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::{DeleteTimelineError, 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 : ///
351 418 : pub fn new(
352 418 : remote_storage: GenericRemoteStorage,
353 418 : deletion_queue_client: DeletionQueueClient,
354 418 : conf: &'static PageServerConf,
355 418 : tenant_shard_id: TenantShardId,
356 418 : timeline_id: TimelineId,
357 418 : generation: Generation,
358 418 : ) -> RemoteTimelineClient {
359 418 : RemoteTimelineClient {
360 418 : conf,
361 418 : runtime: if cfg!(test) {
362 : // remote_timeline_client.rs tests rely on current-thread runtime
363 418 : tokio::runtime::Handle::current()
364 : } else {
365 0 : BACKGROUND_RUNTIME.handle().clone()
366 : },
367 418 : tenant_shard_id,
368 418 : timeline_id,
369 418 : generation,
370 418 : storage_impl: remote_storage,
371 418 : deletion_queue_client,
372 418 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
373 418 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
374 418 : &tenant_shard_id,
375 418 : &timeline_id,
376 418 : )),
377 418 : cancel: CancellationToken::new(),
378 418 : }
379 418 : }
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`.
397 412 : pub fn init_upload_queue_for_empty_remote(
398 412 : &self,
399 412 : local_metadata: &TimelineMetadata,
400 412 : ) -> anyhow::Result<()> {
401 412 : let mut upload_queue = self.upload_queue.lock().unwrap();
402 412 : upload_queue.initialize_empty_remote(local_metadata)?;
403 412 : self.update_remote_physical_size_gauge(None);
404 412 : info!("initialized upload queue as empty");
405 412 : Ok(())
406 412 : }
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 2 : pub(crate) fn is_archived(&self) -> Option<bool> {
470 2 : self.upload_queue
471 2 : .lock()
472 2 : .unwrap()
473 2 : .initialized_mut()
474 2 : .map(|q| q.clean.0.archived_at.is_some())
475 2 : .ok()
476 2 : }
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 2 : pub(crate) fn archived_at_stopped_queue(
482 2 : &self,
483 2 : ) -> Result<Option<NaiveDateTime>, UploadQueueNotReadyError> {
484 2 : self.upload_queue
485 2 : .lock()
486 2 : .unwrap()
487 2 : .stopped_mut()
488 2 : .map(|q| q.upload_queue_for_deletion.clean.0.archived_at)
489 2 : .map_err(|_| UploadQueueNotReadyError)
490 2 : }
491 :
492 1828 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
493 1828 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
494 1416 : current_remote_index_part
495 1416 : .layer_metadata
496 1416 : .values()
497 17536 : .map(|ilmd| ilmd.file_size)
498 1416 : .sum()
499 : } else {
500 412 : 0
501 : };
502 1828 : self.metrics.remote_physical_size_gauge.set(size);
503 1828 : }
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 95 : .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 : // We will refuse to proceed, as there is no reasonable scenario where this should happen, but
578 : // there _is_ a clear bug/corruption scenario where it would happen (controller sets the generation
579 : // backwards).
580 0 : tracing::error!(
581 : ?index_generation,
582 : ?latest_index_generation,
583 : ?latest_index_mtime,
584 0 : "Found a newer index while loading an old one"
585 : );
586 0 : return Err(DownloadError::Fatal(
587 0 : "Index age exceeds threshold and a newer index exists".into(),
588 0 : ));
589 0 : }
590 20 : }
591 :
592 20 : if index_part.deleted_at.is_some() {
593 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
594 : } else {
595 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
596 : }
597 20 : }
598 :
599 : /// Download a (layer) file from `path`, into local filesystem.
600 : ///
601 : /// 'layer_metadata' is the metadata from the remote index file.
602 : ///
603 : /// On success, returns the size of the downloaded file.
604 6 : pub async fn download_layer_file(
605 6 : &self,
606 6 : layer_file_name: &LayerName,
607 6 : layer_metadata: &LayerFileMetadata,
608 6 : local_path: &Utf8Path,
609 6 : cancel: &CancellationToken,
610 6 : ctx: &RequestContext,
611 6 : ) -> Result<u64, DownloadError> {
612 6 : let downloaded_size = {
613 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
614 6 : &RemoteOpFileKind::Layer,
615 6 : &RemoteOpKind::Download,
616 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
617 6 : reason: "no need for a downloads gauge",
618 6 : },
619 6 : );
620 6 : download::download_layer_file(
621 6 : self.conf,
622 6 : &self.storage_impl,
623 6 : self.tenant_shard_id,
624 6 : self.timeline_id,
625 6 : layer_file_name,
626 6 : layer_metadata,
627 6 : local_path,
628 6 : cancel,
629 6 : ctx,
630 6 : )
631 6 : .measure_remote_op(
632 6 : RemoteOpFileKind::Layer,
633 6 : RemoteOpKind::Download,
634 6 : Arc::clone(&self.metrics),
635 6 : )
636 67 : .await?
637 : };
638 :
639 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
640 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
641 6 :
642 6 : Ok(downloaded_size)
643 6 : }
644 :
645 : //
646 : // Upload operations.
647 : //
648 :
649 : /// Launch an index-file upload operation in the background, with
650 : /// fully updated metadata.
651 : ///
652 : /// This should only be used to upload initial metadata to remote storage.
653 : ///
654 : /// The upload will be added to the queue immediately, but it
655 : /// won't be performed until all previously scheduled layer file
656 : /// upload operations have completed successfully. This is to
657 : /// ensure that when the index file claims that layers X, Y and Z
658 : /// exist in remote storage, they really do. To wait for the upload
659 : /// to complete, use `wait_completion`.
660 : ///
661 : /// If there were any changes to the list of files, i.e. if any
662 : /// layer file uploads were scheduled, since the last index file
663 : /// upload, those will be included too.
664 230 : pub fn schedule_index_upload_for_full_metadata_update(
665 230 : self: &Arc<Self>,
666 230 : metadata: &TimelineMetadata,
667 230 : ) -> anyhow::Result<()> {
668 230 : let mut guard = self.upload_queue.lock().unwrap();
669 230 : let upload_queue = guard.initialized_mut()?;
670 :
671 : // As documented in the struct definition, it's ok for latest_metadata to be
672 : // ahead of what's _actually_ on the remote during index upload.
673 230 : upload_queue.dirty.metadata = metadata.clone();
674 230 :
675 230 : self.schedule_index_upload(upload_queue)?;
676 :
677 230 : Ok(())
678 230 : }
679 :
680 : /// Launch an index-file upload operation in the background, with only parts of the metadata
681 : /// updated.
682 : ///
683 : /// This is the regular way of updating metadata on layer flushes or Gc.
684 : ///
685 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
686 : /// `index_part.json`, while being more clear on what values update regularly.
687 1146 : pub(crate) fn schedule_index_upload_for_metadata_update(
688 1146 : self: &Arc<Self>,
689 1146 : update: &MetadataUpdate,
690 1146 : ) -> anyhow::Result<()> {
691 1146 : let mut guard = self.upload_queue.lock().unwrap();
692 1146 : let upload_queue = guard.initialized_mut()?;
693 :
694 1146 : upload_queue.dirty.metadata.apply(update);
695 1146 :
696 1146 : self.schedule_index_upload(upload_queue)?;
697 :
698 1146 : Ok(())
699 1146 : }
700 :
701 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
702 : ///
703 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
704 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
705 : /// been in the queue yet.
706 2 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
707 2 : self: &Arc<Self>,
708 2 : state: TimelineArchivalState,
709 2 : ) -> anyhow::Result<bool> {
710 2 : let mut guard = self.upload_queue.lock().unwrap();
711 2 : let upload_queue = guard.initialized_mut()?;
712 :
713 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
714 : /// change needed to set archived_at.
715 4 : fn need_change(
716 4 : archived_at: &Option<NaiveDateTime>,
717 4 : state: TimelineArchivalState,
718 4 : ) -> Option<bool> {
719 4 : match (archived_at, state) {
720 : (Some(_), TimelineArchivalState::Archived)
721 : | (None, TimelineArchivalState::Unarchived) => {
722 : // Nothing to do
723 0 : tracing::info!("intended state matches present state");
724 0 : None
725 : }
726 4 : (None, TimelineArchivalState::Archived) => Some(true),
727 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
728 : }
729 4 : }
730 2 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
731 :
732 2 : if let Some(archived_at_set) = need_upload_scheduled {
733 2 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
734 2 : upload_queue.dirty.archived_at = intended_archived_at;
735 2 : self.schedule_index_upload(upload_queue)?;
736 0 : }
737 :
738 2 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
739 2 : Ok(need_wait)
740 2 : }
741 :
742 : ///
743 : /// Launch an index-file upload operation in the background, if necessary.
744 : ///
745 : /// Use this function to schedule the update of the index file after
746 : /// scheduling file uploads or deletions. If no file uploads or deletions
747 : /// have been scheduled since the last index file upload, this does
748 : /// nothing.
749 : ///
750 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
751 : /// the upload to the upload queue and returns quickly.
752 370 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
753 370 : let mut guard = self.upload_queue.lock().unwrap();
754 370 : let upload_queue = guard.initialized_mut()?;
755 :
756 370 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
757 14 : self.schedule_index_upload(upload_queue)?;
758 356 : }
759 :
760 370 : Ok(())
761 370 : }
762 :
763 : /// Launch an index-file upload operation in the background (internal function)
764 1452 : fn schedule_index_upload(
765 1452 : self: &Arc<Self>,
766 1452 : upload_queue: &mut UploadQueueInitialized,
767 1452 : ) -> Result<(), NotInitialized> {
768 1452 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
769 1452 : // fix up the duplicated field
770 1452 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
771 1452 :
772 1452 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
773 1452 : // look like a retryable error
774 1452 : let void = std::io::sink();
775 1452 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
776 1452 :
777 1452 : let index_part = &upload_queue.dirty;
778 1452 :
779 1452 : info!(
780 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
781 0 : index_part.layer_metadata.len(),
782 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
783 : );
784 :
785 1452 : let op = UploadOp::UploadMetadata {
786 1452 : uploaded: Box::new(index_part.clone()),
787 1452 : };
788 1452 : self.metric_begin(&op);
789 1452 : upload_queue.queued_operations.push_back(op);
790 1452 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
791 1452 :
792 1452 : // Launch the task immediately, if possible
793 1452 : self.launch_queued_tasks(upload_queue);
794 1452 : Ok(())
795 1452 : }
796 :
797 : /// Reparent this timeline to a new parent.
