 : #[cfg(feature = "testing")]
1101 512 : for (name, metadata) in &with_metadata {
1102 436 : let gen = metadata.generation;
1103 436 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
1104 0 : if unexpected == gen {
1105 0 : tracing::error!("{name} was unlinked twice with same generation");
1106 : } else {
1107 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
1108 : }
1109 436 : }
1110 : }
1111 :
1112 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1113 : // index_part update, because that needs to be uploaded before we can actually delete the
1114 : // files.
1115 76 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1116 60 : self.schedule_index_upload(upload_queue)?;
1117 16 : }
1118 :
1119 76 : Ok(with_metadata)
1120 76 : }
1121 :
1122 : /// Schedules deletion for layer files which have previously been unlinked from the
1123 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1124 475 : pub(crate) fn schedule_deletion_of_unlinked(
1125 475 : self: &Arc<Self>,
1126 475 : layers: Vec<(LayerName, LayerFileMetadata)>,
1127 475 : ) -> anyhow::Result<()> {
1128 475 : let mut guard = self.upload_queue.lock().unwrap();
1129 475 : let upload_queue = guard.initialized_mut()?;
1130 :
1131 475 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1132 475 : self.launch_queued_tasks(upload_queue);
1133 475 : Ok(())
1134 475 : }
1135 :
1136 483 : fn schedule_deletion_of_unlinked0(
1137 483 : self: &Arc<Self>,
1138 483 : upload_queue: &mut UploadQueueInitialized,
1139 483 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1140 483 : ) {
1141 483 : // Filter out any layers which were not created by this tenant shard. These are
1142 483 : // layers that originate from some ancestor shard after a split, and may still
1143 483 : // be referenced by other shards. We are free to delete them locally and remove
1144 483 : // them from our index (and would have already done so when we reach this point
1145 483 : // in the code), but we may not delete them remotely.
1146 483 : with_metadata.retain(|(name, meta)| {
1147 477 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1148 477 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1149 477 : if !retain {
1150 0 : tracing::debug!(
1151 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1152 : meta.shard
1153 : );
1154 477 : }
1155 477 : retain
1156 483 : });
1157 :
1158 960 : for (name, meta) in &with_metadata {
1159 477 : info!(
1160 0 : "scheduling deletion of layer {}{} (shard {})",
1161 0 : name,
1162 0 : meta.generation.get_suffix(),
1163 : meta.shard
1164 : );
1165 : }
1166 :
1167 : #[cfg(feature = "testing")]
1168 960 : for (name, meta) in &with_metadata {
1169 477 : let gen = meta.generation;
1170 477 : match upload_queue.dangling_files.remove(name) {
1171 426 : Some(same) if same == gen => { /* expected */ }
1172 0 : Some(other) => {
1173 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
1174 : }
1175 : None => {
1176 51 : tracing::error!("{name} was unlinked but was not dangling");
1177 : }
1178 : }
1179 : }
1180 :
1181 : // schedule the actual deletions
1182 483 : if with_metadata.is_empty() {
1183 : // avoid scheduling the op & bumping the metric
1184 6 : return;
1185 477 : }
1186 477 : let op = UploadOp::Delete(Delete {
1187 477 : layers: with_metadata,
1188 477 : });
1189 477 : self.metric_begin(&op);
1190 477 : upload_queue.queued_operations.push_back(op);
1191 483 : }
1192 :
1193 : /// Schedules a compaction update to the remote `index_part.json`.
1194 : ///
1195 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1196 64 : pub(crate) fn schedule_compaction_update(
1197 64 : self: &Arc<Self>,
1198 64 : compacted_from: &[Layer],
1199 64 : compacted_to: &[ResidentLayer],
1200 64 : ) -> Result<(), NotInitialized> {
1201 64 : let mut guard = self.upload_queue.lock().unwrap();
1202 64 : let upload_queue = guard.initialized_mut()?;
1203 :
1204 412 : for layer in compacted_to {
1205 348 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1206 348 : }
1207 :
1208 476 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1209 64 :
1210 64 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1211 64 : self.launch_queued_tasks(upload_queue);
1212 64 :
1213 64 : Ok(())
1214 64 : }
1215 :
1216 : /// Wait for all previously scheduled uploads/deletions to complete
1217 1324 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1218 1324 : let receiver = {
1219 1324 : let mut guard = self.upload_queue.lock().unwrap();
1220 1324 : let upload_queue = guard
1221 1324 : .initialized_mut()
1222 1324 : .map_err(WaitCompletionError::NotInitialized)?;
1223 1324 : self.schedule_barrier0(upload_queue)
1224 1324 : };
1225 1324 :
1226 1324 : Self::wait_completion0(receiver).await
1227 1324 : }
1228 :
1229 1324 : async fn wait_completion0(
1230 1324 : mut receiver: tokio::sync::watch::Receiver<()>,
1231 1324 : ) -> Result<(), WaitCompletionError> {
1232 1324 : if receiver.changed().await.is_err() {
1233 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1234 1324 : }
1235 1324 :
1236 1324 : Ok(())
1237 1324 : }
1238 :
1239 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1240 6 : let mut guard = self.upload_queue.lock().unwrap();
1241 6 : let upload_queue = guard.initialized_mut()?;
1242 6 : self.schedule_barrier0(upload_queue);
1243 6 : Ok(())
1244 6 : }
1245 :
1246 1330 : fn schedule_barrier0(
1247 1330 : self: &Arc<Self>,
1248 1330 : upload_queue: &mut UploadQueueInitialized,
1249 1330 : ) -> tokio::sync::watch::Receiver<()> {
1250 1330 : let (sender, receiver) = tokio::sync::watch::channel(());
1251 1330 : let barrier_op = UploadOp::Barrier(sender);
1252 1330 :
1253 1330 : upload_queue.queued_operations.push_back(barrier_op);
1254 1330 : // Don't count this kind of operation!
1255 1330 :
1256 1330 : // Launch the task immediately, if possible
1257 1330 : self.launch_queued_tasks(upload_queue);
1258 1330 :
1259 1330 : receiver
1260 1330 : }
1261 :
1262 : /// Wait for all previously scheduled operations to complete, and then stop.
1263 : ///
1264 : /// Not cancellation safe
1265 6 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1266 6 : // On cancellation the queue is left in ackward state of refusing new operations but
1267 6 : // proper stop is yet to be called. On cancel the original or some later task must call
1268 6 : // `stop` or `shutdown`.
1269 6 : let sg = scopeguard::guard((), |_| {
1270 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1271 6 : });
1272 :
1273 6 : let fut = {
1274 6 : let mut guard = self.upload_queue.lock().unwrap();
1275 6 : let upload_queue = match &mut *guard {
1276 : UploadQueue::Stopped(_) => {
1277 0 : scopeguard::ScopeGuard::into_inner(sg);
1278 0 : return;
1279 : }
1280 : UploadQueue::Uninitialized => {
1281 : // transition into Stopped state
1282 0 : self.stop_impl(&mut guard);
1283 0 : scopeguard::ScopeGuard::into_inner(sg);
1284 0 : return;
1285 : }
1286 6 : UploadQueue::Initialized(ref mut init) => init,
1287 6 : };
1288 6 :
1289 6 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1290 6 : // just don't add more of these as they would never complete.
1291 6 : //
1292 6 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1293 6 : // in every place we would not have to jump through this hoop, and this method could be
1294 6 : // made cancellable.
