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
95 : //! [`UploadQueueInitialized::latest_files`] and [`UploadQueueInitialized::latest_metadata`].
96 : //! It is initialized based on the [`IndexPart`] that was passed during init
97 : //! and updated with every `schedule_*` function call.
98 : //! All this is necessary necessary to compute the future [`IndexPart`]s
99 : //! when scheduling an operation while other operations that also affect the
100 : //! remote [`IndexPart`] are in flight.
101 : //!
102 : //! # Retries & Error Handling
103 : //!
104 : //! The client retries operations indefinitely, using exponential back-off.
105 : //! There is no way to force a retry, i.e., interrupt the back-off.
106 : //! This could be built easily.
107 : //!
108 : //! # Cancellation
109 : //!
110 : //! The operations execute as plain [`task_mgr`] tasks, scoped to
111 : //! the client's tenant and timeline.
112 : //! Dropping the client will drop queued operations but not executing operations.
113 : //! These will complete unless the `task_mgr` tasks are cancelled using `task_mgr`
114 : //! APIs, e.g., during pageserver shutdown, timeline delete, or tenant detach.
115 : //!
116 : //! # Completion
117 : //!
118 : //! Once an operation has completed, we update
119 : //! [`UploadQueueInitialized::projected_remote_consistent_lsn`] immediately,
120 : //! and submit a request through the DeletionQueue to update
121 : //! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
122 : //! validated that our generation is not stale. It is this visible value
123 : //! that is advertized to safekeepers as a signal that that they can
124 : //! delete the WAL up to that LSN.
125 : //!
126 : //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
127 : //! for all pending operations to complete. It does not prevent more
128 : //! operations from getting scheduled.
129 : //!
130 : //! # Crash Consistency
131 : //!
132 : //! We do not persist the upload queue state.
133 : //! If we drop the client, or crash, all unfinished operations are lost.
134 : //!
135 : //! To recover, the following steps need to be taken:
136 : //! - Retrieve the current remote [`IndexPart`]. This gives us a
137 : //! consistent remote state, assuming the user scheduled the operations in
138 : //! the correct order.
139 : //! - Initiate upload queue with that [`IndexPart`].
140 : //! - Reschedule all lost operations by comparing the local filesystem state
141 : //! and remote state as per [`IndexPart`]. This is done in
142 : //! [`Tenant::timeline_init_and_sync`].
143 : //!
144 : //! Note that if we crash during file deletion between the index update
145 : //! that removes the file from the list of files, and deleting the remote file,
146 : //! the file is leaked in the remote storage. Similarly, if a new file is created
147 : //! and uploaded, but the pageserver dies permanently before updating the
148 : //! remote index file, the new file is leaked in remote storage. We accept and
149 : //! tolerate that for now.
150 : //! Note further that we cannot easily fix this by scheduling deletes for every
151 : //! file that is present only on the remote, because we cannot distinguish the
152 : //! following two cases:
153 : //! - (1) We had the file locally, deleted it locally, scheduled a remote delete,
154 : //! but crashed before it finished remotely.
155 : //! - (2) We never had the file locally because we haven't on-demand downloaded
156 : //! it yet.
157 : //!
158 : //! # Downloads
159 : //!
160 : //! In addition to the upload queue, [`RemoteTimelineClient`] has functions for
161 : //! downloading files from the remote storage. Downloads are performed immediately
162 : //! against the `RemoteStorage`, independently of the upload queue.
163 : //!
164 : //! When we attach a tenant, we perform the following steps:
165 : //! - create `Tenant` object in `TenantState::Attaching` state
166 : //! - List timelines that are present in remote storage, and for each:
167 : //! - download their remote [`IndexPart`]s
168 : //! - create `Timeline` struct and a `RemoteTimelineClient`
169 : //! - initialize the client's upload queue with its `IndexPart`
170 : //! - schedule uploads for layers that are only present locally.
171 : //! - After the above is done for each timeline, open the tenant for business by
172 : //! transitioning it from `TenantState::Attaching` to `TenantState::Active` state.
173 : //! This starts the timelines' WAL-receivers and the tenant's GC & Compaction loops.
174 : //!
175 : //! # Operating Without Remote Storage
176 : //!
177 : //! If no remote storage configuration is provided, the [`RemoteTimelineClient`] is
178 : //! not created and the uploads are skipped.
179 : //!
180 : //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
181 : //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
182 :
183 : pub(crate) mod download;
184 : pub mod index;
185 : pub(crate) mod upload;
186 :
187 : use anyhow::Context;
188 : use camino::Utf8Path;
189 : use chrono::{NaiveDateTime, Utc};
190 :
191 : pub(crate) use download::download_initdb_tar_zst;
192 : use pageserver_api::shard::{ShardIndex, TenantShardId};
193 : use scopeguard::ScopeGuard;
194 : use tokio_util::sync::CancellationToken;
195 : pub(crate) use upload::upload_initdb_dir;
196 : use utils::backoff::{
197 : self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
198 : };
199 :
200 : use std::collections::{HashMap, VecDeque};
201 : use std::sync::atomic::{AtomicU32, Ordering};
202 : use std::sync::{Arc, Mutex};
203 :
204 : use remote_storage::{DownloadError, GenericRemoteStorage, RemotePath, TimeoutOrCancel};
205 : use std::ops::DerefMut;
206 : use tracing::{debug, error, info, instrument, warn};
207 : use tracing::{info_span, Instrument};
208 : use utils::lsn::Lsn;
209 :
210 : use crate::deletion_queue::DeletionQueueClient;
211 : use crate::metrics::{
212 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
213 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
214 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
215 : };
216 : use crate::task_mgr::shutdown_token;
217 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
218 : use crate::tenant::remote_timeline_client::download::download_retry;
219 : use crate::tenant::storage_layer::AsLayerDesc;
220 : use crate::tenant::upload_queue::Delete;
221 : use crate::tenant::TIMELINES_SEGMENT_NAME;
222 : use crate::{
223 : config::PageServerConf,
224 : task_mgr,
225 : task_mgr::TaskKind,
226 : task_mgr::BACKGROUND_RUNTIME,
227 : tenant::metadata::TimelineMetadata,
228 : tenant::upload_queue::{
229 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
230 : },
231 : TENANT_HEATMAP_BASENAME,
232 : };
233 :
234 : use utils::id::{TenantId, TimelineId};
235 :
236 : use self::index::IndexPart;
237 :
238 : use super::storage_layer::{Layer, LayerFileName, ResidentLayer};
239 : use super::upload_queue::SetDeletedFlagProgress;
240 : use super::Generation;
241 :
242 : pub(crate) use download::{is_temp_download_file, list_remote_timelines};
243 : pub(crate) use index::LayerFileMetadata;
244 :
245 : // Occasional network issues and such can cause remote operations to fail, and
246 : // that's expected. If a download fails, we log it at info-level, and retry.
247 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
248 : // level instead, as repeated failures can mean a more serious problem. If it
249 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
250 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
251 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
252 :
253 : // Similarly log failed uploads and deletions at WARN level, after this many
254 : // retries. Uploads and deletions are retried forever, though.
255 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
256 :
257 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
258 :
259 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
260 :
261 : /// Default buffer size when interfacing with [`tokio::fs::File`].
262 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
263 :
264 : pub enum MaybeDeletedIndexPart {
265 : IndexPart(IndexPart),
266 : Deleted(IndexPart),
267 : }
268 :
269 : /// Errors that can arise when calling [`RemoteTimelineClient::stop`].
270 0 : #[derive(Debug, thiserror::Error)]
271 : pub enum StopError {
272 : /// Returned if the upload queue was never initialized.
273 : /// See [`RemoteTimelineClient::init_upload_queue`] and [`RemoteTimelineClient::init_upload_queue_for_empty_remote`].
274 : #[error("queue is not initialized")]
275 : QueueUninitialized,
276 : }
277 :
278 0 : #[derive(Debug, thiserror::Error)]
279 : pub enum PersistIndexPartWithDeletedFlagError {
280 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
281 : AlreadyInProgress(NaiveDateTime),
282 : #[error("the deleted_flag was already set, value is {0:?}")]
283 : AlreadyDeleted(NaiveDateTime),
284 : #[error(transparent)]
285 : Other(#[from] anyhow::Error),
286 : }
287 :
288 : /// A client for accessing a timeline's data in remote storage.
289 : ///
290 : /// This takes care of managing the number of connections, and balancing them
291 : /// across tenants. This also handles retries of failed uploads.
292 : ///
293 : /// Upload and delete requests are ordered so that before a deletion is
294 : /// performed, we wait for all preceding uploads to finish. This ensures sure
295 : /// that if you perform a compaction operation that reshuffles data in layer
296 : /// files, we don't have a transient state where the old files have already been
297 : /// deleted, but new files have not yet been uploaded.
298 : ///
299 : /// Similarly, this enforces an order between index-file uploads, and layer
300 : /// uploads. Before an index-file upload is performed, all preceding layer
301 : /// uploads must be finished.
302 : ///
303 : /// This also maintains a list of remote files, and automatically includes that
304 : /// in the index part file, whenever timeline metadata is uploaded.
305 : ///
306 : /// Downloads are not queued, they are performed immediately.
307 : pub struct RemoteTimelineClient {
308 : conf: &'static PageServerConf,
309 :
310 : runtime: tokio::runtime::Handle,
311 :
312 : tenant_shard_id: TenantShardId,
313 : timeline_id: TimelineId,
314 : generation: Generation,
315 :
316 : upload_queue: Mutex<UploadQueue>,
317 :
318 : metrics: Arc<RemoteTimelineClientMetrics>,
319 :
320 : storage_impl: GenericRemoteStorage,
321 :
322 : deletion_queue_client: DeletionQueueClient,
323 :
324 : cancel: CancellationToken,
325 : }
326 :
327 : impl RemoteTimelineClient {
328 : ///
329 : /// Create a remote storage client for given timeline
330 : ///
331 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
332 : /// by calling init_upload_queue.
