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 : use std::time::Duration;
204 :
205 : use remote_storage::{
206 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
207 : };
208 : use std::ops::DerefMut;
209 : use tracing::{debug, error, info, instrument, warn};
210 : use tracing::{info_span, Instrument};
211 : use utils::lsn::Lsn;
212 :
213 : use crate::context::RequestContext;
214 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
215 : use crate::metrics::{
216 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
217 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
218 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
219 : };
220 : use crate::task_mgr::shutdown_token;
221 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
222 : use crate::tenant::remote_timeline_client::download::download_retry;
223 : use crate::tenant::storage_layer::AsLayerDesc;
224 : use crate::tenant::upload_queue::{Delete, UploadQueueStoppedDeletable};
225 : use crate::tenant::TIMELINES_SEGMENT_NAME;
226 : use crate::{
227 : config::PageServerConf,
228 : task_mgr,
229 : task_mgr::TaskKind,
230 : task_mgr::BACKGROUND_RUNTIME,
231 : tenant::metadata::TimelineMetadata,
232 : tenant::upload_queue::{
233 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
234 : },
235 : TENANT_HEATMAP_BASENAME,
236 : };
237 :
238 : use utils::id::{TenantId, TimelineId};
239 :
240 : use self::index::IndexPart;
241 :
242 : use super::metadata::MetadataUpdate;
243 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
244 : use super::upload_queue::SetDeletedFlagProgress;
245 : use super::Generation;
246 :
247 : pub(crate) use download::{
248 : download_index_part, is_temp_download_file, list_remote_tenant_shards, list_remote_timelines,
249 : };
250 : pub(crate) use index::LayerFileMetadata;
251 :
252 : // Occasional network issues and such can cause remote operations to fail, and
253 : // that's expected. If a download fails, we log it at info-level, and retry.
254 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
255 : // level instead, as repeated failures can mean a more serious problem. If it
256 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
257 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
258 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
259 :
260 : // Similarly log failed uploads and deletions at WARN level, after this many
261 : // retries. Uploads and deletions are retried forever, though.
262 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
263 :
264 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
265 :
266 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
267 :
268 : /// Default buffer size when interfacing with [`tokio::fs::File`].
269 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
270 :
271 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
272 : /// which we warn and skip.
273 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
274 :
275 : pub enum MaybeDeletedIndexPart {
276 : IndexPart(IndexPart),
277 : Deleted(IndexPart),
278 : }
279 :
280 0 : #[derive(Debug, thiserror::Error)]
281 : pub enum PersistIndexPartWithDeletedFlagError {
282 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
283 : AlreadyInProgress(NaiveDateTime),
284 : #[error("the deleted_flag was already set, value is {0:?}")]
285 : AlreadyDeleted(NaiveDateTime),
286 : #[error(transparent)]
287 : Other(#[from] anyhow::Error),
288 : }
289 :
290 : /// A client for accessing a timeline's data in remote storage.
291 : ///
292 : /// This takes care of managing the number of connections, and balancing them
293 : /// across tenants. This also handles retries of failed uploads.
294 : ///
295 : /// Upload and delete requests are ordered so that before a deletion is
296 : /// performed, we wait for all preceding uploads to finish. This ensures sure
297 : /// that if you perform a compaction operation that reshuffles data in layer
298 : /// files, we don't have a transient state where the old files have already been
299 : /// deleted, but new files have not yet been uploaded.
300 : ///
301 : /// Similarly, this enforces an order between index-file uploads, and layer
302 : /// uploads. Before an index-file upload is performed, all preceding layer
303 : /// uploads must be finished.
304 : ///
305 : /// This also maintains a list of remote files, and automatically includes that
306 : /// in the index part file, whenever timeline metadata is uploaded.
307 : ///
308 : /// Downloads are not queued, they are performed immediately.
309 : pub struct RemoteTimelineClient {
310 : conf: &'static PageServerConf,
311 :
312 : runtime: tokio::runtime::Handle,
313 :
314 : tenant_shard_id: TenantShardId,
315 : timeline_id: TimelineId,
316 : generation: Generation,
317 :
318 : upload_queue: Mutex<UploadQueue>,
319 :
320 : metrics: Arc<RemoteTimelineClientMetrics>,
321 :
322 : storage_impl: GenericRemoteStorage,
323 :
324 : deletion_queue_client: DeletionQueueClient,
325 :
326 : cancel: CancellationToken,
327 : }
328 :
329 : impl RemoteTimelineClient {
330 : ///
331 : /// Create a remote storage client for given timeline
332 : ///
333 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
334 : /// by calling init_upload_queue.
335 : ///
336 334 : pub fn new(
337 334 : remote_storage: GenericRemoteStorage,
338 334 : deletion_queue_client: DeletionQueueClient,
339 334 : conf: &'static PageServerConf,
340 334 : tenant_shard_id: TenantShardId,
341 334 : timeline_id: TimelineId,
342 334 : generation: Generation,
343 334 : ) -> RemoteTimelineClient {
344 334 : RemoteTimelineClient {
345 334 : conf,
346 334 : runtime: if cfg!(test) {
347 : // remote_timeline_client.rs tests rely on current-thread runtime
348 334 : tokio::runtime::Handle::current()
349 : } else {
350 0 : BACKGROUND_RUNTIME.handle().clone()
351 : },
352 334 : tenant_shard_id,
353 334 : timeline_id,
354 334 : generation,
355 334 : storage_impl: remote_storage,
356 334 : deletion_queue_client,
357 334 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
358 334 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
359 334 : &tenant_shard_id,
360 334 : &timeline_id,
361 334 : )),
362 334 : cancel: CancellationToken::new(),
363 334 : }
364 334 : }
365 :
366 : /// Initialize the upload queue for a remote storage that already received
367 : /// an index file upload, i.e., it's not empty.
368 : /// The given `index_part` must be the one on the remote.
369 6 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
370 6 : let mut upload_queue = self.upload_queue.lock().unwrap();
371 6 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
372 6 : self.update_remote_physical_size_gauge(Some(index_part));
373 6 : info!(
374 0 : "initialized upload queue from remote index with {} layer files",
375 0 : index_part.layer_metadata.len()
376 : );
377 6 : Ok(())
378 6 : }
379 :
380 : /// Initialize the upload queue for the case where the remote storage is empty,
381 : /// i.e., it doesn't have an `IndexPart`.
382 328 : pub fn init_upload_queue_for_empty_remote(
383 328 : &self,
384 328 : local_metadata: &TimelineMetadata,
385 328 : ) -> anyhow::Result<()> {
386 328 : let mut upload_queue = self.upload_queue.lock().unwrap();
387 328 : upload_queue.initialize_empty_remote(local_metadata)?;
388 328 : self.update_remote_physical_size_gauge(None);
389 328 : info!("initialized upload queue as empty");
390 328 : Ok(())
391 328 : }
392 :
393 : /// Initialize the queue in stopped state. Used in startup path
394 : /// to continue deletion operation interrupted by pageserver crash or restart.
395 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
396 0 : &self,
397 0 : index_part: &IndexPart,
398 0 : ) -> anyhow::Result<()> {
399 : // FIXME: consider newtype for DeletedIndexPart.
400 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
401 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
402 0 : ))?;
403 :
404 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
405 0 : upload_queue.initialize_with_current_remote_index_part(index_part)?;
406 0 : self.update_remote_physical_size_gauge(Some(index_part));
407 0 : self.stop_impl(&mut upload_queue);
408 0 :
409 0 : upload_queue
410 0 : .stopped_mut()
411 0 : .expect("stopped above")
412 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
413 0 :
414 0 : Ok(())
415 0 : }
416 :
417 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
418 0 : match &mut *self.upload_queue.lock().unwrap() {
419 0 : UploadQueue::Uninitialized => None,
420 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
421 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
422 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
423 0 : .upload_queue_for_deletion
424 0 : .get_last_remote_consistent_lsn_projected(),
425 : }
426 0 : }
427 :
428 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
429 0 : match &mut *self.upload_queue.lock().unwrap() {
430 0 : UploadQueue::Uninitialized => None,
431 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
432 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
433 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
434 0 : q.upload_queue_for_deletion
435 0 : .get_last_remote_consistent_lsn_visible(),
436 0 : ),
437 : }
438 0 : }
439 :
440 1478 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
441 1478 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
442 1150 : current_remote_index_part
443 1150 : .layer_metadata
444 1150 : .values()
445 1150 : // If we don't have the file size for the layer, don't account for it in the metric.