798 : ///
799 : /// A retryable step of timeline ancestor detach.
800 0 : pub(crate) async fn schedule_reparenting_and_wait(
801 0 : self: &Arc<Self>,
802 0 : new_parent: &TimelineId,
803 0 : ) -> anyhow::Result<()> {
804 0 : let receiver = {
805 0 : let mut guard = self.upload_queue.lock().unwrap();
806 0 : let upload_queue = guard.initialized_mut()?;
807 :
808 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
809 0 : return Err(anyhow::anyhow!(
810 0 : "cannot reparent without a current ancestor"
811 0 : ));
812 : };
813 :
814 0 : let uploaded = &upload_queue.clean.0.metadata;
815 0 :
816 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
817 : // nothing to do
818 0 : None
819 : } else {
820 0 : upload_queue.dirty.metadata.reparent(new_parent);
821 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
822 0 :
823 0 : self.schedule_index_upload(upload_queue)?;
824 :
825 0 : Some(self.schedule_barrier0(upload_queue))
826 : }
827 : };
828 :
829 0 : if let Some(receiver) = receiver {
830 0 : Self::wait_completion0(receiver).await?;
831 0 : }
832 0 : Ok(())
833 0 : }
834 :
835 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
836 : /// detaching from ancestor and waits for it to complete.
837 : ///
838 : /// This is used with `Timeline::detach_ancestor` functionality.
839 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
840 0 : self: &Arc<Self>,
841 0 : layers: &[Layer],
842 0 : adopted: (TimelineId, Lsn),
843 0 : ) -> anyhow::Result<()> {
844 0 : let barrier = {
845 0 : let mut guard = self.upload_queue.lock().unwrap();
846 0 : let upload_queue = guard.initialized_mut()?;
847 :
848 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
849 0 : None
850 : } else {
851 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
852 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
853 :
854 0 : for layer in layers {
855 0 : let prev = upload_queue
856 0 : .dirty
857 0 : .layer_metadata
858 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
859 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
860 : }
861 :
862 0 : self.schedule_index_upload(upload_queue)?;
863 :
864 0 : Some(self.schedule_barrier0(upload_queue))
865 : }
866 : };
867 :
868 0 : if let Some(barrier) = barrier {
869 0 : Self::wait_completion0(barrier).await?;
870 0 : }
871 0 : Ok(())
872 0 : }
873 :
874 : /// Adds a gc blocking reason for this timeline if one does not exist already.
875 : ///
876 : /// A retryable step of timeline detach ancestor.
877 : ///
878 : /// Returns a future which waits until the completion of the upload.
879 0 : pub(crate) fn schedule_insert_gc_block_reason(
880 0 : self: &Arc<Self>,
881 0 : reason: index::GcBlockingReason,
882 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
883 0 : {
884 0 : let maybe_barrier = {
885 0 : let mut guard = self.upload_queue.lock().unwrap();
886 0 : let upload_queue = guard.initialized_mut()?;
887 :
888 0 : if let index::GcBlockingReason::DetachAncestor = reason {
889 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
890 0 : drop(guard);
891 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
892 0 : }
893 0 : }
894 :
895 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
896 :
897 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
898 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
899 0 :
900 0 : match (current, uploaded) {
901 0 : (x, y) if wanted(x) && wanted(y) => None,
902 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
903 : // Usual case: !wanted(x) && !wanted(y)
904 : //
905 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
906 : // turn on and off some reason.
907 0 : (x, y) => {
908 0 : if !wanted(x) && wanted(y) {
909 : // this could be avoided by having external in-memory synchronization, like
910 : // timeline detach ancestor
911 0 : warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
912 0 : }
913 :
914 : // at this point, the metadata must always show that there is a parent
915 0 : upload_queue.dirty.gc_blocking = current
916 0 : .map(|x| x.with_reason(reason))
917 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
918 0 : self.schedule_index_upload(upload_queue)?;
919 0 : Some(self.schedule_barrier0(upload_queue))
920 : }
921 : }
922 : };
923 :
924 0 : Ok(async move {
925 0 : if let Some(barrier) = maybe_barrier {
926 0 : Self::wait_completion0(barrier).await?;
927 0 : }
928 0 : Ok(())
929 0 : })
930 0 : }
931 :
932 : /// Removes a gc blocking reason for this timeline if one exists.
933 : ///
934 : /// A retryable step of timeline detach ancestor.
935 : ///
936 : /// Returns a future which waits until the completion of the upload.
937 0 : pub(crate) fn schedule_remove_gc_block_reason(
938 0 : self: &Arc<Self>,
939 0 : reason: index::GcBlockingReason,
940 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
941 0 : {
942 0 : let maybe_barrier = {
943 0 : let mut guard = self.upload_queue.lock().unwrap();
944 0 : let upload_queue = guard.initialized_mut()?;
945 :
946 0 : if let index::GcBlockingReason::DetachAncestor = reason {
947 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
948 0 : drop(guard);
949 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
950 0 : }
951 0 : }
952 :
953 0 : let wanted = |x: Option<&index::GcBlocking>| {
954 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
955 0 : };
956 :
957 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
958 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
959 0 :
960 0 : match (current, uploaded) {
961 0 : (x, y) if wanted(x) && wanted(y) => None,
962 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
963 0 : (x, y) => {
964 0 : if !wanted(x) && wanted(y) {
965 0 : warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
966 0 : }
967 :
968 0 : upload_queue.dirty.gc_blocking =
969 0 : current.as_ref().and_then(|x| x.without_reason(reason));
970 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
971 : // FIXME: bogus ?
972 0 : self.schedule_index_upload(upload_queue)?;
973 0 : Some(self.schedule_barrier0(upload_queue))
974 : }
975 : }
976 : };
977 :
978 0 : Ok(async move {
979 0 : if let Some(barrier) = maybe_barrier {
980 0 : Self::wait_completion0(barrier).await?;
981 0 : }
982 0 : Ok(())
983 0 : })
984 0 : }
985 :
986 : /// Launch an upload operation in the background; the file is added to be included in next
987 : /// `index_part.json` upload.
988 1190 : pub(crate) fn schedule_layer_file_upload(
989 1190 : self: &Arc<Self>,
990 1190 : layer: ResidentLayer,
991 1190 : ) -> Result<(), NotInitialized> {
992 1190 : let mut guard = self.upload_queue.lock().unwrap();
993 1190 : let upload_queue = guard.initialized_mut()?;
994 :
995 1190 : self.schedule_layer_file_upload0(upload_queue, layer);
996 1190 : self.launch_queued_tasks(upload_queue);
997 1190 : Ok(())
998 1190 : }
999 :
1000 1538 : fn schedule_layer_file_upload0(
1001 1538 : self: &Arc<Self>,
1002 1538 : upload_queue: &mut UploadQueueInitialized,
1003 1538 : layer: ResidentLayer,
1004 1538 : ) {
1005 1538 : let metadata = layer.metadata();
1006 1538 :
1007 1538 : upload_queue
1008 1538 : .dirty
1009 1538 : .layer_metadata
1010 1538 : .insert(layer.layer_desc().layer_name(), metadata.clone());
1011 1538 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1012 1538 :
1013 1538 : info!(
1014 : gen=?metadata.generation,
1015 : shard=?metadata.shard,
1016 0 : "scheduled layer file upload {layer}",
1017 : );
1018 :
1019 1538 : let op = UploadOp::UploadLayer(layer, metadata);
1020 1538 : self.metric_begin(&op);
1021 1538 : upload_queue.queued_operations.push_back(op);
1022 1538 : }
1023 :
1024 : /// Launch a delete operation in the background.
1025 : ///
1026 : /// The operation does not modify local filesystem state.
1027 : ///
1028 : /// Note: This schedules an index file upload before the deletions. The
1029 : /// deletion won't actually be performed, until all previously scheduled
1030 : /// upload operations, and the index file upload, have completed
1031 : /// successfully.
1032 8 : pub fn schedule_layer_file_deletion(
1033 8 : self: &Arc<Self>,
1034 8 : names: &[LayerName],
1035 8 : ) -> anyhow::Result<()> {
1036 8 : let mut guard = self.upload_queue.lock().unwrap();
1037 8 : let upload_queue = guard.initialized_mut()?;
1038 :
1039 8 : let with_metadata = self
1040 8 : .schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned())?;
1041 :
1042 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1043 8 :
1044 8 : // Launch the tasks immediately, if possible
1045 8 : self.launch_queued_tasks(upload_queue);
1046 8 : Ok(())
1047 8 : }
1048 :
1049 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1050 : /// layer files, leaving them dangling.
1051 : ///
1052 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1053 : /// is invoked on them.
1054 4 : pub(crate) fn schedule_gc_update(
1055 4 : self: &Arc<Self>,
1056 4 : gc_layers: &[Layer],
1057 4 : ) -> Result<(), NotInitialized> {
1058 4 : let mut guard = self.upload_queue.lock().unwrap();
1059 4 : let upload_queue = guard.initialized_mut()?;
1060 :
1061 : // just forget the return value; after uploading the next index_part.json, we can consider
1062 : // the layer files as "dangling". this is fine, at worst case we create work for the
1063 : // scrubber.
1064 :
1065 4 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1066 4 :
1067 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1068 :
1069 4 : self.launch_queued_tasks(upload_queue);
1070 4 :
1071 4 : Ok(())
1072 4 : }
1073 :
1074 : /// Update the remote index file, removing the to-be-deleted files from the index,
1075 : /// allowing scheduling of actual deletions later.
1076 76 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1077 76 : self: &Arc<Self>,
1078 76 : upload_queue: &mut UploadQueueInitialized,
1079 76 : names: I,
1080 76 : ) -> Result<Vec<(LayerName, LayerFileMetadata)>, NotInitialized>
1081 76 : where
1082 76 : I: IntoIterator<Item = LayerName>,
1083 76 : {
1084 76 : // Decorate our list of names with each name's metadata, dropping
1085 76 : // names that are unexpectedly missing from our metadata. This metadata
1086 76 : // is later used when physically deleting layers, to construct key paths.