333 : ///
334 296 : pub fn new(
335 296 : remote_storage: GenericRemoteStorage,
336 296 : deletion_queue_client: DeletionQueueClient,
337 296 : conf: &'static PageServerConf,
338 296 : tenant_shard_id: TenantShardId,
339 296 : timeline_id: TimelineId,
340 296 : generation: Generation,
341 296 : ) -> RemoteTimelineClient {
342 296 : RemoteTimelineClient {
343 296 : conf,
344 296 : runtime: if cfg!(test) {
345 : // remote_timeline_client.rs tests rely on current-thread runtime
346 296 : tokio::runtime::Handle::current()
347 : } else {
348 0 : BACKGROUND_RUNTIME.handle().clone()
349 : },
350 296 : tenant_shard_id,
351 296 : timeline_id,
352 296 : generation,
353 296 : storage_impl: remote_storage,
354 296 : deletion_queue_client,
355 296 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
356 296 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
357 296 : &tenant_shard_id,
358 296 : &timeline_id,
359 296 : )),
360 296 : cancel: CancellationToken::new(),
361 296 : }
362 296 : }
363 :
364 : /// Initialize the upload queue for a remote storage that already received
365 : /// an index file upload, i.e., it's not empty.
366 : /// The given `index_part` must be the one on the remote.
367 6 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
368 6 : let mut upload_queue = self.upload_queue.lock().unwrap();
369 6 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
370 6 : self.update_remote_physical_size_gauge(Some(index_part));
371 6 : info!(
372 6 : "initialized upload queue from remote index with {} layer files",
373 6 : index_part.layer_metadata.len()
374 6 : );
375 6 : Ok(())
376 6 : }
377 :
378 : /// Initialize the upload queue for the case where the remote storage is empty,
379 : /// i.e., it doesn't have an `IndexPart`.
380 290 : pub fn init_upload_queue_for_empty_remote(
381 290 : &self,
382 290 : local_metadata: &TimelineMetadata,
383 290 : ) -> anyhow::Result<()> {
384 290 : let mut upload_queue = self.upload_queue.lock().unwrap();
385 290 : upload_queue.initialize_empty_remote(local_metadata)?;
386 290 : self.update_remote_physical_size_gauge(None);
387 290 : info!("initialized upload queue as empty");
388 290 : Ok(())
389 290 : }
390 :
391 : /// Initialize the queue in stopped state. Used in startup path
392 : /// to continue deletion operation interrupted by pageserver crash or restart.
393 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
394 0 : &self,
395 0 : index_part: &IndexPart,
396 0 : ) -> anyhow::Result<()> {
397 : // FIXME: consider newtype for DeletedIndexPart.
398 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
399 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
400 0 : ))?;
401 :
402 : {
403 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
404 0 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
405 0 : self.update_remote_physical_size_gauge(Some(index_part));
406 0 : }
407 0 : // also locks upload queue, without dropping the guard above it will be a deadlock
408 0 : self.stop().expect("initialized line above");
409 0 :
410 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
411 0 :
412 0 : upload_queue
413 0 : .stopped_mut()
414 0 : .expect("stopped above")
415 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
416 0 :
417 0 : Ok(())
418 0 : }
419 :
420 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
421 0 : match &mut *self.upload_queue.lock().unwrap() {
422 0 : UploadQueue::Uninitialized => None,
423 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
424 0 : UploadQueue::Stopped(q) => q
425 0 : .upload_queue_for_deletion
426 0 : .get_last_remote_consistent_lsn_projected(),
427 : }
428 0 : }
429 :
430 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
431 0 : match &mut *self.upload_queue.lock().unwrap() {
432 0 : UploadQueue::Uninitialized => None,
433 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
434 0 : UploadQueue::Stopped(q) => Some(
435 0 : q.upload_queue_for_deletion
436 0 : .get_last_remote_consistent_lsn_visible(),
437 0 : ),
438 : }
439 0 : }
440 :
441 1041 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
442 1041 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
443 751 : current_remote_index_part
444 751 : .layer_metadata
445 751 : .values()
446 751 : // If we don't have the file size for the layer, don't account for it in the metric.
447 2922 : .map(|ilmd| ilmd.file_size)
448 751 : .sum()
449 : } else {
450 290 : 0
451 : };
452 1041 : self.metrics.remote_physical_size_set(size);
453 1041 : }
454 :
455 0 : pub fn get_remote_physical_size(&self) -> u64 {
456 0 : self.metrics.remote_physical_size_get()
457 0 : }
458 :
459 : //
460 : // Download operations.
461 : //
462 : // These don't use the per-timeline queue. They do use the global semaphore in
463 : // S3Bucket, to limit the total number of concurrent operations, though.
464 : //
465 :
466 : /// Download index file
467 20 : pub async fn download_index_file(
468 20 : &self,
469 20 : cancel: &CancellationToken,
470 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
471 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
472 20 : &RemoteOpFileKind::Index,
473 20 : &RemoteOpKind::Download,
474 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
475 20 : reason: "no need for a downloads gauge",
476 20 : },
477 20 : );
478 :
479 20 : let index_part = download::download_index_part(
480 20 : &self.storage_impl,
481 20 : &self.tenant_shard_id,
482 20 : &self.timeline_id,
483 20 : self.generation,
484 20 : cancel,
485 20 : )
486 20 : .measure_remote_op(
487 20 : RemoteOpFileKind::Index,
488 20 : RemoteOpKind::Download,
489 20 : Arc::clone(&self.metrics),
490 20 : )
491 72 : .await?;
492 :
493 20 : if index_part.deleted_at.is_some() {
494 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
495 : } else {
496 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
497 : }
498 20 : }
499 :
500 : /// Download a (layer) file from `path`, into local filesystem.
501 : ///
502 : /// 'layer_metadata' is the metadata from the remote index file.
503 : ///
504 : /// On success, returns the size of the downloaded file.
505 0 : pub async fn download_layer_file(
506 0 : &self,
507 0 : layer_file_name: &LayerFileName,
508 0 : layer_metadata: &LayerFileMetadata,
509 0 : cancel: &CancellationToken,
510 0 : ) -> anyhow::Result<u64> {
511 0 : let downloaded_size = {
512 0 : let _unfinished_gauge_guard = self.metrics.call_begin(
513 0 : &RemoteOpFileKind::Layer,
514 0 : &RemoteOpKind::Download,
515 0 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
516 0 : reason: "no need for a downloads gauge",
517 0 : },
518 0 : );
519 0 : download::download_layer_file(
520 0 : self.conf,
521 0 : &self.storage_impl,
522 0 : self.tenant_shard_id,
523 0 : self.timeline_id,
524 0 : layer_file_name,
525 0 : layer_metadata,
526 0 : cancel,
527 0 : )
528 0 : .measure_remote_op(
529 0 : RemoteOpFileKind::Layer,
530 0 : RemoteOpKind::Download,
531 0 : Arc::clone(&self.metrics),
532 0 : )
533 0 : .await?
534 : };
535 :
536 0 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
537 0 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
538 0 :
539 0 : Ok(downloaded_size)
540 0 : }
541 :
542 : //
543 : // Upload operations.
544 : //
545 :
546 : ///
547 : /// Launch an index-file upload operation in the background, with
548 : /// updated metadata.
549 : ///
550 : /// The upload will be added to the queue immediately, but it
551 : /// won't be performed until all previously scheduled layer file
552 : /// upload operations have completed successfully. This is to
553 : /// ensure that when the index file claims that layers X, Y and Z
554 : /// exist in remote storage, they really do. To wait for the upload
555 : /// to complete, use `wait_completion`.
556 : ///
557 : /// If there were any changes to the list of files, i.e. if any
558 : /// layer file uploads were scheduled, since the last index file
559 : /// upload, those will be included too.
560 736 : pub fn schedule_index_upload_for_metadata_update(
561 736 : self: &Arc<Self>,
562 736 : metadata: &TimelineMetadata,
563 736 : ) -> anyhow::Result<()> {
564 736 : let mut guard = self.upload_queue.lock().unwrap();
565 736 : let upload_queue = guard.initialized_mut()?;
566 :
567 : // As documented in the struct definition, it's ok for latest_metadata to be
568 : // ahead of what's _actually_ on the remote during index upload.
569 736 : upload_queue.latest_metadata = metadata.clone();
570 736 :
571 736 : self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
572 736 :
573 736 : Ok(())
574 736 : }
575 :
576 : ///
577 : /// Launch an index-file upload operation in the background, if necessary.
578 : ///
579 : /// Use this function to schedule the update of the index file after
580 : /// scheduling file uploads or deletions. If no file uploads or deletions
581 : /// have been scheduled since the last index file upload, this does
582 : /// nothing.
583 : ///
584 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
585 : /// the upload to the upload queue and returns quickly.
586 414 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> anyhow::Result<()> {
587 414 : let mut guard = self.upload_queue.lock().unwrap();
588 414 : let upload_queue = guard.initialized_mut()?;
589 :
590 414 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
591 0 : self.schedule_index_upload(upload_queue, upload_queue.latest_metadata.clone());
592 414 : }
593 :
594 414 : Ok(())
595 414 : }
596 :
597 : /// Launch an index-file upload operation in the background (internal function)
598 768 : fn schedule_index_upload(
599 768 : self: &Arc<Self>,
600 768 : upload_queue: &mut UploadQueueInitialized,
601 768 : metadata: TimelineMetadata,
602 768 : ) {
603 768 : info!(
604 768 : "scheduling metadata upload with {} files ({} changed)",
605 768 : upload_queue.latest_files.len(),
606 768 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
607 768 : );
608 :
609 768 : let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();
610 768 :
611 768 : let index_part = IndexPart::new(
612 768 : upload_queue.latest_files.clone(),
613 768 : disk_consistent_lsn,
614 768 : metadata,
615 768 : );
616 768 : let op = UploadOp::UploadMetadata(index_part, disk_consistent_lsn);
617 768 : self.metric_begin(&op);
618 768 : upload_queue.queued_operations.push_back(op);
619 768 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
620 768 :
621 768 : // Launch the task immediately, if possible
622 768 : self.launch_queued_tasks(upload_queue);
623 768 : }
624 :
625 : ///
626 : /// Launch an upload operation in the background.
