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 : use utils::shard::ShardNumber;
201 :
202 : use std::collections::{HashMap, VecDeque};
203 : use std::sync::atomic::{AtomicU32, Ordering};
204 : use std::sync::{Arc, Mutex, OnceLock};
205 : use std::time::Duration;
206 :
207 : use remote_storage::{
208 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
209 : };
210 : use std::ops::DerefMut;
211 : use tracing::{debug, error, info, instrument, warn};
212 : use tracing::{info_span, Instrument};
213 : use utils::lsn::Lsn;
214 :
215 : use crate::context::RequestContext;
216 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
217 : use crate::metrics::{
218 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
219 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
220 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
221 : };
222 : use crate::task_mgr::shutdown_token;
223 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
224 : use crate::tenant::remote_timeline_client::download::download_retry;
225 : use crate::tenant::storage_layer::AsLayerDesc;
226 : use crate::tenant::upload_queue::{Delete, UploadQueueStoppedDeletable};
227 : use crate::tenant::TIMELINES_SEGMENT_NAME;
228 : use crate::{
229 : config::PageServerConf,
230 : task_mgr,
231 : task_mgr::TaskKind,
232 : task_mgr::BACKGROUND_RUNTIME,
233 : tenant::metadata::TimelineMetadata,
234 : tenant::upload_queue::{
235 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
236 : },
237 : TENANT_HEATMAP_BASENAME,
238 : };
239 :
240 : use utils::id::{TenantId, TimelineId};
241 :
242 : use self::index::IndexPart;
243 :
244 : use super::metadata::MetadataUpdate;
245 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
246 : use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
247 : use super::{DeleteTimelineError, Generation};
248 :
249 : pub(crate) use download::{
250 : download_index_part, download_tenant_manifest, is_temp_download_file,
251 : list_remote_tenant_shards, list_remote_timelines,
252 : };
253 : pub(crate) use index::LayerFileMetadata;
254 : pub(crate) use upload::upload_initdb_dir;
255 :
256 : // Occasional network issues and such can cause remote operations to fail, and
257 : // that's expected. If a download fails, we log it at info-level, and retry.
258 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
259 : // level instead, as repeated failures can mean a more serious problem. If it
260 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
261 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
262 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
263 :
264 : // Similarly log failed uploads and deletions at WARN level, after this many
265 : // retries. Uploads and deletions are retried forever, though.
266 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
267 :
268 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
269 :
270 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
271 :
272 : /// Default buffer size when interfacing with [`tokio::fs::File`].
273 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
274 :
275 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
276 : /// which we warn and skip.
277 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
278 :
279 : pub enum MaybeDeletedIndexPart {
280 : IndexPart(IndexPart),
281 : Deleted(IndexPart),
282 : }
283 :
284 0 : #[derive(Debug, thiserror::Error)]
285 : pub enum PersistIndexPartWithDeletedFlagError {
286 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
287 : AlreadyInProgress(NaiveDateTime),
288 : #[error("the deleted_flag was already set, value is {0:?}")]
289 : AlreadyDeleted(NaiveDateTime),
290 : #[error(transparent)]
291 : Other(#[from] anyhow::Error),
292 : }
293 :
294 0 : #[derive(Debug, thiserror::Error)]
295 : pub enum WaitCompletionError {
296 : #[error(transparent)]
297 : NotInitialized(NotInitialized),
298 : #[error("wait_completion aborted because upload queue was stopped")]
299 : UploadQueueShutDownOrStopped,
300 : }
301 :
302 0 : #[derive(Debug, thiserror::Error)]
303 : #[error("Upload queue either in unexpected state or hasn't downloaded manifest yet")]
304 : pub struct UploadQueueNotReadyError;
305 :
306 : /// A client for accessing a timeline's data in remote storage.
307 : ///
308 : /// This takes care of managing the number of connections, and balancing them
309 : /// across tenants. This also handles retries of failed uploads.
310 : ///
311 : /// Upload and delete requests are ordered so that before a deletion is
312 : /// performed, we wait for all preceding uploads to finish. This ensures sure
313 : /// that if you perform a compaction operation that reshuffles data in layer
314 : /// files, we don't have a transient state where the old files have already been
315 : /// deleted, but new files have not yet been uploaded.
316 : ///
317 : /// Similarly, this enforces an order between index-file uploads, and layer
318 : /// uploads. Before an index-file upload is performed, all preceding layer
319 : /// uploads must be finished.
320 : ///
321 : /// This also maintains a list of remote files, and automatically includes that
322 : /// in the index part file, whenever timeline metadata is uploaded.
323 : ///
324 : /// Downloads are not queued, they are performed immediately.
325 : pub struct RemoteTimelineClient {
326 : conf: &'static PageServerConf,
327 :
328 : runtime: tokio::runtime::Handle,
329 :
330 : tenant_shard_id: TenantShardId,
331 : timeline_id: TimelineId,
332 : generation: Generation,
333 :
334 : upload_queue: Mutex<UploadQueue>,
335 :
336 : pub(crate) metrics: Arc<RemoteTimelineClientMetrics>,
337 :
338 : storage_impl: GenericRemoteStorage,
339 :
340 : deletion_queue_client: DeletionQueueClient,
341 :
342 : cancel: CancellationToken,
343 : }
344 :
345 : impl RemoteTimelineClient {
346 : ///
347 : /// Create a remote storage client for given timeline
348 : ///
349 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
350 : /// by calling init_upload_queue.
351 : ///
352 418 : pub fn new(
353 418 : remote_storage: GenericRemoteStorage,
354 418 : deletion_queue_client: DeletionQueueClient,
355 418 : conf: &'static PageServerConf,
356 418 : tenant_shard_id: TenantShardId,
357 418 : timeline_id: TimelineId,
358 418 : generation: Generation,
359 418 : ) -> RemoteTimelineClient {
360 418 : RemoteTimelineClient {
361 418 : conf,
362 418 : runtime: if cfg!(test) {
363 : // remote_timeline_client.rs tests rely on current-thread runtime
364 418 : tokio::runtime::Handle::current()
365 : } else {
366 0 : BACKGROUND_RUNTIME.handle().clone()
367 : },
368 418 : tenant_shard_id,
369 418 : timeline_id,
370 418 : generation,
371 418 : storage_impl: remote_storage,
372 418 : deletion_queue_client,
373 418 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
374 418 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
375 418 : &tenant_shard_id,
376 418 : &timeline_id,
377 418 : )),
378 418 : cancel: CancellationToken::new(),
379 418 : }
380 418 : }
381 :
382 : /// Initialize the upload queue for a remote storage that already received
383 : /// an index file upload, i.e., it's not empty.
384 : /// The given `index_part` must be the one on the remote.
385 6 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
386 6 : let mut upload_queue = self.upload_queue.lock().unwrap();
387 6 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
388 6 : self.update_remote_physical_size_gauge(Some(index_part));
389 6 : info!(
390 0 : "initialized upload queue from remote index with {} layer files",
391 0 : index_part.layer_metadata.len()
392 : );
393 6 : Ok(())
394 6 : }
395 :
396 : /// Initialize the upload queue for the case where the remote storage is empty,
397 : /// i.e., it doesn't have an `IndexPart`.
398 412 : pub fn init_upload_queue_for_empty_remote(
399 412 : &self,
400 412 : local_metadata: &TimelineMetadata,
401 412 : ) -> anyhow::Result<()> {
402 412 : let mut upload_queue = self.upload_queue.lock().unwrap();
403 412 : upload_queue.initialize_empty_remote(local_metadata)?;
404 412 : self.update_remote_physical_size_gauge(None);
405 412 : info!("initialized upload queue as empty");
406 412 : Ok(())
407 412 : }
408 :
409 : /// Initialize the queue in stopped state. Used in startup path
410 : /// to continue deletion operation interrupted by pageserver crash or restart.
411 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
412 0 : &self,
413 0 : index_part: &IndexPart,
414 0 : ) -> anyhow::Result<()> {
415 : // FIXME: consider newtype for DeletedIndexPart.
416 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
417 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
418 0 : ))?;
419 :
420 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
421 0 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
422 0 : self.update_remote_physical_size_gauge(Some(index_part));
423 0 : self.stop_impl(&mut upload_queue);
424 0 :
425 0 : upload_queue
426 0 : .stopped_mut()
427 0 : .expect("stopped above")
428 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
429 0 :
430 0 : Ok(())
431 0 : }
432 :
433 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
434 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
435 0 : match &mut *self.upload_queue.lock().unwrap() {
436 0 : UploadQueue::Uninitialized => None,
437 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
438 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
439 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
440 0 : .upload_queue_for_deletion
441 0 : .get_last_remote_consistent_lsn_projected(),
442 : }
443 0 : }
444 :
445 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
446 0 : match &mut *self.upload_queue.lock().unwrap() {
447 0 : UploadQueue::Uninitialized => None,
448 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
449 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
450 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
451 0 : q.upload_queue_for_deletion
452 0 : .get_last_remote_consistent_lsn_visible(),
453 0 : ),
454 : }
455 0 : }
456 :
457 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
458 : /// client is currently initialized.
459 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
460 0 : self.upload_queue
461 0 : .lock()
462 0 : .unwrap()
463 0 : .initialized_mut()
464 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
465 0 : .unwrap_or(false)
466 0 : }
467 :
468 : /// Returns whether the timeline is archived.
469 : /// Return None if the remote index_part hasn't been downloaded yet.
470 2 : pub(crate) fn is_archived(&self) -> Option<bool> {
471 2 : self.upload_queue
472 2 : .lock()
473 2 : .unwrap()
474 2 : .initialized_mut()
475 2 : .map(|q| q.clean.0.archived_at.is_some())
476 2 : .ok()
477 2 : }
478 :
479 : /// Returns `Ok(Some(timestamp))` if the timeline has been archived, `Ok(None)` if the timeline hasn't been archived.
480 : ///
481 : /// Return Err(_) if the remote index_part hasn't been downloaded yet, or the timeline hasn't been stopped yet.
482 2 : pub(crate) fn archived_at_stopped_queue(
483 2 : &self,
484 2 : ) -> Result<Option<NaiveDateTime>, UploadQueueNotReadyError> {
485 2 : self.upload_queue
486 2 : .lock()
487 2 : .unwrap()
488 2 : .stopped_mut()
489 2 : .map(|q| q.upload_queue_for_deletion.clean.0.archived_at)
490 2 : .map_err(|_| UploadQueueNotReadyError)
491 2 : }
492 :
493 1825 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
494 1825 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
495 1413 : current_remote_index_part
496 1413 : .layer_metadata
497 1413 : .values()
498 17522 : .map(|ilmd| ilmd.file_size)
499 1413 : .sum()
500 : } else {
501 412 : 0
502 : };
503 1825 : self.metrics.remote_physical_size_gauge.set(size);
504 1825 : }
505 :
506 2 : pub fn get_remote_physical_size(&self) -> u64 {
507 2 : self.metrics.remote_physical_size_gauge.get()
508 2 : }
509 :
510 : //
511 : // Download operations.
512 : //
513 : // These don't use the per-timeline queue. They do use the global semaphore in
514 : // S3Bucket, to limit the total number of concurrent operations, though.
515 : //
516 :
517 : /// Download index file
518 20 : pub async fn download_index_file(
519 20 : &self,
520 20 : cancel: &CancellationToken,
521 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
522 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
523 20 : &RemoteOpFileKind::Index,
524 20 : &RemoteOpKind::Download,
525 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
526 20 : reason: "no need for a downloads gauge",
527 20 : },
528 20 : );
529 :
530 20 : let (index_part, index_generation, index_last_modified) = download::download_index_part(
531 20 : &self.storage_impl,
532 20 : &self.tenant_shard_id,
533 20 : &self.timeline_id,
534 20 : self.generation,
535 20 : cancel,
536 20 : )
537 20 : .measure_remote_op(
538 20 : RemoteOpFileKind::Index,
539 20 : RemoteOpKind::Download,
540 20 : Arc::clone(&self.metrics),
541 20 : )
542 98 : .await?;
543 :
544 : // Defense in depth: monotonicity of generation numbers is an important correctness guarantee, so when we see a very
545 : // old index, we do extra checks in case this is the result of backward time-travel of the generation number (e.g.