1295 6 : if !upload_queue.shutting_down {
1296 6 : upload_queue.shutting_down = true;
1297 6 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1298 6 : // this operation is not counted similar to Barrier
1299 6 :
1300 6 : self.launch_queued_tasks(upload_queue);
1301 6 : }
1302 :
1303 6 : upload_queue.shutdown_ready.clone().acquire_owned()
1304 : };
1305 :
1306 6 : let res = fut.await;
1307 :
1308 6 : scopeguard::ScopeGuard::into_inner(sg);
1309 6 :
1310 6 : match res {
1311 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1312 6 : Err(_closed) => {
1313 6 : // expected
1314 6 : }
1315 6 : }
1316 6 :
1317 6 : self.stop();
1318 6 : }
1319 :
1320 : /// Set the deleted_at field in the remote index file.
1321 : ///
1322 : /// This fails if the upload queue has not been `stop()`ed.
1323 : ///
1324 : /// The caller is responsible for calling `stop()` AND for waiting
1325 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1326 : /// Check method [`RemoteTimelineClient::stop`] for details.
1327 0 : #[instrument(skip_all)]
1328 : pub(crate) async fn persist_index_part_with_deleted_flag(
1329 : self: &Arc<Self>,
1330 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1331 : let index_part_with_deleted_at = {
1332 : let mut locked = self.upload_queue.lock().unwrap();
1333 :
1334 : // We must be in stopped state because otherwise
1335 : // we can have inprogress index part upload that can overwrite the file
1336 : // with missing is_deleted flag that we going to set below
1337 : let stopped = locked.stopped_mut()?;
1338 :
1339 : match stopped.deleted_at {
1340 : SetDeletedFlagProgress::NotRunning => (), // proceed
1341 : SetDeletedFlagProgress::InProgress(at) => {
1342 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1343 : }
1344 : SetDeletedFlagProgress::Successful(at) => {
1345 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1346 : }
1347 : };
1348 : let deleted_at = Utc::now().naive_utc();
1349 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1350 :
1351 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1352 : index_part.deleted_at = Some(deleted_at);
1353 : index_part
1354 : };
1355 :
1356 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1357 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1358 0 : let stopped = locked
1359 0 : .stopped_mut()
1360 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1361 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1362 0 : });
1363 :
1364 : pausable_failpoint!("persist_deleted_index_part");
1365 :
1366 : backoff::retry(
1367 0 : || {
1368 0 : upload::upload_index_part(
1369 0 : &self.storage_impl,
1370 0 : &self.tenant_shard_id,
1371 0 : &self.timeline_id,
1372 0 : self.generation,
1373 0 : &index_part_with_deleted_at,
1374 0 : &self.cancel,
1375 0 : )
1376 0 : },
1377 0 : |_e| false,
1378 : 1,
1379 : // have just a couple of attempts
1380 : // when executed as part of timeline deletion this happens in context of api call
1381 : // when executed as part of tenant deletion this happens in the background
1382 : 2,
1383 : "persist_index_part_with_deleted_flag",
1384 : &self.cancel,
1385 : )
1386 : .await
1387 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1388 0 : .and_then(|x| x)?;
1389 :
1390 : // all good, disarm the guard and mark as success
1391 : ScopeGuard::into_inner(undo_deleted_at);
1392 : {
1393 : let mut locked = self.upload_queue.lock().unwrap();
1394 :
1395 : let stopped = locked
1396 : .stopped_mut()
1397 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1398 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1399 : index_part_with_deleted_at
1400 : .deleted_at
1401 : .expect("we set it above"),
1402 : );
1403 : }
1404 :
1405 : Ok(())
1406 : }
1407 :
1408 0 : pub(crate) fn is_deleting(&self) -> bool {
1409 0 : let mut locked = self.upload_queue.lock().unwrap();
1410 0 : locked.stopped_mut().is_ok()
1411 0 : }
1412 :
1413 0 : pub(crate) async fn preserve_initdb_archive(
1414 0 : self: &Arc<Self>,
1415 0 : tenant_id: &TenantId,
1416 0 : timeline_id: &TimelineId,
1417 0 : cancel: &CancellationToken,
1418 0 : ) -> anyhow::Result<()> {
1419 0 : backoff::retry(
1420 0 : || async {
1421 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1422 0 : .await
1423 0 : },
1424 0 : TimeoutOrCancel::caused_by_cancel,
1425 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1426 0 : FAILED_REMOTE_OP_RETRIES,
1427 0 : "preserve_initdb_tar_zst",
1428 0 : &cancel.clone(),
1429 0 : )
1430 0 : .await
1431 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1432 0 : .and_then(|x| x)
1433 0 : .context("backing up initdb archive")?;
1434 0 : Ok(())
1435 0 : }
1436 :
1437 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1438 : ///
1439 : /// This is not normally needed.
1440 0 : pub(crate) async fn upload_layer_file(
1441 0 : self: &Arc<Self>,
1442 0 : uploaded: &ResidentLayer,
1443 0 : cancel: &CancellationToken,
1444 0 : ) -> anyhow::Result<()> {
1445 0 : let remote_path = remote_layer_path(
1446 0 : &self.tenant_shard_id.tenant_id,
1447 0 : &self.timeline_id,
1448 0 : uploaded.metadata().shard,
1449 0 : &uploaded.layer_desc().layer_name(),
1450 0 : uploaded.metadata().generation,
1451 0 : );
1452 0 :
1453 0 : backoff::retry(
1454 0 : || async {
1455 0 : upload::upload_timeline_layer(
1456 0 : &self.storage_impl,
1457 0 : uploaded.local_path(),
1458 0 : &remote_path,
1459 0 : uploaded.metadata().file_size,
1460 0 : cancel,
1461 0 : )
1462 0 : .await
1463 0 : },
1464 0 : TimeoutOrCancel::caused_by_cancel,
1465 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1466 0 : FAILED_REMOTE_OP_RETRIES,
1467 0 : "upload a layer without adding it to latest files",
1468 0 : cancel,
1469 0 : )
1470 0 : .await
1471 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1472 0 : .and_then(|x| x)
1473 0 : .context("upload a layer without adding it to latest files")
1474 0 : }
1475 :
1476 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1477 : /// not added to be part of a future `index_part.json` upload.
1478 0 : pub(crate) async fn copy_timeline_layer(
1479 0 : self: &Arc<Self>,
1480 0 : adopted: &Layer,
1481 0 : adopted_as: &Layer,
1482 0 : cancel: &CancellationToken,
1483 0 : ) -> anyhow::Result<()> {
1484 0 : let source_remote_path = remote_layer_path(
1485 0 : &self.tenant_shard_id.tenant_id,
1486 0 : &adopted
1487 0 : .get_timeline_id()
1488 0 : .expect("Source timeline should be alive"),
1489 0 : adopted.metadata().shard,
1490 0 : &adopted.layer_desc().layer_name(),
1491 0 : adopted.metadata().generation,
1492 0 : );
1493 0 :
1494 0 : let target_remote_path = remote_layer_path(
1495 0 : &self.tenant_shard_id.tenant_id,
1496 0 : &self.timeline_id,
1497 0 : adopted_as.metadata().shard,
1498 0 : &adopted_as.layer_desc().layer_name(),
1499 0 : adopted_as.metadata().generation,
1500 0 : );
1501 0 :
1502 0 : backoff::retry(
1503 0 : || async {
1504 0 : upload::copy_timeline_layer(
1505 0 : &self.storage_impl,
1506 0 : &source_remote_path,
1507 0 : &target_remote_path,
1508 0 : cancel,
1509 0 : )
1510 0 : .await
1511 0 : },
1512 0 : TimeoutOrCancel::caused_by_cancel,
1513 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1514 0 : FAILED_REMOTE_OP_RETRIES,
1515 0 : "copy timeline layer",
1516 0 : cancel,
1517 0 : )
1518 0 : .await
1519 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1520 0 : .and_then(|x| x)
1521 0 : .context("remote copy timeline layer")
1522 0 : }
1523 :
1524 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1525 0 : match tokio::time::timeout(
1526 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1527 0 : self.deletion_queue_client.flush_immediate(),
1528 0 : )
1529 0 : .await
1530 : {
1531 0 : Ok(result) => result,
1532 0 : Err(_timeout) => {
1533 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1534 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1535 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1536 0 : tracing::warn!(
1537 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1538 : );
1539 0 : Ok(())
1540 : }
1541 : }
1542 0 : }
1543 :
1544 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1545 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1546 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1547 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
1548 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1549 :
1550 0 : let layers: Vec<RemotePath> = {
1551 0 : let mut locked = self.upload_queue.lock().unwrap();
1552 0 : let stopped = locked.stopped_mut()?;
1553 :
1554 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1555 0 : anyhow::bail!("deleted_at is not set")
1556 0 : }
1557 0 :
1558 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1559 :
1560 0 : stopped
1561 0 : .upload_queue_for_deletion
1562 0 : .dirty
1563 0 : .layer_metadata
1564 0 : .drain()
1565 0 : .filter(|(_file_name, meta)| {
1566 0 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1567 0 : // all shards anyway, we _could_ delete these, but
1568 0 : // - it creates a potential race if other shards are still
1569 0 : // using the layers while this shard deletes them.