627 : ///
628 532 : pub(crate) fn schedule_layer_file_upload(
629 532 : self: &Arc<Self>,
630 532 : layer: ResidentLayer,
631 532 : ) -> anyhow::Result<()> {
632 532 : let mut guard = self.upload_queue.lock().unwrap();
633 532 : let upload_queue = guard.initialized_mut()?;
634 :
635 532 : self.schedule_layer_file_upload0(upload_queue, layer);
636 532 : self.launch_queued_tasks(upload_queue);
637 532 : Ok(())
638 532 : }
639 :
640 562 : fn schedule_layer_file_upload0(
641 562 : self: &Arc<Self>,
642 562 : upload_queue: &mut UploadQueueInitialized,
643 562 : layer: ResidentLayer,
644 562 : ) {
645 562 : let metadata = layer.metadata();
646 562 :
647 562 : upload_queue
648 562 : .latest_files
649 562 : .insert(layer.layer_desc().filename(), metadata.clone());
650 562 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
651 562 :
652 562 : info!(
653 562 : "scheduled layer file upload {layer} gen={:?} shard={:?}",
654 562 : metadata.generation, metadata.shard
655 562 : );
656 562 : let op = UploadOp::UploadLayer(layer, metadata);
657 562 : self.metric_begin(&op);
658 562 : upload_queue.queued_operations.push_back(op);
659 562 : }
660 :
661 : /// Launch a delete operation in the background.
662 : ///
663 : /// The operation does not modify local filesystem state.
664 : ///
665 : /// Note: This schedules an index file upload before the deletions. The
666 : /// deletion won't actually be performed, until all previously scheduled
667 : /// upload operations, and the index file upload, have completed
668 : /// successfully.
669 8 : pub fn schedule_layer_file_deletion(
670 8 : self: &Arc<Self>,
671 8 : names: &[LayerFileName],
672 8 : ) -> anyhow::Result<()> {
673 8 : let mut guard = self.upload_queue.lock().unwrap();
674 8 : let upload_queue = guard.initialized_mut()?;
675 :
676 8 : let with_metadata =
677 8 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned());
678 8 :
679 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
680 8 :
681 8 : // Launch the tasks immediately, if possible
682 8 : self.launch_queued_tasks(upload_queue);
683 8 : Ok(())
684 8 : }
685 :
686 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
687 : /// layer files, leaving them dangling.
688 : ///
689 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
690 : /// is invoked on them.
691 0 : pub(crate) fn schedule_gc_update(self: &Arc<Self>, gc_layers: &[Layer]) -> anyhow::Result<()> {
692 0 : let mut guard = self.upload_queue.lock().unwrap();
693 0 : let upload_queue = guard.initialized_mut()?;
694 :
695 : // just forget the return value; after uploading the next index_part.json, we can consider
696 : // the layer files as "dangling". this is fine, at worst case we create work for the
697 : // scrubber.
698 :
699 0 : let names = gc_layers.iter().map(|x| x.layer_desc().filename());
700 0 :
701 0 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
702 0 :
703 0 : self.launch_queued_tasks(upload_queue);
704 0 :
705 0 : Ok(())
706 0 : }
707 :
708 : /// Update the remote index file, removing the to-be-deleted files from the index,
709 : /// allowing scheduling of actual deletions later.
710 38 : fn schedule_unlinking_of_layers_from_index_part0<I>(
711 38 : self: &Arc<Self>,
712 38 : upload_queue: &mut UploadQueueInitialized,
713 38 : names: I,
714 38 : ) -> Vec<(LayerFileName, LayerFileMetadata)>
715 38 : where
716 38 : I: IntoIterator<Item = LayerFileName>,
717 38 : {
718 38 : // Deleting layers doesn't affect the values stored in TimelineMetadata,
719 38 : // so we don't need update it. Just serialize it.
720 38 : let metadata = upload_queue.latest_metadata.clone();
721 38 :
722 38 : // Decorate our list of names with each name's metadata, dropping
723 38 : // names that are unexpectedly missing from our metadata. This metadata
724 38 : // is later used when physically deleting layers, to construct key paths.
725 38 : let with_metadata: Vec<_> = names
726 38 : .into_iter()
727 302 : .filter_map(|name| {
728 302 : let meta = upload_queue.latest_files.remove(&name);
729 :
730 302 : if let Some(meta) = meta {
731 302 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
732 302 : Some((name, meta))
733 : } else {
734 : // This can only happen if we forgot to to schedule the file upload
735 : // before scheduling the delete. Log it because it is a rare/strange
736 : // situation, and in case something is misbehaving, we'd like to know which
737 : // layers experienced this.
738 0 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
739 0 : None
740 : }
741 302 : })
742 38 : .collect();
743 :
744 : #[cfg(feature = "testing")]
745 340 : for (name, metadata) in &with_metadata {
746 302 : let gen = metadata.generation;
747 302 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
748 0 : if unexpected == gen {
749 0 : tracing::error!("{name} was unlinked twice with same generation");
750 : } else {
751 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
752 : }
753 302 : }
754 : }
755 :
756 : // after unlinking files from the upload_queue.latest_files we must always schedule an
757 : // index_part update, because that needs to be uploaded before we can actually delete the
758 : // files.
759 38 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
760 32 : self.schedule_index_upload(upload_queue, metadata);
761 32 : }
762 :
763 38 : with_metadata
764 38 : }
765 :
766 : /// Schedules deletion for layer files which have previously been unlinked from the
767 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
768 302 : pub(crate) fn schedule_deletion_of_unlinked(
769 302 : self: &Arc<Self>,
770 302 : layers: Vec<(LayerFileName, LayerFileMetadata)>,
771 302 : ) -> anyhow::Result<()> {
772 302 : let mut guard = self.upload_queue.lock().unwrap();
773 302 : let upload_queue = guard.initialized_mut()?;
774 :
775 302 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
776 302 : self.launch_queued_tasks(upload_queue);
777 302 : Ok(())
778 302 : }
779 :
780 310 : fn schedule_deletion_of_unlinked0(
781 310 : self: &Arc<Self>,
782 310 : upload_queue: &mut UploadQueueInitialized,
783 310 : mut with_metadata: Vec<(LayerFileName, LayerFileMetadata)>,
784 310 : ) {
785 310 : // Filter out any layers which were not created by this tenant shard. These are
786 310 : // layers that originate from some ancestor shard after a split, and may still
787 310 : // be referenced by other shards. We are free to delete them locally and remove
788 310 : // them from our index (and would have already done so when we reach this point
789 310 : // in the code), but we may not delete them remotely.
790 310 : with_metadata.retain(|(name, meta)| {
791 304 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
792 304 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
793 304 : if !retain {
794 0 : tracing::debug!(
795 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
796 0 : meta.shard
797 0 : );
798 304 : }
799 304 : retain
800 310 : });
801 :
802 614 : for (name, meta) in &with_metadata {
803 304 : info!(
804 304 : "scheduling deletion of layer {}{} (shard {})",
805 304 : name,
806 304 : meta.generation.get_suffix(),
807 304 : meta.shard
808 304 : );
809 : }
810 :
811 : #[cfg(feature = "testing")]
812 614 : for (name, meta) in &with_metadata {
813 304 : let gen = meta.generation;
814 304 : match upload_queue.dangling_files.remove(name) {
815 302 : Some(same) if same == gen => { /* expected */ }
816 0 : Some(other) => {
817 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
818 : }
819 : None => {
820 2 : tracing::error!("{name} was unlinked but was not dangling");
821 : }
822 : }
823 : }
824 :
825 : // schedule the actual deletions
826 310 : if with_metadata.is_empty() {
827 : // avoid scheduling the op & bumping the metric
828 6 : return;
829 304 : }
830 304 : let op = UploadOp::Delete(Delete {
831 304 : layers: with_metadata,
832 304 : });
833 304 : self.metric_begin(&op);
834 304 : upload_queue.queued_operations.push_back(op);
835 310 : }
836 :
837 : /// Schedules a compaction update to the remote `index_part.json`.
838 : ///
839 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
840 30 : pub(crate) fn schedule_compaction_update(
841 30 : self: &Arc<Self>,
842 30 : compacted_from: &[Layer],
843 30 : compacted_to: &[ResidentLayer],
844 30 : ) -> anyhow::Result<()> {
845 30 : let mut guard = self.upload_queue.lock().unwrap();
846 30 : let upload_queue = guard.initialized_mut()?;
847 :
848 60 : for layer in compacted_to {
849 30 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
850 30 : }
851 :
852 300 : let names = compacted_from.iter().map(|x| x.layer_desc().filename());
853 30 :
854 30 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
855 30 : self.launch_queued_tasks(upload_queue);
856 30 :
857 30 : Ok(())
858 30 : }
859 :
860 : /// Wait for all previously scheduled uploads/deletions to complete
861 78 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
862 78 : let mut receiver = {
863 78 : let mut guard = self.upload_queue.lock().unwrap();
864 78 : let upload_queue = guard.initialized_mut()?;
865 78 : self.schedule_barrier0(upload_queue)
866 78 : };
867 78 :
868 78 : if receiver.changed().await.is_err() {
869 0 : anyhow::bail!("wait_completion aborted because upload queue was stopped");
870 78 : }
871 78 :
872 78 : Ok(())
873 78 : }
874 :
875 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
876 6 : let mut guard = self.upload_queue.lock().unwrap();
877 6 : let upload_queue = guard.initialized_mut()?;
878 6 : self.schedule_barrier0(upload_queue);
879 6 : Ok(())
880 6 : }
881 :
882 84 : fn schedule_barrier0(
883 84 : self: &Arc<Self>,
884 84 : upload_queue: &mut UploadQueueInitialized,
885 84 : ) -> tokio::sync::watch::Receiver<()> {
886 84 : let (sender, receiver) = tokio::sync::watch::channel(());
887 84 : let barrier_op = UploadOp::Barrier(sender);
888 84 :
889 84 : upload_queue.queued_operations.push_back(barrier_op);
890 84 : // Don't count this kind of operation!