546 : // in case of a bug in the service that issues generation numbers). Indices are allowed to be old, but we expect that
547 : // when we load an old index we are loading the _latest_ index: if we are asked to load an old index and there is
548 : // also a newer index available, that is surprising.
549 : const INDEX_AGE_CHECKS_THRESHOLD: Duration = Duration::from_secs(14 * 24 * 3600);
550 20 : let index_age = index_last_modified.elapsed().unwrap_or_else(|e| {
551 0 : if e.duration() > Duration::from_secs(5) {
552 : // We only warn if the S3 clock and our local clock are >5s out: because this is a low resolution
553 : // timestamp, it is common to be out by at least 1 second.
554 0 : tracing::warn!("Index has modification time in the future: {e}");
555 0 : }
556 0 : Duration::ZERO
557 20 : });
558 20 : if index_age > INDEX_AGE_CHECKS_THRESHOLD {
559 0 : tracing::info!(
560 : ?index_generation,
561 0 : age = index_age.as_secs_f64(),
562 0 : "Loaded an old index, checking for other indices..."
563 : );
564 :
565 : // Find the highest-generation index
566 0 : let (_latest_index_part, latest_index_generation, latest_index_mtime) =
567 0 : download::download_index_part(
568 0 : &self.storage_impl,
569 0 : &self.tenant_shard_id,
570 0 : &self.timeline_id,
571 0 : Generation::MAX,
572 0 : cancel,
573 0 : )
574 0 : .await?;
575 :
576 0 : if latest_index_generation > index_generation {
577 : // Unexpected! Why are we loading such an old index if a more recent one exists?
578 : // We will refuse to proceed, as there is no reasonable scenario where this should happen, but
579 : // there _is_ a clear bug/corruption scenario where it would happen (controller sets the generation
580 : // backwards).
581 0 : tracing::error!(
582 : ?index_generation,
583 : ?latest_index_generation,
584 : ?latest_index_mtime,
585 0 : "Found a newer index while loading an old one"
586 : );
587 0 : return Err(DownloadError::Fatal(
588 0 : "Index age exceeds threshold and a newer index exists".into(),
589 0 : ));
590 0 : }
591 20 : }
592 :
593 20 : if index_part.deleted_at.is_some() {
594 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
595 : } else {
596 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
597 : }
598 20 : }
599 :
600 : /// Download a (layer) file from `path`, into local filesystem.
601 : ///
602 : /// 'layer_metadata' is the metadata from the remote index file.
603 : ///
604 : /// On success, returns the size of the downloaded file.
605 6 : pub async fn download_layer_file(
606 6 : &self,
607 6 : layer_file_name: &LayerName,
608 6 : layer_metadata: &LayerFileMetadata,
609 6 : local_path: &Utf8Path,
610 6 : cancel: &CancellationToken,
611 6 : ctx: &RequestContext,
612 6 : ) -> Result<u64, DownloadError> {
613 6 : let downloaded_size = {
614 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
615 6 : &RemoteOpFileKind::Layer,
616 6 : &RemoteOpKind::Download,
617 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
618 6 : reason: "no need for a downloads gauge",
619 6 : },
620 6 : );
621 6 : download::download_layer_file(
622 6 : self.conf,
623 6 : &self.storage_impl,
624 6 : self.tenant_shard_id,
625 6 : self.timeline_id,
626 6 : layer_file_name,
627 6 : layer_metadata,
628 6 : local_path,
629 6 : cancel,
630 6 : ctx,
631 6 : )
632 6 : .measure_remote_op(
633 6 : RemoteOpFileKind::Layer,
634 6 : RemoteOpKind::Download,
635 6 : Arc::clone(&self.metrics),
636 6 : )
637 74 : .await?
638 : };
639 :
640 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
641 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
642 6 :
643 6 : Ok(downloaded_size)
644 6 : }
645 :
646 : //
647 : // Upload operations.
648 : //
649 :
650 : /// Launch an index-file upload operation in the background, with
651 : /// fully updated metadata.
652 : ///
653 : /// This should only be used to upload initial metadata to remote storage.
654 : ///
655 : /// The upload will be added to the queue immediately, but it
656 : /// won't be performed until all previously scheduled layer file
657 : /// upload operations have completed successfully. This is to
658 : /// ensure that when the index file claims that layers X, Y and Z
659 : /// exist in remote storage, they really do. To wait for the upload
660 : /// to complete, use `wait_completion`.
661 : ///
662 : /// If there were any changes to the list of files, i.e. if any
663 : /// layer file uploads were scheduled, since the last index file
664 : /// upload, those will be included too.
665 230 : pub fn schedule_index_upload_for_full_metadata_update(
666 230 : self: &Arc<Self>,
667 230 : metadata: &TimelineMetadata,
668 230 : ) -> anyhow::Result<()> {
669 230 : let mut guard = self.upload_queue.lock().unwrap();
670 230 : let upload_queue = guard.initialized_mut()?;
671 :
672 : // As documented in the struct definition, it's ok for latest_metadata to be
673 : // ahead of what's _actually_ on the remote during index upload.
674 230 : upload_queue.dirty.metadata = metadata.clone();
675 230 :
676 230 : self.schedule_index_upload(upload_queue)?;
677 :
678 230 : Ok(())
679 230 : }
680 :
681 : /// Launch an index-file upload operation in the background, with only parts of the metadata
682 : /// updated.
683 : ///
684 : /// This is the regular way of updating metadata on layer flushes or Gc.
685 : ///
686 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
687 : /// `index_part.json`, while being more clear on what values update regularly.
688 1146 : pub(crate) fn schedule_index_upload_for_metadata_update(
689 1146 : self: &Arc<Self>,
690 1146 : update: &MetadataUpdate,
691 1146 : ) -> anyhow::Result<()> {
692 1146 : let mut guard = self.upload_queue.lock().unwrap();
693 1146 : let upload_queue = guard.initialized_mut()?;
694 :
695 1146 : upload_queue.dirty.metadata.apply(update);
696 1146 :
697 1146 : self.schedule_index_upload(upload_queue)?;
698 :
699 1146 : Ok(())
700 1146 : }
701 :
702 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
703 : ///
704 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
705 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
706 : /// been in the queue yet.
707 2 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
708 2 : self: &Arc<Self>,
709 2 : state: TimelineArchivalState,
710 2 : ) -> anyhow::Result<bool> {
711 2 : let mut guard = self.upload_queue.lock().unwrap();
712 2 : let upload_queue = guard.initialized_mut()?;
713 :
714 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
715 : /// change needed to set archived_at.
716 4 : fn need_change(
717 4 : archived_at: &Option<NaiveDateTime>,
718 4 : state: TimelineArchivalState,
719 4 : ) -> Option<bool> {
720 4 : match (archived_at, state) {
721 : (Some(_), TimelineArchivalState::Archived)
722 : | (None, TimelineArchivalState::Unarchived) => {
723 : // Nothing to do
724 0 : tracing::info!("intended state matches present state");
725 0 : None
726 : }
727 4 : (None, TimelineArchivalState::Archived) => Some(true),
728 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
729 : }
730 4 : }
731 2 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
732 :
733 2 : if let Some(archived_at_set) = need_upload_scheduled {
734 2 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
735 2 : upload_queue.dirty.archived_at = intended_archived_at;
736 2 : self.schedule_index_upload(upload_queue)?;
737 0 : }
738 :
739 2 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
740 2 : Ok(need_wait)
741 2 : }
742 :
743 : ///
744 : /// Launch an index-file upload operation in the background, if necessary.
745 : ///
746 : /// Use this function to schedule the update of the index file after
747 : /// scheduling file uploads or deletions. If no file uploads or deletions
748 : /// have been scheduled since the last index file upload, this does
749 : /// nothing.
750 : ///
751 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
752 : /// the upload to the upload queue and returns quickly.
753 370 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
754 370 : let mut guard = self.upload_queue.lock().unwrap();
755 370 : let upload_queue = guard.initialized_mut()?;
756 :
757 370 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
758 14 : self.schedule_index_upload(upload_queue)?;
759 356 : }
760 :
761 370 : Ok(())
762 370 : }
763 :
764 : /// Launch an index-file upload operation in the background (internal function)
765 1452 : fn schedule_index_upload(
766 1452 : self: &Arc<Self>,
767 1452 : upload_queue: &mut UploadQueueInitialized,
768 1452 : ) -> Result<(), NotInitialized> {
769 1452 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
770 1452 : // fix up the duplicated field
771 1452 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
772 1452 :
773 1452 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
774 1452 : // look like a retryable error
775 1452 : let void = std::io::sink();
776 1452 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
777 1452 :
778 1452 : let index_part = &upload_queue.dirty;
779 1452 :
780 1452 : info!(
781 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
782 0 : index_part.layer_metadata.len(),
783 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
784 : );
785 :
786 1452 : let op = UploadOp::UploadMetadata {
787 1452 : uploaded: Box::new(index_part.clone()),
788 1452 : };
789 1452 : self.metric_begin(&op);
790 1452 : upload_queue.queued_operations.push_back(op);
791 1452 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
792 1452 :
793 1452 : // Launch the task immediately, if possible
794 1452 : self.launch_queued_tasks(upload_queue);
795 1452 : Ok(())
796 1452 : }
797 :
798 : /// Reparent this timeline to a new parent.
799 : ///
800 : /// A retryable step of timeline ancestor detach.
801 0 : pub(crate) async fn schedule_reparenting_and_wait(
802 0 : self: &Arc<Self>,
803 0 : new_parent: &TimelineId,
804 0 : ) -> anyhow::Result<()> {
805 0 : let receiver = {
806 0 : let mut guard = self.upload_queue.lock().unwrap();
807 0 : let upload_queue = guard.initialized_mut()?;
808 :
809 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
810 0 : return Err(anyhow::anyhow!(
811 0 : "cannot reparent without a current ancestor"
812 0 : ));
813 : };
814 :
815 0 : let uploaded = &upload_queue.clean.0.metadata;
816 0 :
817 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
818 : // nothing to do
819 0 : None
820 : } else {
821 0 : upload_queue.dirty.metadata.reparent(new_parent);
822 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
823 0 :
824 0 : self.schedule_index_upload(upload_queue)?;
825 :
826 0 : Some(self.schedule_barrier0(upload_queue))
827 : }
828 : };
829 :
830 0 : if let Some(receiver) = receiver {
831 0 : Self::wait_completion0(receiver).await?;
832 0 : }
833 0 : Ok(())
834 0 : }
835 :
836 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
837 : /// detaching from ancestor and waits for it to complete.
838 : ///
839 : /// This is used with `Timeline::detach_ancestor` functionality.
840 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
841 0 : self: &Arc<Self>,
842 0 : layers: &[Layer],
843 0 : adopted: (TimelineId, Lsn),
844 0 : ) -> anyhow::Result<()> {
845 0 : let barrier = {
846 0 : let mut guard = self.upload_queue.lock().unwrap();
847 0 : let upload_queue = guard.initialized_mut()?;
848 :
849 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
850 0 : None
851 : } else {
852 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
853 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
854 :
855 0 : for layer in layers {
856 0 : let prev = upload_queue
857 0 : .dirty
858 0 : .layer_metadata
859 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
860 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
861 : }
862 :
863 0 : self.schedule_index_upload(upload_queue)?;
864 :
865 0 : Some(self.schedule_barrier0(upload_queue))
866 : }
867 : };
868 :
869 0 : if let Some(barrier) = barrier {
870 0 : Self::wait_completion0(barrier).await?;
871 0 : }
872 0 : Ok(())
873 0 : }
874 :
875 : /// Adds a gc blocking reason for this timeline if one does not exist already.