446 13167 : .map(|ilmd| ilmd.file_size)
447 1150 : .sum()
448 : } else {
449 328 : 0
450 : };
451 1478 : self.metrics.remote_physical_size_set(size);
452 1478 : }
453 :
454 2 : pub fn get_remote_physical_size(&self) -> u64 {
455 2 : self.metrics.remote_physical_size_get()
456 2 : }
457 :
458 : //
459 : // Download operations.
460 : //
461 : // These don't use the per-timeline queue. They do use the global semaphore in
462 : // S3Bucket, to limit the total number of concurrent operations, though.
463 : //
464 :
465 : /// Download index file
466 20 : pub async fn download_index_file(
467 20 : &self,
468 20 : cancel: &CancellationToken,
469 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
470 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
471 20 : &RemoteOpFileKind::Index,
472 20 : &RemoteOpKind::Download,
473 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
474 20 : reason: "no need for a downloads gauge",
475 20 : },
476 20 : );
477 :
478 20 : let (index_part, _index_generation) = download::download_index_part(
479 20 : &self.storage_impl,
480 20 : &self.tenant_shard_id,
481 20 : &self.timeline_id,
482 20 : self.generation,
483 20 : cancel,
484 20 : )
485 20 : .measure_remote_op(
486 20 : RemoteOpFileKind::Index,
487 20 : RemoteOpKind::Download,
488 20 : Arc::clone(&self.metrics),
489 20 : )
490 90 : .await?;
491 :
492 20 : if index_part.deleted_at.is_some() {
493 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
494 : } else {
495 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
496 : }
497 20 : }
498 :
499 : /// Download a (layer) file from `path`, into local filesystem.
500 : ///
501 : /// 'layer_metadata' is the metadata from the remote index file.
502 : ///
503 : /// On success, returns the size of the downloaded file.
504 6 : pub async fn download_layer_file(
505 6 : &self,
506 6 : layer_file_name: &LayerName,
507 6 : layer_metadata: &LayerFileMetadata,
508 6 : cancel: &CancellationToken,
509 6 : ctx: &RequestContext,
510 6 : ) -> anyhow::Result<u64> {
511 6 : let downloaded_size = {
512 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
513 6 : &RemoteOpFileKind::Layer,
514 6 : &RemoteOpKind::Download,
515 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
516 6 : reason: "no need for a downloads gauge",
517 6 : },
518 6 : );
519 6 : download::download_layer_file(
520 6 : self.conf,
521 6 : &self.storage_impl,
522 6 : self.tenant_shard_id,
523 6 : self.timeline_id,
524 6 : layer_file_name,
525 6 : layer_metadata,
526 6 : cancel,
527 6 : ctx,
528 6 : )
529 6 : .measure_remote_op(
530 6 : RemoteOpFileKind::Layer,
531 6 : RemoteOpKind::Download,
532 6 : Arc::clone(&self.metrics),
533 6 : )
534 88 : .await?
535 : };
536 :
537 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
538 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
539 6 :
540 6 : Ok(downloaded_size)
541 6 : }
542 :
543 : //
544 : // Upload operations.
545 : //
546 :
547 : /// Launch an index-file upload operation in the background, with
548 : /// fully updated metadata.
549 : ///
550 : /// This should only be used to upload initial metadata to remote storage.
551 : ///
552 : /// The upload will be added to the queue immediately, but it
553 : /// won't be performed until all previously scheduled layer file
554 : /// upload operations have completed successfully. This is to
555 : /// ensure that when the index file claims that layers X, Y and Z
556 : /// exist in remote storage, they really do. To wait for the upload
557 : /// to complete, use `wait_completion`.
558 : ///
559 : /// If there were any changes to the list of files, i.e. if any
560 : /// layer file uploads were scheduled, since the last index file
561 : /// upload, those will be included too.
562 220 : pub fn schedule_index_upload_for_full_metadata_update(
563 220 : self: &Arc<Self>,
564 220 : metadata: &TimelineMetadata,
565 220 : ) -> anyhow::Result<()> {
566 220 : let mut guard = self.upload_queue.lock().unwrap();
567 220 : let upload_queue = guard.initialized_mut()?;
568 :
569 : // As documented in the struct definition, it's ok for latest_metadata to be
570 : // ahead of what's _actually_ on the remote during index upload.
571 220 : upload_queue.latest_metadata = metadata.clone();
572 220 :
573 220 : self.schedule_index_upload(upload_queue);
574 220 :
575 220 : Ok(())
576 220 : }
577 :
578 : /// Launch an index-file upload operation in the background, with only parts of the metadata
579 : /// updated.
580 : ///
581 : /// This is the regular way of updating metadata on layer flushes or Gc.
582 : ///
583 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
584 : /// `index_part.json`, while being more clear on what values update regularly.
585 928 : pub(crate) fn schedule_index_upload_for_metadata_update(
586 928 : self: &Arc<Self>,
587 928 : update: &MetadataUpdate,
588 928 : ) -> anyhow::Result<()> {
589 928 : let mut guard = self.upload_queue.lock().unwrap();
590 928 : let upload_queue = guard.initialized_mut()?;
591 :
592 928 : upload_queue.latest_metadata.apply(update);
593 928 :
594 928 : self.schedule_index_upload(upload_queue);
595 928 :
596 928 : Ok(())
597 928 : }
598 :
599 : ///
600 : /// Launch an index-file upload operation in the background, if necessary.
601 : ///
602 : /// Use this function to schedule the update of the index file after
603 : /// scheduling file uploads or deletions. If no file uploads or deletions
604 : /// have been scheduled since the last index file upload, this does
605 : /// nothing.
606 : ///
607 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
608 : /// the upload to the upload queue and returns quickly.
609 336 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> anyhow::Result<()> {
610 336 : let mut guard = self.upload_queue.lock().unwrap();
611 336 : let upload_queue = guard.initialized_mut()?;
612 :
613 336 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
614 0 : self.schedule_index_upload(upload_queue);
615 336 : }
616 :
617 336 : Ok(())
618 336 : }
619 :
620 : /// Launch an index-file upload operation in the background (internal function)
621 1176 : fn schedule_index_upload(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
622 1176 : let disk_consistent_lsn = upload_queue.latest_metadata.disk_consistent_lsn();
623 1176 :
624 1176 : info!(
625 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
626 0 : upload_queue.latest_files.len(),
627 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
628 : );
629 :
630 1176 : let index_part = IndexPart::from(&*upload_queue);
631 1176 : let op = UploadOp::UploadMetadata(index_part, disk_consistent_lsn);
632 1176 : self.metric_begin(&op);
633 1176 : upload_queue.queued_operations.push_back(op);
634 1176 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
635 1176 :
636 1176 : // Launch the task immediately, if possible
637 1176 : self.launch_queued_tasks(upload_queue);
638 1176 : }
639 :
640 0 : pub(crate) async fn schedule_reparenting_and_wait(
641 0 : self: &Arc<Self>,
642 0 : new_parent: &TimelineId,
643 0 : ) -> anyhow::Result<()> {
644 : // FIXME: because of how Timeline::schedule_uploads works when called from layer flushing
645 : // and reads the in-memory part we cannot do the detaching like this
646 0 : let receiver = {
647 0 : let mut guard = self.upload_queue.lock().unwrap();
648 0 : let upload_queue = guard.initialized_mut()?;
649 :
650 0 : upload_queue.latest_metadata.reparent(new_parent);
651 0 :
652 0 : self.schedule_index_upload(upload_queue);
653 0 :
654 0 : self.schedule_barrier0(upload_queue)
655 0 : };
656 0 :
657 0 : Self::wait_completion0(receiver).await
658 0 : }
659 :
660 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
661 : /// detaching from ancestor and waits for it to complete.