891 84 :
892 84 : // Launch the task immediately, if possible
893 84 : self.launch_queued_tasks(upload_queue);
894 84 :
895 84 : receiver
896 84 : }
897 :
898 : /// Wait for all previously scheduled operations to complete, and then stop.
899 : ///
900 : /// Not cancellation safe
901 6 : pub(crate) async fn shutdown(self: &Arc<Self>) -> Result<(), StopError> {
902 6 : // On cancellation the queue is left in ackward state of refusing new operations but
903 6 : // proper stop is yet to be called. On cancel the original or some later task must call
904 6 : // `stop` or `shutdown`.
905 6 : let sg = scopeguard::guard((), |_| {
906 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
907 6 : });
908 :
909 6 : let fut = {
910 6 : let mut guard = self.upload_queue.lock().unwrap();
911 6 : let upload_queue = match &mut *guard {
912 0 : UploadQueue::Stopped(_) => return Ok(()),
913 0 : UploadQueue::Uninitialized => return Err(StopError::QueueUninitialized),
914 6 : UploadQueue::Initialized(ref mut init) => init,
915 6 : };
916 6 :
917 6 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
918 6 : // just don't add more of these as they would never complete.
919 6 : //
920 6 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
921 6 : // in every place we would not have to jump through this hoop, and this method could be
922 6 : // made cancellable.
923 6 : if !upload_queue.shutting_down {
924 6 : upload_queue.shutting_down = true;
925 6 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
926 6 : // this operation is not counted similar to Barrier
927 6 :
928 6 : self.launch_queued_tasks(upload_queue);
929 6 : }
930 :
931 6 : upload_queue.shutdown_ready.clone().acquire_owned()
932 : };
933 :
934 6 : let res = fut.await;
935 :
936 6 : scopeguard::ScopeGuard::into_inner(sg);
937 6 :
938 6 : match res {
939 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
940 6 : Err(_closed) => {
941 6 : // expected
942 6 : }
943 6 : }
944 6 :
945 6 : self.stop()
946 6 : }
947 :
948 : /// Set the deleted_at field in the remote index file.
949 : ///
950 : /// This fails if the upload queue has not been `stop()`ed.
951 : ///
952 : /// The caller is responsible for calling `stop()` AND for waiting
953 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
954 : /// Check method [`RemoteTimelineClient::stop`] for details.
955 0 : #[instrument(skip_all)]
956 : pub(crate) async fn persist_index_part_with_deleted_flag(
957 : self: &Arc<Self>,
958 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
959 : let index_part_with_deleted_at = {
960 : let mut locked = self.upload_queue.lock().unwrap();
961 :
962 : // We must be in stopped state because otherwise
963 : // we can have inprogress index part upload that can overwrite the file
964 : // with missing is_deleted flag that we going to set below
965 : let stopped = locked.stopped_mut()?;
966 :
967 : match stopped.deleted_at {
968 : SetDeletedFlagProgress::NotRunning => (), // proceed
969 : SetDeletedFlagProgress::InProgress(at) => {
970 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
971 : }
972 : SetDeletedFlagProgress::Successful(at) => {
973 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
974 : }
975 : };
976 : let deleted_at = Utc::now().naive_utc();
977 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
978 :
979 : let mut index_part = IndexPart::try_from(&stopped.upload_queue_for_deletion)
980 : .context("IndexPart serialize")?;
981 : index_part.deleted_at = Some(deleted_at);
982 : index_part
983 : };
984 :
985 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
986 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
987 0 : let stopped = locked
988 0 : .stopped_mut()
989 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
990 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
991 0 : });
992 :
993 0 : pausable_failpoint!("persist_deleted_index_part");
994 :
995 : backoff::retry(
996 0 : || {
997 0 : upload::upload_index_part(
998 0 : &self.storage_impl,
999 0 : &self.tenant_shard_id,
1000 0 : &self.timeline_id,
1001 0 : self.generation,
1002 0 : &index_part_with_deleted_at,
1003 0 : &self.cancel,
1004 0 : )
1005 0 : },
1006 0 : |_e| false,
1007 : 1,
1008 : // have just a couple of attempts
1009 : // when executed as part of timeline deletion this happens in context of api call
1010 : // when executed as part of tenant deletion this happens in the background
1011 : 2,
1012 : "persist_index_part_with_deleted_flag",
1013 : &self.cancel,
1014 : )
1015 : .await
1016 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1017 0 : .and_then(|x| x)?;
1018 :
1019 : // all good, disarm the guard and mark as success
1020 : ScopeGuard::into_inner(undo_deleted_at);
1021 : {
1022 : let mut locked = self.upload_queue.lock().unwrap();
1023 :
1024 : let stopped = locked
1025 : .stopped_mut()
1026 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1027 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1028 : index_part_with_deleted_at
1029 : .deleted_at
1030 : .expect("we set it above"),
1031 : );
1032 : }
1033 :
1034 : Ok(())
1035 : }
1036 :
1037 0 : pub(crate) async fn preserve_initdb_archive(
1038 0 : self: &Arc<Self>,
1039 0 : tenant_id: &TenantId,
1040 0 : timeline_id: &TimelineId,
1041 0 : cancel: &CancellationToken,
1042 0 : ) -> anyhow::Result<()> {
1043 0 : backoff::retry(
1044 0 : || async {
1045 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1046 0 : .await
1047 0 : },
1048 0 : TimeoutOrCancel::caused_by_cancel,
1049 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1050 0 : FAILED_REMOTE_OP_RETRIES,
1051 0 : "preserve_initdb_tar_zst",
1052 0 : &cancel.clone(),
1053 0 : )
1054 0 : .await
1055 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1056 0 : .and_then(|x| x)
1057 0 : .context("backing up initdb archive")?;
1058 0 : Ok(())
1059 0 : }
1060 :
1061 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1062 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1063 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1064 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
1065 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1066 :
1067 0 : let layers: Vec<RemotePath> = {
1068 0 : let mut locked = self.upload_queue.lock().unwrap();
1069 0 : let stopped = locked.stopped_mut()?;
1070 :
1071 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1072 0 : anyhow::bail!("deleted_at is not set")
1073 0 : }
1074 :
1075 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1076 :
1077 0 : stopped
1078 0 : .upload_queue_for_deletion
1079 0 : .latest_files
1080 0 : .drain()
1081 0 : .map(|(file_name, meta)| {
1082 0 : remote_layer_path(
1083 0 : &self.tenant_shard_id.tenant_id,
1084 0 : &self.timeline_id,
1085 0 : meta.shard,
1086 0 : &file_name,
1087 0 : meta.generation,
1088 0 : )
1089 0 : })
1090 0 : .collect()
1091 0 : };
1092 0 :
1093 0 : let layer_deletion_count = layers.len();
1094 0 : self.deletion_queue_client.push_immediate(layers).await?;
1095 :
1096 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1097 : // inexistant objects are not considered errors.
1098 0 : let initdb_path =
1099 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1100 0 : self.deletion_queue_client
1101 0 : .push_immediate(vec![initdb_path])
1102 0 : .await?;
1103 :
1104 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1105 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1106 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1107 0 :
1108 0 : // Execute all pending deletions, so that when we proceed to do a list_prefixes below, we aren't
1109 0 : // taking the burden of listing all the layers that we already know we should delete.
1110 0 : self.deletion_queue_client.flush_immediate().await?;
1111 :
1112 0 : let cancel = shutdown_token();
1113 :
1114 0 : let remaining = download_retry(
1115 0 : || async {
1116 0 : self.storage_impl
1117 0 : .list_files(Some(&timeline_storage_path), None, &cancel)
1118 0 : .await
1119 0 : },
1120 0 : "list remaining files",
1121 0 : &cancel,
1122 0 : )
1123 0 : .await
1124 0 : .context("list files remaining files")?;
1125 :
1126 : // We will delete the current index_part object last, since it acts as a deletion
1127 : // marker via its deleted_at attribute
1128 0 : let latest_index = remaining
1129 0 : .iter()
1130 0 : .filter(|p| {
1131 0 : p.object_name()
1132 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1133 0 : .unwrap_or(false)
1134 0 : })
1135 0 : .filter_map(|path| parse_remote_index_path(path.clone()).map(|gen| (path, gen)))
1136 0 : .max_by_key(|i| i.1)
1137 0 : .map(|i| i.0.clone())
1138 0 : .unwrap_or(
1139 0 : // No generation-suffixed indices, assume we are dealing with
1140 0 : // a legacy index.
1141 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1142 0 : );
1143 0 :
1144 0 : let remaining_layers: Vec<RemotePath> = remaining
1145 0 : .into_iter()
1146 0 : .filter(|p| {
1147 0 : if p == &latest_index {
1148 0 : return false;
1149 0 : }
1150 0 : if p.object_name() == Some(INITDB_PRESERVED_PATH) {
1151 0 : return false;
1152 0 : }
1153 0 : true
1154 0 : })
1155 0 : .inspect(|path| {
1156 0 : if let Some(name) = path.object_name() {
1157 0 : info!(%name, "deleting a file not referenced from index_part.json");
1158 : } else {
1159 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1160 : }
1161 0 : })
1162 0 : .collect();
1163 0 :
1164 0 : let not_referenced_count = remaining_layers.len();
1165 0 : if !remaining_layers.is_empty() {
1166 0 : self.deletion_queue_client
1167 0 : .push_immediate(remaining_layers)
1168 0 : .await?;
1169 0 : }
1170 :
1171 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1172 0 : Err(anyhow::anyhow!(
1173 0 : "failpoint: timeline-delete-before-index-delete"
1174 0 : ))?