876 : ///
877 : /// A retryable step of timeline detach ancestor.
878 : ///
879 : /// Returns a future which waits until the completion of the upload.
880 0 : pub(crate) fn schedule_insert_gc_block_reason(
881 0 : self: &Arc<Self>,
882 0 : reason: index::GcBlockingReason,
883 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
884 0 : {
885 0 : let maybe_barrier = {
886 0 : let mut guard = self.upload_queue.lock().unwrap();
887 0 : let upload_queue = guard.initialized_mut()?;
888 :
889 0 : if let index::GcBlockingReason::DetachAncestor = reason {
890 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
891 0 : drop(guard);
892 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
893 0 : }
894 0 : }
895 :
896 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
897 :
898 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
899 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
900 0 :
901 0 : match (current, uploaded) {
902 0 : (x, y) if wanted(x) && wanted(y) => None,
903 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
904 : // Usual case: !wanted(x) && !wanted(y)
905 : //
906 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
907 : // turn on and off some reason.
908 0 : (x, y) => {
909 0 : if !wanted(x) && wanted(y) {
910 : // this could be avoided by having external in-memory synchronization, like
911 : // timeline detach ancestor
912 0 : warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
913 0 : }
914 :
915 : // at this point, the metadata must always show that there is a parent
916 0 : upload_queue.dirty.gc_blocking = current
917 0 : .map(|x| x.with_reason(reason))
918 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
919 0 : self.schedule_index_upload(upload_queue)?;
920 0 : Some(self.schedule_barrier0(upload_queue))
921 : }
922 : }
923 : };
924 :
925 0 : Ok(async move {
926 0 : if let Some(barrier) = maybe_barrier {
927 0 : Self::wait_completion0(barrier).await?;
928 0 : }
929 0 : Ok(())
930 0 : })
931 0 : }
932 :
933 : /// Removes a gc blocking reason for this timeline if one exists.
934 : ///
935 : /// A retryable step of timeline detach ancestor.
936 : ///
937 : /// Returns a future which waits until the completion of the upload.
938 0 : pub(crate) fn schedule_remove_gc_block_reason(
939 0 : self: &Arc<Self>,
940 0 : reason: index::GcBlockingReason,
941 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
942 0 : {
943 0 : let maybe_barrier = {
944 0 : let mut guard = self.upload_queue.lock().unwrap();
945 0 : let upload_queue = guard.initialized_mut()?;
946 :
947 0 : if let index::GcBlockingReason::DetachAncestor = reason {
948 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
949 0 : drop(guard);
950 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
951 0 : }
952 0 : }
953 :
954 0 : let wanted = |x: Option<&index::GcBlocking>| {
955 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
956 0 : };
957 :
958 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
959 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
960 0 :
961 0 : match (current, uploaded) {
962 0 : (x, y) if wanted(x) && wanted(y) => None,
963 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
964 0 : (x, y) => {
965 0 : if !wanted(x) && wanted(y) {
966 0 : warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
967 0 : }
968 :
969 0 : upload_queue.dirty.gc_blocking =
970 0 : current.as_ref().and_then(|x| x.without_reason(reason));
971 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
972 : // FIXME: bogus ?
973 0 : self.schedule_index_upload(upload_queue)?;
974 0 : Some(self.schedule_barrier0(upload_queue))
975 : }
976 : }
977 : };
978 :
979 0 : Ok(async move {
980 0 : if let Some(barrier) = maybe_barrier {
981 0 : Self::wait_completion0(barrier).await?;
982 0 : }
983 0 : Ok(())
984 0 : })
985 0 : }
986 :
987 : /// Launch an upload operation in the background; the file is added to be included in next
988 : /// `index_part.json` upload.
989 1190 : pub(crate) fn schedule_layer_file_upload(
990 1190 : self: &Arc<Self>,
991 1190 : layer: ResidentLayer,
992 1190 : ) -> Result<(), NotInitialized> {
993 1190 : let mut guard = self.upload_queue.lock().unwrap();
994 1190 : let upload_queue = guard.initialized_mut()?;
995 :
996 1190 : self.schedule_layer_file_upload0(upload_queue, layer);
997 1190 : self.launch_queued_tasks(upload_queue);
998 1190 : Ok(())
999 1190 : }
1000 :
1001 1538 : fn schedule_layer_file_upload0(
1002 1538 : self: &Arc<Self>,
1003 1538 : upload_queue: &mut UploadQueueInitialized,
1004 1538 : layer: ResidentLayer,
1005 1538 : ) {
1006 1538 : let metadata = layer.metadata();
1007 1538 :
1008 1538 : upload_queue
1009 1538 : .dirty
1010 1538 : .layer_metadata
1011 1538 : .insert(layer.layer_desc().layer_name(), metadata.clone());
1012 1538 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1013 1538 :
1014 1538 : info!(
1015 : gen=?metadata.generation,
1016 : shard=?metadata.shard,
1017 0 : "scheduled layer file upload {layer}",
1018 : );
1019 :
1020 1538 : let op = UploadOp::UploadLayer(layer, metadata);
1021 1538 : self.metric_begin(&op);
1022 1538 : upload_queue.queued_operations.push_back(op);
1023 1538 : }
1024 :
1025 : /// Launch a delete operation in the background.
1026 : ///
1027 : /// The operation does not modify local filesystem state.
1028 : ///
1029 : /// Note: This schedules an index file upload before the deletions. The
1030 : /// deletion won't actually be performed, until all previously scheduled
1031 : /// upload operations, and the index file upload, have completed
1032 : /// successfully.
1033 8 : pub fn schedule_layer_file_deletion(
1034 8 : self: &Arc<Self>,
1035 8 : names: &[LayerName],
1036 8 : ) -> anyhow::Result<()> {
1037 8 : let mut guard = self.upload_queue.lock().unwrap();
1038 8 : let upload_queue = guard.initialized_mut()?;
1039 :
1040 8 : let with_metadata = self
1041 8 : .schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned())?;
1042 :
1043 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1044 8 :
1045 8 : // Launch the tasks immediately, if possible
1046 8 : self.launch_queued_tasks(upload_queue);
1047 8 : Ok(())
1048 8 : }
1049 :
1050 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1051 : /// layer files, leaving them dangling.
1052 : ///
1053 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1054 : /// is invoked on them.
1055 4 : pub(crate) fn schedule_gc_update(
1056 4 : self: &Arc<Self>,
1057 4 : gc_layers: &[Layer],
1058 4 : ) -> Result<(), NotInitialized> {
1059 4 : let mut guard = self.upload_queue.lock().unwrap();
1060 4 : let upload_queue = guard.initialized_mut()?;
1061 :
1062 : // just forget the return value; after uploading the next index_part.json, we can consider
1063 : // the layer files as "dangling". this is fine, at worst case we create work for the
1064 : // scrubber.
1065 :
1066 4 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1067 4 :
1068 4 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1069 :
1070 4 : self.launch_queued_tasks(upload_queue);
1071 4 :
1072 4 : Ok(())
1073 4 : }
1074 :
1075 : /// Update the remote index file, removing the to-be-deleted files from the index,
1076 : /// allowing scheduling of actual deletions later.
1077 76 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1078 76 : self: &Arc<Self>,
1079 76 : upload_queue: &mut UploadQueueInitialized,
1080 76 : names: I,
1081 76 : ) -> Result<Vec<(LayerName, LayerFileMetadata)>, NotInitialized>
1082 76 : where
1083 76 : I: IntoIterator<Item = LayerName>,
1084 76 : {
1085 76 : // Decorate our list of names with each name's metadata, dropping
1086 76 : // names that are unexpectedly missing from our metadata. This metadata
1087 76 : // is later used when physically deleting layers, to construct key paths.
1088 76 : let with_metadata: Vec<_> = names
1089 76 : .into_iter()
1090 482 : .filter_map(|name| {
1091 482 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1092 :
1093 482 : if let Some(meta) = meta {
1094 436 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1095 436 : Some((name, meta))
1096 : } else {
1097 : // This can only happen if we forgot to to schedule the file upload
1098 : // before scheduling the delete. Log it because it is a rare/strange
1099 : // situation, and in case something is misbehaving, we'd like to know which
1100 : // layers experienced this.
1101 46 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1102 46 : None
1103 : }
1104 482 : })
1105 76 : .collect();
1106 :
1107 : #[cfg(feature = "testing")]
1108 512 : for (name, metadata) in &with_metadata {
1109 436 : let gen = metadata.generation;
1110 436 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
1111 0 : if unexpected == gen {
1112 0 : tracing::error!("{name} was unlinked twice with same generation");
1113 : } else {
1114 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
1115 : }
1116 436 : }
1117 : }
1118 :
1119 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1120 : // index_part update, because that needs to be uploaded before we can actually delete the
1121 : // files.
1122 76 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1123 60 : self.schedule_index_upload(upload_queue)?;
1124 16 : }
1125 :
1126 76 : Ok(with_metadata)
1127 76 : }
1128 :
1129 : /// Schedules deletion for layer files which have previously been unlinked from the
1130 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1131 481 : pub(crate) fn schedule_deletion_of_unlinked(
1132 481 : self: &Arc<Self>,
1133 481 : layers: Vec<(LayerName, LayerFileMetadata)>,
1134 481 : ) -> anyhow::Result<()> {
1135 481 : let mut guard = self.upload_queue.lock().unwrap();
1136 481 : let upload_queue = guard.initialized_mut()?;
1137 :
1138 481 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1139 481 : self.launch_queued_tasks(upload_queue);
1140 481 : Ok(())
1141 481 : }
1142 :
1143 489 : fn schedule_deletion_of_unlinked0(
1144 489 : self: &Arc<Self>,
1145 489 : upload_queue: &mut UploadQueueInitialized,
1146 489 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1147 489 : ) {
1148 489 : // Filter out any layers which were not created by this tenant shard. These are
1149 489 : // layers that originate from some ancestor shard after a split, and may still
1150 489 : // be referenced by other shards. We are free to delete them locally and remove
1151 489 : // them from our index (and would have already done so when we reach this point
1152 489 : // in the code), but we may not delete them remotely.
1153 489 : with_metadata.retain(|(name, meta)| {
1154 483 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1155 483 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1156 483 : if !retain {
1157 0 : tracing::debug!(
1158 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1159 : meta.shard
1160 : );
1161 483 : }
1162 483 : retain
1163 489 : });
1164 :
1165 972 : for (name, meta) in &with_metadata {
1166 483 : info!(
1167 0 : "scheduling deletion of layer {}{} (shard {})",
1168 0 : name,
1169 0 : meta.generation.get_suffix(),
1170 : meta.shard
1171 : );
1172 : }
1173 :
1174 : #[cfg(feature = "testing")]
1175 972 : for (name, meta) in &with_metadata {
1176 483 : let gen = meta.generation;
1177 483 : match upload_queue.dangling_files.remove(name) {
1178 432 : Some(same) if same == gen => { /* expected */ }
1179 0 : Some(other) => {
1180 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
1181 : }
1182 : None => {
1183 51 : tracing::error!("{name} was unlinked but was not dangling");
1184 : }
1185 : }
1186 : }
1187 :
1188 : // schedule the actual deletions
1189 489 : if with_metadata.is_empty() {
1190 : // avoid scheduling the op & bumping the metric
1191 6 : return;
1192 483 : }
1193 483 : let op = UploadOp::Delete(Delete {
1194 483 : layers: with_metadata,
1195 483 : });
1196 483 : self.metric_begin(&op);
1197 483 : upload_queue.queued_operations.push_back(op);
1198 489 : }
1199 :
1200 : /// Schedules a compaction update to the remote `index_part.json`.