662 : ///
663 : /// This is used with `Timeline::detach_ancestor` functionality.
664 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
665 0 : self: &Arc<Self>,
666 0 : layers: &[Layer],
667 0 : adopted: (TimelineId, Lsn),
668 0 : ) -> anyhow::Result<()> {
669 0 : let barrier = {
670 0 : let mut guard = self.upload_queue.lock().unwrap();
671 0 : let upload_queue = guard.initialized_mut()?;
672 :
673 0 : upload_queue
674 0 : .latest_metadata
675 0 : .detach_from_ancestor(&adopted.0, &adopted.1);
676 :
677 0 : for layer in layers {
678 0 : upload_queue
679 0 : .latest_files
680 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
681 0 : }
682 :
683 0 : self.schedule_index_upload(upload_queue);
684 0 :
685 0 : let barrier = self.schedule_barrier0(upload_queue);
686 0 : self.launch_queued_tasks(upload_queue);
687 0 : barrier
688 0 : };
689 0 :
690 0 : Self::wait_completion0(barrier).await
691 0 : }
692 :
693 : /// Launch an upload operation in the background; the file is added to be included in next
694 : /// `index_part.json` upload.
695 950 : pub(crate) fn schedule_layer_file_upload(
696 950 : self: &Arc<Self>,
697 950 : layer: ResidentLayer,
698 950 : ) -> anyhow::Result<()> {
699 950 : let mut guard = self.upload_queue.lock().unwrap();
700 950 : let upload_queue = guard.initialized_mut()?;
701 :
702 950 : self.schedule_layer_file_upload0(upload_queue, layer);
703 950 : self.launch_queued_tasks(upload_queue);
704 950 : Ok(())
705 950 : }
706 :
707 1174 : fn schedule_layer_file_upload0(
708 1174 : self: &Arc<Self>,
709 1174 : upload_queue: &mut UploadQueueInitialized,
710 1174 : layer: ResidentLayer,
711 1174 : ) {
712 1174 : let metadata = layer.metadata();
713 1174 :
714 1174 : upload_queue
715 1174 : .latest_files
716 1174 : .insert(layer.layer_desc().layer_name(), metadata.clone());
717 1174 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
718 1174 :
719 1174 : info!(
720 : gen=?metadata.generation,
721 : shard=?metadata.shard,
722 0 : "scheduled layer file upload {layer}",
723 : );
724 :
725 1174 : let op = UploadOp::UploadLayer(layer, metadata);
726 1174 : self.metric_begin(&op);
727 1174 : upload_queue.queued_operations.push_back(op);
728 1174 : }
729 :
730 : /// Launch a delete operation in the background.
731 : ///
732 : /// The operation does not modify local filesystem state.
733 : ///
734 : /// Note: This schedules an index file upload before the deletions. The
735 : /// deletion won't actually be performed, until all previously scheduled
736 : /// upload operations, and the index file upload, have completed
737 : /// successfully.
738 8 : pub fn schedule_layer_file_deletion(
739 8 : self: &Arc<Self>,
740 8 : names: &[LayerName],
741 8 : ) -> anyhow::Result<()> {
742 8 : let mut guard = self.upload_queue.lock().unwrap();
743 8 : let upload_queue = guard.initialized_mut()?;
744 :
745 8 : let with_metadata =
746 8 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned());
747 8 :
748 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
749 8 :
750 8 : // Launch the tasks immediately, if possible
751 8 : self.launch_queued_tasks(upload_queue);
752 8 : Ok(())
753 8 : }
754 :
755 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
756 : /// layer files, leaving them dangling.
757 : ///
758 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
759 : /// is invoked on them.
760 2 : pub(crate) fn schedule_gc_update(self: &Arc<Self>, gc_layers: &[Layer]) -> anyhow::Result<()> {
761 2 : let mut guard = self.upload_queue.lock().unwrap();
762 2 : let upload_queue = guard.initialized_mut()?;
763 :
764 : // just forget the return value; after uploading the next index_part.json, we can consider
765 : // the layer files as "dangling". this is fine, at worst case we create work for the
766 : // scrubber.
767 :
768 2 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
769 2 :
770 2 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
771 2 :
772 2 : self.launch_queued_tasks(upload_queue);
773 2 :
774 2 : Ok(())
775 2 : }
776 :
777 : /// Update the remote index file, removing the to-be-deleted files from the index,
778 : /// allowing scheduling of actual deletions later.
779 34 : fn schedule_unlinking_of_layers_from_index_part0<I>(
780 34 : self: &Arc<Self>,
781 34 : upload_queue: &mut UploadQueueInitialized,
782 34 : names: I,
783 34 : ) -> Vec<(LayerName, LayerFileMetadata)>
784 34 : where
785 34 : I: IntoIterator<Item = LayerName>,
786 34 : {
787 34 : // Decorate our list of names with each name's metadata, dropping
788 34 : // names that are unexpectedly missing from our metadata. This metadata
789 34 : // is later used when physically deleting layers, to construct key paths.
790 34 : let with_metadata: Vec<_> = names
791 34 : .into_iter()
792 266 : .filter_map(|name| {
793 266 : let meta = upload_queue.latest_files.remove(&name);
794 :
795 266 : if let Some(meta) = meta {
796 266 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
797 266 : Some((name, meta))
798 : } else {
799 : // This can only happen if we forgot to to schedule the file upload
800 : // before scheduling the delete. Log it because it is a rare/strange
801 : // situation, and in case something is misbehaving, we'd like to know which
802 : // layers experienced this.
803 0 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
804 0 : None
805 : }
806 266 : })
807 34 : .collect();
808 :
809 : #[cfg(feature = "testing")]
810 300 : for (name, metadata) in &with_metadata {
811 266 : let gen = metadata.generation;
812 266 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
813 0 : if unexpected == gen {
814 0 : tracing::error!("{name} was unlinked twice with same generation");
815 : } else {
816 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
817 : }
818 266 : }
819 : }
820 :
821 : // after unlinking files from the upload_queue.latest_files we must always schedule an
822 : // index_part update, because that needs to be uploaded before we can actually delete the
823 : // files.
824 34 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
825 28 : self.schedule_index_upload(upload_queue);
826 28 : }
827 :
828 34 : with_metadata
829 34 : }
830 :
831 : /// Schedules deletion for layer files which have previously been unlinked from the
832 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
833 270 : pub(crate) fn schedule_deletion_of_unlinked(
834 270 : self: &Arc<Self>,
835 270 : layers: Vec<(LayerName, LayerFileMetadata)>,
836 270 : ) -> anyhow::Result<()> {
837 270 : let mut guard = self.upload_queue.lock().unwrap();
838 270 : let upload_queue = guard.initialized_mut()?;
839 :
840 270 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
841 270 : self.launch_queued_tasks(upload_queue);
842 270 : Ok(())
843 270 : }
844 :
845 278 : fn schedule_deletion_of_unlinked0(
846 278 : self: &Arc<Self>,
847 278 : upload_queue: &mut UploadQueueInitialized,
848 278 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
849 278 : ) {
850 278 : // Filter out any layers which were not created by this tenant shard. These are
851 278 : // layers that originate from some ancestor shard after a split, and may still
852 278 : // be referenced by other shards. We are free to delete them locally and remove
853 278 : // them from our index (and would have already done so when we reach this point
854 278 : // in the code), but we may not delete them remotely.