1175 0 : });
1176 :
1177 0 : debug!("enqueuing index part deletion");
1178 0 : self.deletion_queue_client
1179 0 : .push_immediate([latest_index].to_vec())
1180 0 : .await?;
1181 :
1182 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1183 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1184 0 : self.deletion_queue_client.flush_immediate().await?;
1185 :
1186 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1187 0 : Err(anyhow::anyhow!(
1188 0 : "failpoint: timeline-delete-after-index-delete"
1189 0 : ))?
1190 0 : });
1191 :
1192 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1193 :
1194 0 : Ok(())
1195 0 : }
1196 :
1197 : ///
1198 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1199 : /// the ordering constraints.
1200 : ///
1201 : /// The caller needs to already hold the `upload_queue` lock.
1202 3262 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1203 4899 : while let Some(next_op) = upload_queue.queued_operations.front() {
1204 : // Can we run this task now?
1205 2720 : let can_run_now = match next_op {
1206 : UploadOp::UploadLayer(_, _) => {
1207 : // Can always be scheduled.
1208 560 : true
1209 : }
1210 : UploadOp::UploadMetadata(_, _) => {
1211 : // These can only be performed after all the preceding operations
1212 : // have finished.
1213 1720 : upload_queue.inprogress_tasks.is_empty()
1214 : }
1215 : UploadOp::Delete(_) => {
1216 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1217 270 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1218 : }
1219 :
1220 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1221 170 : upload_queue.inprogress_tasks.is_empty()
1222 : }
1223 : };
1224 :
1225 : // If we cannot launch this task, don't look any further.
1226 : //
1227 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1228 : // them now, but we don't try to do that currently. For example, if the frontmost task
1229 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1230 : // could still start layer uploads that were scheduled later.
1231 2720 : if !can_run_now {
1232 1077 : break;
1233 1643 : }
1234 1643 :
1235 1643 : if let UploadOp::Shutdown = next_op {
1236 : // leave the op in the queue but do not start more tasks; it will be dropped when
1237 : // the stop is called.
1238 6 : upload_queue.shutdown_ready.close();
1239 6 : break;
1240 1637 : }
1241 1637 :
1242 1637 : // We can launch this task. Remove it from the queue first.
1243 1637 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1244 1637 :
1245 1637 : debug!("starting op: {}", next_op);
1246 :
1247 : // Update the counters
1248 1637 : match next_op {
1249 560 : UploadOp::UploadLayer(_, _) => {
1250 560 : upload_queue.num_inprogress_layer_uploads += 1;
1251 560 : }
1252 749 : UploadOp::UploadMetadata(_, _) => {
1253 749 : upload_queue.num_inprogress_metadata_uploads += 1;
1254 749 : }
1255 244 : UploadOp::Delete(_) => {
1256 244 : upload_queue.num_inprogress_deletions += 1;
1257 244 : }
1258 84 : UploadOp::Barrier(sender) => {
1259 84 : sender.send_replace(());
1260 84 : continue;
1261 : }
1262 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1263 : };
1264 :
1265 : // Assign unique ID to this task
1266 1553 : upload_queue.task_counter += 1;
1267 1553 : let upload_task_id = upload_queue.task_counter;
1268 1553 :
1269 1553 : // Add it to the in-progress map
1270 1553 : let task = Arc::new(UploadTask {
1271 1553 : task_id: upload_task_id,
1272 1553 : op: next_op,
1273 1553 : retries: AtomicU32::new(0),
1274 1553 : });
1275 1553 : upload_queue
1276 1553 : .inprogress_tasks
1277 1553 : .insert(task.task_id, Arc::clone(&task));
1278 1553 :
1279 1553 : // Spawn task to perform the task
1280 1553 : let self_rc = Arc::clone(self);
1281 1553 : let tenant_shard_id = self.tenant_shard_id;
1282 1553 : let timeline_id = self.timeline_id;
1283 1553 : task_mgr::spawn(
1284 1553 : &self.runtime,
1285 1553 : TaskKind::RemoteUploadTask,
1286 1553 : Some(self.tenant_shard_id),
1287 1553 : Some(self.timeline_id),
1288 1553 : "remote upload",
1289 : false,
1290 1546 : async move {
1291 22306 : self_rc.perform_upload_task(task).await;
1292 1532 : Ok(())
1293 1532 : }
1294 1553 : .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)),
1295 : );
1296 :
1297 : // Loop back to process next task
1298 : }
1299 3262 : }
1300 :
1301 : ///
1302 : /// Perform an upload task.
1303 : ///
1304 : /// The task is in the `inprogress_tasks` list. This function will try to
1305 : /// execute it, retrying forever. On successful completion, the task is
1306 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1307 : /// queue that were waiting by the completion are launched.
1308 : ///
1309 : /// The task can be shut down, however. That leads to stopping the whole
1310 : /// queue.
1311 : ///
1312 1546 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1313 1546 : let cancel = shutdown_token();
1314 : // Loop to retry until it completes.
1315 1546 : loop {
1316 1546 : // If we're requested to shut down, close up shop and exit.
1317 1546 : //
1318 1546 : // Note: We only check for the shutdown requests between retries, so
1319 1546 : // if a shutdown request arrives while we're busy uploading, in the
1320 1546 : // upload::upload:*() call below, we will wait not exit until it has
1321 1546 : // finished. We probably could cancel the upload by simply dropping
1322 1546 : // the Future, but we're not 100% sure if the remote storage library
1323 1546 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1324 1546 : // upload finishes or times out soon enough.
1325 1546 : if cancel.is_cancelled() {
1326 0 : info!("upload task cancelled by shutdown request");
1327 0 : match self.stop() {
1328 0 : Ok(()) => {}
1329 : Err(StopError::QueueUninitialized) => {
1330 0 : unreachable!("we never launch an upload task if the queue is uninitialized, and once it is initialized, we never go back")
1331 : }
1332 : }
1333 0 : return;
1334 1546 : }
1335 :
1336 1546 : let upload_result: anyhow::Result<()> = match &task.op {
1337 555 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1338 555 : let path = layer.local_path();
1339 555 : upload::upload_timeline_layer(
1340 555 : self.conf,
1341 555 : &self.storage_impl,
1342 555 : path,
1343 555 : layer_metadata,
1344 555 : self.generation,
1345 555 : &self.cancel,
1346 555 : )
1347 555 : .measure_remote_op(
1348 555 : RemoteOpFileKind::Layer,
1349 555 : RemoteOpKind::Upload,
1350 555 : Arc::clone(&self.metrics),
1351 555 : )
1352 18891 : .await
1353 : }
1354 747 : UploadOp::UploadMetadata(ref index_part, _lsn) => {
1355 747 : let mention_having_future_layers = if cfg!(feature = "testing") {
1356 747 : index_part
1357 747 : .layer_metadata
1358 747 : .keys()
1359 2900 : .any(|x| x.is_in_future(*_lsn))
1360 : } else {
1361 0 : false
1362 : };
1363 :
1364 747 : let res = upload::upload_index_part(
1365 747 : &self.storage_impl,
1366 747 : &self.tenant_shard_id,
1367 747 : &self.timeline_id,
1368 747 : self.generation,
1369 747 : index_part,
1370 747 : &self.cancel,
1371 747 : )
1372 747 : .measure_remote_op(
1373 747 : RemoteOpFileKind::Index,
1374 747 : RemoteOpKind::Upload,
1375 747 : Arc::clone(&self.metrics),
1376 747 : )
1377 3171 : .await;
1378 745 : if res.is_ok() {
1379 745 : self.update_remote_physical_size_gauge(Some(index_part));
1380 745 : if mention_having_future_layers {
1381 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1382 4 : tracing::info!(disk_consistent_lsn=%_lsn, "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup");
1383 741 : }
1384 0 : }
1385 745 : res
1386 : }
1387 244 : UploadOp::Delete(delete) => {
1388 244 : pausable_failpoint!("before-delete-layer-pausable");
1389 244 : self.deletion_queue_client
1390 244 : .push_layers(
1391 244 : self.tenant_shard_id,
1392 244 : self.timeline_id,
1393 244 : self.generation,
1394 244 : delete.layers.clone(),
1395 244 : )
1396 0 : .await
1397 244 : .map_err(|e| anyhow::anyhow!(e))
1398 : }
1399 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1400 : // unreachable. Barrier operations are handled synchronously in
1401 : // launch_queued_tasks
1402 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1403 0 : break;
1404 : }
1405 : };
1406 :
1407 0 : match upload_result {
1408 : Ok(()) => {
1409 1532 : break;
1410 : }
1411 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1412 0 : // loop around to do the proper stopping
1413 0 : continue;
1414 : }
1415 0 : Err(e) => {
1416 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1417 0 :
1418 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1419 0 : // or other external reasons. Such issues are relatively regular, so log them
1420 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1421 0 : //
1422 0 : // (See similar logic for downloads in `download::download_retry`)
1423 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1424 0 : info!(
1425 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1426 0 : task.op, retries, e
1427 0 : );
1428 : } else {
1429 0 : warn!(
1430 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1431 0 : task.op, retries, e
1432 0 : );
1433 : }
1434 :
1435 : // sleep until it's time to retry, or we're cancelled
1436 0 : exponential_backoff(
1437 0 : retries,
1438 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1439 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1440 0 : &cancel,
1441 0 : )
1442 0 : .await;
1443 : }
1444 : }
1445 : }
1446 :
1447 1532 : let retries = task.retries.load(Ordering::SeqCst);
1448 1532 : if retries > 0 {
1449 0 : info!(
1450 0 : "remote task {} completed successfully after {} retries",
1451 0 : task.op, retries
1452 0 : );
1453 : } else {
1454 0 : debug!("remote task {} completed successfully", task.op);
1455 : }
1456 :
1457 : // The task has completed successfully. Remove it from the in-progress list.