1201 : ///
1202 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1203 64 : pub(crate) fn schedule_compaction_update(
1204 64 : self: &Arc<Self>,
1205 64 : compacted_from: &[Layer],
1206 64 : compacted_to: &[ResidentLayer],
1207 64 : ) -> Result<(), NotInitialized> {
1208 64 : let mut guard = self.upload_queue.lock().unwrap();
1209 64 : let upload_queue = guard.initialized_mut()?;
1210 :
1211 412 : for layer in compacted_to {
1212 348 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1213 348 : }
1214 :
1215 476 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1216 64 :
1217 64 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
1218 64 : self.launch_queued_tasks(upload_queue);
1219 64 :
1220 64 : Ok(())
1221 64 : }
1222 :
1223 : /// Wait for all previously scheduled uploads/deletions to complete
1224 1330 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1225 1330 : let receiver = {
1226 1330 : let mut guard = self.upload_queue.lock().unwrap();
1227 1330 : let upload_queue = guard
1228 1330 : .initialized_mut()
1229 1330 : .map_err(WaitCompletionError::NotInitialized)?;
1230 1330 : self.schedule_barrier0(upload_queue)
1231 1330 : };
1232 1330 :
1233 1330 : Self::wait_completion0(receiver).await
1234 1330 : }
1235 :
1236 1330 : async fn wait_completion0(
1237 1330 : mut receiver: tokio::sync::watch::Receiver<()>,
1238 1330 : ) -> Result<(), WaitCompletionError> {
1239 1330 : if receiver.changed().await.is_err() {
1240 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1241 1330 : }
1242 1330 :
1243 1330 : Ok(())
1244 1330 : }
1245 :
1246 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1247 6 : let mut guard = self.upload_queue.lock().unwrap();
1248 6 : let upload_queue = guard.initialized_mut()?;
1249 6 : self.schedule_barrier0(upload_queue);
1250 6 : Ok(())
1251 6 : }
1252 :
1253 1336 : fn schedule_barrier0(
1254 1336 : self: &Arc<Self>,
1255 1336 : upload_queue: &mut UploadQueueInitialized,
1256 1336 : ) -> tokio::sync::watch::Receiver<()> {
1257 1336 : let (sender, receiver) = tokio::sync::watch::channel(());
1258 1336 : let barrier_op = UploadOp::Barrier(sender);
1259 1336 :
1260 1336 : upload_queue.queued_operations.push_back(barrier_op);
1261 1336 : // Don't count this kind of operation!
1262 1336 :
1263 1336 : // Launch the task immediately, if possible
1264 1336 : self.launch_queued_tasks(upload_queue);
1265 1336 :
1266 1336 : receiver
1267 1336 : }
1268 :
1269 : /// Wait for all previously scheduled operations to complete, and then stop.
1270 : ///
1271 : /// Not cancellation safe
1272 8 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1273 8 : // On cancellation the queue is left in ackward state of refusing new operations but
1274 8 : // proper stop is yet to be called. On cancel the original or some later task must call
1275 8 : // `stop` or `shutdown`.
1276 8 : let sg = scopeguard::guard((), |_| {
1277 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1278 8 : });
1279 :
1280 8 : let fut = {
1281 8 : let mut guard = self.upload_queue.lock().unwrap();
1282 8 : let upload_queue = match &mut *guard {
1283 : UploadQueue::Stopped(_) => {
1284 0 : scopeguard::ScopeGuard::into_inner(sg);
1285 0 : return;
1286 : }
1287 : UploadQueue::Uninitialized => {
1288 : // transition into Stopped state
1289 0 : self.stop_impl(&mut guard);
1290 0 : scopeguard::ScopeGuard::into_inner(sg);
1291 0 : return;
1292 : }
1293 8 : UploadQueue::Initialized(ref mut init) => init,
1294 8 : };
1295 8 :
1296 8 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1297 8 : // just don't add more of these as they would never complete.
1298 8 : //
1299 8 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1300 8 : // in every place we would not have to jump through this hoop, and this method could be
1301 8 : // made cancellable.
1302 8 : if !upload_queue.shutting_down {
1303 8 : upload_queue.shutting_down = true;
1304 8 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1305 8 : // this operation is not counted similar to Barrier
1306 8 :
1307 8 : self.launch_queued_tasks(upload_queue);
1308 8 : }
1309 :
1310 8 : upload_queue.shutdown_ready.clone().acquire_owned()
1311 : };
1312 :
1313 8 : let res = fut.await;
1314 :
1315 8 : scopeguard::ScopeGuard::into_inner(sg);
1316 8 :
1317 8 : match res {
1318 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1319 8 : Err(_closed) => {
1320 8 : // expected
1321 8 : }
1322 8 : }
1323 8 :
1324 8 : self.stop();
1325 8 : }
1326 :
1327 : /// Set the deleted_at field in the remote index file.
1328 : ///
1329 : /// This fails if the upload queue has not been `stop()`ed.
1330 : ///
1331 : /// The caller is responsible for calling `stop()` AND for waiting
1332 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1333 : /// Check method [`RemoteTimelineClient::stop`] for details.
1334 0 : #[instrument(skip_all)]
1335 : pub(crate) async fn persist_index_part_with_deleted_flag(
1336 : self: &Arc<Self>,
1337 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1338 : let index_part_with_deleted_at = {
1339 : let mut locked = self.upload_queue.lock().unwrap();
1340 :
1341 : // We must be in stopped state because otherwise
1342 : // we can have inprogress index part upload that can overwrite the file
1343 : // with missing is_deleted flag that we going to set below
1344 : let stopped = locked.stopped_mut()?;
1345 :
1346 : match stopped.deleted_at {
1347 : SetDeletedFlagProgress::NotRunning => (), // proceed
1348 : SetDeletedFlagProgress::InProgress(at) => {
1349 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1350 : }
1351 : SetDeletedFlagProgress::Successful(at) => {
1352 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1353 : }
1354 : };
1355 : let deleted_at = Utc::now().naive_utc();
1356 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1357 :
1358 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1359 : index_part.deleted_at = Some(deleted_at);
1360 : index_part
1361 : };
1362 :
1363 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1364 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1365 0 : let stopped = locked
1366 0 : .stopped_mut()
1367 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1368 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1369 0 : });
1370 :
1371 : pausable_failpoint!("persist_deleted_index_part");
1372 :
1373 : backoff::retry(
1374 0 : || {
1375 0 : upload::upload_index_part(
1376 0 : &self.storage_impl,
1377 0 : &self.tenant_shard_id,
1378 0 : &self.timeline_id,
1379 0 : self.generation,
1380 0 : &index_part_with_deleted_at,
1381 0 : &self.cancel,
1382 0 : )
1383 0 : },
1384 0 : |_e| false,
1385 : 1,
1386 : // have just a couple of attempts
1387 : // when executed as part of timeline deletion this happens in context of api call
1388 : // when executed as part of tenant deletion this happens in the background
1389 : 2,
1390 : "persist_index_part_with_deleted_flag",
1391 : &self.cancel,
1392 : )
1393 : .await
1394 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1395 0 : .and_then(|x| x)?;
1396 :
1397 : // all good, disarm the guard and mark as success
1398 : ScopeGuard::into_inner(undo_deleted_at);
1399 : {
1400 : let mut locked = self.upload_queue.lock().unwrap();
1401 :
1402 : let stopped = locked
1403 : .stopped_mut()
1404 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1405 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1406 : index_part_with_deleted_at
1407 : .deleted_at
1408 : .expect("we set it above"),
1409 : );
1410 : }
1411 :
1412 : Ok(())
1413 : }
1414 :
1415 0 : pub(crate) fn is_deleting(&self) -> bool {
1416 0 : let mut locked = self.upload_queue.lock().unwrap();
1417 0 : locked.stopped_mut().is_ok()
1418 0 : }
1419 :
1420 0 : pub(crate) async fn preserve_initdb_archive(
1421 0 : self: &Arc<Self>,
1422 0 : tenant_id: &TenantId,
1423 0 : timeline_id: &TimelineId,
1424 0 : cancel: &CancellationToken,
1425 0 : ) -> anyhow::Result<()> {
1426 0 : backoff::retry(
1427 0 : || async {
1428 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1429 0 : .await
1430 0 : },
1431 0 : TimeoutOrCancel::caused_by_cancel,
1432 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1433 0 : FAILED_REMOTE_OP_RETRIES,
1434 0 : "preserve_initdb_tar_zst",
1435 0 : &cancel.clone(),
1436 0 : )
1437 0 : .await
1438 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1439 0 : .and_then(|x| x)
1440 0 : .context("backing up initdb archive")?;
1441 0 : Ok(())
1442 0 : }
1443 :
1444 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1445 : ///
1446 : /// This is not normally needed.
1447 0 : pub(crate) async fn upload_layer_file(
1448 0 : self: &Arc<Self>,
1449 0 : uploaded: &ResidentLayer,
1450 0 : cancel: &CancellationToken,
1451 0 : ) -> anyhow::Result<()> {
1452 0 : let remote_path = remote_layer_path(
1453 0 : &self.tenant_shard_id.tenant_id,
1454 0 : &self.timeline_id,
1455 0 : uploaded.metadata().shard,
1456 0 : &uploaded.layer_desc().layer_name(),
1457 0 : uploaded.metadata().generation,
1458 0 : );
1459 0 :
1460 0 : backoff::retry(
1461 0 : || async {
1462 0 : upload::upload_timeline_layer(
1463 0 : &self.storage_impl,
1464 0 : uploaded.local_path(),
1465 0 : &remote_path,
1466 0 : uploaded.metadata().file_size,
1467 0 : cancel,
1468 0 : )
1469 0 : .await
1470 0 : },
1471 0 : TimeoutOrCancel::caused_by_cancel,
1472 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1473 0 : FAILED_REMOTE_OP_RETRIES,
1474 0 : "upload a layer without adding it to latest files",
1475 0 : cancel,
1476 0 : )
1477 0 : .await
1478 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1479 0 : .and_then(|x| x)
1480 0 : .context("upload a layer without adding it to latest files")
1481 0 : }
1482 :
1483 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1484 : /// not added to be part of a future `index_part.json` upload.