855 278 : with_metadata.retain(|(name, meta)| {
856 272 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
857 272 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
858 272 : if !retain {
859 0 : tracing::debug!(
860 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
861 : meta.shard
862 : );
863 272 : }
864 272 : retain
865 278 : });
866 :
867 550 : for (name, meta) in &with_metadata {
868 272 : info!(
869 0 : "scheduling deletion of layer {}{} (shard {})",
870 0 : name,
871 0 : meta.generation.get_suffix(),
872 : meta.shard
873 : );
874 : }
875 :
876 : #[cfg(feature = "testing")]
877 550 : for (name, meta) in &with_metadata {
878 272 : let gen = meta.generation;
879 272 : match upload_queue.dangling_files.remove(name) {
880 266 : Some(same) if same == gen => { /* expected */ }
881 0 : Some(other) => {
882 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
883 : }
884 : None => {
885 6 : tracing::error!("{name} was unlinked but was not dangling");
886 : }
887 : }
888 : }
889 :
890 : // schedule the actual deletions
891 278 : if with_metadata.is_empty() {
892 : // avoid scheduling the op & bumping the metric
893 6 : return;
894 272 : }
895 272 : let op = UploadOp::Delete(Delete {
896 272 : layers: with_metadata,
897 272 : });
898 272 : self.metric_begin(&op);
899 272 : upload_queue.queued_operations.push_back(op);
900 278 : }
901 :
902 : /// Schedules a compaction update to the remote `index_part.json`.
903 : ///
904 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
905 24 : pub(crate) fn schedule_compaction_update(
906 24 : self: &Arc<Self>,
907 24 : compacted_from: &[Layer],
908 24 : compacted_to: &[ResidentLayer],
909 24 : ) -> anyhow::Result<()> {
910 24 : let mut guard = self.upload_queue.lock().unwrap();
911 24 : let upload_queue = guard.initialized_mut()?;
912 :
913 248 : for layer in compacted_to {
914 224 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
915 224 : }
916 :
917 262 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
918 24 :
919 24 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
920 24 : self.launch_queued_tasks(upload_queue);
921 24 :
922 24 : Ok(())
923 24 : }
924 :
925 : /// Wait for all previously scheduled uploads/deletions to complete
926 110 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
927 110 : let receiver = {
928 110 : let mut guard = self.upload_queue.lock().unwrap();
929 110 : let upload_queue = guard.initialized_mut()?;
930 110 : self.schedule_barrier0(upload_queue)
931 110 : };
932 110 :
933 110 : Self::wait_completion0(receiver).await
934 110 : }
935 :
936 110 : async fn wait_completion0(
937 110 : mut receiver: tokio::sync::watch::Receiver<()>,
938 110 : ) -> anyhow::Result<()> {
939 110 : if receiver.changed().await.is_err() {
940 0 : anyhow::bail!("wait_completion aborted because upload queue was stopped");
941 110 : }
942 110 :
943 110 : Ok(())
944 110 : }
945 :
946 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
947 6 : let mut guard = self.upload_queue.lock().unwrap();
948 6 : let upload_queue = guard.initialized_mut()?;
949 6 : self.schedule_barrier0(upload_queue);
950 6 : Ok(())
951 6 : }
952 :
953 116 : fn schedule_barrier0(
954 116 : self: &Arc<Self>,
955 116 : upload_queue: &mut UploadQueueInitialized,
956 116 : ) -> tokio::sync::watch::Receiver<()> {
957 116 : let (sender, receiver) = tokio::sync::watch::channel(());
958 116 : let barrier_op = UploadOp::Barrier(sender);
959 116 :
960 116 : upload_queue.queued_operations.push_back(barrier_op);
961 116 : // Don't count this kind of operation!
962 116 :
963 116 : // Launch the task immediately, if possible
964 116 : self.launch_queued_tasks(upload_queue);
965 116 :
966 116 : receiver
967 116 : }
968 :
969 : /// Wait for all previously scheduled operations to complete, and then stop.
970 : ///
971 : /// Not cancellation safe
972 6 : pub(crate) async fn shutdown(self: &Arc<Self>) {
973 6 : // On cancellation the queue is left in ackward state of refusing new operations but
974 6 : // proper stop is yet to be called. On cancel the original or some later task must call
975 6 : // `stop` or `shutdown`.
976 6 : let sg = scopeguard::guard((), |_| {
977 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
978 6 : });
979 :
980 6 : let fut = {
981 6 : let mut guard = self.upload_queue.lock().unwrap();
982 6 : let upload_queue = match &mut *guard {
983 0 : UploadQueue::Stopped(_) => return,
984 : UploadQueue::Uninitialized => {
985 : // transition into Stopped state
986 0 : self.stop_impl(&mut guard);
987 0 : return;
988 : }
989 6 : UploadQueue::Initialized(ref mut init) => init,
990 6 : };
991 6 :
992 6 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
993 6 : // just don't add more of these as they would never complete.
994 6 : //
995 6 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
996 6 : // in every place we would not have to jump through this hoop, and this method could be
997 6 : // made cancellable.
998 6 : if !upload_queue.shutting_down {
999 6 : upload_queue.shutting_down = true;
1000 6 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1001 6 : // this operation is not counted similar to Barrier
1002 6 :
1003 6 : self.launch_queued_tasks(upload_queue);
1004 6 : }
1005 :
1006 6 : upload_queue.shutdown_ready.clone().acquire_owned()
1007 : };
1008 :
1009 6 : let res = fut.await;
1010 :
1011 6 : scopeguard::ScopeGuard::into_inner(sg);
1012 6 :
1013 6 : match res {
1014 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1015 6 : Err(_closed) => {
1016 6 : // expected
1017 6 : }
1018 6 : }
1019 6 :
1020 6 : self.stop();
1021 6 : }
1022 :
1023 : /// Set the deleted_at field in the remote index file.
1024 : ///
1025 : /// This fails if the upload queue has not been `stop()`ed.
1026 : ///
1027 : /// The caller is responsible for calling `stop()` AND for waiting
1028 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1029 : /// Check method [`RemoteTimelineClient::stop`] for details.
1030 0 : #[instrument(skip_all)]
1031 : pub(crate) async fn persist_index_part_with_deleted_flag(
1032 : self: &Arc<Self>,
1033 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1034 : let index_part_with_deleted_at = {
1035 : let mut locked = self.upload_queue.lock().unwrap();
1036 :
1037 : // We must be in stopped state because otherwise
1038 : // we can have inprogress index part upload that can overwrite the file
1039 : // with missing is_deleted flag that we going to set below
1040 : let stopped = locked.stopped_mut()?;
1041 :
1042 : match stopped.deleted_at {
1043 : SetDeletedFlagProgress::NotRunning => (), // proceed
1044 : SetDeletedFlagProgress::InProgress(at) => {
1045 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1046 : }
1047 : SetDeletedFlagProgress::Successful(at) => {
1048 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1049 : }
1050 : };
1051 : let deleted_at = Utc::now().naive_utc();
1052 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1053 :
1054 : let mut index_part = IndexPart::from(&stopped.upload_queue_for_deletion);
1055 : index_part.deleted_at = Some(deleted_at);
1056 : index_part
1057 : };
1058 :
1059 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1060 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1061 0 : let stopped = locked
1062 0 : .stopped_mut()
1063 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1064 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1065 0 : });
1066 :
1067 : pausable_failpoint!("persist_deleted_index_part");
1068 :
1069 : backoff::retry(
1070 0 : || {
1071 0 : upload::upload_index_part(
1072 0 : &self.storage_impl,
1073 0 : &self.tenant_shard_id,
1074 0 : &self.timeline_id,
1075 0 : self.generation,
1076 0 : &index_part_with_deleted_at,
1077 0 : &self.cancel,
1078 0 : )
1079 0 : },
1080 0 : |_e| false,
1081 : 1,
1082 : // have just a couple of attempts
1083 : // when executed as part of timeline deletion this happens in context of api call
1084 : // when executed as part of tenant deletion this happens in the background
1085 : 2,
1086 : "persist_index_part_with_deleted_flag",
1087 : &self.cancel,
1088 : )
1089 : .await
1090 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1091 0 : .and_then(|x| x)?;
1092 :
1093 : // all good, disarm the guard and mark as success
1094 : ScopeGuard::into_inner(undo_deleted_at);
1095 : {
1096 : let mut locked = self.upload_queue.lock().unwrap();
1097 :
1098 : let stopped = locked
1099 : .stopped_mut()
1100 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1101 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1102 : index_part_with_deleted_at
1103 : .deleted_at
1104 : .expect("we set it above"),
1105 : );
1106 : }
1107 :
1108 : Ok(())
1109 : }
1110 :
1111 0 : pub(crate) async fn preserve_initdb_archive(
1112 0 : self: &Arc<Self>,
1113 0 : tenant_id: &TenantId,
1114 0 : timeline_id: &TimelineId,
1115 0 : cancel: &CancellationToken,
1116 0 : ) -> anyhow::Result<()> {
1117 0 : backoff::retry(
1118 0 : || async {
1119 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1120 0 : .await
1121 0 : },
1122 0 : TimeoutOrCancel::caused_by_cancel,
1123 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1124 0 : FAILED_REMOTE_OP_RETRIES,
1125 0 : "preserve_initdb_tar_zst",
1126 0 : &cancel.clone(),
1127 0 : )
1128 0 : .await
1129 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1130 0 : .and_then(|x| x)
1131 0 : .context("backing up initdb archive")?;
1132 0 : Ok(())
1133 0 : }
1134 :
1135 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1136 : ///
1137 : /// This is not normally needed.