1570 0 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1571 0 : // these timelines are present but corrupt (their index exists but some layers don't)
1572 0 : //
1573 0 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1574 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1575 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1576 0 : })
1577 0 : .map(|(file_name, meta)| {
1578 0 : remote_layer_path(
1579 0 : &self.tenant_shard_id.tenant_id,
1580 0 : &self.timeline_id,
1581 0 : meta.shard,
1582 0 : &file_name,
1583 0 : meta.generation,
1584 0 : )
1585 0 : })
1586 0 : .collect()
1587 0 : };
1588 0 :
1589 0 : let layer_deletion_count = layers.len();
1590 0 : self.deletion_queue_client.push_immediate(layers).await?;
1591 :
1592 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1593 : // inexistant objects are not considered errors.
1594 0 : let initdb_path =
1595 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1596 0 : self.deletion_queue_client
1597 0 : .push_immediate(vec![initdb_path])
1598 0 : .await?;
1599 :
1600 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1601 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1602 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1603 0 :
1604 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1605 0 : // taking the burden of listing all the layers that we already know we should delete.
1606 0 : self.flush_deletion_queue().await?;
1607 :
1608 0 : let cancel = shutdown_token();
1609 :
1610 0 : let remaining = download_retry(
1611 0 : || async {
1612 0 : self.storage_impl
1613 0 : .list(
1614 0 : Some(&timeline_storage_path),
1615 0 : ListingMode::NoDelimiter,
1616 0 : None,
1617 0 : &cancel,
1618 0 : )
1619 0 : .await
1620 0 : },
1621 0 : "list remaining files",
1622 0 : &cancel,
1623 0 : )
1624 0 : .await
1625 0 : .context("list files remaining files")?
1626 : .keys;
1627 :
1628 : // We will delete the current index_part object last, since it acts as a deletion
1629 : // marker via its deleted_at attribute
1630 0 : let latest_index = remaining
1631 0 : .iter()
1632 0 : .filter(|o| {
1633 0 : o.key
1634 0 : .object_name()
1635 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1636 0 : .unwrap_or(false)
1637 0 : })
1638 0 : .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
1639 0 : .max_by_key(|i| i.1)
1640 0 : .map(|i| i.0.clone())
1641 0 : .unwrap_or(
1642 0 : // No generation-suffixed indices, assume we are dealing with
1643 0 : // a legacy index.
1644 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1645 0 : );
1646 0 :
1647 0 : let remaining_layers: Vec<RemotePath> = remaining
1648 0 : .into_iter()
1649 0 : .filter_map(|o| {
1650 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1651 0 : None
1652 : } else {
1653 0 : Some(o.key)
1654 : }
1655 0 : })
1656 0 : .inspect(|path| {
1657 0 : if let Some(name) = path.object_name() {
1658 0 : info!(%name, "deleting a file not referenced from index_part.json");
1659 : } else {
1660 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1661 : }
1662 0 : })
1663 0 : .collect();
1664 0 :
1665 0 : let not_referenced_count = remaining_layers.len();
1666 0 : if !remaining_layers.is_empty() {
1667 0 : self.deletion_queue_client
1668 0 : .push_immediate(remaining_layers)
1669 0 : .await?;
1670 0 : }
1671 :
1672 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1673 0 : Err(anyhow::anyhow!(
1674 0 : "failpoint: timeline-delete-before-index-delete"
1675 0 : ))?
1676 0 : });
1677 :
1678 0 : debug!("enqueuing index part deletion");
1679 0 : self.deletion_queue_client
1680 0 : .push_immediate([latest_index].to_vec())
1681 0 : .await?;
1682 :
1683 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1684 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1685 0 : self.flush_deletion_queue().await?;
1686 :
1687 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1688 0 : Err(anyhow::anyhow!(
1689 0 : "failpoint: timeline-delete-after-index-delete"
1690 0 : ))?
1691 0 : });
1692 :
1693 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1694 :
1695 0 : Ok(())
1696 0 : }
1697 :
1698 : ///
1699 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1700 : /// the ordering constraints.
1701 : ///
1702 : /// The caller needs to already hold the `upload_queue` lock.
1703 7567 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1704 12129 : while let Some(next_op) = upload_queue.queued_operations.front() {
1705 : // Can we run this task now?
1706 8812 : let can_run_now = match next_op {
1707 : UploadOp::UploadLayer(..) => {
1708 : // Can always be scheduled.
1709 1527 : true
1710 : }
1711 : UploadOp::UploadMetadata { .. } => {
1712 : // These can only be performed after all the preceding operations
1713 : // have finished.
1714 4467 : upload_queue.inprogress_tasks.is_empty()
1715 : }
1716 : UploadOp::Delete(_) => {
1717 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1718 334 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1719 : }
1720 :
1721 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1722 2484 : upload_queue.inprogress_tasks.is_empty()
1723 : }
1724 : };
1725 :
1726 : // If we cannot launch this task, don't look any further.
1727 : //
1728 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1729 : // them now, but we don't try to do that currently. For example, if the frontmost task
1730 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1731 : // could still start layer uploads that were scheduled later.
1732 8812 : if !can_run_now {
1733 4244 : break;
1734 4568 : }
1735 4568 :
1736 4568 : if let UploadOp::Shutdown = next_op {
1737 : // leave the op in the queue but do not start more tasks; it will be dropped when
1738 : // the stop is called.
1739 6 : upload_queue.shutdown_ready.close();
1740 6 : break;
1741 4562 : }
1742 4562 :
1743 4562 : // We can launch this task. Remove it from the queue first.