1485 0 : pub(crate) async fn copy_timeline_layer(
1486 0 : self: &Arc<Self>,
1487 0 : adopted: &Layer,
1488 0 : adopted_as: &Layer,
1489 0 : cancel: &CancellationToken,
1490 0 : ) -> anyhow::Result<()> {
1491 0 : let source_remote_path = remote_layer_path(
1492 0 : &self.tenant_shard_id.tenant_id,
1493 0 : &adopted
1494 0 : .get_timeline_id()
1495 0 : .expect("Source timeline should be alive"),
1496 0 : adopted.metadata().shard,
1497 0 : &adopted.layer_desc().layer_name(),
1498 0 : adopted.metadata().generation,
1499 0 : );
1500 0 :
1501 0 : let target_remote_path = remote_layer_path(
1502 0 : &self.tenant_shard_id.tenant_id,
1503 0 : &self.timeline_id,
1504 0 : adopted_as.metadata().shard,
1505 0 : &adopted_as.layer_desc().layer_name(),
1506 0 : adopted_as.metadata().generation,
1507 0 : );
1508 0 :
1509 0 : backoff::retry(
1510 0 : || async {
1511 0 : upload::copy_timeline_layer(
1512 0 : &self.storage_impl,
1513 0 : &source_remote_path,
1514 0 : &target_remote_path,
1515 0 : cancel,
1516 0 : )
1517 0 : .await
1518 0 : },
1519 0 : TimeoutOrCancel::caused_by_cancel,
1520 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1521 0 : FAILED_REMOTE_OP_RETRIES,
1522 0 : "copy timeline layer",
1523 0 : cancel,
1524 0 : )
1525 0 : .await
1526 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1527 0 : .and_then(|x| x)
1528 0 : .context("remote copy timeline layer")
1529 0 : }
1530 :
1531 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1532 0 : match tokio::time::timeout(
1533 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1534 0 : self.deletion_queue_client.flush_immediate(),
1535 0 : )
1536 0 : .await
1537 : {
1538 0 : Ok(result) => result,
1539 0 : Err(_timeout) => {
1540 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1541 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1542 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1543 0 : tracing::warn!(
1544 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1545 : );
1546 0 : Ok(())
1547 : }
1548 : }
1549 0 : }
1550 :
1551 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1552 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1553 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1554 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> Result<(), DeleteTimelineError> {
1555 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1556 :
1557 0 : let layers: Vec<RemotePath> = {
1558 0 : let mut locked = self.upload_queue.lock().unwrap();
1559 0 : let stopped = locked.stopped_mut().map_err(DeleteTimelineError::Other)?;
1560 :
1561 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1562 0 : return Err(DeleteTimelineError::Other(anyhow::anyhow!(
1563 0 : "deleted_at is not set"
1564 0 : )));
1565 0 : }
1566 0 :
1567 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1568 :
1569 0 : stopped
1570 0 : .upload_queue_for_deletion
1571 0 : .dirty
1572 0 : .layer_metadata
1573 0 : .drain()
1574 0 : .filter(|(_file_name, meta)| {
1575 0 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1576 0 : // all shards anyway, we _could_ delete these, but
1577 0 : // - it creates a potential race if other shards are still
1578 0 : // using the layers while this shard deletes them.
1579 0 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1580 0 : // these timelines are present but corrupt (their index exists but some layers don't)
1581 0 : //
1582 0 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1583 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1584 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1585 0 : })
1586 0 : .map(|(file_name, meta)| {
1587 0 : remote_layer_path(
1588 0 : &self.tenant_shard_id.tenant_id,
1589 0 : &self.timeline_id,
1590 0 : meta.shard,
1591 0 : &file_name,
1592 0 : meta.generation,
1593 0 : )
1594 0 : })
1595 0 : .collect()
1596 0 : };
1597 0 :
1598 0 : let layer_deletion_count = layers.len();
1599 0 : self.deletion_queue_client
1600 0 : .push_immediate(layers)
1601 0 : .await
1602 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1603 :
1604 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1605 : // inexistant objects are not considered errors.
1606 0 : let initdb_path =
1607 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1608 0 : self.deletion_queue_client
1609 0 : .push_immediate(vec![initdb_path])
1610 0 : .await
1611 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1612 :
1613 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1614 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1615 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1616 0 :
1617 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1618 0 : // taking the burden of listing all the layers that we already know we should delete.
1619 0 : self.flush_deletion_queue()
1620 0 : .await
1621 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1622 :
1623 0 : let cancel = shutdown_token();
1624 :
1625 0 : let remaining = download_retry(
1626 0 : || async {
1627 0 : self.storage_impl
1628 0 : .list(
1629 0 : Some(&timeline_storage_path),
1630 0 : ListingMode::NoDelimiter,
1631 0 : None,
1632 0 : &cancel,
1633 0 : )
1634 0 : .await
1635 0 : },
1636 0 : "list remaining files",
1637 0 : &cancel,
1638 0 : )
1639 0 : .await
1640 0 : .context("list files remaining files")?
1641 : .keys;
1642 :
1643 : // We will delete the current index_part object last, since it acts as a deletion
1644 : // marker via its deleted_at attribute
1645 0 : let latest_index = remaining
1646 0 : .iter()
1647 0 : .filter(|o| {
1648 0 : o.key
1649 0 : .object_name()
1650 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1651 0 : .unwrap_or(false)
1652 0 : })
1653 0 : .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
1654 0 : .max_by_key(|i| i.1)
1655 0 : .map(|i| i.0.clone())
1656 0 : .unwrap_or(
1657 0 : // No generation-suffixed indices, assume we are dealing with
1658 0 : // a legacy index.
1659 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1660 0 : );
1661 0 :
1662 0 : let remaining_layers: Vec<RemotePath> = remaining
1663 0 : .into_iter()
1664 0 : .filter_map(|o| {
1665 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1666 0 : None
1667 : } else {
1668 0 : Some(o.key)
1669 : }
1670 0 : })
1671 0 : .inspect(|path| {
1672 0 : if let Some(name) = path.object_name() {
1673 0 : info!(%name, "deleting a file not referenced from index_part.json");
1674 : } else {
1675 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1676 : }
1677 0 : })
1678 0 : .collect();
1679 0 :
1680 0 : let not_referenced_count = remaining_layers.len();
1681 0 : if !remaining_layers.is_empty() {
1682 0 : self.deletion_queue_client
1683 0 : .push_immediate(remaining_layers)
1684 0 : .await
1685 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1686 0 : }
1687 :
1688 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1689 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1690 0 : "failpoint: timeline-delete-before-index-delete"
1691 0 : )))?
1692 0 : });
1693 :
1694 0 : debug!("enqueuing index part deletion");
1695 0 : self.deletion_queue_client
1696 0 : .push_immediate([latest_index].to_vec())
1697 0 : .await
1698 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1699 :
1700 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1701 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1702 0 : self.flush_deletion_queue()
1703 0 : .await
1704 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1705 :
1706 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1707 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1708 0 : "failpoint: timeline-delete-after-index-delete"
1709 0 : )))?
1710 0 : });
1711 :
1712 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1713 :
1714 0 : Ok(())
1715 0 : }
1716 :
1717 : ///
1718 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1719 : /// the ordering constraints.
1720 : ///
1721 : /// The caller needs to already hold the `upload_queue` lock.
1722 7448 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1723 11980 : while let Some(next_op) = upload_queue.queued_operations.front() {
1724 : // Can we run this task now?
1725 8696 : let can_run_now = match next_op {
1726 : UploadOp::UploadLayer(..) => {
1727 : // Can always be scheduled.
1728 1531 : true
1729 : }
1730 : UploadOp::UploadMetadata { .. } => {
1731 : // These can only be performed after all the preceding operations
1732 : // have finished.
1733 4372 : upload_queue.inprogress_tasks.is_empty()
1734 : }
1735 : UploadOp::Delete(_) => {
1736 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1737 297 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1738 : }
1739 :
1740 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1741 2496 : upload_queue.inprogress_tasks.is_empty()
1742 : }
1743 : };
1744 :
1745 : // If we cannot launch this task, don't look any further.
1746 : //
1747 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1748 : // them now, but we don't try to do that currently. For example, if the frontmost task
1749 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1750 : // could still start layer uploads that were scheduled later.
1751 8696 : if !can_run_now {
1752 4156 : break;
1753 4540 : }
1754 4540 :
1755 4540 : if let UploadOp::Shutdown = next_op {
1756 : // leave the op in the queue but do not start more tasks; it will be dropped when
1757 : // the stop is called.
1758 8 : upload_queue.shutdown_ready.close();
1759 8 : break;
1760 4532 : }
1761 4532 :
1762 4532 : // We can launch this task. Remove it from the queue first.
1763 4532 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1764 4532 :
1765 4532 : debug!("starting op: {}", next_op);
1766 :
1767 : // Update the counters
1768 4532 : match next_op {
1769 1531 : UploadOp::UploadLayer(_, _) => {
1770 1531 : upload_queue.num_inprogress_layer_uploads += 1;
1771 1531 : }
1772 1425 : UploadOp::UploadMetadata { .. } => {
1773 1425 : upload_queue.num_inprogress_metadata_uploads += 1;
1774 1425 : }
1775 240 : UploadOp::Delete(_) => {
1776 240 : upload_queue.num_inprogress_deletions += 1;
1777 240 : }
1778 1336 : UploadOp::Barrier(sender) => {
1779 1336 : sender.send_replace(());
1780 1336 : continue;
1781 : }
1782 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1783 : };
1784 :
1785 : // Assign unique ID to this task
1786 3196 : upload_queue.task_counter += 1;
1787 3196 : let upload_task_id = upload_queue.task_counter;
1788 3196 :
1789 3196 : // Add it to the in-progress map
1790 3196 : let task = Arc::new(UploadTask {
1791 3196 : task_id: upload_task_id,
1792 3196 : op: next_op,
1793 3196 : retries: AtomicU32::new(0),
1794 3196 : });
1795 3196 : upload_queue
1796 3196 : .inprogress_tasks
1797 3196 : .insert(task.task_id, Arc::clone(&task));
1798 3196 :
1799 3196 : // Spawn task to perform the task
1800 3196 : let self_rc = Arc::clone(self);
1801 3196 : let tenant_shard_id = self.tenant_shard_id;
1802 3196 : let timeline_id = self.timeline_id;
1803 3196 : task_mgr::spawn(
1804 3196 : &self.runtime,
1805 3196 : TaskKind::RemoteUploadTask,
1806 3196 : self.tenant_shard_id,
1807 3196 : Some(self.timeline_id),
1808 3196 : "remote upload",
1809 3091 : async move {
1810 43300 : self_rc.perform_upload_task(task).await;
1811 2905 : Ok(())
1812 2905 : }
1813 3196 : .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)),
1814 : );
1815 :
1816 : // Loop back to process next task
1817 : }
1818 7448 : }
1819 :
1820 : ///
1821 : /// Perform an upload task.
1822 : ///
1823 : /// The task is in the `inprogress_tasks` list. This function will try to
1824 : /// execute it, retrying forever. On successful completion, the task is
1825 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1826 : /// queue that were waiting by the completion are launched.
1827 : ///
1828 : /// The task can be shut down, however. That leads to stopping the whole
1829 : /// queue.
1830 : ///
1831 3091 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1832 3091 : let cancel = shutdown_token();
1833 : // Loop to retry until it completes.
1834 : loop {
1835 : // If we're requested to shut down, close up shop and exit.
1836 : //
1837 : // Note: We only check for the shutdown requests between retries, so
1838 : // if a shutdown request arrives while we're busy uploading, in the
1839 : // upload::upload:*() call below, we will wait not exit until it has
1840 : // finished. We probably could cancel the upload by simply dropping
1841 : // the Future, but we're not 100% sure if the remote storage library
1842 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1843 : // upload finishes or times out soon enough.
1844 3091 : if cancel.is_cancelled() {
1845 0 : info!("upload task cancelled by shutdown request");
1846 0 : self.stop();
1847 0 : return;
1848 3091 : }
1849 :
1850 3091 : let upload_result: anyhow::Result<()> = match &task.op {
1851 1436 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1852 1436 : let local_path = layer.local_path();
1853 1436 :
1854 1436 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1855 1436 : // the metadata in the upload should always match our current generation.