1458 1532 : let lsn_update = {
1459 1532 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1460 1532 : let upload_queue = match upload_queue_guard.deref_mut() {
1461 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1462 0 : UploadQueue::Stopped(_stopped) => {
1463 0 : None
1464 : },
1465 1532 : UploadQueue::Initialized(qi) => { Some(qi) }
1466 : };
1467 :
1468 1532 : let upload_queue = match upload_queue {
1469 1532 : Some(upload_queue) => upload_queue,
1470 : None => {
1471 0 : info!("another concurrent task already stopped the queue");
1472 0 : return;
1473 : }
1474 : };
1475 :
1476 1532 : upload_queue.inprogress_tasks.remove(&task.task_id);
1477 :
1478 1532 : let lsn_update = match task.op {
1479 : UploadOp::UploadLayer(_, _) => {
1480 543 : upload_queue.num_inprogress_layer_uploads -= 1;
1481 543 : None
1482 : }
1483 745 : UploadOp::UploadMetadata(_, lsn) => {
1484 745 : upload_queue.num_inprogress_metadata_uploads -= 1;
1485 745 : // XXX monotonicity check?
1486 745 :
1487 745 : upload_queue.projected_remote_consistent_lsn = Some(lsn);
1488 745 : if self.generation.is_none() {
1489 : // Legacy mode: skip validating generation
1490 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
1491 0 : None
1492 : } else {
1493 745 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
1494 : }
1495 : }
1496 : UploadOp::Delete(_) => {
1497 244 : upload_queue.num_inprogress_deletions -= 1;
1498 244 : None
1499 : }
1500 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
1501 : };
1502 :
1503 : // Launch any queued tasks that were unblocked by this one.
1504 1532 : self.launch_queued_tasks(upload_queue);
1505 1532 : lsn_update
1506 : };
1507 :
1508 1532 : if let Some((lsn, slot)) = lsn_update {
1509 : // Updates to the remote_consistent_lsn we advertise to pageservers
1510 : // are all routed through the DeletionQueue, to enforce important
1511 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
1512 745 : self.deletion_queue_client
1513 745 : .update_remote_consistent_lsn(
1514 745 : self.tenant_shard_id,
1515 745 : self.timeline_id,
1516 745 : self.generation,
1517 745 : lsn,
1518 745 : slot,
1519 745 : )
1520 0 : .await;
1521 787 : }
1522 :
1523 1532 : self.metric_end(&task.op);
1524 1532 : }
1525 :
1526 3172 : fn metric_impl(
1527 3172 : &self,
1528 3172 : op: &UploadOp,
1529 3172 : ) -> Option<(
1530 3172 : RemoteOpFileKind,
1531 3172 : RemoteOpKind,
1532 3172 : RemoteTimelineClientMetricsCallTrackSize,
1533 3172 : )> {
1534 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
1535 3172 : let res = match op {
1536 1105 : UploadOp::UploadLayer(_, m) => (
1537 1105 : RemoteOpFileKind::Layer,
1538 1105 : RemoteOpKind::Upload,
1539 1105 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size()),
1540 1105 : ),
1541 1513 : UploadOp::UploadMetadata(_, _) => (
1542 1513 : RemoteOpFileKind::Index,
1543 1513 : RemoteOpKind::Upload,
1544 1513 : DontTrackSize {
1545 1513 : reason: "metadata uploads are tiny",
1546 1513 : },
1547 1513 : ),
1548 548 : UploadOp::Delete(_delete) => (
1549 548 : RemoteOpFileKind::Layer,
1550 548 : RemoteOpKind::Delete,
1551 548 : DontTrackSize {
1552 548 : reason: "should we track deletes? positive or negative sign?",
1553 548 : },
1554 548 : ),
1555 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
1556 : // we do not account these
1557 6 : return None;
1558 : }
1559 : };
1560 3166 : Some(res)
1561 3172 : }
1562 :
1563 1634 : fn metric_begin(&self, op: &UploadOp) {
1564 1634 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1565 1634 : Some(x) => x,
1566 0 : None => return,
1567 : };
1568 1634 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
1569 1634 : guard.will_decrement_manually(); // in metric_end(), see right below
1570 1634 : }
1571 :
1572 1538 : fn metric_end(&self, op: &UploadOp) {
1573 1538 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1574 1532 : Some(x) => x,
1575 6 : None => return,
1576 : };
1577 1532 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
1578 1538 : }
1579 :
1580 : /// Close the upload queue for new operations and cancel queued operations.
1581 : ///
1582 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
1583 : ///
1584 : /// In-progress operations will still be running after this function returns.
1585 : /// Use `task_mgr::shutdown_tasks(None, Some(self.tenant_id), Some(timeline_id))`
1586 : /// to wait for them to complete, after calling this function.
1587 14 : pub(crate) fn stop(&self) -> Result<(), StopError> {
1588 14 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
1589 14 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
1590 14 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
1591 14 : let mut guard = self.upload_queue.lock().unwrap();
1592 14 : match &mut *guard {
1593 0 : UploadQueue::Uninitialized => Err(StopError::QueueUninitialized),
1594 : UploadQueue::Stopped(_) => {
1595 : // nothing to do
1596 6 : info!("another concurrent task already shut down the queue");
1597 6 : Ok(())
1598 : }
1599 8 : UploadQueue::Initialized(initialized) => {
1600 8 : info!("shutting down upload queue");
1601 :
1602 : // Replace the queue with the Stopped state, taking ownership of the old
1603 : // Initialized queue. We will do some checks on it, and then drop it.
1604 8 : let qi = {
1605 : // Here we preserve working version of the upload queue for possible use during deletions.
1606 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
1607 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
1608 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
1609 8 : let upload_queue_for_deletion = UploadQueueInitialized {
1610 8 : task_counter: 0,
1611 8 : latest_files: initialized.latest_files.clone(),
1612 8 : latest_files_changes_since_metadata_upload_scheduled: 0,
1613 8 : latest_metadata: initialized.latest_metadata.clone(),
1614 8 : projected_remote_consistent_lsn: None,
1615 8 : visible_remote_consistent_lsn: initialized
1616 8 : .visible_remote_consistent_lsn
1617 8 : .clone(),
1618 8 : num_inprogress_layer_uploads: 0,
1619 8 : num_inprogress_metadata_uploads: 0,
1620 8 : num_inprogress_deletions: 0,
1621 8 : inprogress_tasks: HashMap::default(),
1622 8 : queued_operations: VecDeque::default(),
1623 8 : #[cfg(feature = "testing")]
1624 8 : dangling_files: HashMap::default(),
1625 8 : shutting_down: false,
1626 8 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
1627 8 : };
1628 8 :
1629 8 : let upload_queue = std::mem::replace(
1630 8 : &mut *guard,
1631 8 : UploadQueue::Stopped(UploadQueueStopped {
1632 8 : upload_queue_for_deletion,
1633 8 : deleted_at: SetDeletedFlagProgress::NotRunning,
1634 8 : }),
1635 8 : );
1636 8 : if let UploadQueue::Initialized(qi) = upload_queue {
1637 8 : qi
1638 : } else {
1639 0 : unreachable!("we checked in the match above that it is Initialized");
1640 : }
1641 : };
1642 :
1643 : // consistency check
1644 8 : assert_eq!(
1645 8 : qi.num_inprogress_layer_uploads
1646 8 : + qi.num_inprogress_metadata_uploads
1647 8 : + qi.num_inprogress_deletions,
1648 8 : qi.inprogress_tasks.len()
1649 8 : );
1650 :
1651 : // We don't need to do anything here for in-progress tasks. They will finish
1652 : // on their own, decrement the unfinished-task counter themselves, and observe
1653 : // that the queue is Stopped.
1654 8 : drop(qi.inprogress_tasks);
1655 :
1656 : // Tear down queued ops
1657 8 : for op in qi.queued_operations.into_iter() {
1658 6 : self.metric_end(&op);
1659 6 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
1660 6 : // which is exactly what we want to happen.
1661 6 : drop(op);
1662 6 : }
1663 :
1664 : // We're done.