1087 76 : let with_metadata: Vec<_> = names
1088 76 : .into_iter()
1089 482 : .filter_map(|name| {
1090 482 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1091 :
1092 482 : if let Some(meta) = meta {
1093 436 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1094 436 : Some((name, meta))
1095 : } else {
1096 : // This can only happen if we forgot to to schedule the file upload
1097 : // before scheduling the delete. Log it because it is a rare/strange
1098 : // situation, and in case something is misbehaving, we'd like to know which
1099 : // layers experienced this.
1100 46 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1101 46 : None
1102 : }
1103 482 : })
1104 76 : .collect();
1105 :
1106 : #[cfg(feature = "testing")]
1107 512 : for (name, metadata) in &with_metadata {
1108 436 : let gen = metadata.generation;
1109 436 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
1110 0 : if unexpected == gen {
1111 0 : tracing::error!("{name} was unlinked twice with same generation");
1112 : } else {
1113 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
1114 : }
1115 436 : }
1116 : }
1117 :
1118 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1119 : // index_part update, because that needs to be uploaded before we can actually delete the
1120 : // files.
1121 76 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1122 60 : self.schedule_index_upload(upload_queue)?;
1123 16 : }
1124 :
1125 76 : Ok(with_metadata)
1126 76 : }
1127 :
1128 : /// Schedules deletion for layer files which have previously been unlinked from the
1129 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1130 475 : pub(crate) fn schedule_deletion_of_unlinked(
1131 475 : self: &Arc<Self>,
1132 475 : layers: Vec<(LayerName, LayerFileMetadata)>,
1133 475 : ) -> anyhow::Result<()> {
1134 475 : let mut guard = self.upload_queue.lock().unwrap();
1135 475 : let upload_queue = guard.initialized_mut()?;
1136 :
1137 475 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1138 475 : self.launch_queued_tasks(upload_queue);
1139 475 : Ok(())
1140 475 : }
1141 :
1142 483 : fn schedule_deletion_of_unlinked0(
1143 483 : self: &Arc<Self>,
1144 483 : upload_queue: &mut UploadQueueInitialized,
1145 483 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1146 483 : ) {
1147 483 : // Filter out any layers which were not created by this tenant shard. These are
1148 483 : // layers that originate from some ancestor shard after a split, and may still
1149 483 : // be referenced by other shards. We are free to delete them locally and remove
1150 483 : // them from our index (and would have already done so when we reach this point
1151 483 : // in the code), but we may not delete them remotely.
1152 483 : with_metadata.retain(|(name, meta)| {
1153 477 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1154 477 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1155 477 : if !retain {
1156 0 : tracing::debug!(
1157 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1158 : meta.shard
1159 : );
1160 477 : }
1161 477 : retain
1162 483 : });
1163 :
1164 960 : for (name, meta) in &with_metadata {
1165 477 : info!(
1166 0 : "scheduling deletion of layer {}{} (shard {})",
1167 0 : name,
1168 0 : meta.generation.get_suffix(),
1169 : meta.shard
1170 : );
1171 : }
1172 :
1173 : #[cfg(feature = "testing")]
1174 960 : for (name, meta) in &with_metadata {
1175 477 : let gen = meta.generation;
1176 477 : match upload_queue.dangling_files.remove(name) {
1177 425 : Some(same) if same == gen => { /* expected */ }
1178 0 : Some(other) => {
1179 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
1180 : }
1181 : None => {
1182 52 : tracing::error!("{name} was unlinked but was not dangling");
1183 : }
1184 : }
1185 : }
1186 :
1187 : // schedule the actual deletions
1188 483 : if with_metadata.is_empty() {
1189 : // avoid scheduling the op & bumping the metric
1190 6 : return;
1191 477 : }
1192 477 : let op = UploadOp::Delete(Delete {
1193 477 : layers: with_metadata,
1194 477 : });
1195 477 : self.metric_begin(&op);
1196 477 : upload_queue.queued_operations.push_back(op);
1197 483 : }
1198 :
1199 : /// Schedules a compaction update to the remote `index_part.json`.
1200 : ///
1201 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1202 64 : pub(crate) fn schedule_compaction_update(
1203 64 : self: &Arc<Self>,
1204 64 : compacted_from: &[Layer],
1205 64 : compacted_to: &[ResidentLayer],
1206 64 : ) -> Result<(), NotInitialized> {
1207 64 : let mut guard = self.upload_queue.lock().unwrap();
1208 64 : let upload_queue = guard.initialized_mut()?;
1209 :
1210 412 : for layer in compacted_to {
1211 348 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1212 348 : }
1213 :
1214 476 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1215 64 :
1216 64 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1217 64 : self.launch_queued_tasks(upload_queue);
1218 64 :
1219 64 : Ok(())
1220 64 : }
1221 :
1222 : /// Wait for all previously scheduled uploads/deletions to complete
1223 1330 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1224 1330 : let receiver = {
1225 1330 : let mut guard = self.upload_queue.lock().unwrap();
1226 1330 : let upload_queue = guard
1227 1330 : .initialized_mut()
1228 1330 : .map_err(WaitCompletionError::NotInitialized)?;
1229 1330 : self.schedule_barrier0(upload_queue)
1230 1330 : };
1231 1330 :
1232 1330 : Self::wait_completion0(receiver).await
1233 1330 : }
1234 :
1235 1330 : async fn wait_completion0(
1236 1330 : mut receiver: tokio::sync::watch::Receiver<()>,
1237 1330 : ) -> Result<(), WaitCompletionError> {
1238 1330 : if receiver.changed().await.is_err() {
1239 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1240 1330 : }
1241 1330 :
1242 1330 : Ok(())
1243 1330 : }
1244 :
1245 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1246 6 : let mut guard = self.upload_queue.lock().unwrap();
1247 6 : let upload_queue = guard.initialized_mut()?;
1248 6 : self.schedule_barrier0(upload_queue);
1249 6 : Ok(())
1250 6 : }
1251 :
1252 1336 : fn schedule_barrier0(
1253 1336 : self: &Arc<Self>,
1254 1336 : upload_queue: &mut UploadQueueInitialized,
1255 1336 : ) -> tokio::sync::watch::Receiver<()> {
1256 1336 : let (sender, receiver) = tokio::sync::watch::channel(());
1257 1336 : let barrier_op = UploadOp::Barrier(sender);
1258 1336 :
1259 1336 : upload_queue.queued_operations.push_back(barrier_op);
1260 1336 : // Don't count this kind of operation!
1261 1336 :
1262 1336 : // Launch the task immediately, if possible
1263 1336 : self.launch_queued_tasks(upload_queue);
1264 1336 :
1265 1336 : receiver
1266 1336 : }
1267 :
1268 : /// Wait for all previously scheduled operations to complete, and then stop.
1269 : ///
1270 : /// Not cancellation safe
1271 8 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1272 8 : // On cancellation the queue is left in ackward state of refusing new operations but
1273 8 : // proper stop is yet to be called. On cancel the original or some later task must call
1274 8 : // `stop` or `shutdown`.
1275 8 : let sg = scopeguard::guard((), |_| {
1276 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1277 8 : });
1278 :
1279 8 : let fut = {
1280 8 : let mut guard = self.upload_queue.lock().unwrap();
1281 8 : let upload_queue = match &mut *guard {
1282 : UploadQueue::Stopped(_) => {
1283 0 : scopeguard::ScopeGuard::into_inner(sg);
1284 0 : return;
1285 : }
1286 : UploadQueue::Uninitialized => {
1287 : // transition into Stopped state
1288 0 : self.stop_impl(&mut guard);
1289 0 : scopeguard::ScopeGuard::into_inner(sg);
1290 0 : return;
1291 : }
1292 8 : UploadQueue::Initialized(ref mut init) => init,
1293 8 : };
1294 8 :
1295 8 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1296 8 : // just don't add more of these as they would never complete.
1297 8 : //
1298 8 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1299 8 : // in every place we would not have to jump through this hoop, and this method could be
1300 8 : // made cancellable.
1301 8 : if !upload_queue.shutting_down {
1302 8 : upload_queue.shutting_down = true;
1303 8 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1304 8 : // this operation is not counted similar to Barrier
1305 8 :
1306 8 : self.launch_queued_tasks(upload_queue);
1307 8 : }
1308 :
1309 8 : upload_queue.shutdown_ready.clone().acquire_owned()
1310 : };
1311 :
1312 8 : let res = fut.await;
1313 :
1314 8 : scopeguard::ScopeGuard::into_inner(sg);
1315 8 :
1316 8 : match res {
1317 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1318 8 : Err(_closed) => {
1319 8 : // expected
1320 8 : }
1321 8 : }
1322 8 :
1323 8 : self.stop();
1324 8 : }
1325 :
1326 : /// Set the deleted_at field in the remote index file.
1327 : ///
1328 : /// This fails if the upload queue has not been `stop()`ed.
1329 : ///
1330 : /// The caller is responsible for calling `stop()` AND for waiting
1331 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1332 : /// Check method [`RemoteTimelineClient::stop`] for details.