1856 1436 : assert_eq!(layer_metadata.generation, self.generation);
1857 :
1858 1436 : let remote_path = remote_layer_path(
1859 1436 : &self.tenant_shard_id.tenant_id,
1860 1436 : &self.timeline_id,
1861 1436 : layer_metadata.shard,
1862 1436 : &layer.layer_desc().layer_name(),
1863 1436 : layer_metadata.generation,
1864 1436 : );
1865 1436 :
1866 1436 : upload::upload_timeline_layer(
1867 1436 : &self.storage_impl,
1868 1436 : local_path,
1869 1436 : &remote_path,
1870 1436 : layer_metadata.file_size,
1871 1436 : &self.cancel,
1872 1436 : )
1873 1436 : .measure_remote_op(
1874 1436 : RemoteOpFileKind::Layer,
1875 1436 : RemoteOpKind::Upload,
1876 1436 : Arc::clone(&self.metrics),
1877 1436 : )
1878 37896 : .await
1879 : }
1880 1415 : UploadOp::UploadMetadata { ref uploaded } => {
1881 1415 : let res = upload::upload_index_part(
1882 1415 : &self.storage_impl,
1883 1415 : &self.tenant_shard_id,
1884 1415 : &self.timeline_id,
1885 1415 : self.generation,
1886 1415 : uploaded,
1887 1415 : &self.cancel,
1888 1415 : )
1889 1415 : .measure_remote_op(
1890 1415 : RemoteOpFileKind::Index,
1891 1415 : RemoteOpKind::Upload,
1892 1415 : Arc::clone(&self.metrics),
1893 1415 : )
1894 5189 : .await;
1895 1407 : if res.is_ok() {
1896 1407 : self.update_remote_physical_size_gauge(Some(uploaded));
1897 1407 : let mention_having_future_layers = if cfg!(feature = "testing") {
1898 1407 : uploaded
1899 1407 : .layer_metadata
1900 1407 : .keys()
1901 17483 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
1902 : } else {
1903 0 : false
1904 : };
1905 1407 : if mention_having_future_layers {
1906 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1907 16 : tracing::info!(
1908 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
1909 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
1910 : );
1911 1391 : }
1912 0 : }
1913 1407 : res
1914 : }
1915 240 : UploadOp::Delete(delete) => {
1916 240 : pausable_failpoint!("before-delete-layer-pausable");
1917 240 : self.deletion_queue_client
1918 240 : .push_layers(
1919 240 : self.tenant_shard_id,
1920 240 : self.timeline_id,
1921 240 : self.generation,
1922 240 : delete.layers.clone(),
1923 240 : )
1924 0 : .await
1925 240 : .map_err(|e| anyhow::anyhow!(e))
1926 : }
1927 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1928 : // unreachable. Barrier operations are handled synchronously in
1929 : // launch_queued_tasks
1930 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1931 0 : break;
1932 : }
1933 : };
1934 :
1935 0 : match upload_result {
1936 : Ok(()) => {
1937 2905 : break;
1938 : }
1939 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1940 0 : // loop around to do the proper stopping
1941 0 : continue;
1942 : }
1943 0 : Err(e) => {
1944 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1945 0 :
1946 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1947 0 : // or other external reasons. Such issues are relatively regular, so log them
1948 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1949 0 : //
1950 0 : // (See similar logic for downloads in `download::download_retry`)
1951 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1952 0 : info!(
1953 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1954 0 : task.op, retries, e
1955 : );
1956 : } else {
1957 0 : warn!(
1958 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1959 0 : task.op, retries, e
1960 : );
1961 : }
1962 :
1963 : // sleep until it's time to retry, or we're cancelled
1964 0 : exponential_backoff(
1965 0 : retries,
1966 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1967 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1968 0 : &cancel,
1969 0 : )
1970 0 : .await;
1971 : }
1972 : }
1973 : }
1974 :
1975 2905 : let retries = task.retries.load(Ordering::SeqCst);
1976 2905 : if retries > 0 {
1977 0 : info!(
1978 0 : "remote task {} completed successfully after {} retries",
1979 0 : task.op, retries
1980 : );
1981 : } else {
1982 2905 : debug!("remote task {} completed successfully", task.op);
1983 : }
1984 :
1985 : // The task has completed successfully. Remove it from the in-progress list.
1986 2905 : let lsn_update = {
1987 2905 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1988 2905 : let upload_queue = match upload_queue_guard.deref_mut() {
1989 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1990 0 : UploadQueue::Stopped(_stopped) => {
1991 0 : None
1992 : },
1993 2905 : UploadQueue::Initialized(qi) => { Some(qi) }
1994 : };
1995 :
1996 2905 : let upload_queue = match upload_queue {
1997 2905 : Some(upload_queue) => upload_queue,
1998 : None => {
1999 0 : info!("another concurrent task already stopped the queue");
2000 0 : return;
2001 : }
2002 : };
2003 :
2004 2905 : upload_queue.inprogress_tasks.remove(&task.task_id);
2005 :
2006 2905 : let lsn_update = match task.op {
2007 : UploadOp::UploadLayer(_, _) => {
2008 1258 : upload_queue.num_inprogress_layer_uploads -= 1;
2009 1258 : None
2010 : }
2011 1407 : UploadOp::UploadMetadata { ref uploaded } => {
2012 1407 : upload_queue.num_inprogress_metadata_uploads -= 1;
2013 1407 :
2014 1407 : // the task id is reused as a monotonicity check for storing the "clean"
2015 1407 : // IndexPart.
2016 1407 : let last_updater = upload_queue.clean.1;
2017 1407 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
2018 1407 : let monotone = is_later || last_updater.is_none();
2019 :
2020 1407 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
2021 :
2022 : // not taking ownership is wasteful
2023 1407 : upload_queue.clean.0.clone_from(uploaded);
2024 1407 : upload_queue.clean.1 = Some(task.task_id);
2025 1407 :
2026 1407 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
2027 1407 :
2028 1407 : if self.generation.is_none() {
2029 : // Legacy mode: skip validating generation
2030 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
2031 0 : None
2032 : } else {
2033 1407 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
2034 : }
2035 : }
2036 : UploadOp::Delete(_) => {
2037 240 : upload_queue.num_inprogress_deletions -= 1;
2038 240 : None
2039 : }
2040 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
2041 : };
2042 :
2043 : // Launch any queued tasks that were unblocked by this one.
2044 2905 : self.launch_queued_tasks(upload_queue);
2045 2905 : lsn_update
2046 : };
2047 :
2048 2905 : if let Some((lsn, slot)) = lsn_update {
2049 : // Updates to the remote_consistent_lsn we advertise to pageservers
2050 : // are all routed through the DeletionQueue, to enforce important
2051 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2052 1407 : self.deletion_queue_client
2053 1407 : .update_remote_consistent_lsn(
2054 1407 : self.tenant_shard_id,
2055 1407 : self.timeline_id,
2056 1407 : self.generation,
2057 1407 : lsn,
2058 1407 : slot,
2059 1407 : )
2060 0 : .await;
2061 1498 : }
2062 :
2063 2905 : self.metric_end(&task.op);
2064 2905 : }
2065 :
2066 6386 : fn metric_impl(
2067 6386 : &self,
2068 6386 : op: &UploadOp,
2069 6386 : ) -> Option<(
2070 6386 : RemoteOpFileKind,
2071 6386 : RemoteOpKind,
2072 6386 : RemoteTimelineClientMetricsCallTrackSize,
2073 6386 : )> {
2074 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2075 6386 : let res = match op {
2076 2796 : UploadOp::UploadLayer(_, m) => (
2077 2796 : RemoteOpFileKind::Layer,
2078 2796 : RemoteOpKind::Upload,
2079 2796 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2080 2796 : ),
2081 2859 : UploadOp::UploadMetadata { .. } => (
2082 2859 : RemoteOpFileKind::Index,
2083 2859 : RemoteOpKind::Upload,
2084 2859 : DontTrackSize {
2085 2859 : reason: "metadata uploads are tiny",
2086 2859 : },
2087 2859 : ),
2088 723 : UploadOp::Delete(_delete) => (
2089 723 : RemoteOpFileKind::Layer,
2090 723 : RemoteOpKind::Delete,
2091 723 : DontTrackSize {
2092 723 : reason: "should we track deletes? positive or negative sign?",
2093 723 : },
2094 723 : ),
2095 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2096 : // we do not account these
2097 8 : return None;
2098 : }
2099 : };
2100 6378 : Some(res)
2101 6386 : }
2102 :
2103 3473 : fn metric_begin(&self, op: &UploadOp) {
2104 3473 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2105 3473 : Some(x) => x,
2106 0 : None => return,
2107 : };
2108 3473 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2109 3473 : guard.will_decrement_manually(); // in metric_end(), see right below
2110 3473 : }
2111 :
2112 2913 : fn metric_end(&self, op: &UploadOp) {
2113 2913 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2114 2905 : Some(x) => x,
2115 8 : None => return,
2116 : };
2117 2905 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2118 2913 : }
2119 :
2120 : /// Close the upload queue for new operations and cancel queued operations.
2121 : ///
2122 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2123 : ///
2124 : /// In-progress operations will still be running after this function returns.
2125 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2126 : /// to wait for them to complete, after calling this function.
2127 18 : pub(crate) fn stop(&self) {
2128 18 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2129 18 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2130 18 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2131 18 : let mut guard = self.upload_queue.lock().unwrap();
2132 18 : self.stop_impl(&mut guard);
2133 18 : }
2134 :
2135 18 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2136 18 : match &mut **guard {
2137 : UploadQueue::Uninitialized => {
2138 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2139 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2140 : }
2141 : UploadQueue::Stopped(_) => {
2142 : // nothing to do
2143 8 : info!("another concurrent task already shut down the queue");
2144 : }
2145 10 : UploadQueue::Initialized(initialized) => {
2146 10 : info!("shutting down upload queue");
2147 :
2148 : // Replace the queue with the Stopped state, taking ownership of the old
2149 : // Initialized queue. We will do some checks on it, and then drop it.
2150 10 : let qi = {
2151 : // Here we preserve working version of the upload queue for possible use during deletions.
2152 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2153 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2154 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2155 10 : let upload_queue_for_deletion = UploadQueueInitialized {
2156 10 : task_counter: 0,
2157 10 : dirty: initialized.dirty.clone(),
2158 10 : clean: initialized.clean.clone(),
2159 10 : latest_files_changes_since_metadata_upload_scheduled: 0,
2160 10 : visible_remote_consistent_lsn: initialized
2161 10 : .visible_remote_consistent_lsn
2162 10 : .clone(),
2163 10 : num_inprogress_layer_uploads: 0,
2164 10 : num_inprogress_metadata_uploads: 0,
2165 10 : num_inprogress_deletions: 0,
2166 10 : inprogress_tasks: HashMap::default(),
2167 10 : queued_operations: VecDeque::default(),
2168 10 : #[cfg(feature = "testing")]
2169 10 : dangling_files: HashMap::default(),
2170 10 : shutting_down: false,
2171 10 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2172 10 : };
2173 10 :
2174 10 : let upload_queue = std::mem::replace(
2175 10 : &mut **guard,
2176 10 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2177 10 : UploadQueueStoppedDeletable {
2178 10 : upload_queue_for_deletion,
2179 10 : deleted_at: SetDeletedFlagProgress::NotRunning,
2180 10 : },
2181 10 : )),
2182 10 : );
2183 10 : if let UploadQueue::Initialized(qi) = upload_queue {
2184 10 : qi
2185 : } else {
2186 0 : unreachable!("we checked in the match above that it is Initialized");
2187 : }
2188 : };
2189 :
2190 : // consistency check
2191 10 : assert_eq!(
2192 10 : qi.num_inprogress_layer_uploads
2193 10 : + qi.num_inprogress_metadata_uploads
2194 10 : + qi.num_inprogress_deletions,
2195 10 : qi.inprogress_tasks.len()
2196 10 : );
2197 :
2198 : // We don't need to do anything here for in-progress tasks. They will finish
2199 : // on their own, decrement the unfinished-task counter themselves, and observe
2200 : // that the queue is Stopped.
2201 10 : drop(qi.inprogress_tasks);
2202 :
2203 : // Tear down queued ops
2204 10 : for op in qi.queued_operations.into_iter() {
2205 8 : self.metric_end(&op);
2206 8 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2207 8 : // which is exactly what we want to happen.
2208 8 : drop(op);
2209 8 : }
2210 : }
2211 : }
2212 18 : }
2213 :
2214 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2215 : /// externally to RemoteTimelineClient.