1138 0 : pub(crate) async fn upload_layer_file(
1139 0 : self: &Arc<Self>,
1140 0 : uploaded: &ResidentLayer,
1141 0 : cancel: &CancellationToken,
1142 0 : ) -> anyhow::Result<()> {
1143 0 : let remote_path = remote_layer_path(
1144 0 : &self.tenant_shard_id.tenant_id,
1145 0 : &self.timeline_id,
1146 0 : self.tenant_shard_id.to_index(),
1147 0 : &uploaded.layer_desc().layer_name(),
1148 0 : uploaded.metadata().generation,
1149 0 : );
1150 0 :
1151 0 : backoff::retry(
1152 0 : || async {
1153 0 : upload::upload_timeline_layer(
1154 0 : &self.storage_impl,
1155 0 : uploaded.local_path(),
1156 0 : &remote_path,
1157 0 : uploaded.metadata().file_size(),
1158 0 : cancel,
1159 0 : )
1160 0 : .await
1161 0 : },
1162 0 : TimeoutOrCancel::caused_by_cancel,
1163 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1164 0 : FAILED_REMOTE_OP_RETRIES,
1165 0 : "upload a layer without adding it to latest files",
1166 0 : cancel,
1167 0 : )
1168 0 : .await
1169 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1170 0 : .and_then(|x| x)
1171 0 : .context("upload a layer without adding it to latest files")
1172 0 : }
1173 :
1174 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1175 : /// not added to be part of a future `index_part.json` upload.
1176 0 : pub(crate) async fn copy_timeline_layer(
1177 0 : self: &Arc<Self>,
1178 0 : adopted: &Layer,
1179 0 : adopted_as: &Layer,
1180 0 : cancel: &CancellationToken,
1181 0 : ) -> anyhow::Result<()> {
1182 0 : let source_remote_path = remote_layer_path(
1183 0 : &self.tenant_shard_id.tenant_id,
1184 0 : &adopted
1185 0 : .get_timeline_id()
1186 0 : .expect("Source timeline should be alive"),
1187 0 : self.tenant_shard_id.to_index(),
1188 0 : &adopted.layer_desc().layer_name(),
1189 0 : adopted.metadata().generation,
1190 0 : );
1191 0 :
1192 0 : let target_remote_path = remote_layer_path(
1193 0 : &self.tenant_shard_id.tenant_id,
1194 0 : &self.timeline_id,
1195 0 : self.tenant_shard_id.to_index(),
1196 0 : &adopted_as.layer_desc().layer_name(),
1197 0 : adopted_as.metadata().generation,
1198 0 : );
1199 0 :
1200 0 : backoff::retry(
1201 0 : || async {
1202 0 : upload::copy_timeline_layer(
1203 0 : &self.storage_impl,
1204 0 : &source_remote_path,
1205 0 : &target_remote_path,
1206 0 : cancel,
1207 0 : )
1208 0 : .await
1209 0 : },
1210 0 : TimeoutOrCancel::caused_by_cancel,
1211 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1212 0 : FAILED_REMOTE_OP_RETRIES,
1213 0 : "copy timeline layer",
1214 0 : cancel,
1215 0 : )
1216 0 : .await
1217 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1218 0 : .and_then(|x| x)
1219 0 : .context("remote copy timeline layer")
1220 0 : }
1221 :
1222 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1223 0 : match tokio::time::timeout(
1224 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1225 0 : self.deletion_queue_client.flush_immediate(),
1226 0 : )
1227 0 : .await
1228 : {
1229 0 : Ok(result) => result,
1230 0 : Err(_timeout) => {
1231 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1232 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1233 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1234 0 : tracing::warn!(
1235 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1236 : );
1237 0 : Ok(())
1238 : }
1239 : }
1240 0 : }
1241 :
1242 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1243 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1244 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1245 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
1246 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1247 :
1248 0 : let layers: Vec<RemotePath> = {
1249 0 : let mut locked = self.upload_queue.lock().unwrap();
1250 0 : let stopped = locked.stopped_mut()?;
1251 :
1252 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1253 0 : anyhow::bail!("deleted_at is not set")
1254 0 : }
1255 0 :
1256 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1257 :
1258 0 : stopped
1259 0 : .upload_queue_for_deletion
1260 0 : .latest_files
1261 0 : .drain()
1262 0 : .map(|(file_name, meta)| {
1263 0 : remote_layer_path(
1264 0 : &self.tenant_shard_id.tenant_id,
1265 0 : &self.timeline_id,
1266 0 : meta.shard,
1267 0 : &file_name,
1268 0 : meta.generation,
1269 0 : )
1270 0 : })
1271 0 : .collect()
1272 0 : };
1273 0 :
1274 0 : let layer_deletion_count = layers.len();
1275 0 : self.deletion_queue_client.push_immediate(layers).await?;
1276 :
1277 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1278 : // inexistant objects are not considered errors.
1279 0 : let initdb_path =
1280 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1281 0 : self.deletion_queue_client
1282 0 : .push_immediate(vec![initdb_path])
1283 0 : .await?;
1284 :
1285 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1286 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1287 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1288 0 :
1289 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1290 0 : // taking the burden of listing all the layers that we already know we should delete.
1291 0 : self.flush_deletion_queue().await?;
1292 :
1293 0 : let cancel = shutdown_token();
1294 :
1295 0 : let remaining = download_retry(
1296 0 : || async {
1297 0 : self.storage_impl
1298 0 : .list(
1299 0 : Some(&timeline_storage_path),
1300 0 : ListingMode::NoDelimiter,
1301 0 : None,
1302 0 : &cancel,
1303 0 : )
1304 0 : .await
1305 0 : },
1306 0 : "list remaining files",
1307 0 : &cancel,
1308 0 : )
1309 0 : .await
1310 0 : .context("list files remaining files")?