1333 0 : #[instrument(skip_all)]
1334 : pub(crate) async fn persist_index_part_with_deleted_flag(
1335 : self: &Arc<Self>,
1336 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1337 : let index_part_with_deleted_at = {
1338 : let mut locked = self.upload_queue.lock().unwrap();
1339 :
1340 : // We must be in stopped state because otherwise
1341 : // we can have inprogress index part upload that can overwrite the file
1342 : // with missing is_deleted flag that we going to set below
1343 : let stopped = locked.stopped_mut()?;
1344 :
1345 : match stopped.deleted_at {
1346 : SetDeletedFlagProgress::NotRunning => (), // proceed
1347 : SetDeletedFlagProgress::InProgress(at) => {
1348 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1349 : }
1350 : SetDeletedFlagProgress::Successful(at) => {
1351 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1352 : }
1353 : };
1354 : let deleted_at = Utc::now().naive_utc();
1355 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1356 :
1357 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1358 : index_part.deleted_at = Some(deleted_at);
1359 : index_part
1360 : };
1361 :
1362 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1363 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1364 0 : let stopped = locked
1365 0 : .stopped_mut()
1366 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1367 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1368 0 : });
1369 :
1370 : pausable_failpoint!("persist_deleted_index_part");
1371 :
1372 : backoff::retry(
1373 0 : || {
1374 0 : upload::upload_index_part(
1375 0 : &self.storage_impl,
1376 0 : &self.tenant_shard_id,
1377 0 : &self.timeline_id,
1378 0 : self.generation,
1379 0 : &index_part_with_deleted_at,
1380 0 : &self.cancel,
1381 0 : )
1382 0 : },
1383 0 : |_e| false,
1384 : 1,
1385 : // have just a couple of attempts
1386 : // when executed as part of timeline deletion this happens in context of api call
1387 : // when executed as part of tenant deletion this happens in the background
1388 : 2,
1389 : "persist_index_part_with_deleted_flag",
1390 : &self.cancel,
1391 : )
1392 : .await
1393 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1394 0 : .and_then(|x| x)?;
1395 :
1396 : // all good, disarm the guard and mark as success
1397 : ScopeGuard::into_inner(undo_deleted_at);
1398 : {
1399 : let mut locked = self.upload_queue.lock().unwrap();
1400 :
1401 : let stopped = locked
1402 : .stopped_mut()
1403 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1404 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1405 : index_part_with_deleted_at
1406 : .deleted_at
1407 : .expect("we set it above"),
1408 : );
1409 : }
1410 :
1411 : Ok(())
1412 : }
1413 :
1414 0 : pub(crate) fn is_deleting(&self) -> bool {
1415 0 : let mut locked = self.upload_queue.lock().unwrap();
1416 0 : locked.stopped_mut().is_ok()
1417 0 : }
1418 :
1419 0 : pub(crate) async fn preserve_initdb_archive(
1420 0 : self: &Arc<Self>,
1421 0 : tenant_id: &TenantId,
1422 0 : timeline_id: &TimelineId,
1423 0 : cancel: &CancellationToken,
1424 0 : ) -> anyhow::Result<()> {
1425 0 : backoff::retry(
1426 0 : || async {
1427 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1428 0 : .await
1429 0 : },
1430 0 : TimeoutOrCancel::caused_by_cancel,
1431 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1432 0 : FAILED_REMOTE_OP_RETRIES,
1433 0 : "preserve_initdb_tar_zst",
1434 0 : &cancel.clone(),
1435 0 : )
1436 0 : .await
1437 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1438 0 : .and_then(|x| x)
1439 0 : .context("backing up initdb archive")?;
1440 0 : Ok(())
1441 0 : }
1442 :
1443 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1444 : ///
1445 : /// This is not normally needed.
1446 0 : pub(crate) async fn upload_layer_file(
1447 0 : self: &Arc<Self>,
1448 0 : uploaded: &ResidentLayer,
1449 0 : cancel: &CancellationToken,
1450 0 : ) -> anyhow::Result<()> {
1451 0 : let remote_path = remote_layer_path(
1452 0 : &self.tenant_shard_id.tenant_id,
1453 0 : &self.timeline_id,
1454 0 : uploaded.metadata().shard,
1455 0 : &uploaded.layer_desc().layer_name(),
1456 0 : uploaded.metadata().generation,
1457 0 : );
1458 0 :
1459 0 : backoff::retry(
1460 0 : || async {
1461 0 : upload::upload_timeline_layer(
1462 0 : &self.storage_impl,
1463 0 : uploaded.local_path(),
1464 0 : &remote_path,
1465 0 : uploaded.metadata().file_size,
1466 0 : cancel,
1467 0 : )
1468 0 : .await
1469 0 : },
1470 0 : TimeoutOrCancel::caused_by_cancel,
1471 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1472 0 : FAILED_REMOTE_OP_RETRIES,
1473 0 : "upload a layer without adding it to latest files",
1474 0 : cancel,
1475 0 : )
1476 0 : .await
1477 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1478 0 : .and_then(|x| x)
1479 0 : .context("upload a layer without adding it to latest files")
1480 0 : }
1481 :
1482 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1483 : /// not added to be part of a future `index_part.json` upload.
1484 0 : pub(crate) async fn copy_timeline_layer(
1485 0 : self: &Arc<Self>,
1486 0 : adopted: &Layer,
1487 0 : adopted_as: &Layer,
1488 0 : cancel: &CancellationToken,
1489 0 : ) -> anyhow::Result<()> {
1490 0 : let source_remote_path = remote_layer_path(
1491 0 : &self.tenant_shard_id.tenant_id,
1492 0 : &adopted
1493 0 : .get_timeline_id()
1494 0 : .expect("Source timeline should be alive"),
1495 0 : adopted.metadata().shard,
1496 0 : &adopted.layer_desc().layer_name(),
1497 0 : adopted.metadata().generation,
1498 0 : );
1499 0 :
1500 0 : let target_remote_path = remote_layer_path(
1501 0 : &self.tenant_shard_id.tenant_id,
1502 0 : &self.timeline_id,
1503 0 : adopted_as.metadata().shard,
1504 0 : &adopted_as.layer_desc().layer_name(),
1505 0 : adopted_as.metadata().generation,
1506 0 : );
1507 0 :
1508 0 : backoff::retry(
1509 0 : || async {
1510 0 : upload::copy_timeline_layer(
1511 0 : &self.storage_impl,
1512 0 : &source_remote_path,
1513 0 : &target_remote_path,
1514 0 : cancel,
1515 0 : )
1516 0 : .await
1517 0 : },
1518 0 : TimeoutOrCancel::caused_by_cancel,
1519 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1520 0 : FAILED_REMOTE_OP_RETRIES,
1521 0 : "copy timeline layer",
1522 0 : cancel,
1523 0 : )
1524 0 : .await
1525 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1526 0 : .and_then(|x| x)
1527 0 : .context("remote copy timeline layer")
1528 0 : }
1529 :
1530 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1531 0 : match tokio::time::timeout(
1532 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1533 0 : self.deletion_queue_client.flush_immediate(),
1534 0 : )
1535 0 : .await
1536 : {
1537 0 : Ok(result) => result,
1538 0 : Err(_timeout) => {
1539 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1540 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1541 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1542 0 : tracing::warn!(
1543 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1544 : );
1545 0 : Ok(())
1546 : }
1547 : }
1548 0 : }
1549 :
1550 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1551 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1552 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1553 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> Result<(), DeleteTimelineError> {
1554 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1555 :
1556 0 : let layers: Vec<RemotePath> = {
1557 0 : let mut locked = self.upload_queue.lock().unwrap();
1558 0 : let stopped = locked.stopped_mut().map_err(DeleteTimelineError::Other)?;
1559 :
1560 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1561 0 : return Err(DeleteTimelineError::Other(anyhow::anyhow!(
1562 0 : "deleted_at is not set"
1563 0 : )));
1564 0 : }
1565 0 :
1566 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1567 :
1568 0 : stopped
1569 0 : .upload_queue_for_deletion
1570 0 : .dirty
1571 0 : .layer_metadata
1572 0 : .drain()
1573 0 : .filter(|(_file_name, meta)| {
1574 0 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1575 0 : // all shards anyway, we _could_ delete these, but
1576 0 : // - it creates a potential race if other shards are still
1577 0 : // using the layers while this shard deletes them.
1578 0 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1579 0 : // these timelines are present but corrupt (their index exists but some layers don't)
1580 0 : //
1581 0 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1582 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1583 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1584 0 : })
1585 0 : .map(|(file_name, meta)| {
1586 0 : remote_layer_path(
1587 0 : &self.tenant_shard_id.tenant_id,
1588 0 : &self.timeline_id,
1589 0 : meta.shard,
1590 0 : &file_name,
1591 0 : meta.generation,
1592 0 : )
1593 0 : })
1594 0 : .collect()
1595 0 : };
1596 0 :
1597 0 : let layer_deletion_count = layers.len();
1598 0 : self.deletion_queue_client
1599 0 : .push_immediate(layers)
1600 0 : .await
1601 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1602 :
1603 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1604 : // inexistant objects are not considered errors.
1605 0 : let initdb_path =
1606 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1607 0 : self.deletion_queue_client
1608 0 : .push_immediate(vec![initdb_path])
1609 0 : .await
1610 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1611 :
1612 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1613 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1614 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1615 0 :
1616 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1617 0 : // taking the burden of listing all the layers that we already know we should delete.
1618 0 : self.flush_deletion_queue()
1619 0 : .await
1620 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1621 :
1622 0 : let cancel = shutdown_token();
1623 :
1624 0 : let remaining = download_retry(
1625 0 : || async {
1626 0 : self.storage_impl
1627 0 : .list(
1628 0 : Some(&timeline_storage_path),
1629 0 : ListingMode::NoDelimiter,
1630 0 : None,
1631 0 : &cancel,
1632 0 : )
1633 0 : .await
1634 0 : },
1635 0 : "list remaining files",
1636 0 : &cancel,
1637 0 : )
1638 0 : .await
1639 0 : .context("list files remaining files")?
1640 : .keys;
1641 :
1642 : // We will delete the current index_part object last, since it acts as a deletion
1643 : // marker via its deleted_at attribute
1644 0 : let latest_index = remaining
1645 0 : .iter()
1646 0 : .filter(|o| {
1647 0 : o.key
1648 0 : .object_name()
1649 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1650 0 : .unwrap_or(false)
1651 0 : })
1652 0 : .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
1653 0 : .max_by_key(|i| i.1)
1654 0 : .map(|i| i.0.clone())
1655 0 : .unwrap_or(
1656 0 : // No generation-suffixed indices, assume we are dealing with
1657 0 : // a legacy index.
1658 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1659 0 : );
1660 0 :
1661 0 : let remaining_layers: Vec<RemotePath> = remaining
1662 0 : .into_iter()
1663 0 : .filter_map(|o| {
1664 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1665 0 : None
1666 : } else {
1667 0 : Some(o.key)
1668 : }
1669 0 : })
1670 0 : .inspect(|path| {
1671 0 : if let Some(name) = path.object_name() {
1672 0 : info!(%name, "deleting a file not referenced from index_part.json");
1673 : } else {
1674 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1675 : }
1676 0 : })
1677 0 : .collect();
1678 0 :
1679 0 : let not_referenced_count = remaining_layers.len();
1680 0 : if !remaining_layers.is_empty() {
1681 0 : self.deletion_queue_client
1682 0 : .push_immediate(remaining_layers)
1683 0 : .await
1684 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1685 0 : }
1686 :
1687 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1688 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1689 0 : "failpoint: timeline-delete-before-index-delete"
1690 0 : )))?