1744 4562 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1745 4562 :
1746 4562 : debug!("starting op: {}", next_op);
1747 :
1748 : // Update the counters
1749 4562 : match next_op {
1750 1527 : UploadOp::UploadLayer(_, _) => {
1751 1527 : upload_queue.num_inprogress_layer_uploads += 1;
1752 1527 : }
1753 1418 : UploadOp::UploadMetadata { .. } => {
1754 1418 : upload_queue.num_inprogress_metadata_uploads += 1;
1755 1418 : }
1756 287 : UploadOp::Delete(_) => {
1757 287 : upload_queue.num_inprogress_deletions += 1;
1758 287 : }
1759 1330 : UploadOp::Barrier(sender) => {
1760 1330 : sender.send_replace(());
1761 1330 : continue;
1762 : }
1763 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1764 : };
1765 :
1766 : // Assign unique ID to this task
1767 3232 : upload_queue.task_counter += 1;
1768 3232 : let upload_task_id = upload_queue.task_counter;
1769 3232 :
1770 3232 : // Add it to the in-progress map
1771 3232 : let task = Arc::new(UploadTask {
1772 3232 : task_id: upload_task_id,
1773 3232 : op: next_op,
1774 3232 : retries: AtomicU32::new(0),
1775 3232 : });
1776 3232 : upload_queue
1777 3232 : .inprogress_tasks
1778 3232 : .insert(task.task_id, Arc::clone(&task));
1779 3232 :
1780 3232 : // Spawn task to perform the task
1781 3232 : let self_rc = Arc::clone(self);
1782 3232 : let tenant_shard_id = self.tenant_shard_id;
1783 3232 : let timeline_id = self.timeline_id;
1784 3232 : task_mgr::spawn(
1785 3232 : &self.runtime,
1786 3232 : TaskKind::RemoteUploadTask,
1787 3232 : self.tenant_shard_id,
1788 3232 : Some(self.timeline_id),
1789 3232 : "remote upload",
1790 3129 : async move {
1791 43274 : self_rc.perform_upload_task(task).await;
1792 3046 : Ok(())
1793 3046 : }
1794 3232 : .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)),
1795 : );
1796 :
1797 : // Loop back to process next task
1798 : }
1799 7567 : }
1800 :
1801 : ///
1802 : /// Perform an upload task.
1803 : ///
1804 : /// The task is in the `inprogress_tasks` list. This function will try to
1805 : /// execute it, retrying forever. On successful completion, the task is
1806 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1807 : /// queue that were waiting by the completion are launched.
1808 : ///
1809 : /// The task can be shut down, however. That leads to stopping the whole
1810 : /// queue.
1811 : ///
1812 3129 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1813 3129 : let cancel = shutdown_token();
1814 : // Loop to retry until it completes.
1815 : loop {
1816 : // If we're requested to shut down, close up shop and exit.
1817 : //
1818 : // Note: We only check for the shutdown requests between retries, so
1819 : // if a shutdown request arrives while we're busy uploading, in the
1820 : // upload::upload:*() call below, we will wait not exit until it has
1821 : // finished. We probably could cancel the upload by simply dropping
1822 : // the Future, but we're not 100% sure if the remote storage library
1823 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1824 : // upload finishes or times out soon enough.
1825 3129 : if cancel.is_cancelled() {
1826 0 : info!("upload task cancelled by shutdown request");
1827 0 : self.stop();
1828 0 : return;
1829 3129 : }
1830 :
1831 3129 : let upload_result: anyhow::Result<()> = match &task.op {
1832 1434 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1833 1434 : let local_path = layer.local_path();
1834 1434 :
1835 1434 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1836 1434 : // the metadata in the upload should always match our current generation.
1837 1434 : assert_eq!(layer_metadata.generation, self.generation);
1838 :
1839 1434 : let remote_path = remote_layer_path(
1840 1434 : &self.tenant_shard_id.tenant_id,
1841 1434 : &self.timeline_id,
1842 1434 : layer_metadata.shard,
1843 1434 : &layer.layer_desc().layer_name(),
1844 1434 : layer_metadata.generation,
1845 1434 : );
1846 1434 :
1847 1434 : upload::upload_timeline_layer(
1848 1434 : &self.storage_impl,
1849 1434 : local_path,
1850 1434 : &remote_path,
1851 1434 : layer_metadata.file_size,
1852 1434 : &self.cancel,
1853 1434 : )
1854 1434 : .measure_remote_op(
1855 1434 : RemoteOpFileKind::Layer,
1856 1434 : RemoteOpKind::Upload,
1857 1434 : Arc::clone(&self.metrics),
1858 1434 : )
1859 37722 : .await
1860 : }
1861 1408 : UploadOp::UploadMetadata { ref uploaded } => {
1862 1408 : let res = upload::upload_index_part(
1863 1408 : &self.storage_impl,
1864 1408 : &self.tenant_shard_id,
1865 1408 : &self.timeline_id,
1866 1408 : self.generation,
1867 1408 : uploaded,
1868 1408 : &self.cancel,
1869 1408 : )
1870 1408 : .measure_remote_op(
1871 1408 : RemoteOpFileKind::Index,
1872 1408 : RemoteOpKind::Upload,
1873 1408 : Arc::clone(&self.metrics),
1874 1408 : )
1875 5331 : .await;
1876 1398 : if res.is_ok() {
1877 1398 : self.update_remote_physical_size_gauge(Some(uploaded));
1878 1398 : let mention_having_future_layers = if cfg!(feature = "testing") {
1879 1398 : uploaded
1880 1398 : .layer_metadata
1881 1398 : .keys()
1882 17611 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
1883 : } else {
1884 0 : false
1885 : };
1886 1398 : if mention_having_future_layers {
1887 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1888 14 : tracing::info!(
1889 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
1890 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
1891 : );
1892 1384 : }
1893 0 : }
1894 1398 : res
1895 : }
1896 287 : UploadOp::Delete(delete) => {
1897 287 : pausable_failpoint!("before-delete-layer-pausable");
1898 287 : self.deletion_queue_client
1899 287 : .push_layers(
1900 287 : self.tenant_shard_id,
1901 287 : self.timeline_id,
1902 287 : self.generation,
1903 287 : delete.layers.clone(),
1904 287 : )
1905 0 : .await
1906 287 : .map_err(|e| anyhow::anyhow!(e))
1907 : }
1908 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1909 : // unreachable. Barrier operations are handled synchronously in
1910 : // launch_queued_tasks
1911 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1912 0 : break;
1913 : }
1914 : };
1915 :
1916 0 : match upload_result {
1917 : Ok(()) => {
1918 3046 : break;
1919 : }
1920 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1921 0 : // loop around to do the proper stopping
1922 0 : continue;
1923 : }
1924 0 : Err(e) => {
1925 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1926 0 :
1927 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1928 0 : // or other external reasons. Such issues are relatively regular, so log them
1929 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1930 0 : //
1931 0 : // (See similar logic for downloads in `download::download_retry`)
1932 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1933 0 : info!(
1934 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1935 0 : task.op, retries, e
1936 : );
1937 : } else {
1938 0 : warn!(
1939 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1940 0 : task.op, retries, e
1941 : );
1942 : }
1943 :
1944 : // sleep until it's time to retry, or we're cancelled
1945 0 : exponential_backoff(
1946 0 : retries,
1947 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1948 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1949 0 : &cancel,
1950 0 : )
1951 0 : .await;
1952 : }
1953 : }
1954 : }
1955 :
1956 3046 : let retries = task.retries.load(Ordering::SeqCst);
1957 3046 : if retries > 0 {
1958 0 : info!(
1959 0 : "remote task {} completed successfully after {} retries",
1960 0 : task.op, retries
1961 : );
1962 : } else {
1963 3046 : debug!("remote task {} completed successfully", task.op);
1964 : }
1965 :
1966 : // The task has completed successfully. Remove it from the in-progress list.