1311 : .keys;
1312 :
1313 : // We will delete the current index_part object last, since it acts as a deletion
1314 : // marker via its deleted_at attribute
1315 0 : let latest_index = remaining
1316 0 : .iter()
1317 0 : .filter(|p| {
1318 0 : p.object_name()
1319 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1320 0 : .unwrap_or(false)
1321 0 : })
1322 0 : .filter_map(|path| parse_remote_index_path(path.clone()).map(|gen| (path, gen)))
1323 0 : .max_by_key(|i| i.1)
1324 0 : .map(|i| i.0.clone())
1325 0 : .unwrap_or(
1326 0 : // No generation-suffixed indices, assume we are dealing with
1327 0 : // a legacy index.
1328 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1329 0 : );
1330 0 :
1331 0 : let remaining_layers: Vec<RemotePath> = remaining
1332 0 : .into_iter()
1333 0 : .filter(|p| {
1334 0 : if p == &latest_index {
1335 0 : return false;
1336 0 : }
1337 0 : if p.object_name() == Some(INITDB_PRESERVED_PATH) {
1338 0 : return false;
1339 0 : }
1340 0 : true
1341 0 : })
1342 0 : .inspect(|path| {
1343 0 : if let Some(name) = path.object_name() {
1344 0 : info!(%name, "deleting a file not referenced from index_part.json");
1345 : } else {
1346 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1347 : }
1348 0 : })
1349 0 : .collect();
1350 0 :
1351 0 : let not_referenced_count = remaining_layers.len();
1352 0 : if !remaining_layers.is_empty() {
1353 0 : self.deletion_queue_client
1354 0 : .push_immediate(remaining_layers)
1355 0 : .await?;
1356 0 : }
1357 :
1358 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1359 0 : Err(anyhow::anyhow!(
1360 0 : "failpoint: timeline-delete-before-index-delete"
1361 0 : ))?
1362 0 : });
1363 :
1364 0 : debug!("enqueuing index part deletion");
1365 0 : self.deletion_queue_client
1366 0 : .push_immediate([latest_index].to_vec())
1367 0 : .await?;
1368 :
1369 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1370 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1371 0 : self.flush_deletion_queue().await?;
1372 :
1373 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1374 0 : Err(anyhow::anyhow!(
1375 0 : "failpoint: timeline-delete-after-index-delete"
1376 0 : ))?
1377 0 : });
1378 :
1379 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1380 :
1381 0 : Ok(())
1382 0 : }
1383 :
1384 : ///
1385 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1386 : /// the ordering constraints.
1387 : ///
1388 : /// The caller needs to already hold the `upload_queue` lock.
1389 4825 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1390 7438 : while let Some(next_op) = upload_queue.queued_operations.front() {
1391 : // Can we run this task now?
1392 4115 : let can_run_now = match next_op {
1393 : UploadOp::UploadLayer(..) => {
1394 : // Can always be scheduled.
1395 1173 : true
1396 : }
1397 : UploadOp::UploadMetadata(_, _) => {
1398 : // These can only be performed after all the preceding operations
1399 : // have finished.
1400 2527 : upload_queue.inprogress_tasks.is_empty()
1401 : }
1402 : UploadOp::Delete(_) => {
1403 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1404 192 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1405 : }
1406 :
1407 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1408 223 : upload_queue.inprogress_tasks.is_empty()
1409 : }
1410 : };
1411 :
1412 : // If we cannot launch this task, don't look any further.
1413 : //
1414 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1415 : // them now, but we don't try to do that currently. For example, if the frontmost task
1416 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1417 : // could still start layer uploads that were scheduled later.
1418 4115 : if !can_run_now {
1419 1496 : break;
1420 2619 : }
1421 2619 :
1422 2619 : if let UploadOp::Shutdown = next_op {
1423 : // leave the op in the queue but do not start more tasks; it will be dropped when
1424 : // the stop is called.
1425 6 : upload_queue.shutdown_ready.close();
1426 6 : break;
1427 2613 : }
1428 2613 :
1429 2613 : // We can launch this task. Remove it from the queue first.
1430 2613 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1431 2613 :
1432 2613 : debug!("starting op: {}", next_op);
1433 :
1434 : // Update the counters
1435 2613 : match next_op {
1436 1173 : UploadOp::UploadLayer(_, _) => {
1437 1173 : upload_queue.num_inprogress_layer_uploads += 1;
1438 1173 : }
1439 1154 : UploadOp::UploadMetadata(_, _) => {
1440 1154 : upload_queue.num_inprogress_metadata_uploads += 1;
1441 1154 : }
1442 170 : UploadOp::Delete(_) => {
1443 170 : upload_queue.num_inprogress_deletions += 1;
1444 170 : }
1445 116 : UploadOp::Barrier(sender) => {
1446 116 : sender.send_replace(());
1447 116 : continue;
1448 : }
1449 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1450 : };
1451 :
1452 : // Assign unique ID to this task
1453 2497 : upload_queue.task_counter += 1;
1454 2497 : let upload_task_id = upload_queue.task_counter;
1455 2497 :
1456 2497 : // Add it to the in-progress map
1457 2497 : let task = Arc::new(UploadTask {
1458 2497 : task_id: upload_task_id,
1459 2497 : op: next_op,
1460 2497 : retries: AtomicU32::new(0),
1461 2497 : });
1462 2497 : upload_queue
1463 2497 : .inprogress_tasks
1464 2497 : .insert(task.task_id, Arc::clone(&task));
1465 2497 :
1466 2497 : // Spawn task to perform the task
1467 2497 : let self_rc = Arc::clone(self);
1468 2497 : let tenant_shard_id = self.tenant_shard_id;
1469 2497 : let timeline_id = self.timeline_id;
1470 2497 : task_mgr::spawn(
1471 2497 : &self.runtime,
1472 2497 : TaskKind::RemoteUploadTask,
1473 2497 : Some(self.tenant_shard_id),
1474 2497 : Some(self.timeline_id),
1475 2497 : "remote upload",
1476 : false,
1477 2393 : async move {
1478 35492 : self_rc.perform_upload_task(task).await;
1479 2273 : Ok(())
1480 2273 : }
1481 2497 : .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)),
1482 : );
1483 :
1484 : // Loop back to process next task
1485 : }
1486 4825 : }
1487 :
1488 : ///
1489 : /// Perform an upload task.
1490 : ///
1491 : /// The task is in the `inprogress_tasks` list. This function will try to
1492 : /// execute it, retrying forever. On successful completion, the task is
1493 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1494 : /// queue that were waiting by the completion are launched.
1495 : ///
1496 : /// The task can be shut down, however. That leads to stopping the whole
1497 : /// queue.
1498 : ///
1499 2393 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1500 2393 : let cancel = shutdown_token();
1501 : // Loop to retry until it completes.
1502 2393 : loop {
1503 2393 : // If we're requested to shut down, close up shop and exit.
1504 2393 : //
1505 2393 : // Note: We only check for the shutdown requests between retries, so
1506 2393 : // if a shutdown request arrives while we're busy uploading, in the
1507 2393 : // upload::upload:*() call below, we will wait not exit until it has
1508 2393 : // finished. We probably could cancel the upload by simply dropping
1509 2393 : // the Future, but we're not 100% sure if the remote storage library
1510 2393 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1511 2393 : // upload finishes or times out soon enough.
1512 2393 : if cancel.is_cancelled() {
1513 0 : info!("upload task cancelled by shutdown request");
1514 0 : self.stop();
1515 0 : return;
1516 2393 : }
1517 :
1518 2393 : let upload_result: anyhow::Result<()> = match &task.op {
1519 1072 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1520 1072 : let local_path = layer.local_path();
1521 1072 :
1522 1072 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1523 1072 : // the metadata in the upload should always match our current generation.