1691 0 : });
1692 :
1693 0 : debug!("enqueuing index part deletion");
1694 0 : self.deletion_queue_client
1695 0 : .push_immediate([latest_index].to_vec())
1696 0 : .await
1697 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1698 :
1699 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1700 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1701 0 : self.flush_deletion_queue()
1702 0 : .await
1703 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1704 :
1705 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1706 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1707 0 : "failpoint: timeline-delete-after-index-delete"
1708 0 : )))?
1709 0 : });
1710 :
1711 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1712 :
1713 0 : Ok(())
1714 0 : }
1715 :
1716 : ///
1717 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1718 : /// the ordering constraints.
1719 : ///
1720 : /// The caller needs to already hold the `upload_queue` lock.
1721 7445 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1722 11978 : while let Some(next_op) = upload_queue.queued_operations.front() {
1723 : // Can we run this task now?
1724 8688 : let can_run_now = match next_op {
1725 : UploadOp::UploadLayer(..) => {
1726 : // Can always be scheduled.
1727 1532 : true
1728 : }
1729 : UploadOp::UploadMetadata { .. } => {
1730 : // These can only be performed after all the preceding operations
1731 : // have finished.
1732 4380 : upload_queue.inprogress_tasks.is_empty()
1733 : }
1734 : UploadOp::Delete(_) => {
1735 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1736 281 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1737 : }
1738 :
1739 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1740 2495 : upload_queue.inprogress_tasks.is_empty()
1741 : }
1742 : };
1743 :
1744 : // If we cannot launch this task, don't look any further.
1745 : //
1746 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1747 : // them now, but we don't try to do that currently. For example, if the frontmost task
1748 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1749 : // could still start layer uploads that were scheduled later.
1750 8688 : if !can_run_now {
1751 4147 : break;
1752 4541 : }
1753 4541 :
1754 4541 : if let UploadOp::Shutdown = next_op {
1755 : // leave the op in the queue but do not start more tasks; it will be dropped when
1756 : // the stop is called.
1757 8 : upload_queue.shutdown_ready.close();
1758 8 : break;
1759 4533 : }
1760 4533 :
1761 4533 : // We can launch this task. Remove it from the queue first.
1762 4533 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1763 4533 :
1764 4533 : debug!("starting op: {}", next_op);
1765 :
1766 : // Update the counters
1767 4533 : match next_op {
1768 1532 : UploadOp::UploadLayer(_, _) => {
1769 1532 : upload_queue.num_inprogress_layer_uploads += 1;
1770 1532 : }
1771 1424 : UploadOp::UploadMetadata { .. } => {
1772 1424 : upload_queue.num_inprogress_metadata_uploads += 1;
1773 1424 : }
1774 241 : UploadOp::Delete(_) => {
1775 241 : upload_queue.num_inprogress_deletions += 1;
1776 241 : }
1777 1336 : UploadOp::Barrier(sender) => {
1778 1336 : sender.send_replace(());
1779 1336 : continue;
1780 : }
1781 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1782 : };
1783 :
1784 : // Assign unique ID to this task
1785 3197 : upload_queue.task_counter += 1;
1786 3197 : let upload_task_id = upload_queue.task_counter;
1787 3197 :
1788 3197 : // Add it to the in-progress map
1789 3197 : let task = Arc::new(UploadTask {
1790 3197 : task_id: upload_task_id,
1791 3197 : op: next_op,
1792 3197 : retries: AtomicU32::new(0),
1793 3197 : });
1794 3197 : upload_queue
1795 3197 : .inprogress_tasks
1796 3197 : .insert(task.task_id, Arc::clone(&task));
1797 3197 :
1798 3197 : // Spawn task to perform the task
1799 3197 : let self_rc = Arc::clone(self);
1800 3197 : let tenant_shard_id = self.tenant_shard_id;
1801 3197 : let timeline_id = self.timeline_id;
1802 3197 : task_mgr::spawn(
1803 3197 : &self.runtime,
1804 3197 : TaskKind::RemoteUploadTask,
1805 3197 : self.tenant_shard_id,
1806 3197 : Some(self.timeline_id),
1807 3197 : "remote upload",
1808 3093 : async move {
1809 43244 : self_rc.perform_upload_task(task).await;
1810 2908 : Ok(())
1811 2908 : }
1812 3197 : .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)),
1813 : );
1814 :
1815 : // Loop back to process next task
1816 : }
1817 7445 : }
1818 :
1819 : ///
1820 : /// Perform an upload task.
1821 : ///
1822 : /// The task is in the `inprogress_tasks` list. This function will try to
1823 : /// execute it, retrying forever. On successful completion, the task is
1824 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1825 : /// queue that were waiting by the completion are launched.
1826 : ///
1827 : /// The task can be shut down, however. That leads to stopping the whole
1828 : /// queue.
1829 : ///
1830 3093 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1831 3093 : let cancel = shutdown_token();
1832 : // Loop to retry until it completes.
1833 : loop {
1834 : // If we're requested to shut down, close up shop and exit.
1835 : //
1836 : // Note: We only check for the shutdown requests between retries, so
1837 : // if a shutdown request arrives while we're busy uploading, in the
1838 : // upload::upload:*() call below, we will wait not exit until it has
1839 : // finished. We probably could cancel the upload by simply dropping
1840 : // the Future, but we're not 100% sure if the remote storage library
1841 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1842 : // upload finishes or times out soon enough.
1843 3093 : if cancel.is_cancelled() {
1844 0 : info!("upload task cancelled by shutdown request");
1845 0 : self.stop();
1846 0 : return;
1847 3093 : }
1848 :
1849 3093 : let upload_result: anyhow::Result<()> = match &task.op {
1850 1438 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1851 1438 : let local_path = layer.local_path();
1852 1438 :
1853 1438 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1854 1438 : // the metadata in the upload should always match our current generation.
1855 1438 : assert_eq!(layer_metadata.generation, self.generation);
1856 :
1857 1438 : let remote_path = remote_layer_path(
1858 1438 : &self.tenant_shard_id.tenant_id,
1859 1438 : &self.timeline_id,
1860 1438 : layer_metadata.shard,
1861 1438 : &layer.layer_desc().layer_name(),
1862 1438 : layer_metadata.generation,
1863 1438 : );
1864 1438 :
1865 1438 : upload::upload_timeline_layer(
1866 1438 : &self.storage_impl,
1867 1438 : local_path,
1868 1438 : &remote_path,
1869 1438 : layer_metadata.file_size,
1870 1438 : &self.cancel,
1871 1438 : )
1872 1438 : .measure_remote_op(
1873 1438 : RemoteOpFileKind::Layer,
1874 1438 : RemoteOpKind::Upload,
1875 1438 : Arc::clone(&self.metrics),
1876 1438 : )
1877 37942 : .await
1878 : }
1879 1414 : UploadOp::UploadMetadata { ref uploaded } => {
1880 1414 : let res = upload::upload_index_part(
1881 1414 : &self.storage_impl,
1882 1414 : &self.tenant_shard_id,
1883 1414 : &self.timeline_id,
1884 1414 : self.generation,
1885 1414 : uploaded,
1886 1414 : &self.cancel,
1887 1414 : )
1888 1414 : .measure_remote_op(
1889 1414 : RemoteOpFileKind::Index,
1890 1414 : RemoteOpKind::Upload,
1891 1414 : Arc::clone(&self.metrics),
1892 1414 : )
1893 5112 : .await;
1894 1410 : if res.is_ok() {
1895 1410 : self.update_remote_physical_size_gauge(Some(uploaded));
1896 1410 : let mention_having_future_layers = if cfg!(feature = "testing") {
1897 1410 : uploaded
1898 1410 : .layer_metadata
1899 1410 : .keys()
1900 17492 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
1901 : } else {
1902 0 : false
1903 : };
1904 1410 : if mention_having_future_layers {
1905 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1906 18 : tracing::info!(
1907 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
1908 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
1909 : );
1910 1392 : }
1911 0 : }
1912 1410 : res
1913 : }
1914 241 : UploadOp::Delete(delete) => {
1915 241 : pausable_failpoint!("before-delete-layer-pausable");
1916 241 : self.deletion_queue_client
1917 241 : .push_layers(
1918 241 : self.tenant_shard_id,
1919 241 : self.timeline_id,
1920 241 : self.generation,
1921 241 : delete.layers.clone(),
1922 241 : )
1923 0 : .await
1924 241 : .map_err(|e| anyhow::anyhow!(e))
1925 : }
1926 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1927 : // unreachable. Barrier operations are handled synchronously in
1928 : // launch_queued_tasks
1929 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1930 0 : break;
1931 : }
1932 : };
1933 :
1934 0 : match upload_result {
1935 : Ok(()) => {
1936 2908 : break;
1937 : }
1938 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1939 0 : // loop around to do the proper stopping
1940 0 : continue;
1941 : }
1942 0 : Err(e) => {
1943 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1944 0 :
1945 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1946 0 : // or other external reasons. Such issues are relatively regular, so log them
1947 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1948 0 : //
1949 0 : // (See similar logic for downloads in `download::download_retry`)
1950 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1951 0 : info!(
1952 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1953 0 : task.op, retries, e
1954 : );
1955 : } else {
1956 0 : warn!(
1957 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1958 0 : task.op, retries, e
1959 : );
1960 : }
1961 :
1962 : // sleep until it's time to retry, or we're cancelled
1963 0 : exponential_backoff(
1964 0 : retries,
1965 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1966 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1967 0 : &cancel,
1968 0 : )
1969 0 : .await;
1970 : }
1971 : }
1972 : }
1973 :
1974 2908 : let retries = task.retries.load(Ordering::SeqCst);
1975 2908 : if retries > 0 {
1976 0 : info!(
1977 0 : "remote task {} completed successfully after {} retries",
1978 0 : task.op, retries
1979 : );
1980 : } else {
1981 2908 : debug!("remote task {} completed successfully", task.op);
1982 : }
1983 :
1984 : // The task has completed successfully. Remove it from the in-progress list.