1665 8 : drop(guard);
1666 8 : Ok(())
1667 : }
1668 : }
1669 14 : }
1670 : }
1671 :
1672 121 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1673 121 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
1674 121 : RemotePath::from_string(&path).expect("Failed to construct path")
1675 121 : }
1676 :
1677 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
1678 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
1679 0 : RemotePath::from_string(&path).expect("Failed to construct path")
1680 0 : }
1681 :
1682 33 : pub fn remote_timeline_path(
1683 33 : tenant_shard_id: &TenantShardId,
1684 33 : timeline_id: &TimelineId,
1685 33 : ) -> RemotePath {
1686 33 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
1687 33 : }
1688 :
1689 : /// Note that the shard component of a remote layer path is _not_ always the same
1690 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
1691 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
1692 13 : pub fn remote_layer_path(
1693 13 : tenant_id: &TenantId,
1694 13 : timeline_id: &TimelineId,
1695 13 : shard: ShardIndex,
1696 13 : layer_file_name: &LayerFileName,
1697 13 : generation: Generation,
1698 13 : ) -> RemotePath {
1699 13 : // Generation-aware key format
1700 13 : let path = format!(
1701 13 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
1702 13 : shard.get_suffix(),
1703 13 : layer_file_name.file_name(),
1704 13 : generation.get_suffix()
1705 13 : );
1706 13 :
1707 13 : RemotePath::from_string(&path).expect("Failed to construct path")
1708 13 : }
1709 :
1710 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
1711 4 : RemotePath::from_string(&format!(
1712 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
1713 4 : ))
1714 4 : .expect("Failed to construct path")
1715 4 : }
1716 :
1717 2 : pub fn remote_initdb_preserved_archive_path(
1718 2 : tenant_id: &TenantId,
1719 2 : timeline_id: &TimelineId,
1720 2 : ) -> RemotePath {
1721 2 : RemotePath::from_string(&format!(
1722 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
1723 2 : ))
1724 2 : .expect("Failed to construct path")
1725 2 : }
1726 :
1727 799 : pub fn remote_index_path(
1728 799 : tenant_shard_id: &TenantShardId,
1729 799 : timeline_id: &TimelineId,
1730 799 : generation: Generation,
1731 799 : ) -> RemotePath {
1732 799 : RemotePath::from_string(&format!(
1733 799 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
1734 799 : IndexPart::FILE_NAME,
1735 799 : generation.get_suffix()
1736 799 : ))
1737 799 : .expect("Failed to construct path")
1738 799 : }
1739 :
1740 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1741 0 : RemotePath::from_string(&format!(
1742 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
1743 0 : ))
1744 0 : .expect("Failed to construct path")
1745 0 : }
1746 :
1747 : /// Given the key of an index, parse out the generation part of the name
1748 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
1749 18 : let file_name = match path.get_path().file_name() {
1750 18 : Some(f) => f,
1751 : None => {
1752 : // Unexpected: we should be seeing index_part.json paths only
1753 0 : tracing::warn!("Malformed index key {}", path);
1754 0 : return None;
1755 : }
1756 : };
1757 :
1758 18 : match file_name.split_once('-') {
1759 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
1760 6 : None => None,
1761 : }
1762 18 : }
1763 :
1764 : /// Files on the remote storage are stored with paths, relative to the workdir.
1765 : /// That path includes in itself both tenant and timeline ids, allowing to have a unique remote storage path.
1766 : ///
1767 : /// Errors if the path provided does not start from pageserver's workdir.
1768 554 : pub fn remote_path(
1769 554 : conf: &PageServerConf,
1770 554 : local_path: &Utf8Path,
1771 554 : generation: Generation,
1772 554 : ) -> anyhow::Result<RemotePath> {
1773 554 : let stripped = local_path
1774 554 : .strip_prefix(&conf.workdir)
1775 554 : .context("Failed to strip workdir prefix")?;
1776 :
1777 554 : let suffixed = format!("{0}{1}", stripped, generation.get_suffix());
1778 554 :
1779 554 : RemotePath::new(Utf8Path::new(&suffixed)).with_context(|| {
1780 0 : format!(
1781 0 : "to resolve remote part of path {:?} for base {:?}",
1782 0 : local_path, conf.workdir
1783 0 : )
1784 554 : })
1785 554 : }
1786 :
1787 : #[cfg(test)]
1788 : mod tests {
1789 : use super::*;
1790 : use crate::{
1791 : context::RequestContext,
1792 : tenant::{
1793 : harness::{TenantHarness, TIMELINE_ID},
1794 : storage_layer::Layer,
1795 : Generation, Tenant, Timeline,
1796 : },
1797 : DEFAULT_PG_VERSION,
1798 : };
1799 :
1800 : use std::collections::HashSet;
1801 : use utils::lsn::Lsn;
1802 :
1803 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
1804 8 : format!("contents for {name}").into()
1805 8 : }
1806 :
1807 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
1808 2 : let metadata = TimelineMetadata::new(
1809 2 : disk_consistent_lsn,
1810 2 : None,
1811 2 : None,
1812 2 : Lsn(0),
1813 2 : Lsn(0),
1814 2 : Lsn(0),
1815 2 : // Any version will do
1816 2 : // but it should be consistent with the one in the tests
1817 2 : crate::DEFAULT_PG_VERSION,
1818 2 : );
1819 2 :
1820 2 : // go through serialize + deserialize to fix the header, including checksum
1821 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
1822 2 : }
1823 :
1824 2 : fn assert_file_list(a: &HashSet<LayerFileName>, b: &[&str]) {
1825 6 : let mut avec: Vec<String> = a.iter().map(|x| x.file_name()).collect();
1826 2 : avec.sort();
1827 2 :
1828 2 : let mut bvec = b.to_vec();
1829 2 : bvec.sort_unstable();
1830 2 :
1831 2 : assert_eq!(avec, bvec);
1832 2 : }
1833 :
1834 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
1835 4 : let mut expected: Vec<String> = expected
1836 4 : .iter()
1837 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
1838 4 : .collect();
1839 4 : expected.sort();
1840 4 :
1841 4 : let mut found: Vec<String> = Vec::new();
1842 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
1843 16 : let entry_name = entry.file_name();
1844 16 : let fname = entry_name.to_str().unwrap();
1845 16 : found.push(String::from(fname));
1846 16 : }
1847 4 : found.sort();
1848 4 :
1849 4 : assert_eq!(found, expected);
1850 4 : }
1851 :
1852 : struct TestSetup {
1853 : harness: TenantHarness,
1854 : tenant: Arc<Tenant>,
1855 : timeline: Arc<Timeline>,
1856 : tenant_ctx: RequestContext,
1857 : }
1858 :
1859 : impl TestSetup {
1860 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
1861 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
1862 8 : let harness = TenantHarness::create(test_name)?;
1863 8 : let (tenant, ctx) = harness.load().await;
1864 :
1865 8 : let timeline = tenant
1866 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
1867 24 : .await?;
1868 :
1869 8 : Ok(Self {
1870 8 : harness,
1871 8 : tenant,
1872 8 : timeline,
1873 8 : tenant_ctx: ctx,
1874 8 : })
1875 8 : }
1876 :
1877 : /// Construct a RemoteTimelineClient in an arbitrary generation
1878 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
1879 10 : Arc::new(RemoteTimelineClient {
1880 10 : conf: self.harness.conf,
1881 10 : runtime: tokio::runtime::Handle::current(),
1882 10 : tenant_shard_id: self.harness.tenant_shard_id,
1883 10 : timeline_id: TIMELINE_ID,
1884 10 : generation,
1885 10 : storage_impl: self.harness.remote_storage.clone(),
1886 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
1887 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
1888 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
1889 10 : &self.harness.tenant_shard_id,
1890 10 : &TIMELINE_ID,
1891 10 : )),
1892 10 : cancel: CancellationToken::new(),
1893 10 : })
1894 10 : }
1895 :
1896 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
1897 : /// and timeline_id are present.
1898 6 : fn span(&self) -> tracing::Span {
1899 6 : tracing::info_span!(
1900 : "test",
1901 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
1902 6 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
1903 : timeline_id = %TIMELINE_ID
1904 : )
1905 6 : }
1906 : }
1907 :
1908 : // Test scheduling
1909 2 : #[tokio::test]
1910 2 : async fn upload_scheduling() {
1911 2 : // Test outline:
1912 2 : //
1913 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
1914 2 : // Schedule upload of index. Check that it is queued
1915 2 : // let the layer file uploads finish. Check that the index-upload is now started
1916 2 : // let the index-upload finish.
1917 2 : //
1918 2 : // Download back the index.json. Check that the list of files is correct
1919 2 : //
1920 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
1921 2 : // let upload finish. Check that deletion is now started
1922 2 : // Schedule another deletion. Check that it's launched immediately.
1923 2 : // Schedule index upload. Check that it's queued
1924 2 :
1925 6 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
1926 2 : let span = test_setup.span();
1927 2 : let _guard = span.enter();
1928 2 :
1929 2 : let TestSetup {
1930 2 : harness,
1931 2 : tenant: _tenant,
1932 2 : timeline,
1933 2 : tenant_ctx: _tenant_ctx,
1934 2 : } = test_setup;
1935 2 :
1936 2 : let client = timeline.remote_client.as_ref().unwrap();
1937 2 :
1938 2 : // Download back the index.json, and check that the list of files is correct
1939 2 : let initial_index_part = match client
1940 2 : .download_index_file(&CancellationToken::new())
1941 6 : .await
1942 2 : .unwrap()
1943 2 : {
1944 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
1945 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
1946 2 : };
1947 2 : let initial_layers = initial_index_part
1948 2 : .layer_metadata
1949 2 : .keys()
1950 2 : .map(|f| f.to_owned())
1951 2 : .collect::<HashSet<LayerFileName>>();
1952 2 : let initial_layer = {
1953 2 : assert!(initial_layers.len() == 1);
1954 2 : initial_layers.into_iter().next().unwrap()
1955 2 : };
1956 2 :
1957 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
1958 2 :
1959 2 : println!("workdir: {}", harness.conf.workdir);
1960 2 :
1961 2 : let remote_timeline_dir = harness
1962 2 : .remote_fs_dir
1963 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
1964 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
1965 2 :
1966 2 : let generation = harness.generation;
1967 2 : let shard = harness.shard;
1968 2 :
1969 2 : // Create a couple of dummy files, schedule upload for them
1970 2 :
1971 2 : let layers = [
1972 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
1973 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
1974 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
1975 2 : ]
1976 2 : .into_iter()
1977 6 : .map(|(name, contents): (LayerFileName, Vec<u8>)| {
1978 6 : std::fs::write(timeline_path.join(name.file_name()), &contents).unwrap();
1979 6 :
1980 6 : Layer::for_resident(
1981 6 : harness.conf,
1982 6 : &timeline,
1983 6 : name,
1984 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
1985 6 : )
1986 6 : }).collect::<Vec<_>>();
1987 2 :
1988 2 : client
1989 2 : .schedule_layer_file_upload(layers[0].clone())
1990 2 : .unwrap();
1991 2 : client
1992 2 : .schedule_layer_file_upload(layers[1].clone())
1993 2 : .unwrap();
1994 2 :
1995 2 : // Check that they are started immediately, not queued
1996 2 : //
1997 2 : // this works because we running within block_on, so any futures are now queued up until
1998 2 : // our next await point.