1967 3046 : let lsn_update = {
1968 3046 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1969 3046 : let upload_queue = match upload_queue_guard.deref_mut() {
1970 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1971 0 : UploadQueue::Stopped(_stopped) => {
1972 0 : None
1973 : },
1974 3046 : UploadQueue::Initialized(qi) => { Some(qi) }
1975 : };
1976 :
1977 3046 : let upload_queue = match upload_queue {
1978 3046 : Some(upload_queue) => upload_queue,
1979 : None => {
1980 0 : info!("another concurrent task already stopped the queue");
1981 0 : return;
1982 : }
1983 : };
1984 :
1985 3046 : upload_queue.inprogress_tasks.remove(&task.task_id);
1986 :
1987 3046 : let lsn_update = match task.op {
1988 : UploadOp::UploadLayer(_, _) => {
1989 1361 : upload_queue.num_inprogress_layer_uploads -= 1;
1990 1361 : None
1991 : }
1992 1398 : UploadOp::UploadMetadata { ref uploaded } => {
1993 1398 : upload_queue.num_inprogress_metadata_uploads -= 1;
1994 1398 :
1995 1398 : // the task id is reused as a monotonicity check for storing the "clean"
1996 1398 : // IndexPart.
1997 1398 : let last_updater = upload_queue.clean.1;
1998 1398 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
1999 1398 : let monotone = is_later || last_updater.is_none();
2000 :
2001 1398 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
2002 :
2003 : // not taking ownership is wasteful
2004 1398 : upload_queue.clean.0.clone_from(uploaded);
2005 1398 : upload_queue.clean.1 = Some(task.task_id);
2006 1398 :
2007 1398 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
2008 1398 :
2009 1398 : if self.generation.is_none() {
2010 : // Legacy mode: skip validating generation
2011 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
2012 0 : None
2013 : } else {
2014 1398 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
2015 : }
2016 : }
2017 : UploadOp::Delete(_) => {
2018 287 : upload_queue.num_inprogress_deletions -= 1;
2019 287 : None
2020 : }
2021 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
2022 : };
2023 :
2024 : // Launch any queued tasks that were unblocked by this one.
2025 3046 : self.launch_queued_tasks(upload_queue);
2026 3046 : lsn_update
2027 : };
2028 :
2029 3046 : if let Some((lsn, slot)) = lsn_update {
2030 : // Updates to the remote_consistent_lsn we advertise to pageservers
2031 : // are all routed through the DeletionQueue, to enforce important
2032 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2033 1398 : self.deletion_queue_client
2034 1398 : .update_remote_consistent_lsn(
2035 1398 : self.tenant_shard_id,
2036 1398 : self.timeline_id,
2037 1398 : self.generation,
2038 1398 : lsn,
2039 1398 : slot,
2040 1398 : )
2041 0 : .await;
2042 1648 : }
2043 :
2044 3046 : self.metric_end(&task.op);
2045 3046 : }
2046 :
2047 6511 : fn metric_impl(
2048 6511 : &self,
2049 6511 : op: &UploadOp,
2050 6511 : ) -> Option<(
2051 6511 : RemoteOpFileKind,
2052 6511 : RemoteOpKind,
2053 6511 : RemoteTimelineClientMetricsCallTrackSize,
2054 6511 : )> {
2055 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2056 6511 : let res = match op {
2057 2897 : UploadOp::UploadLayer(_, m) => (
2058 2897 : RemoteOpFileKind::Layer,
2059 2897 : RemoteOpKind::Upload,
2060 2897 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2061 2897 : ),
2062 2844 : UploadOp::UploadMetadata { .. } => (
2063 2844 : RemoteOpFileKind::Index,
2064 2844 : RemoteOpKind::Upload,
2065 2844 : DontTrackSize {
2066 2844 : reason: "metadata uploads are tiny",
2067 2844 : },
2068 2844 : ),
2069 764 : UploadOp::Delete(_delete) => (
2070 764 : RemoteOpFileKind::Layer,
2071 764 : RemoteOpKind::Delete,
2072 764 : DontTrackSize {
2073 764 : reason: "should we track deletes? positive or negative sign?",
2074 764 : },
2075 764 : ),
2076 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2077 : // we do not account these
2078 6 : return None;
2079 : }
2080 : };
2081 6505 : Some(res)
2082 6511 : }
2083 :
2084 3459 : fn metric_begin(&self, op: &UploadOp) {
2085 3459 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2086 3459 : Some(x) => x,
2087 0 : None => return,
2088 : };
2089 3459 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2090 3459 : guard.will_decrement_manually(); // in metric_end(), see right below
2091 3459 : }
2092 :
2093 3052 : fn metric_end(&self, op: &UploadOp) {
2094 3052 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2095 3046 : Some(x) => x,
2096 6 : None => return,
2097 : };
2098 3046 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2099 3052 : }
2100 :
2101 : /// Close the upload queue for new operations and cancel queued operations.
2102 : ///
2103 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2104 : ///
2105 : /// In-progress operations will still be running after this function returns.
2106 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2107 : /// to wait for them to complete, after calling this function.
2108 14 : pub(crate) fn stop(&self) {
2109 14 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2110 14 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2111 14 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2112 14 : let mut guard = self.upload_queue.lock().unwrap();
2113 14 : self.stop_impl(&mut guard);
2114 14 : }
2115 :
2116 14 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2117 14 : match &mut **guard {
2118 : UploadQueue::Uninitialized => {
2119 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2120 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2121 : }
2122 : UploadQueue::Stopped(_) => {
2123 : // nothing to do
2124 6 : info!("another concurrent task already shut down the queue");
2125 : }
2126 8 : UploadQueue::Initialized(initialized) => {
2127 8 : info!("shutting down upload queue");
2128 :
2129 : // Replace the queue with the Stopped state, taking ownership of the old
2130 : // Initialized queue. We will do some checks on it, and then drop it.
2131 8 : let qi = {
2132 : // Here we preserve working version of the upload queue for possible use during deletions.
2133 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2134 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2135 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2136 8 : let upload_queue_for_deletion = UploadQueueInitialized {
2137 8 : task_counter: 0,
2138 8 : dirty: initialized.dirty.clone(),
2139 8 : clean: initialized.clean.clone(),
2140 8 : latest_files_changes_since_metadata_upload_scheduled: 0,
2141 8 : visible_remote_consistent_lsn: initialized
2142 8 : .visible_remote_consistent_lsn
2143 8 : .clone(),
2144 8 : num_inprogress_layer_uploads: 0,
2145 8 : num_inprogress_metadata_uploads: 0,
2146 8 : num_inprogress_deletions: 0,
2147 8 : inprogress_tasks: HashMap::default(),
2148 8 : queued_operations: VecDeque::default(),
2149 8 : #[cfg(feature = "testing")]
2150 8 : dangling_files: HashMap::default(),
2151 8 : shutting_down: false,
2152 8 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2153 8 : };
2154 8 :
2155 8 : let upload_queue = std::mem::replace(
2156 8 : &mut **guard,
2157 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2158 8 : UploadQueueStoppedDeletable {
2159 8 : upload_queue_for_deletion,
2160 8 : deleted_at: SetDeletedFlagProgress::NotRunning,
2161 8 : },
2162 8 : )),
2163 8 : );
2164 8 : if let UploadQueue::Initialized(qi) = upload_queue {
2165 8 : qi
2166 : } else {
2167 0 : unreachable!("we checked in the match above that it is Initialized");
2168 : }
2169 : };
2170 :
2171 : // consistency check
2172 8 : assert_eq!(
2173 8 : qi.num_inprogress_layer_uploads
2174 8 : + qi.num_inprogress_metadata_uploads
2175 8 : + qi.num_inprogress_deletions,
2176 8 : qi.inprogress_tasks.len()
2177 8 : );
2178 :
2179 : // We don't need to do anything here for in-progress tasks. They will finish
2180 : // on their own, decrement the unfinished-task counter themselves, and observe
2181 : // that the queue is Stopped.
2182 8 : drop(qi.inprogress_tasks);
2183 :
2184 : // Tear down queued ops
2185 8 : for op in qi.queued_operations.into_iter() {
2186 6 : self.metric_end(&op);
2187 6 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2188 6 : // which is exactly what we want to happen.