1524 1072 : assert_eq!(layer_metadata.generation, self.generation);
1525 :
1526 1072 : let remote_path = remote_layer_path(
1527 1072 : &self.tenant_shard_id.tenant_id,
1528 1072 : &self.timeline_id,
1529 1072 : layer_metadata.shard,
1530 1072 : &layer.layer_desc().layer_name(),
1531 1072 : layer_metadata.generation,
1532 1072 : );
1533 1072 :
1534 1072 : upload::upload_timeline_layer(
1535 1072 : &self.storage_impl,
1536 1072 : local_path,
1537 1072 : &remote_path,
1538 1072 : layer_metadata.file_size(),
1539 1072 : &self.cancel,
1540 1072 : )
1541 1072 : .measure_remote_op(
1542 1072 : RemoteOpFileKind::Layer,
1543 1072 : RemoteOpKind::Upload,
1544 1072 : Arc::clone(&self.metrics),
1545 1072 : )
1546 30597 : .await
1547 : }
1548 1151 : UploadOp::UploadMetadata(ref index_part, _lsn) => {
1549 1151 : let mention_having_future_layers = if cfg!(feature = "testing") {
1550 1151 : index_part
1551 1151 : .layer_metadata
1552 1151 : .keys()
1553 13200 : .any(|x| x.is_in_future(*_lsn))
1554 : } else {
1555 0 : false
1556 : };
1557 :
1558 1151 : let res = upload::upload_index_part(
1559 1151 : &self.storage_impl,
1560 1151 : &self.tenant_shard_id,
1561 1151 : &self.timeline_id,
1562 1151 : self.generation,
1563 1151 : index_part,
1564 1151 : &self.cancel,
1565 1151 : )
1566 1151 : .measure_remote_op(
1567 1151 : RemoteOpFileKind::Index,
1568 1151 : RemoteOpKind::Upload,
1569 1151 : Arc::clone(&self.metrics),
1570 1151 : )
1571 4725 : .await;
1572 1144 : if res.is_ok() {
1573 1144 : self.update_remote_physical_size_gauge(Some(index_part));
1574 1144 : if mention_having_future_layers {
1575 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1576 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");
1577 1140 : }
1578 0 : }
1579 1144 : res
1580 : }
1581 170 : UploadOp::Delete(delete) => {
1582 : pausable_failpoint!("before-delete-layer-pausable");
1583 170 : self.deletion_queue_client
1584 170 : .push_layers(
1585 170 : self.tenant_shard_id,
1586 170 : self.timeline_id,
1587 170 : self.generation,
1588 170 : delete.layers.clone(),
1589 170 : )
1590 0 : .await
1591 170 : .map_err(|e| anyhow::anyhow!(e))
1592 : }
1593 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1594 : // unreachable. Barrier operations are handled synchronously in
1595 : // launch_queued_tasks
1596 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1597 0 : break;
1598 : }
1599 : };
1600 :
1601 0 : match upload_result {
1602 : Ok(()) => {
1603 2273 : break;
1604 : }
1605 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1606 0 : // loop around to do the proper stopping
1607 0 : continue;
1608 : }
1609 0 : Err(e) => {
1610 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1611 0 :
1612 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1613 0 : // or other external reasons. Such issues are relatively regular, so log them
1614 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1615 0 : //
1616 0 : // (See similar logic for downloads in `download::download_retry`)
1617 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1618 0 : info!(
1619 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1620 0 : task.op, retries, e
1621 : );
1622 : } else {
1623 0 : warn!(
1624 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1625 0 : task.op, retries, e
1626 : );
1627 : }
1628 :
1629 : // sleep until it's time to retry, or we're cancelled
1630 0 : exponential_backoff(
1631 0 : retries,
1632 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1633 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1634 0 : &cancel,
1635 0 : )
1636 0 : .await;
1637 : }
1638 : }
1639 : }
1640 :
1641 2273 : let retries = task.retries.load(Ordering::SeqCst);
1642 2273 : if retries > 0 {
1643 0 : info!(
1644 0 : "remote task {} completed successfully after {} retries",
1645 0 : task.op, retries
1646 : );
1647 : } else {
1648 2273 : debug!("remote task {} completed successfully", task.op);
1649 : }
1650 :
1651 : // The task has completed successfully. Remove it from the in-progress list.
1652 2273 : let lsn_update = {
1653 2273 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1654 2273 : let upload_queue = match upload_queue_guard.deref_mut() {
1655 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1656 0 : UploadQueue::Stopped(_stopped) => {
1657 0 : None
1658 : },
1659 2273 : UploadQueue::Initialized(qi) => { Some(qi) }
1660 : };
1661 :
1662 2273 : let upload_queue = match upload_queue {
1663 2273 : Some(upload_queue) => upload_queue,
1664 : None => {
1665 0 : info!("another concurrent task already stopped the queue");
1666 0 : return;
1667 : }
1668 : };
1669 :
1670 2273 : upload_queue.inprogress_tasks.remove(&task.task_id);
1671 :
1672 2273 : let lsn_update = match task.op {
1673 : UploadOp::UploadLayer(_, _) => {
1674 959 : upload_queue.num_inprogress_layer_uploads -= 1;
1675 959 : None
1676 : }
1677 1144 : UploadOp::UploadMetadata(_, lsn) => {
1678 1144 : upload_queue.num_inprogress_metadata_uploads -= 1;
1679 1144 : // XXX monotonicity check?
1680 1144 :
1681 1144 : upload_queue.projected_remote_consistent_lsn = Some(lsn);
1682 1144 : if self.generation.is_none() {
1683 : // Legacy mode: skip validating generation
1684 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
1685 0 : None
1686 : } else {
1687 1144 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
1688 : }
1689 : }
1690 : UploadOp::Delete(_) => {
1691 170 : upload_queue.num_inprogress_deletions -= 1;
1692 170 : None
1693 : }
1694 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
1695 : };
1696 :
1697 : // Launch any queued tasks that were unblocked by this one.
1698 2273 : self.launch_queued_tasks(upload_queue);
1699 2273 : lsn_update
1700 : };
1701 :
1702 2273 : if let Some((lsn, slot)) = lsn_update {
1703 : // Updates to the remote_consistent_lsn we advertise to pageservers
1704 : // are all routed through the DeletionQueue, to enforce important
1705 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
1706 1144 : self.deletion_queue_client
1707 1144 : .update_remote_consistent_lsn(
1708 1144 : self.tenant_shard_id,
1709 1144 : self.timeline_id,
1710 1144 : self.generation,
1711 1144 : lsn,
1712 1144 : slot,
1713 1144 : )
1714 0 : .await;
1715 1129 : }
1716 :
1717 2273 : self.metric_end(&task.op);
1718 2273 : }
1719 :
1720 4901 : fn metric_impl(
1721 4901 : &self,
1722 4901 : op: &UploadOp,
1723 4901 : ) -> Option<(
1724 4901 : RemoteOpFileKind,
1725 4901 : RemoteOpKind,
1726 4901 : RemoteTimelineClientMetricsCallTrackSize,
1727 4901 : )> {
1728 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
1729 4901 : let res = match op {
1730 2133 : UploadOp::UploadLayer(_, m) => (
1731 2133 : RemoteOpFileKind::Layer,
1732 2133 : RemoteOpKind::Upload,
1733 2133 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size()),
1734 2133 : ),
1735 2320 : UploadOp::UploadMetadata(_, _) => (
1736 2320 : RemoteOpFileKind::Index,
1737 2320 : RemoteOpKind::Upload,
1738 2320 : DontTrackSize {
1739 2320 : reason: "metadata uploads are tiny",
1740 2320 : },
1741 2320 : ),
1742 442 : UploadOp::Delete(_delete) => (
1743 442 : RemoteOpFileKind::Layer,
1744 442 : RemoteOpKind::Delete,
1745 442 : DontTrackSize {
1746 442 : reason: "should we track deletes? positive or negative sign?",
1747 442 : },
1748 442 : ),
1749 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
1750 : // we do not account these
1751 6 : return None;
1752 : }
1753 : };
1754 4895 : Some(res)
1755 4901 : }
1756 :
1757 2622 : fn metric_begin(&self, op: &UploadOp) {
1758 2622 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1759 2622 : Some(x) => x,
1760 0 : None => return,
1761 : };
1762 2622 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
1763 2622 : guard.will_decrement_manually(); // in metric_end(), see right below
1764 2622 : }
1765 :
1766 2279 : fn metric_end(&self, op: &UploadOp) {
1767 2279 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1768 2273 : Some(x) => x,
1769 6 : None => return,
1770 : };
1771 2273 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
1772 2279 : }
1773 :
1774 : /// Close the upload queue for new operations and cancel queued operations.