1985 2908 : let lsn_update = {
1986 2908 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1987 2908 : let upload_queue = match upload_queue_guard.deref_mut() {
1988 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1989 0 : UploadQueue::Stopped(_stopped) => {
1990 0 : None
1991 : },
1992 2908 : UploadQueue::Initialized(qi) => { Some(qi) }
1993 : };
1994 :
1995 2908 : let upload_queue = match upload_queue {
1996 2908 : Some(upload_queue) => upload_queue,
1997 : None => {
1998 0 : info!("another concurrent task already stopped the queue");
1999 0 : return;
2000 : }
2001 : };
2002 :
2003 2908 : upload_queue.inprogress_tasks.remove(&task.task_id);
2004 :
2005 2908 : let lsn_update = match task.op {
2006 : UploadOp::UploadLayer(_, _) => {
2007 1257 : upload_queue.num_inprogress_layer_uploads -= 1;
2008 1257 : None
2009 : }
2010 1410 : UploadOp::UploadMetadata { ref uploaded } => {
2011 1410 : upload_queue.num_inprogress_metadata_uploads -= 1;
2012 1410 :
2013 1410 : // the task id is reused as a monotonicity check for storing the "clean"
2014 1410 : // IndexPart.
2015 1410 : let last_updater = upload_queue.clean.1;
2016 1410 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
2017 1410 : let monotone = is_later || last_updater.is_none();
2018 :
2019 1410 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
2020 :
2021 : // not taking ownership is wasteful
2022 1410 : upload_queue.clean.0.clone_from(uploaded);
2023 1410 : upload_queue.clean.1 = Some(task.task_id);
2024 1410 :
2025 1410 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
2026 1410 :
2027 1410 : if self.generation.is_none() {
2028 : // Legacy mode: skip validating generation
2029 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
2030 0 : None
2031 : } else {
2032 1410 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
2033 : }
2034 : }
2035 : UploadOp::Delete(_) => {
2036 241 : upload_queue.num_inprogress_deletions -= 1;
2037 241 : None
2038 : }
2039 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
2040 : };
2041 :
2042 : // Launch any queued tasks that were unblocked by this one.
2043 2908 : self.launch_queued_tasks(upload_queue);
2044 2908 : lsn_update
2045 : };
2046 :
2047 2908 : if let Some((lsn, slot)) = lsn_update {
2048 : // Updates to the remote_consistent_lsn we advertise to pageservers
2049 : // are all routed through the DeletionQueue, to enforce important
2050 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2051 1410 : self.deletion_queue_client
2052 1410 : .update_remote_consistent_lsn(
2053 1410 : self.tenant_shard_id,
2054 1410 : self.timeline_id,
2055 1410 : self.generation,
2056 1410 : lsn,
2057 1410 : slot,
2058 1410 : )
2059 0 : .await;
2060 1498 : }
2061 :
2062 2908 : self.metric_end(&task.op);
2063 2908 : }
2064 :
2065 6383 : fn metric_impl(
2066 6383 : &self,
2067 6383 : op: &UploadOp,
2068 6383 : ) -> Option<(
2069 6383 : RemoteOpFileKind,
2070 6383 : RemoteOpKind,
2071 6383 : RemoteTimelineClientMetricsCallTrackSize,
2072 6383 : )> {
2073 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2074 6383 : let res = match op {
2075 2795 : UploadOp::UploadLayer(_, m) => (
2076 2795 : RemoteOpFileKind::Layer,
2077 2795 : RemoteOpKind::Upload,
2078 2795 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2079 2795 : ),
2080 2862 : UploadOp::UploadMetadata { .. } => (
2081 2862 : RemoteOpFileKind::Index,
2082 2862 : RemoteOpKind::Upload,
2083 2862 : DontTrackSize {
2084 2862 : reason: "metadata uploads are tiny",
2085 2862 : },
2086 2862 : ),
2087 718 : UploadOp::Delete(_delete) => (
2088 718 : RemoteOpFileKind::Layer,
2089 718 : RemoteOpKind::Delete,
2090 718 : DontTrackSize {
2091 718 : reason: "should we track deletes? positive or negative sign?",
2092 718 : },
2093 718 : ),
2094 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2095 : // we do not account these
2096 8 : return None;
2097 : }
2098 : };
2099 6375 : Some(res)
2100 6383 : }
2101 :
2102 3467 : fn metric_begin(&self, op: &UploadOp) {
2103 3467 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2104 3467 : Some(x) => x,
2105 0 : None => return,
2106 : };
2107 3467 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2108 3467 : guard.will_decrement_manually(); // in metric_end(), see right below
2109 3467 : }
2110 :
2111 2916 : fn metric_end(&self, op: &UploadOp) {
2112 2916 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2113 2908 : Some(x) => x,
2114 8 : None => return,
2115 : };
2116 2908 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2117 2916 : }
2118 :
2119 : /// Close the upload queue for new operations and cancel queued operations.
2120 : ///
2121 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2122 : ///
2123 : /// In-progress operations will still be running after this function returns.
2124 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2125 : /// to wait for them to complete, after calling this function.
2126 18 : pub(crate) fn stop(&self) {
2127 18 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2128 18 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2129 18 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2130 18 : let mut guard = self.upload_queue.lock().unwrap();
2131 18 : self.stop_impl(&mut guard);
2132 18 : }
2133 :
2134 18 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2135 18 : match &mut **guard {
2136 : UploadQueue::Uninitialized => {
2137 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2138 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2139 : }
2140 : UploadQueue::Stopped(_) => {
2141 : // nothing to do
2142 8 : info!("another concurrent task already shut down the queue");
2143 : }
2144 10 : UploadQueue::Initialized(initialized) => {
2145 10 : info!("shutting down upload queue");
2146 :
2147 : // Replace the queue with the Stopped state, taking ownership of the old
2148 : // Initialized queue. We will do some checks on it, and then drop it.
2149 10 : let qi = {
2150 : // Here we preserve working version of the upload queue for possible use during deletions.
2151 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2152 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2153 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2154 10 : let upload_queue_for_deletion = UploadQueueInitialized {
2155 10 : task_counter: 0,
2156 10 : dirty: initialized.dirty.clone(),
2157 10 : clean: initialized.clean.clone(),
2158 10 : latest_files_changes_since_metadata_upload_scheduled: 0,
2159 10 : visible_remote_consistent_lsn: initialized
2160 10 : .visible_remote_consistent_lsn
2161 10 : .clone(),
2162 10 : num_inprogress_layer_uploads: 0,
2163 10 : num_inprogress_metadata_uploads: 0,
2164 10 : num_inprogress_deletions: 0,
2165 10 : inprogress_tasks: HashMap::default(),
2166 10 : queued_operations: VecDeque::default(),
2167 10 : #[cfg(feature = "testing")]
2168 10 : dangling_files: HashMap::default(),
2169 10 : shutting_down: false,
2170 10 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2171 10 : };
2172 10 :
2173 10 : let upload_queue = std::mem::replace(
2174 10 : &mut **guard,
2175 10 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2176 10 : UploadQueueStoppedDeletable {
2177 10 : upload_queue_for_deletion,
2178 10 : deleted_at: SetDeletedFlagProgress::NotRunning,
2179 10 : },
2180 10 : )),
2181 10 : );
2182 10 : if let UploadQueue::Initialized(qi) = upload_queue {
2183 10 : qi
2184 : } else {
2185 0 : unreachable!("we checked in the match above that it is Initialized");
2186 : }
2187 : };
2188 :
2189 : // consistency check
2190 10 : assert_eq!(
2191 10 : qi.num_inprogress_layer_uploads
2192 10 : + qi.num_inprogress_metadata_uploads
2193 10 : + qi.num_inprogress_deletions,
2194 10 : qi.inprogress_tasks.len()
2195 10 : );
2196 :
2197 : // We don't need to do anything here for in-progress tasks. They will finish
2198 : // on their own, decrement the unfinished-task counter themselves, and observe
2199 : // that the queue is Stopped.
2200 10 : drop(qi.inprogress_tasks);
2201 :
2202 : // Tear down queued ops
2203 10 : for op in qi.queued_operations.into_iter() {
2204 8 : self.metric_end(&op);
2205 8 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2206 8 : // which is exactly what we want to happen.
2207 8 : drop(op);
2208 8 : }
2209 : }
2210 : }
2211 18 : }
2212 :
2213 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2214 : /// externally to RemoteTimelineClient.