2216 0 : pub(crate) fn initialized_upload_queue(
2217 0 : &self,
2218 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2219 0 : let mut inner = self.upload_queue.lock().unwrap();
2220 0 : inner.initialized_mut()?;
2221 0 : Ok(UploadQueueAccessor { inner })
2222 0 : }
2223 :
2224 8 : pub(crate) fn no_pending_work(&self) -> bool {
2225 8 : let inner = self.upload_queue.lock().unwrap();
2226 8 : match &*inner {
2227 : UploadQueue::Uninitialized
2228 0 : | UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => true,
2229 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(x)) => {
2230 8 : x.upload_queue_for_deletion.no_pending_work()
2231 : }
2232 0 : UploadQueue::Initialized(x) => x.no_pending_work(),
2233 : }
2234 8 : }
2235 :
2236 : /// 'foreign' in the sense that it does not belong to this tenant shard. This method
2237 : /// is used during GC for other shards to get the index of shard zero.
2238 0 : pub(crate) async fn download_foreign_index(
2239 0 : &self,
2240 0 : shard_number: ShardNumber,
2241 0 : cancel: &CancellationToken,
2242 0 : ) -> Result<(IndexPart, Generation, std::time::SystemTime), DownloadError> {
2243 0 : let foreign_shard_id = TenantShardId {
2244 0 : shard_number,
2245 0 : shard_count: self.tenant_shard_id.shard_count,
2246 0 : tenant_id: self.tenant_shard_id.tenant_id,
2247 0 : };
2248 0 : download_index_part(
2249 0 : &self.storage_impl,
2250 0 : &foreign_shard_id,
2251 0 : &self.timeline_id,
2252 0 : Generation::MAX,
2253 0 : cancel,
2254 0 : )
2255 0 : .await
2256 0 : }
2257 : }
2258 :
2259 : pub(crate) struct UploadQueueAccessor<'a> {
2260 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2261 : }
2262 :
2263 : impl UploadQueueAccessor<'_> {
2264 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2265 0 : match &*self.inner {
2266 0 : UploadQueue::Initialized(x) => &x.clean.0,
2267 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2268 0 : unreachable!("checked before constructing")
2269 : }
2270 : }
2271 0 : }
2272 : }
2273 :
2274 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2275 0 : let path = format!("tenants/{tenant_shard_id}");
2276 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2277 0 : }
2278 :
2279 578 : pub fn remote_tenant_manifest_path(
2280 578 : tenant_shard_id: &TenantShardId,
2281 578 : generation: Generation,
2282 578 : ) -> RemotePath {
2283 578 : let path = format!(
2284 578 : "tenants/{tenant_shard_id}/tenant-manifest{}.json",
2285 578 : generation.get_suffix()
2286 578 : );
2287 578 : RemotePath::from_string(&path).expect("Failed to construct path")
2288 578 : }
2289 :
2290 : /// Prefix to all generations' manifest objects in a tenant shard
2291 192 : pub fn remote_tenant_manifest_prefix(tenant_shard_id: &TenantShardId) -> RemotePath {
2292 192 : let path = format!("tenants/{tenant_shard_id}/tenant-manifest",);
2293 192 : RemotePath::from_string(&path).expect("Failed to construct path")
2294 192 : }
2295 :
2296 222 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2297 222 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2298 222 : RemotePath::from_string(&path).expect("Failed to construct path")
2299 222 : }
2300 :
2301 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2302 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2303 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2304 0 : }
2305 :
2306 30 : pub fn remote_timeline_path(
2307 30 : tenant_shard_id: &TenantShardId,
2308 30 : timeline_id: &TimelineId,
2309 30 : ) -> RemotePath {
2310 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2311 30 : }
2312 :
2313 : /// Obtains the path of the given Layer in the remote
2314 : ///
2315 : /// Note that the shard component of a remote layer path is _not_ always the same
2316 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2317 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2318 1454 : pub fn remote_layer_path(
2319 1454 : tenant_id: &TenantId,
2320 1454 : timeline_id: &TimelineId,
2321 1454 : shard: ShardIndex,
2322 1454 : layer_file_name: &LayerName,
2323 1454 : generation: Generation,
2324 1454 : ) -> RemotePath {
2325 1454 : // Generation-aware key format
2326 1454 : let path = format!(
2327 1454 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2328 1454 : shard.get_suffix(),
2329 1454 : layer_file_name,
2330 1454 : generation.get_suffix()
2331 1454 : );
2332 1454 :
2333 1454 : RemotePath::from_string(&path).expect("Failed to construct path")
2334 1454 : }
2335 :
2336 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2337 4 : RemotePath::from_string(&format!(
2338 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2339 4 : ))
2340 4 : .expect("Failed to construct path")
2341 4 : }
2342 :
2343 2 : pub fn remote_initdb_preserved_archive_path(
2344 2 : tenant_id: &TenantId,
2345 2 : timeline_id: &TimelineId,
2346 2 : ) -> RemotePath {
2347 2 : RemotePath::from_string(&format!(
2348 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2349 2 : ))
2350 2 : .expect("Failed to construct path")
2351 2 : }
2352 :
2353 1480 : pub fn remote_index_path(
2354 1480 : tenant_shard_id: &TenantShardId,
2355 1480 : timeline_id: &TimelineId,
2356 1480 : generation: Generation,
2357 1480 : ) -> RemotePath {
2358 1480 : RemotePath::from_string(&format!(
2359 1480 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2360 1480 : IndexPart::FILE_NAME,
2361 1480 : generation.get_suffix()
2362 1480 : ))
2363 1480 : .expect("Failed to construct path")
2364 1480 : }
2365 :
2366 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2367 0 : RemotePath::from_string(&format!(
2368 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2369 0 : ))
2370 0 : .expect("Failed to construct path")
2371 0 : }
2372 :
2373 : /// Given the key of an index, parse out the generation part of the name
2374 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2375 18 : let file_name = match path.get_path().file_name() {
2376 18 : Some(f) => f,
2377 : None => {
2378 : // Unexpected: we should be seeing index_part.json paths only
2379 0 : tracing::warn!("Malformed index key {}", path);
2380 0 : return None;
2381 : }
2382 : };
2383 :
2384 18 : match file_name.split_once('-') {
2385 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2386 6 : None => None,
2387 : }
2388 18 : }
2389 :
2390 : /// Given the key of a tenant manifest, parse out the generation number
2391 0 : pub(crate) fn parse_remote_tenant_manifest_path(path: RemotePath) -> Option<Generation> {
2392 : static RE: OnceLock<Regex> = OnceLock::new();
2393 0 : let re = RE.get_or_init(|| Regex::new(r".+tenant-manifest-([0-9a-f]{8}).json").unwrap());
2394 0 : re.captures(path.get_path().as_str())
2395 0 : .and_then(|c| c.get(1))
2396 0 : .and_then(|m| Generation::parse_suffix(m.as_str()))
2397 0 : }
2398 :
2399 : #[cfg(test)]
2400 : mod tests {
2401 : use super::*;
2402 : use crate::{
2403 : context::RequestContext,
2404 : tenant::{
2405 : harness::{TenantHarness, TIMELINE_ID},
2406 : storage_layer::layer::local_layer_path,
2407 : Tenant, Timeline,
2408 : },
2409 : DEFAULT_PG_VERSION,
2410 : };
2411 :
2412 : use std::collections::HashSet;
2413 :
2414 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2415 8 : format!("contents for {name}").into()
2416 8 : }
2417 :
2418 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2419 2 : let metadata = TimelineMetadata::new(
2420 2 : disk_consistent_lsn,
2421 2 : None,
2422 2 : None,
2423 2 : Lsn(0),
2424 2 : Lsn(0),
2425 2 : Lsn(0),
2426 2 : // Any version will do
2427 2 : // but it should be consistent with the one in the tests
2428 2 : crate::DEFAULT_PG_VERSION,
2429 2 : );
2430 2 :
2431 2 : // go through serialize + deserialize to fix the header, including checksum
2432 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2433 2 : }
2434 :
2435 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2436 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2437 2 : avec.sort();
2438 2 :
2439 2 : let mut bvec = b.to_vec();
2440 2 : bvec.sort_unstable();
2441 2 :
2442 2 : assert_eq!(avec, bvec);
2443 2 : }
2444 :
2445 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2446 4 : let mut expected: Vec<String> = expected
2447 4 : .iter()
2448 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2449 4 : .collect();
2450 4 : expected.sort();
2451 4 :
2452 4 : let mut found: Vec<String> = Vec::new();
2453 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2454 16 : let entry_name = entry.file_name();
2455 16 : let fname = entry_name.to_str().unwrap();
2456 16 : found.push(String::from(fname));
2457 16 : }
2458 4 : found.sort();
2459 4 :
2460 4 : assert_eq!(found, expected);
2461 4 : }
2462 :
2463 : struct TestSetup {
2464 : harness: TenantHarness,
2465 : tenant: Arc<Tenant>,
2466 : timeline: Arc<Timeline>,
2467 : tenant_ctx: RequestContext,
2468 : }
2469 :
2470 : impl TestSetup {
2471 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2472 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2473 8 : let harness = TenantHarness::create(test_name).await?;
2474 80 : let (tenant, ctx) = harness.load().await;
2475 :
2476 8 : let timeline = tenant
2477 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2478 24 : .await?;
2479 :
2480 8 : Ok(Self {
2481 8 : harness,
2482 8 : tenant,
2483 8 : timeline,
2484 8 : tenant_ctx: ctx,
2485 8 : })
2486 8 : }
2487 :
2488 : /// Construct a RemoteTimelineClient in an arbitrary generation
2489 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2490 10 : Arc::new(RemoteTimelineClient {
2491 10 : conf: self.harness.conf,
2492 10 : runtime: tokio::runtime::Handle::current(),
2493 10 : tenant_shard_id: self.harness.tenant_shard_id,
2494 10 : timeline_id: TIMELINE_ID,
2495 10 : generation,
2496 10 : storage_impl: self.harness.remote_storage.clone(),
2497 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2498 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2499 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2500 10 : &self.harness.tenant_shard_id,
2501 10 : &TIMELINE_ID,
2502 10 : )),
2503 10 : cancel: CancellationToken::new(),
2504 10 : })
2505 10 : }
2506 :
2507 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2508 : /// and timeline_id are present.
2509 6 : fn span(&self) -> tracing::Span {
2510 6 : tracing::info_span!(
2511 : "test",
2512 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2513 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2514 : timeline_id = %TIMELINE_ID
2515 : )
2516 6 : }
2517 : }
2518 :
2519 : // Test scheduling
2520 : #[tokio::test]
2521 2 : async fn upload_scheduling() {
2522 2 : // Test outline:
2523 2 : //
2524 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2525 2 : // Schedule upload of index. Check that it is queued
2526 2 : // let the layer file uploads finish. Check that the index-upload is now started
2527 2 : // let the index-upload finish.
2528 2 : //
2529 2 : // Download back the index.json. Check that the list of files is correct
2530 2 : //
2531 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2532 2 : // let upload finish. Check that deletion is now started
2533 2 : // Schedule another deletion. Check that it's launched immediately.