2189 6 : drop(op);
2190 6 : }
2191 : }
2192 : }
2193 14 : }
2194 :
2195 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2196 : /// externally to RemoteTimelineClient.
2197 0 : pub(crate) fn initialized_upload_queue(
2198 0 : &self,
2199 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2200 0 : let mut inner = self.upload_queue.lock().unwrap();
2201 0 : inner.initialized_mut()?;
2202 0 : Ok(UploadQueueAccessor { inner })
2203 0 : }
2204 :
2205 6 : pub(crate) fn no_pending_work(&self) -> bool {
2206 6 : let inner = self.upload_queue.lock().unwrap();
2207 6 : match &*inner {
2208 : UploadQueue::Uninitialized
2209 0 : | UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => true,
2210 6 : UploadQueue::Stopped(UploadQueueStopped::Deletable(x)) => {
2211 6 : x.upload_queue_for_deletion.no_pending_work()
2212 : }
2213 0 : UploadQueue::Initialized(x) => x.no_pending_work(),
2214 : }
2215 6 : }
2216 : }
2217 :
2218 : pub(crate) struct UploadQueueAccessor<'a> {
2219 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2220 : }
2221 :
2222 : impl UploadQueueAccessor<'_> {
2223 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2224 0 : match &*self.inner {
2225 0 : UploadQueue::Initialized(x) => &x.clean.0,
2226 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2227 0 : unreachable!("checked before constructing")
2228 : }
2229 : }
2230 0 : }
2231 : }
2232 :
2233 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2234 0 : let path = format!("tenants/{tenant_shard_id}");
2235 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2236 0 : }
2237 :
2238 570 : pub fn remote_tenant_manifest_path(
2239 570 : tenant_shard_id: &TenantShardId,
2240 570 : generation: Generation,
2241 570 : ) -> RemotePath {
2242 570 : let path = format!(
2243 570 : "tenants/{tenant_shard_id}/tenant-manifest{}.json",
2244 570 : generation.get_suffix()
2245 570 : );
2246 570 : RemotePath::from_string(&path).expect("Failed to construct path")
2247 570 : }
2248 :
2249 : /// Prefix to all generations' manifest objects in a tenant shard
2250 190 : pub fn remote_tenant_manifest_prefix(tenant_shard_id: &TenantShardId) -> RemotePath {
2251 190 : let path = format!("tenants/{tenant_shard_id}/tenant-manifest",);
2252 190 : RemotePath::from_string(&path).expect("Failed to construct path")
2253 190 : }
2254 :
2255 220 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2256 220 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2257 220 : RemotePath::from_string(&path).expect("Failed to construct path")
2258 220 : }
2259 :
2260 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2261 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2262 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2263 0 : }
2264 :
2265 30 : pub fn remote_timeline_path(
2266 30 : tenant_shard_id: &TenantShardId,
2267 30 : timeline_id: &TimelineId,
2268 30 : ) -> RemotePath {
2269 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2270 30 : }
2271 :
2272 : /// Obtains the path of the given Layer in the remote
2273 : ///
2274 : /// Note that the shard component of a remote layer path is _not_ always the same
2275 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2276 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2277 1452 : pub fn remote_layer_path(
2278 1452 : tenant_id: &TenantId,
2279 1452 : timeline_id: &TimelineId,
2280 1452 : shard: ShardIndex,
2281 1452 : layer_file_name: &LayerName,
2282 1452 : generation: Generation,
2283 1452 : ) -> RemotePath {
2284 1452 : // Generation-aware key format
2285 1452 : let path = format!(
2286 1452 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2287 1452 : shard.get_suffix(),
2288 1452 : layer_file_name,
2289 1452 : generation.get_suffix()
2290 1452 : );
2291 1452 :
2292 1452 : RemotePath::from_string(&path).expect("Failed to construct path")
2293 1452 : }
2294 :
2295 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2296 4 : RemotePath::from_string(&format!(
2297 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2298 4 : ))
2299 4 : .expect("Failed to construct path")
2300 4 : }
2301 :
2302 2 : pub fn remote_initdb_preserved_archive_path(
2303 2 : tenant_id: &TenantId,
2304 2 : timeline_id: &TimelineId,
2305 2 : ) -> RemotePath {
2306 2 : RemotePath::from_string(&format!(
2307 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2308 2 : ))
2309 2 : .expect("Failed to construct path")
2310 2 : }
2311 :
2312 1473 : pub fn remote_index_path(
2313 1473 : tenant_shard_id: &TenantShardId,
2314 1473 : timeline_id: &TimelineId,
2315 1473 : generation: Generation,
2316 1473 : ) -> RemotePath {
2317 1473 : RemotePath::from_string(&format!(
2318 1473 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2319 1473 : IndexPart::FILE_NAME,
2320 1473 : generation.get_suffix()
2321 1473 : ))
2322 1473 : .expect("Failed to construct path")
2323 1473 : }
2324 :
2325 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2326 0 : RemotePath::from_string(&format!(
2327 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2328 0 : ))
2329 0 : .expect("Failed to construct path")
2330 0 : }
2331 :
2332 : /// Given the key of an index, parse out the generation part of the name
2333 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2334 18 : let file_name = match path.get_path().file_name() {
2335 18 : Some(f) => f,
2336 : None => {
2337 : // Unexpected: we should be seeing index_part.json paths only
2338 0 : tracing::warn!("Malformed index key {}", path);
2339 0 : return None;
2340 : }
2341 : };
2342 :
2343 18 : match file_name.split_once('-') {
2344 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2345 6 : None => None,
2346 : }
2347 18 : }
2348 :
2349 : /// Given the key of a tenant manifest, parse out the generation number
2350 0 : pub(crate) fn parse_remote_tenant_manifest_path(path: RemotePath) -> Option<Generation> {
2351 : static RE: OnceLock<Regex> = OnceLock::new();
2352 0 : let re = RE.get_or_init(|| Regex::new(r".+tenant-manifest-([0-9a-f]{8}).json").unwrap());
2353 0 : re.captures(path.get_path().as_str())
2354 0 : .and_then(|c| c.get(1))
2355 0 : .and_then(|m| Generation::parse_suffix(m.as_str()))
2356 0 : }
2357 :
2358 : #[cfg(test)]
2359 : mod tests {
2360 : use super::*;
2361 : use crate::{
2362 : context::RequestContext,
2363 : tenant::{
2364 : harness::{TenantHarness, TIMELINE_ID},
2365 : storage_layer::layer::local_layer_path,
2366 : Tenant, Timeline,
2367 : },
2368 : DEFAULT_PG_VERSION,
2369 : };
2370 :
2371 : use std::collections::HashSet;
2372 :
2373 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2374 8 : format!("contents for {name}").into()
2375 8 : }
2376 :
2377 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2378 2 : let metadata = TimelineMetadata::new(
2379 2 : disk_consistent_lsn,
2380 2 : None,
2381 2 : None,
2382 2 : Lsn(0),
2383 2 : Lsn(0),
2384 2 : Lsn(0),
2385 2 : // Any version will do
2386 2 : // but it should be consistent with the one in the tests
2387 2 : crate::DEFAULT_PG_VERSION,
2388 2 : );
2389 2 :
2390 2 : // go through serialize + deserialize to fix the header, including checksum
2391 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2392 2 : }
2393 :
2394 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2395 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2396 2 : avec.sort();
2397 2 :
2398 2 : let mut bvec = b.to_vec();
2399 2 : bvec.sort_unstable();
2400 2 :
2401 2 : assert_eq!(avec, bvec);
2402 2 : }
2403 :
2404 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2405 4 : let mut expected: Vec<String> = expected
2406 4 : .iter()
2407 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2408 4 : .collect();
2409 4 : expected.sort();
2410 4 :
2411 4 : let mut found: Vec<String> = Vec::new();
2412 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2413 16 : let entry_name = entry.file_name();
2414 16 : let fname = entry_name.to_str().unwrap();
2415 16 : found.push(String::from(fname));
2416 16 : }
2417 4 : found.sort();
2418 4 :
2419 4 : assert_eq!(found, expected);
2420 4 : }
2421 :
2422 : struct TestSetup {
2423 : harness: TenantHarness,
2424 : tenant: Arc<Tenant>,
2425 : timeline: Arc<Timeline>,
2426 : tenant_ctx: RequestContext,
2427 : }
2428 :
2429 : impl TestSetup {
2430 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2431 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2432 8 : let harness = TenantHarness::create(test_name).await?;
2433 80 : let (tenant, ctx) = harness.load().await;
2434 :
2435 8 : let timeline = tenant
2436 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2437 24 : .await?;
2438 :
2439 8 : Ok(Self {
2440 8 : harness,
2441 8 : tenant,
2442 8 : timeline,
2443 8 : tenant_ctx: ctx,
2444 8 : })
2445 8 : }
2446 :
2447 : /// Construct a RemoteTimelineClient in an arbitrary generation
2448 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2449 10 : Arc::new(RemoteTimelineClient {
2450 10 : conf: self.harness.conf,
2451 10 : runtime: tokio::runtime::Handle::current(),
2452 10 : tenant_shard_id: self.harness.tenant_shard_id,
2453 10 : timeline_id: TIMELINE_ID,
2454 10 : generation,
2455 10 : storage_impl: self.harness.remote_storage.clone(),
2456 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2457 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2458 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2459 10 : &self.harness.tenant_shard_id,
2460 10 : &TIMELINE_ID,
2461 10 : )),
2462 10 : cancel: CancellationToken::new(),
2463 10 : })
2464 10 : }
2465 :
2466 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2467 : /// and timeline_id are present.