2215 0 : pub(crate) fn initialized_upload_queue(
2216 0 : &self,
2217 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2218 0 : let mut inner = self.upload_queue.lock().unwrap();
2219 0 : inner.initialized_mut()?;
2220 0 : Ok(UploadQueueAccessor { inner })
2221 0 : }
2222 :
2223 8 : pub(crate) fn no_pending_work(&self) -> bool {
2224 8 : let inner = self.upload_queue.lock().unwrap();
2225 8 : match &*inner {
2226 : UploadQueue::Uninitialized
2227 0 : | UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => true,
2228 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(x)) => {
2229 8 : x.upload_queue_for_deletion.no_pending_work()
2230 : }
2231 0 : UploadQueue::Initialized(x) => x.no_pending_work(),
2232 : }
2233 8 : }
2234 : }
2235 :
2236 : pub(crate) struct UploadQueueAccessor<'a> {
2237 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2238 : }
2239 :
2240 : impl UploadQueueAccessor<'_> {
2241 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2242 0 : match &*self.inner {
2243 0 : UploadQueue::Initialized(x) => &x.clean.0,
2244 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2245 0 : unreachable!("checked before constructing")
2246 : }
2247 : }
2248 0 : }
2249 : }
2250 :
2251 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2252 0 : let path = format!("tenants/{tenant_shard_id}");
2253 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2254 0 : }
2255 :
2256 578 : pub fn remote_tenant_manifest_path(
2257 578 : tenant_shard_id: &TenantShardId,
2258 578 : generation: Generation,
2259 578 : ) -> RemotePath {
2260 578 : let path = format!(
2261 578 : "tenants/{tenant_shard_id}/tenant-manifest{}.json",
2262 578 : generation.get_suffix()
2263 578 : );
2264 578 : RemotePath::from_string(&path).expect("Failed to construct path")
2265 578 : }
2266 :
2267 : /// Prefix to all generations' manifest objects in a tenant shard
2268 192 : pub fn remote_tenant_manifest_prefix(tenant_shard_id: &TenantShardId) -> RemotePath {
2269 192 : let path = format!("tenants/{tenant_shard_id}/tenant-manifest",);
2270 192 : RemotePath::from_string(&path).expect("Failed to construct path")
2271 192 : }
2272 :
2273 222 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2274 222 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2275 222 : RemotePath::from_string(&path).expect("Failed to construct path")
2276 222 : }
2277 :
2278 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2279 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2280 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2281 0 : }
2282 :
2283 30 : pub fn remote_timeline_path(
2284 30 : tenant_shard_id: &TenantShardId,
2285 30 : timeline_id: &TimelineId,
2286 30 : ) -> RemotePath {
2287 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2288 30 : }
2289 :
2290 : /// Obtains the path of the given Layer in the remote
2291 : ///
2292 : /// Note that the shard component of a remote layer path is _not_ always the same
2293 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2294 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2295 1456 : pub fn remote_layer_path(
2296 1456 : tenant_id: &TenantId,
2297 1456 : timeline_id: &TimelineId,
2298 1456 : shard: ShardIndex,
2299 1456 : layer_file_name: &LayerName,
2300 1456 : generation: Generation,
2301 1456 : ) -> RemotePath {
2302 1456 : // Generation-aware key format
2303 1456 : let path = format!(
2304 1456 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2305 1456 : shard.get_suffix(),
2306 1456 : layer_file_name,
2307 1456 : generation.get_suffix()
2308 1456 : );
2309 1456 :
2310 1456 : RemotePath::from_string(&path).expect("Failed to construct path")
2311 1456 : }
2312 :
2313 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2314 4 : RemotePath::from_string(&format!(
2315 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2316 4 : ))
2317 4 : .expect("Failed to construct path")
2318 4 : }
2319 :
2320 2 : pub fn remote_initdb_preserved_archive_path(
2321 2 : tenant_id: &TenantId,
2322 2 : timeline_id: &TimelineId,
2323 2 : ) -> RemotePath {
2324 2 : RemotePath::from_string(&format!(
2325 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2326 2 : ))
2327 2 : .expect("Failed to construct path")
2328 2 : }
2329 :
2330 1480 : pub fn remote_index_path(
2331 1480 : tenant_shard_id: &TenantShardId,
2332 1480 : timeline_id: &TimelineId,
2333 1480 : generation: Generation,
2334 1480 : ) -> RemotePath {
2335 1480 : RemotePath::from_string(&format!(
2336 1480 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2337 1480 : IndexPart::FILE_NAME,
2338 1480 : generation.get_suffix()
2339 1480 : ))
2340 1480 : .expect("Failed to construct path")
2341 1480 : }
2342 :
2343 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2344 0 : RemotePath::from_string(&format!(
2345 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2346 0 : ))
2347 0 : .expect("Failed to construct path")
2348 0 : }
2349 :
2350 : /// Given the key of an index, parse out the generation part of the name
2351 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2352 18 : let file_name = match path.get_path().file_name() {
2353 18 : Some(f) => f,
2354 : None => {
2355 : // Unexpected: we should be seeing index_part.json paths only
2356 0 : tracing::warn!("Malformed index key {}", path);
2357 0 : return None;
2358 : }
2359 : };
2360 :
2361 18 : match file_name.split_once('-') {
2362 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2363 6 : None => None,
2364 : }
2365 18 : }
2366 :
2367 : /// Given the key of a tenant manifest, parse out the generation number
2368 0 : pub(crate) fn parse_remote_tenant_manifest_path(path: RemotePath) -> Option<Generation> {
2369 : static RE: OnceLock<Regex> = OnceLock::new();
2370 0 : let re = RE.get_or_init(|| Regex::new(r".+tenant-manifest-([0-9a-f]{8}).json").unwrap());
2371 0 : re.captures(path.get_path().as_str())
2372 0 : .and_then(|c| c.get(1))
2373 0 : .and_then(|m| Generation::parse_suffix(m.as_str()))
2374 0 : }
2375 :
2376 : #[cfg(test)]
2377 : mod tests {
2378 : use super::*;
2379 : use crate::{
2380 : context::RequestContext,
2381 : tenant::{
2382 : harness::{TenantHarness, TIMELINE_ID},
2383 : storage_layer::layer::local_layer_path,
2384 : Tenant, Timeline,
2385 : },
2386 : DEFAULT_PG_VERSION,
2387 : };
2388 :
2389 : use std::collections::HashSet;
2390 :
2391 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2392 8 : format!("contents for {name}").into()
2393 8 : }
2394 :
2395 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2396 2 : let metadata = TimelineMetadata::new(
2397 2 : disk_consistent_lsn,
2398 2 : None,
2399 2 : None,
2400 2 : Lsn(0),
2401 2 : Lsn(0),
2402 2 : Lsn(0),
2403 2 : // Any version will do
2404 2 : // but it should be consistent with the one in the tests
2405 2 : crate::DEFAULT_PG_VERSION,
2406 2 : );
2407 2 :
2408 2 : // go through serialize + deserialize to fix the header, including checksum
2409 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2410 2 : }
2411 :
2412 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2413 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2414 2 : avec.sort();
2415 2 :
2416 2 : let mut bvec = b.to_vec();
2417 2 : bvec.sort_unstable();
2418 2 :
2419 2 : assert_eq!(avec, bvec);
2420 2 : }
2421 :
2422 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2423 4 : let mut expected: Vec<String> = expected
2424 4 : .iter()
2425 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2426 4 : .collect();
2427 4 : expected.sort();
2428 4 :
2429 4 : let mut found: Vec<String> = Vec::new();
2430 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2431 16 : let entry_name = entry.file_name();
2432 16 : let fname = entry_name.to_str().unwrap();
2433 16 : found.push(String::from(fname));
2434 16 : }
2435 4 : found.sort();
2436 4 :
2437 4 : assert_eq!(found, expected);
2438 4 : }
2439 :
2440 : struct TestSetup {
2441 : harness: TenantHarness,
2442 : tenant: Arc<Tenant>,
2443 : timeline: Arc<Timeline>,
2444 : tenant_ctx: RequestContext,
2445 : }
2446 :
2447 : impl TestSetup {
2448 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2449 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2450 8 : let harness = TenantHarness::create(test_name).await?;
2451 80 : let (tenant, ctx) = harness.load().await;
2452 :
2453 8 : let timeline = tenant
2454 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2455 24 : .await?;
2456 :
2457 8 : Ok(Self {
2458 8 : harness,
2459 8 : tenant,
2460 8 : timeline,
2461 8 : tenant_ctx: ctx,
2462 8 : })
2463 8 : }
2464 :
2465 : /// Construct a RemoteTimelineClient in an arbitrary generation
2466 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2467 10 : Arc::new(RemoteTimelineClient {
2468 10 : conf: self.harness.conf,
2469 10 : runtime: tokio::runtime::Handle::current(),
2470 10 : tenant_shard_id: self.harness.tenant_shard_id,
2471 10 : timeline_id: TIMELINE_ID,
2472 10 : generation,
2473 10 : storage_impl: self.harness.remote_storage.clone(),
2474 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2475 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2476 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2477 10 : &self.harness.tenant_shard_id,
2478 10 : &TIMELINE_ID,
2479 10 : )),
2480 10 : cancel: CancellationToken::new(),
2481 10 : })
2482 10 : }
2483 :
2484 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2485 : /// and timeline_id are present.
2486 6 : fn span(&self) -> tracing::Span {
2487 6 : tracing::info_span!(
2488 : "test",
2489 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2490 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2491 : timeline_id = %TIMELINE_ID
2492 : )
2493 6 : }
2494 : }
2495 :
2496 : // Test scheduling
2497 : #[tokio::test]
2498 2 : async fn upload_scheduling() {
2499 2 : // Test outline:
2500 2 : //
2501 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2502 2 : // Schedule upload of index. Check that it is queued
2503 2 : // let the layer file uploads finish. Check that the index-upload is now started
2504 2 : // let the index-upload finish.
2505 2 : //
2506 2 : // Download back the index.json. Check that the list of files is correct
2507 2 : //
2508 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2509 2 : // let upload finish. Check that deletion is now started
2510 2 : // Schedule another deletion. Check that it's launched immediately.
2511 2 : // Schedule index upload. Check that it's queued
2512 2 :
2513 26 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2514 2 : let span = test_setup.span();
2515 2 : let _guard = span.enter();
2516 2 :
2517 2 : let TestSetup {
2518 2 : harness,
2519 2 : tenant: _tenant,
2520 2 : timeline,
2521 2 : tenant_ctx: _tenant_ctx,
2522 2 : } = test_setup;
2523 2 :
2524 2 : let client = &timeline.remote_client;
2525 2 :
2526 2 : // Download back the index.json, and check that the list of files is correct
2527 2 : let initial_index_part = match client
2528 2 : .download_index_file(&CancellationToken::new())
2529 6 : .await
2530 2 : .unwrap()
2531 2 : {
2532 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2533 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2534 2 : };
2535 2 : let initial_layers = initial_index_part
2536 2 : .layer_metadata
2537 2 : .keys()
2538 2 : .map(|f| f.to_owned())
2539 2 : .collect::<HashSet<LayerName>>();
2540 2 : let initial_layer = {
2541 2 : assert!(initial_layers.len() == 1);
2542 2 : initial_layers.into_iter().next().unwrap()
2543 2 : };
2544 2 :
2545 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2546 2 :
2547 2 : println!("workdir: {}", harness.conf.workdir);
2548 2 :
2549 2 : let remote_timeline_dir = harness
2550 2 : .remote_fs_dir
2551 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2552 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2553 2 :
2554 2 : let generation = harness.generation;
2555 2 : let shard = harness.shard;
2556 2 :
2557 2 : // Create a couple of dummy files, schedule upload for them
2558 2 :
2559 2 : let layers = [
2560 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2561 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2562 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2563 2 : ]
2564 2 : .into_iter()
2565 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2566 6 :
2567 6 : let local_path = local_layer_path(
2568 6 : harness.conf,
2569 6 : &timeline.tenant_shard_id,
2570 6 : &timeline.timeline_id,
2571 6 : &name,
2572 6 : &generation,
2573 6 : );
2574 6 : std::fs::write(&local_path, &contents).unwrap();
2575 6 :
2576 6 : Layer::for_resident(
2577 6 : harness.conf,
2578 6 : &timeline,
2579 6 : local_path,
2580 6 : name,
2581 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2582 6 : )
2583 6 : }).collect::<Vec<_>>();
2584 2 :
2585 2 : client
2586 2 : .schedule_layer_file_upload(layers[0].clone())
2587 2 : .unwrap();
2588 2 : client
2589 2 : .schedule_layer_file_upload(layers[1].clone())
2590 2 : .unwrap();
2591 2 :
2592 2 : // Check that they are started immediately, not queued
2593 2 : //
2594 2 : // this works because we running within block_on, so any futures are now queued up until
2595 2 : // our next await point.