2534 2 : // Schedule index upload. Check that it's queued
2535 2 :
2536 26 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2537 2 : let span = test_setup.span();
2538 2 : let _guard = span.enter();
2539 2 :
2540 2 : let TestSetup {
2541 2 : harness,
2542 2 : tenant: _tenant,
2543 2 : timeline,
2544 2 : tenant_ctx: _tenant_ctx,
2545 2 : } = test_setup;
2546 2 :
2547 2 : let client = &timeline.remote_client;
2548 2 :
2549 2 : // Download back the index.json, and check that the list of files is correct
2550 2 : let initial_index_part = match client
2551 2 : .download_index_file(&CancellationToken::new())
2552 6 : .await
2553 2 : .unwrap()
2554 2 : {
2555 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2556 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2557 2 : };
2558 2 : let initial_layers = initial_index_part
2559 2 : .layer_metadata
2560 2 : .keys()
2561 2 : .map(|f| f.to_owned())
2562 2 : .collect::<HashSet<LayerName>>();
2563 2 : let initial_layer = {
2564 2 : assert!(initial_layers.len() == 1);
2565 2 : initial_layers.into_iter().next().unwrap()
2566 2 : };
2567 2 :
2568 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2569 2 :
2570 2 : println!("workdir: {}", harness.conf.workdir);
2571 2 :
2572 2 : let remote_timeline_dir = harness
2573 2 : .remote_fs_dir
2574 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2575 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2576 2 :
2577 2 : let generation = harness.generation;
2578 2 : let shard = harness.shard;
2579 2 :
2580 2 : // Create a couple of dummy files, schedule upload for them
2581 2 :
2582 2 : let layers = [
2583 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2584 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2585 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2586 2 : ]
2587 2 : .into_iter()
2588 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2589 6 :
2590 6 : let local_path = local_layer_path(
2591 6 : harness.conf,
2592 6 : &timeline.tenant_shard_id,
2593 6 : &timeline.timeline_id,
2594 6 : &name,
2595 6 : &generation,
2596 6 : );
2597 6 : std::fs::write(&local_path, &contents).unwrap();
2598 6 :
2599 6 : Layer::for_resident(
2600 6 : harness.conf,
2601 6 : &timeline,
2602 6 : local_path,
2603 6 : name,
2604 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2605 6 : )
2606 6 : }).collect::<Vec<_>>();
2607 2 :
2608 2 : client
2609 2 : .schedule_layer_file_upload(layers[0].clone())
2610 2 : .unwrap();
2611 2 : client
2612 2 : .schedule_layer_file_upload(layers[1].clone())
2613 2 : .unwrap();
2614 2 :
2615 2 : // Check that they are started immediately, not queued
2616 2 : //
2617 2 : // this works because we running within block_on, so any futures are now queued up until
2618 2 : // our next await point.
2619 2 : {
2620 2 : let mut guard = client.upload_queue.lock().unwrap();
2621 2 : let upload_queue = guard.initialized_mut().unwrap();
2622 2 : assert!(upload_queue.queued_operations.is_empty());
2623 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2624 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2625 2 :
2626 2 : // also check that `latest_file_changes` was updated
2627 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2628 2 : }
2629 2 :
2630 2 : // Schedule upload of index. Check that it is queued
2631 2 : let metadata = dummy_metadata(Lsn(0x20));
2632 2 : client
2633 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2634 2 : .unwrap();
2635 2 : {
2636 2 : let mut guard = client.upload_queue.lock().unwrap();
2637 2 : let upload_queue = guard.initialized_mut().unwrap();
2638 2 : assert!(upload_queue.queued_operations.len() == 1);
2639 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2640 2 : }
2641 2 :
2642 2 : // Wait for the uploads to finish
2643 2 : client.wait_completion().await.unwrap();
2644 2 : {
2645 2 : let mut guard = client.upload_queue.lock().unwrap();
2646 2 : let upload_queue = guard.initialized_mut().unwrap();
2647 2 :
2648 2 : assert!(upload_queue.queued_operations.is_empty());
2649 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2650 2 : }
2651 2 :
2652 2 : // Download back the index.json, and check that the list of files is correct
2653 2 : let index_part = match client
2654 2 : .download_index_file(&CancellationToken::new())
2655 6 : .await
2656 2 : .unwrap()
2657 2 : {
2658 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2659 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2660 2 : };
2661 2 :
2662 2 : assert_file_list(
2663 2 : &index_part
2664 2 : .layer_metadata
2665 2 : .keys()
2666 6 : .map(|f| f.to_owned())
2667 2 : .collect(),
2668 2 : &[
2669 2 : &initial_layer.to_string(),
2670 2 : &layers[0].layer_desc().layer_name().to_string(),
2671 2 : &layers[1].layer_desc().layer_name().to_string(),
2672 2 : ],
2673 2 : );
2674 2 : assert_eq!(index_part.metadata, metadata);
2675 2 :
2676 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2677 2 : client
2678 2 : .schedule_layer_file_upload(layers[2].clone())
2679 2 : .unwrap();
2680 2 :
2681 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2682 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2683 2 : // spawn_blocking started by the drop.
2684 2 : client
2685 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2686 2 : .unwrap();
2687 2 : {
2688 2 : let mut guard = client.upload_queue.lock().unwrap();
2689 2 : let upload_queue = guard.initialized_mut().unwrap();
2690 2 :
2691 2 : // Deletion schedules upload of the index file, and the file deletion itself
2692 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2693 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2694 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2695 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2696 2 : assert_eq!(
2697 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2698 2 : 0
2699 2 : );
2700 2 : }
2701 2 : assert_remote_files(
2702 2 : &[
2703 2 : &initial_layer.to_string(),
2704 2 : &layers[0].layer_desc().layer_name().to_string(),
2705 2 : &layers[1].layer_desc().layer_name().to_string(),
2706 2 : "index_part.json",
2707 2 : ],
2708 2 : &remote_timeline_dir,
2709 2 : generation,
2710 2 : );
2711 2 :
2712 2 : // Finish them
2713 2 : client.wait_completion().await.unwrap();
2714 2 : harness.deletion_queue.pump().await;
2715 2 :
2716 2 : assert_remote_files(
2717 2 : &[
2718 2 : &initial_layer.to_string(),
2719 2 : &layers[1].layer_desc().layer_name().to_string(),
2720 2 : &layers[2].layer_desc().layer_name().to_string(),
2721 2 : "index_part.json",
2722 2 : ],
2723 2 : &remote_timeline_dir,
2724 2 : generation,
2725 2 : );
2726 2 : }
2727 :
2728 : #[tokio::test]
2729 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2730 2 : // Setup
2731 2 :
2732 2 : let TestSetup {
2733 2 : harness,
2734 2 : tenant: _tenant,
2735 2 : timeline,
2736 2 : ..
2737 26 : } = TestSetup::new("metrics").await.unwrap();
2738 2 : let client = &timeline.remote_client;
2739 2 :
2740 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2741 2 : let local_path = local_layer_path(
2742 2 : harness.conf,
2743 2 : &timeline.tenant_shard_id,
2744 2 : &timeline.timeline_id,
2745 2 : &layer_file_name_1,
2746 2 : &harness.generation,
2747 2 : );
2748 2 : let content_1 = dummy_contents("foo");
2749 2 : std::fs::write(&local_path, &content_1).unwrap();
2750 2 :
2751 2 : let layer_file_1 = Layer::for_resident(
2752 2 : harness.conf,
2753 2 : &timeline,
2754 2 : local_path,
2755 2 : layer_file_name_1.clone(),
2756 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2757 2 : );
2758 2 :
2759 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2760 2 : struct BytesStartedFinished {
2761 2 : started: Option<usize>,
2762 2 : finished: Option<usize>,
2763 2 : }
2764 2 : impl std::ops::Add for BytesStartedFinished {
2765 2 : type Output = Self;
2766 4 : fn add(self, rhs: Self) -> Self::Output {
2767 4 : Self {
2768 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2769 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2770 4 : }
2771 4 : }
2772 2 : }
2773 6 : let get_bytes_started_stopped = || {
2774 6 : let started = client
2775 6 : .metrics
2776 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2777 6 : .map(|v| v.try_into().unwrap());
2778 6 : let stopped = client
2779 6 : .metrics
2780 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2781 6 : .map(|v| v.try_into().unwrap());
2782 6 : BytesStartedFinished {
2783 6 : started,
2784 6 : finished: stopped,
2785 6 : }
2786 6 : };
2787 2 :
2788 2 : // Test
2789 2 : tracing::info!("now doing actual test");
2790 2 :
2791 2 : let actual_a = get_bytes_started_stopped();
2792 2 :
2793 2 : client
2794 2 : .schedule_layer_file_upload(layer_file_1.clone())
2795 2 : .unwrap();
2796 2 :
2797 2 : let actual_b = get_bytes_started_stopped();
2798 2 :
2799 2 : client.wait_completion().await.unwrap();
2800 2 :
2801 2 : let actual_c = get_bytes_started_stopped();
2802 2 :
2803 2 : // Validate
2804 2 :
2805 2 : let expected_b = actual_a
2806 2 : + BytesStartedFinished {
2807 2 : started: Some(content_1.len()),
2808 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2809 2 : finished: Some(0),
2810 2 : };
2811 2 : assert_eq!(actual_b, expected_b);
2812 2 :
2813 2 : let expected_c = actual_a
2814 2 : + BytesStartedFinished {
2815 2 : started: Some(content_1.len()),
2816 2 : finished: Some(content_1.len()),
2817 2 : };
2818 2 : assert_eq!(actual_c, expected_c);
2819 2 : }
2820 :
2821 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2822 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2823 12 : let example_index_part = IndexPart::example();
2824 12 :
2825 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2826 12 :
2827 12 : let index_path = test_state.harness.remote_fs_dir.join(
2828 12 : remote_index_path(
2829 12 : &test_state.harness.tenant_shard_id,
2830 12 : &TIMELINE_ID,
2831 12 : generation,
2832 12 : )
2833 12 : .get_path(),
2834 12 : );
2835 12 :
2836 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2837 12 : .expect("creating test dir should work");
2838 12 :
2839 12 : eprintln!("Writing {index_path}");
2840 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2841 12 : example_index_part
2842 12 : }
2843 :
2844 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2845 : /// index, the IndexPart returned is equal to `expected`
2846 10 : async fn assert_got_index_part(
2847 10 : test_state: &TestSetup,
2848 10 : get_generation: Generation,
2849 10 : expected: &IndexPart,
2850 10 : ) {
2851 10 : let client = test_state.build_client(get_generation);
2852 :
2853 10 : let download_r = client
2854 10 : .download_index_file(&CancellationToken::new())
2855 68 : .await
2856 10 : .expect("download should always succeed");
2857 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2858 10 : match download_r {
2859 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2860 10 : assert_eq!(&index_part, expected);
2861 : }
2862 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2863 : }
2864 10 : }
2865 :
2866 : #[tokio::test]
2867 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2868 26 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2869 2 : let span = test_state.span();
2870 2 : let _guard = span.enter();
2871 2 :
2872 2 : // Simple case: we are in generation N, load the index from generation N - 1
2873 2 : let generation_n = 5;
2874 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2875 2 :
2876 8 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2877 2 :
2878 2 : Ok(())
2879 2 : }
2880 :
2881 : #[tokio::test]
2882 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2883 2 : let test_state = TestSetup::new("index_part_download_ordering")
2884 26 : .await
2885 2 : .unwrap();
2886 2 :
2887 2 : let span = test_state.span();
2888 2 : let _guard = span.enter();
2889 2 :
2890 2 : // A generation-less IndexPart exists in the bucket, we should find it
2891 2 : let generation_n = 5;
2892 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2893 16 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2894 2 :
2895 2 : // If a more recent-than-none generation exists, we should prefer to load that
2896 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2897 18 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2898 2 :
2899 2 : // If a more-recent-than-me generation exists, we should ignore it.
2900 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2901 20 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2902 2 :
2903 2 : // If a directly previous generation exists, _and_ an index exists in my own
2904 2 : // generation, I should prefer my own generation.
2905 2 : let _injected_prev =
2906 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2907 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2908 2 : assert_got_index_part(
2909 2 : &test_state,
2910 2 : Generation::new(generation_n),
2911 2 : &injected_current,
2912 2 : )
2913 6 : .await;
2914 2 :
2915 2 : Ok(())
2916 2 : }
2917 : }
|