1999 2 : {
2000 2 : let mut guard = client.upload_queue.lock().unwrap();
2001 2 : let upload_queue = guard.initialized_mut().unwrap();
2002 2 : assert!(upload_queue.queued_operations.is_empty());
2003 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2004 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2005 2 :
2006 2 : // also check that `latest_file_changes` was updated
2007 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2008 2 : }
2009 2 :
2010 2 : // Schedule upload of index. Check that it is queued
2011 2 : let metadata = dummy_metadata(Lsn(0x20));
2012 2 : client
2013 2 : .schedule_index_upload_for_metadata_update(&metadata)
2014 2 : .unwrap();
2015 2 : {
2016 2 : let mut guard = client.upload_queue.lock().unwrap();
2017 2 : let upload_queue = guard.initialized_mut().unwrap();
2018 2 : assert!(upload_queue.queued_operations.len() == 1);
2019 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2020 2 : }
2021 2 :
2022 2 : // Wait for the uploads to finish
2023 2 : client.wait_completion().await.unwrap();
2024 2 : {
2025 2 : let mut guard = client.upload_queue.lock().unwrap();
2026 2 : let upload_queue = guard.initialized_mut().unwrap();
2027 2 :
2028 2 : assert!(upload_queue.queued_operations.is_empty());
2029 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2030 2 : }
2031 2 :
2032 2 : // Download back the index.json, and check that the list of files is correct
2033 2 : let index_part = match client
2034 2 : .download_index_file(&CancellationToken::new())
2035 6 : .await
2036 2 : .unwrap()
2037 2 : {
2038 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2039 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2040 2 : };
2041 2 :
2042 2 : assert_file_list(
2043 2 : &index_part
2044 2 : .layer_metadata
2045 2 : .keys()
2046 6 : .map(|f| f.to_owned())
2047 2 : .collect(),
2048 2 : &[
2049 2 : &initial_layer.file_name(),
2050 2 : &layers[0].layer_desc().filename().file_name(),
2051 2 : &layers[1].layer_desc().filename().file_name(),
2052 2 : ],
2053 2 : );
2054 2 : assert_eq!(index_part.metadata, metadata);
2055 2 :
2056 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2057 2 : client
2058 2 : .schedule_layer_file_upload(layers[2].clone())
2059 2 : .unwrap();
2060 2 :
2061 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2062 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2063 2 : // spawn_blocking started by the drop.
2064 2 : client
2065 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().filename()])
2066 2 : .unwrap();
2067 2 : {
2068 2 : let mut guard = client.upload_queue.lock().unwrap();
2069 2 : let upload_queue = guard.initialized_mut().unwrap();
2070 2 :
2071 2 : // Deletion schedules upload of the index file, and the file deletion itself
2072 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2073 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2074 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2075 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2076 2 : assert_eq!(
2077 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2078 2 : 0
2079 2 : );
2080 2 : }
2081 2 : assert_remote_files(
2082 2 : &[
2083 2 : &initial_layer.file_name(),
2084 2 : &layers[0].layer_desc().filename().file_name(),
2085 2 : &layers[1].layer_desc().filename().file_name(),
2086 2 : "index_part.json",
2087 2 : ],
2088 2 : &remote_timeline_dir,
2089 2 : generation,
2090 2 : );
2091 2 :
2092 2 : // Finish them
2093 2 : client.wait_completion().await.unwrap();
2094 2 : harness.deletion_queue.pump().await;
2095 2 :
2096 2 : assert_remote_files(
2097 2 : &[
2098 2 : &initial_layer.file_name(),
2099 2 : &layers[1].layer_desc().filename().file_name(),
2100 2 : &layers[2].layer_desc().filename().file_name(),
2101 2 : "index_part.json",
2102 2 : ],
2103 2 : &remote_timeline_dir,
2104 2 : generation,
2105 2 : );
2106 2 : }
2107 :
2108 2 : #[tokio::test]
2109 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2110 2 : // Setup
2111 2 :
2112 2 : let TestSetup {
2113 2 : harness,
2114 2 : tenant: _tenant,
2115 2 : timeline,
2116 2 : ..
2117 6 : } = TestSetup::new("metrics").await.unwrap();
2118 2 : let client = timeline.remote_client.as_ref().unwrap();
2119 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2120 2 :
2121 2 : let layer_file_name_1: LayerFileName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2122 2 : let content_1 = dummy_contents("foo");
2123 2 : std::fs::write(
2124 2 : timeline_path.join(layer_file_name_1.file_name()),
2125 2 : &content_1,
2126 2 : )
2127 2 : .unwrap();
2128 2 :
2129 2 : let layer_file_1 = Layer::for_resident(
2130 2 : harness.conf,
2131 2 : &timeline,
2132 2 : layer_file_name_1.clone(),
2133 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2134 2 : );
2135 2 :
2136 4 : #[derive(Debug, PartialEq, Clone, Copy)]
2137 2 : struct BytesStartedFinished {
2138 2 : started: Option<usize>,
2139 2 : finished: Option<usize>,
2140 2 : }
2141 2 : impl std::ops::Add for BytesStartedFinished {
2142 2 : type Output = Self;
2143 4 : fn add(self, rhs: Self) -> Self::Output {
2144 4 : Self {
2145 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2146 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2147 4 : }
2148 4 : }
2149 2 : }
2150 6 : let get_bytes_started_stopped = || {
2151 6 : let started = client
2152 6 : .metrics
2153 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2154 6 : .map(|v| v.try_into().unwrap());
2155 6 : let stopped = client
2156 6 : .metrics
2157 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2158 6 : .map(|v| v.try_into().unwrap());
2159 6 : BytesStartedFinished {
2160 6 : started,
2161 6 : finished: stopped,
2162 6 : }
2163 6 : };
2164 2 :
2165 2 : // Test
2166 2 : tracing::info!("now doing actual test");
2167 2 :
2168 2 : let actual_a = get_bytes_started_stopped();
2169 2 :
2170 2 : client
2171 2 : .schedule_layer_file_upload(layer_file_1.clone())
2172 2 : .unwrap();
2173 2 :
2174 2 : let actual_b = get_bytes_started_stopped();
2175 2 :
2176 2 : client.wait_completion().await.unwrap();
2177 2 :
2178 2 : let actual_c = get_bytes_started_stopped();
2179 2 :
2180 2 : // Validate
2181 2 :
2182 2 : let expected_b = actual_a
2183 2 : + BytesStartedFinished {
2184 2 : started: Some(content_1.len()),
2185 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2186 2 : finished: Some(0),
2187 2 : };
2188 2 : assert_eq!(actual_b, expected_b);
2189 2 :
2190 2 : let expected_c = actual_a
2191 2 : + BytesStartedFinished {
2192 2 : started: Some(content_1.len()),
2193 2 : finished: Some(content_1.len()),
2194 2 : };
2195 2 : assert_eq!(actual_c, expected_c);
2196 2 : }
2197 :
2198 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2199 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2200 12 : let example_metadata = TimelineMetadata::example();
2201 12 : let example_index_part = IndexPart::new(
2202 12 : HashMap::new(),
2203 12 : example_metadata.disk_consistent_lsn(),
2204 12 : example_metadata,
2205 12 : );
2206 12 :
2207 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2208 12 :
2209 12 : let index_path = test_state.harness.remote_fs_dir.join(
2210 12 : remote_index_path(
2211 12 : &test_state.harness.tenant_shard_id,
2212 12 : &TIMELINE_ID,
2213 12 : generation,
2214 12 : )
2215 12 : .get_path(),
2216 12 : );
2217 12 :
2218 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2219 12 : .expect("creating test dir should work");
2220 12 :
2221 12 : eprintln!("Writing {index_path}");
2222 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2223 12 : example_index_part
2224 12 : }
2225 :
2226 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2227 : /// index, the IndexPart returned is equal to `expected`
2228 10 : async fn assert_got_index_part(
2229 10 : test_state: &TestSetup,
2230 10 : get_generation: Generation,
2231 10 : expected: &IndexPart,
2232 10 : ) {
2233 10 : let client = test_state.build_client(get_generation);
2234 :
2235 10 : let download_r = client
2236 10 : .download_index_file(&CancellationToken::new())
2237 42 : .await
2238 10 : .expect("download should always succeed");
2239 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2240 10 : match download_r {
2241 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2242 10 : assert_eq!(&index_part, expected);
2243 : }
2244 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2245 : }
2246 10 : }
2247 :
2248 2 : #[tokio::test]
2249 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2250 6 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2251 2 : let span = test_state.span();
2252 2 : let _guard = span.enter();
2253 2 :
2254 2 : // Simple case: we are in generation N, load the index from generation N - 1
2255 2 : let generation_n = 5;
2256 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2257 2 :
2258 6 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2259 2 :
2260 2 : Ok(())
2261 2 : }
2262 :
2263 2 : #[tokio::test]
2264 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2265 2 : let test_state = TestSetup::new("index_part_download_ordering")
2266 6 : .await
2267 2 : .unwrap();
2268 2 :
2269 2 : let span = test_state.span();
2270 2 : let _guard = span.enter();
2271 2 :
2272 2 : // A generation-less IndexPart exists in the bucket, we should find it
2273 2 : let generation_n = 5;
2274 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2275 10 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2276 2 :
2277 2 : // If a more recent-than-none generation exists, we should prefer to load that
2278 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2279 10 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2280 2 :
2281 2 : // If a more-recent-than-me generation exists, we should ignore it.
2282 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2283 10 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2284 2 :
2285 2 : // If a directly previous generation exists, _and_ an index exists in my own
2286 2 : // generation, I should prefer my own generation.
2287 2 : let _injected_prev =
2288 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2289 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2290 2 : assert_got_index_part(
2291 2 : &test_state,
2292 2 : Generation::new(generation_n),
2293 2 : &injected_current,
2294 2 : )
2295 6 : .await;
2296 2 :
2297 2 : Ok(())
2298 2 : }
2299 : }
|