2596 2 : {
2597 2 : let mut guard = client.upload_queue.lock().unwrap();
2598 2 : let upload_queue = guard.initialized_mut().unwrap();
2599 2 : assert!(upload_queue.queued_operations.is_empty());
2600 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2601 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2602 2 :
2603 2 : // also check that `latest_file_changes` was updated
2604 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2605 2 : }
2606 2 :
2607 2 : // Schedule upload of index. Check that it is queued
2608 2 : let metadata = dummy_metadata(Lsn(0x20));
2609 2 : client
2610 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2611 2 : .unwrap();
2612 2 : {
2613 2 : let mut guard = client.upload_queue.lock().unwrap();
2614 2 : let upload_queue = guard.initialized_mut().unwrap();
2615 2 : assert!(upload_queue.queued_operations.len() == 1);
2616 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2617 2 : }
2618 2 :
2619 2 : // Wait for the uploads to finish
2620 2 : client.wait_completion().await.unwrap();
2621 2 : {
2622 2 : let mut guard = client.upload_queue.lock().unwrap();
2623 2 : let upload_queue = guard.initialized_mut().unwrap();
2624 2 :
2625 2 : assert!(upload_queue.queued_operations.is_empty());
2626 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2627 2 : }
2628 2 :
2629 2 : // Download back the index.json, and check that the list of files is correct
2630 2 : let index_part = match client
2631 2 : .download_index_file(&CancellationToken::new())
2632 6 : .await
2633 2 : .unwrap()
2634 2 : {
2635 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2636 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2637 2 : };
2638 2 :
2639 2 : assert_file_list(
2640 2 : &index_part
2641 2 : .layer_metadata
2642 2 : .keys()
2643 6 : .map(|f| f.to_owned())
2644 2 : .collect(),
2645 2 : &[
2646 2 : &initial_layer.to_string(),
2647 2 : &layers[0].layer_desc().layer_name().to_string(),
2648 2 : &layers[1].layer_desc().layer_name().to_string(),
2649 2 : ],
2650 2 : );
2651 2 : assert_eq!(index_part.metadata, metadata);
2652 2 :
2653 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2654 2 : client
2655 2 : .schedule_layer_file_upload(layers[2].clone())
2656 2 : .unwrap();
2657 2 :
2658 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2659 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2660 2 : // spawn_blocking started by the drop.
2661 2 : client
2662 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2663 2 : .unwrap();
2664 2 : {
2665 2 : let mut guard = client.upload_queue.lock().unwrap();
2666 2 : let upload_queue = guard.initialized_mut().unwrap();
2667 2 :
2668 2 : // Deletion schedules upload of the index file, and the file deletion itself
2669 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2670 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2671 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2672 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2673 2 : assert_eq!(
2674 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2675 2 : 0
2676 2 : );
2677 2 : }
2678 2 : assert_remote_files(
2679 2 : &[
2680 2 : &initial_layer.to_string(),
2681 2 : &layers[0].layer_desc().layer_name().to_string(),
2682 2 : &layers[1].layer_desc().layer_name().to_string(),
2683 2 : "index_part.json",
2684 2 : ],
2685 2 : &remote_timeline_dir,
2686 2 : generation,
2687 2 : );
2688 2 :
2689 2 : // Finish them
2690 2 : client.wait_completion().await.unwrap();
2691 2 : harness.deletion_queue.pump().await;
2692 2 :
2693 2 : assert_remote_files(
2694 2 : &[
2695 2 : &initial_layer.to_string(),
2696 2 : &layers[1].layer_desc().layer_name().to_string(),
2697 2 : &layers[2].layer_desc().layer_name().to_string(),
2698 2 : "index_part.json",
2699 2 : ],
2700 2 : &remote_timeline_dir,
2701 2 : generation,
2702 2 : );
2703 2 : }
2704 :
2705 : #[tokio::test]
2706 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2707 2 : // Setup
2708 2 :
2709 2 : let TestSetup {
2710 2 : harness,
2711 2 : tenant: _tenant,
2712 2 : timeline,
2713 2 : ..
2714 26 : } = TestSetup::new("metrics").await.unwrap();
2715 2 : let client = &timeline.remote_client;
2716 2 :
2717 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2718 2 : let local_path = local_layer_path(
2719 2 : harness.conf,
2720 2 : &timeline.tenant_shard_id,
2721 2 : &timeline.timeline_id,
2722 2 : &layer_file_name_1,
2723 2 : &harness.generation,
2724 2 : );
2725 2 : let content_1 = dummy_contents("foo");
2726 2 : std::fs::write(&local_path, &content_1).unwrap();
2727 2 :
2728 2 : let layer_file_1 = Layer::for_resident(
2729 2 : harness.conf,
2730 2 : &timeline,
2731 2 : local_path,
2732 2 : layer_file_name_1.clone(),
2733 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2734 2 : );
2735 2 :
2736 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2737 2 : struct BytesStartedFinished {
2738 2 : started: Option<usize>,
2739 2 : finished: Option<usize>,
2740 2 : }
2741 2 : impl std::ops::Add for BytesStartedFinished {
2742 2 : type Output = Self;
2743 4 : fn add(self, rhs: Self) -> Self::Output {
2744 4 : Self {
2745 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2746 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2747 4 : }
2748 4 : }
2749 2 : }
2750 6 : let get_bytes_started_stopped = || {
2751 6 : let started = client
2752 6 : .metrics
2753 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2754 6 : .map(|v| v.try_into().unwrap());
2755 6 : let stopped = client
2756 6 : .metrics
2757 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2758 6 : .map(|v| v.try_into().unwrap());
2759 6 : BytesStartedFinished {
2760 6 : started,
2761 6 : finished: stopped,
2762 6 : }
2763 6 : };
2764 2 :
2765 2 : // Test
2766 2 : tracing::info!("now doing actual test");
2767 2 :
2768 2 : let actual_a = get_bytes_started_stopped();
2769 2 :
2770 2 : client
2771 2 : .schedule_layer_file_upload(layer_file_1.clone())
2772 2 : .unwrap();
2773 2 :
2774 2 : let actual_b = get_bytes_started_stopped();
2775 2 :
2776 2 : client.wait_completion().await.unwrap();
2777 2 :
2778 2 : let actual_c = get_bytes_started_stopped();
2779 2 :
2780 2 : // Validate
2781 2 :
2782 2 : let expected_b = actual_a
2783 2 : + BytesStartedFinished {
2784 2 : started: Some(content_1.len()),
2785 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2786 2 : finished: Some(0),
2787 2 : };
2788 2 : assert_eq!(actual_b, expected_b);
2789 2 :
2790 2 : let expected_c = actual_a
2791 2 : + BytesStartedFinished {
2792 2 : started: Some(content_1.len()),
2793 2 : finished: Some(content_1.len()),
2794 2 : };
2795 2 : assert_eq!(actual_c, expected_c);
2796 2 : }
2797 :
2798 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2799 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2800 12 : let example_index_part = IndexPart::example();
2801 12 :
2802 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2803 12 :
2804 12 : let index_path = test_state.harness.remote_fs_dir.join(
2805 12 : remote_index_path(
2806 12 : &test_state.harness.tenant_shard_id,
2807 12 : &TIMELINE_ID,
2808 12 : generation,
2809 12 : )
2810 12 : .get_path(),
2811 12 : );
2812 12 :
2813 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2814 12 : .expect("creating test dir should work");
2815 12 :
2816 12 : eprintln!("Writing {index_path}");
2817 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2818 12 : example_index_part
2819 12 : }
2820 :
2821 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2822 : /// index, the IndexPart returned is equal to `expected`
2823 10 : async fn assert_got_index_part(
2824 10 : test_state: &TestSetup,
2825 10 : get_generation: Generation,
2826 10 : expected: &IndexPart,
2827 10 : ) {
2828 10 : let client = test_state.build_client(get_generation);
2829 :
2830 10 : let download_r = client
2831 10 : .download_index_file(&CancellationToken::new())
2832 68 : .await
2833 10 : .expect("download should always succeed");
2834 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2835 10 : match download_r {
2836 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2837 10 : assert_eq!(&index_part, expected);
2838 : }
2839 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2840 : }
2841 10 : }
2842 :
2843 : #[tokio::test]
2844 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2845 26 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2846 2 : let span = test_state.span();
2847 2 : let _guard = span.enter();
2848 2 :
2849 2 : // Simple case: we are in generation N, load the index from generation N - 1
2850 2 : let generation_n = 5;
2851 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2852 2 :
2853 8 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2854 2 :
2855 2 : Ok(())
2856 2 : }
2857 :
2858 : #[tokio::test]
2859 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2860 2 : let test_state = TestSetup::new("index_part_download_ordering")
2861 26 : .await
2862 2 : .unwrap();
2863 2 :
2864 2 : let span = test_state.span();
2865 2 : let _guard = span.enter();
2866 2 :
2867 2 : // A generation-less IndexPart exists in the bucket, we should find it
2868 2 : let generation_n = 5;
2869 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2870 16 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2871 2 :
2872 2 : // If a more recent-than-none generation exists, we should prefer to load that
2873 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2874 18 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2875 2 :
2876 2 : // If a more-recent-than-me generation exists, we should ignore it.
2877 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2878 20 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2879 2 :
2880 2 : // If a directly previous generation exists, _and_ an index exists in my own
2881 2 : // generation, I should prefer my own generation.
2882 2 : let _injected_prev =
2883 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2884 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2885 2 : assert_got_index_part(
2886 2 : &test_state,
2887 2 : Generation::new(generation_n),
2888 2 : &injected_current,
2889 2 : )
2890 6 : .await;
2891 2 :
2892 2 : Ok(())
2893 2 : }
2894 : }
|