2468 6 : fn span(&self) -> tracing::Span {
2469 6 : tracing::info_span!(
2470 : "test",
2471 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2472 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2473 : timeline_id = %TIMELINE_ID
2474 : )
2475 6 : }
2476 : }
2477 :
2478 : // Test scheduling
2479 : #[tokio::test]
2480 2 : async fn upload_scheduling() {
2481 2 : // Test outline:
2482 2 : //
2483 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2484 2 : // Schedule upload of index. Check that it is queued
2485 2 : // let the layer file uploads finish. Check that the index-upload is now started
2486 2 : // let the index-upload finish.
2487 2 : //
2488 2 : // Download back the index.json. Check that the list of files is correct
2489 2 : //
2490 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2491 2 : // let upload finish. Check that deletion is now started
2492 2 : // Schedule another deletion. Check that it's launched immediately.
2493 2 : // Schedule index upload. Check that it's queued
2494 2 :
2495 26 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2496 2 : let span = test_setup.span();
2497 2 : let _guard = span.enter();
2498 2 :
2499 2 : let TestSetup {
2500 2 : harness,
2501 2 : tenant: _tenant,
2502 2 : timeline,
2503 2 : tenant_ctx: _tenant_ctx,
2504 2 : } = test_setup;
2505 2 :
2506 2 : let client = &timeline.remote_client;
2507 2 :
2508 2 : // Download back the index.json, and check that the list of files is correct
2509 2 : let initial_index_part = match client
2510 2 : .download_index_file(&CancellationToken::new())
2511 6 : .await
2512 2 : .unwrap()
2513 2 : {
2514 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2515 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2516 2 : };
2517 2 : let initial_layers = initial_index_part
2518 2 : .layer_metadata
2519 2 : .keys()
2520 2 : .map(|f| f.to_owned())
2521 2 : .collect::<HashSet<LayerName>>();
2522 2 : let initial_layer = {
2523 2 : assert!(initial_layers.len() == 1);
2524 2 : initial_layers.into_iter().next().unwrap()
2525 2 : };
2526 2 :
2527 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2528 2 :
2529 2 : println!("workdir: {}", harness.conf.workdir);
2530 2 :
2531 2 : let remote_timeline_dir = harness
2532 2 : .remote_fs_dir
2533 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2534 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2535 2 :
2536 2 : let generation = harness.generation;
2537 2 : let shard = harness.shard;
2538 2 :
2539 2 : // Create a couple of dummy files, schedule upload for them
2540 2 :
2541 2 : let layers = [
2542 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2543 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2544 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2545 2 : ]
2546 2 : .into_iter()
2547 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2548 6 :
2549 6 : let local_path = local_layer_path(
2550 6 : harness.conf,
2551 6 : &timeline.tenant_shard_id,
2552 6 : &timeline.timeline_id,
2553 6 : &name,
2554 6 : &generation,
2555 6 : );
2556 6 : std::fs::write(&local_path, &contents).unwrap();
2557 6 :
2558 6 : Layer::for_resident(
2559 6 : harness.conf,
2560 6 : &timeline,
2561 6 : local_path,
2562 6 : name,
2563 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2564 6 : )
2565 6 : }).collect::<Vec<_>>();
2566 2 :
2567 2 : client
2568 2 : .schedule_layer_file_upload(layers[0].clone())
2569 2 : .unwrap();
2570 2 : client
2571 2 : .schedule_layer_file_upload(layers[1].clone())
2572 2 : .unwrap();
2573 2 :
2574 2 : // Check that they are started immediately, not queued
2575 2 : //
2576 2 : // this works because we running within block_on, so any futures are now queued up until
2577 2 : // our next await point.
2578 2 : {
2579 2 : let mut guard = client.upload_queue.lock().unwrap();
2580 2 : let upload_queue = guard.initialized_mut().unwrap();
2581 2 : assert!(upload_queue.queued_operations.is_empty());
2582 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2583 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2584 2 :
2585 2 : // also check that `latest_file_changes` was updated
2586 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2587 2 : }
2588 2 :
2589 2 : // Schedule upload of index. Check that it is queued
2590 2 : let metadata = dummy_metadata(Lsn(0x20));
2591 2 : client
2592 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2593 2 : .unwrap();
2594 2 : {
2595 2 : let mut guard = client.upload_queue.lock().unwrap();
2596 2 : let upload_queue = guard.initialized_mut().unwrap();
2597 2 : assert!(upload_queue.queued_operations.len() == 1);
2598 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2599 2 : }
2600 2 :
2601 2 : // Wait for the uploads to finish
2602 2 : client.wait_completion().await.unwrap();
2603 2 : {
2604 2 : let mut guard = client.upload_queue.lock().unwrap();
2605 2 : let upload_queue = guard.initialized_mut().unwrap();
2606 2 :
2607 2 : assert!(upload_queue.queued_operations.is_empty());
2608 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2609 2 : }
2610 2 :
2611 2 : // Download back the index.json, and check that the list of files is correct
2612 2 : let index_part = match client
2613 2 : .download_index_file(&CancellationToken::new())
2614 6 : .await
2615 2 : .unwrap()
2616 2 : {
2617 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2618 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2619 2 : };
2620 2 :
2621 2 : assert_file_list(
2622 2 : &index_part
2623 2 : .layer_metadata
2624 2 : .keys()
2625 6 : .map(|f| f.to_owned())
2626 2 : .collect(),
2627 2 : &[
2628 2 : &initial_layer.to_string(),
2629 2 : &layers[0].layer_desc().layer_name().to_string(),
2630 2 : &layers[1].layer_desc().layer_name().to_string(),
2631 2 : ],
2632 2 : );
2633 2 : assert_eq!(index_part.metadata, metadata);
2634 2 :
2635 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2636 2 : client
2637 2 : .schedule_layer_file_upload(layers[2].clone())
2638 2 : .unwrap();
2639 2 :
2640 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2641 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2642 2 : // spawn_blocking started by the drop.
2643 2 : client
2644 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2645 2 : .unwrap();
2646 2 : {
2647 2 : let mut guard = client.upload_queue.lock().unwrap();
2648 2 : let upload_queue = guard.initialized_mut().unwrap();
2649 2 :
2650 2 : // Deletion schedules upload of the index file, and the file deletion itself
2651 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2652 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2653 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2654 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2655 2 : assert_eq!(
2656 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2657 2 : 0
2658 2 : );
2659 2 : }
2660 2 : assert_remote_files(
2661 2 : &[
2662 2 : &initial_layer.to_string(),
2663 2 : &layers[0].layer_desc().layer_name().to_string(),
2664 2 : &layers[1].layer_desc().layer_name().to_string(),
2665 2 : "index_part.json",
2666 2 : ],
2667 2 : &remote_timeline_dir,
2668 2 : generation,
2669 2 : );
2670 2 :
2671 2 : // Finish them
2672 2 : client.wait_completion().await.unwrap();
2673 2 : harness.deletion_queue.pump().await;
2674 2 :
2675 2 : assert_remote_files(
2676 2 : &[
2677 2 : &initial_layer.to_string(),
2678 2 : &layers[1].layer_desc().layer_name().to_string(),
2679 2 : &layers[2].layer_desc().layer_name().to_string(),
2680 2 : "index_part.json",
2681 2 : ],
2682 2 : &remote_timeline_dir,
2683 2 : generation,
2684 2 : );
2685 2 : }
2686 :
2687 : #[tokio::test]
2688 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2689 2 : // Setup
2690 2 :
2691 2 : let TestSetup {
2692 2 : harness,
2693 2 : tenant: _tenant,
2694 2 : timeline,
2695 2 : ..
2696 26 : } = TestSetup::new("metrics").await.unwrap();
2697 2 : let client = &timeline.remote_client;
2698 2 :
2699 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2700 2 : let local_path = local_layer_path(
2701 2 : harness.conf,
2702 2 : &timeline.tenant_shard_id,
2703 2 : &timeline.timeline_id,
2704 2 : &layer_file_name_1,
2705 2 : &harness.generation,
2706 2 : );
2707 2 : let content_1 = dummy_contents("foo");
2708 2 : std::fs::write(&local_path, &content_1).unwrap();
2709 2 :
2710 2 : let layer_file_1 = Layer::for_resident(
2711 2 : harness.conf,
2712 2 : &timeline,
2713 2 : local_path,
2714 2 : layer_file_name_1.clone(),
2715 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2716 2 : );
2717 2 :
2718 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2719 2 : struct BytesStartedFinished {
2720 2 : started: Option<usize>,
2721 2 : finished: Option<usize>,
2722 2 : }
2723 2 : impl std::ops::Add for BytesStartedFinished {
2724 2 : type Output = Self;
2725 4 : fn add(self, rhs: Self) -> Self::Output {
2726 4 : Self {
2727 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2728 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2729 4 : }
2730 4 : }
2731 2 : }
2732 6 : let get_bytes_started_stopped = || {
2733 6 : let started = client
2734 6 : .metrics
2735 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2736 6 : .map(|v| v.try_into().unwrap());
2737 6 : let stopped = client
2738 6 : .metrics
2739 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2740 6 : .map(|v| v.try_into().unwrap());
2741 6 : BytesStartedFinished {
2742 6 : started,
2743 6 : finished: stopped,
2744 6 : }
2745 6 : };
2746 2 :
2747 2 : // Test
2748 2 : tracing::info!("now doing actual test");
2749 2 :
2750 2 : let actual_a = get_bytes_started_stopped();
2751 2 :
2752 2 : client
2753 2 : .schedule_layer_file_upload(layer_file_1.clone())
2754 2 : .unwrap();
2755 2 :
2756 2 : let actual_b = get_bytes_started_stopped();
2757 2 :
2758 2 : client.wait_completion().await.unwrap();
2759 2 :
2760 2 : let actual_c = get_bytes_started_stopped();
2761 2 :
2762 2 : // Validate
2763 2 :
2764 2 : let expected_b = actual_a
2765 2 : + BytesStartedFinished {
2766 2 : started: Some(content_1.len()),
2767 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2768 2 : finished: Some(0),
2769 2 : };
2770 2 : assert_eq!(actual_b, expected_b);
2771 2 :
2772 2 : let expected_c = actual_a
2773 2 : + BytesStartedFinished {
2774 2 : started: Some(content_1.len()),
2775 2 : finished: Some(content_1.len()),
2776 2 : };
2777 2 : assert_eq!(actual_c, expected_c);
2778 2 : }
2779 :
2780 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2781 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2782 12 : let example_index_part = IndexPart::example();
2783 12 :
2784 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2785 12 :
2786 12 : let index_path = test_state.harness.remote_fs_dir.join(
2787 12 : remote_index_path(
2788 12 : &test_state.harness.tenant_shard_id,
2789 12 : &TIMELINE_ID,
2790 12 : generation,
2791 12 : )
2792 12 : .get_path(),
2793 12 : );
2794 12 :
2795 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2796 12 : .expect("creating test dir should work");
2797 12 :
2798 12 : eprintln!("Writing {index_path}");
2799 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2800 12 : example_index_part
2801 12 : }
2802 :
2803 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2804 : /// index, the IndexPart returned is equal to `expected`
2805 10 : async fn assert_got_index_part(
2806 10 : test_state: &TestSetup,
2807 10 : get_generation: Generation,
2808 10 : expected: &IndexPart,
2809 10 : ) {
2810 10 : let client = test_state.build_client(get_generation);
2811 :
2812 10 : let download_r = client
2813 10 : .download_index_file(&CancellationToken::new())
2814 68 : .await
2815 10 : .expect("download should always succeed");
2816 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2817 10 : match download_r {
2818 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2819 10 : assert_eq!(&index_part, expected);
2820 : }
2821 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2822 : }
2823 10 : }
2824 :
2825 : #[tokio::test]
2826 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2827 26 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2828 2 : let span = test_state.span();
2829 2 : let _guard = span.enter();
2830 2 :
2831 2 : // Simple case: we are in generation N, load the index from generation N - 1
2832 2 : let generation_n = 5;
2833 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2834 2 :
2835 8 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2836 2 :
2837 2 : Ok(())
2838 2 : }
2839 :
2840 : #[tokio::test]
2841 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2842 2 : let test_state = TestSetup::new("index_part_download_ordering")
2843 26 : .await
2844 2 : .unwrap();
2845 2 :
2846 2 : let span = test_state.span();
2847 2 : let _guard = span.enter();
2848 2 :
2849 2 : // A generation-less IndexPart exists in the bucket, we should find it
2850 2 : let generation_n = 5;
2851 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2852 16 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2853 2 :
2854 2 : // If a more recent-than-none generation exists, we should prefer to load that
2855 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2856 18 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2857 2 :
2858 2 : // If a more-recent-than-me generation exists, we should ignore it.
2859 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2860 20 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2861 2 :
2862 2 : // If a directly previous generation exists, _and_ an index exists in my own
2863 2 : // generation, I should prefer my own generation.
2864 2 : let _injected_prev =
2865 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2866 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2867 2 : assert_got_index_part(
2868 2 : &test_state,
2869 2 : Generation::new(generation_n),
2870 2 : &injected_current,
2871 2 : )
2872 6 : .await;
2873 2 :
2874 2 : Ok(())
2875 2 : }
2876 : }
|