1775 : ///
1776 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
1777 : ///
1778 : /// In-progress operations will still be running after this function returns.
1779 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
1780 : /// to wait for them to complete, after calling this function.
1781 14 : pub(crate) fn stop(&self) {
1782 14 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
1783 14 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
1784 14 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
1785 14 : let mut guard = self.upload_queue.lock().unwrap();
1786 14 : self.stop_impl(&mut guard);
1787 14 : }
1788 :
1789 14 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
1790 14 : match &mut **guard {
1791 : UploadQueue::Uninitialized => {
1792 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
1793 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
1794 : }
1795 : UploadQueue::Stopped(_) => {
1796 : // nothing to do
1797 6 : info!("another concurrent task already shut down the queue");
1798 : }
1799 8 : UploadQueue::Initialized(initialized) => {
1800 8 : info!("shutting down upload queue");
1801 :
1802 : // Replace the queue with the Stopped state, taking ownership of the old
1803 : // Initialized queue. We will do some checks on it, and then drop it.
1804 8 : let qi = {
1805 : // Here we preserve working version of the upload queue for possible use during deletions.
1806 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
1807 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
1808 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
1809 8 : let upload_queue_for_deletion = UploadQueueInitialized {
1810 8 : task_counter: 0,
1811 8 : latest_files: initialized.latest_files.clone(),
1812 8 : latest_files_changes_since_metadata_upload_scheduled: 0,
1813 8 : latest_metadata: initialized.latest_metadata.clone(),
1814 8 : projected_remote_consistent_lsn: None,
1815 8 : visible_remote_consistent_lsn: initialized
1816 8 : .visible_remote_consistent_lsn
1817 8 : .clone(),
1818 8 : num_inprogress_layer_uploads: 0,
1819 8 : num_inprogress_metadata_uploads: 0,
1820 8 : num_inprogress_deletions: 0,
1821 8 : inprogress_tasks: HashMap::default(),
1822 8 : queued_operations: VecDeque::default(),
1823 8 : #[cfg(feature = "testing")]
1824 8 : dangling_files: HashMap::default(),
1825 8 : shutting_down: false,
1826 8 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
1827 8 : };
1828 8 :
1829 8 : let upload_queue = std::mem::replace(
1830 8 : &mut **guard,
1831 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
1832 8 : UploadQueueStoppedDeletable {
1833 8 : upload_queue_for_deletion,
1834 8 : deleted_at: SetDeletedFlagProgress::NotRunning,
1835 8 : },
1836 8 : )),
1837 8 : );
1838 8 : if let UploadQueue::Initialized(qi) = upload_queue {
1839 8 : qi
1840 : } else {
1841 0 : unreachable!("we checked in the match above that it is Initialized");
1842 : }
1843 : };
1844 :
1845 : // consistency check
1846 8 : assert_eq!(
1847 8 : qi.num_inprogress_layer_uploads
1848 8 : + qi.num_inprogress_metadata_uploads
1849 8 : + qi.num_inprogress_deletions,
1850 8 : qi.inprogress_tasks.len()
1851 8 : );
1852 :
1853 : // We don't need to do anything here for in-progress tasks. They will finish
1854 : // on their own, decrement the unfinished-task counter themselves, and observe
1855 : // that the queue is Stopped.
1856 8 : drop(qi.inprogress_tasks);
1857 :
1858 : // Tear down queued ops
1859 8 : for op in qi.queued_operations.into_iter() {
1860 6 : self.metric_end(&op);
1861 6 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
1862 6 : // which is exactly what we want to happen.
1863 6 : drop(op);
1864 6 : }
1865 : }
1866 : }
1867 14 : }
1868 : }
1869 :
1870 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1871 0 : let path = format!("tenants/{tenant_shard_id}");
1872 0 : RemotePath::from_string(&path).expect("Failed to construct path")
1873 0 : }
1874 :
1875 148 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1876 148 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
1877 148 : RemotePath::from_string(&path).expect("Failed to construct path")
1878 148 : }
1879 :
1880 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
1881 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
1882 0 : RemotePath::from_string(&path).expect("Failed to construct path")
1883 0 : }
1884 :
1885 30 : pub fn remote_timeline_path(
1886 30 : tenant_shard_id: &TenantShardId,
1887 30 : timeline_id: &TimelineId,
1888 30 : ) -> RemotePath {
1889 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
1890 30 : }
1891 :
1892 : /// Note that the shard component of a remote layer path is _not_ always the same
1893 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
1894 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
1895 1090 : pub fn remote_layer_path(
1896 1090 : tenant_id: &TenantId,
1897 1090 : timeline_id: &TimelineId,
1898 1090 : shard: ShardIndex,
1899 1090 : layer_file_name: &LayerName,
1900 1090 : generation: Generation,
1901 1090 : ) -> RemotePath {
1902 1090 : // Generation-aware key format
1903 1090 : let path = format!(
1904 1090 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
1905 1090 : shard.get_suffix(),
1906 1090 : layer_file_name,
1907 1090 : generation.get_suffix()
1908 1090 : );
1909 1090 :
1910 1090 : RemotePath::from_string(&path).expect("Failed to construct path")
1911 1090 : }
1912 :
1913 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
1914 4 : RemotePath::from_string(&format!(
1915 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
1916 4 : ))
1917 4 : .expect("Failed to construct path")
1918 4 : }
1919 :
1920 2 : pub fn remote_initdb_preserved_archive_path(
1921 2 : tenant_id: &TenantId,
1922 2 : timeline_id: &TimelineId,
1923 2 : ) -> RemotePath {
1924 2 : RemotePath::from_string(&format!(
1925 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
1926 2 : ))
1927 2 : .expect("Failed to construct path")
1928 2 : }
1929 :
1930 1202 : pub fn remote_index_path(
1931 1202 : tenant_shard_id: &TenantShardId,
1932 1202 : timeline_id: &TimelineId,
1933 1202 : generation: Generation,
1934 1202 : ) -> RemotePath {
1935 1202 : RemotePath::from_string(&format!(
1936 1202 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
1937 1202 : IndexPart::FILE_NAME,
1938 1202 : generation.get_suffix()
1939 1202 : ))
1940 1202 : .expect("Failed to construct path")
1941 1202 : }
1942 :
1943 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1944 0 : RemotePath::from_string(&format!(
1945 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
1946 0 : ))
1947 0 : .expect("Failed to construct path")
1948 0 : }
1949 :
1950 : /// Given the key of an index, parse out the generation part of the name
1951 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
1952 18 : let file_name = match path.get_path().file_name() {
1953 18 : Some(f) => f,
1954 : None => {
1955 : // Unexpected: we should be seeing index_part.json paths only
1956 0 : tracing::warn!("Malformed index key {}", path);
1957 0 : return None;
1958 : }
1959 : };
1960 :
1961 18 : match file_name.split_once('-') {
1962 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
1963 6 : None => None,
1964 : }
1965 18 : }
1966 :
1967 : #[cfg(test)]
1968 : mod tests {
1969 : use super::*;
1970 : use crate::{
1971 : context::RequestContext,
1972 : tenant::{
1973 : harness::{TenantHarness, TIMELINE_ID},
1974 : storage_layer::layer::local_layer_path,
1975 : Tenant, Timeline,
1976 : },
1977 : DEFAULT_PG_VERSION,
1978 : };
1979 :
1980 : use std::collections::HashSet;
1981 :
1982 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
1983 8 : format!("contents for {name}").into()
1984 8 : }
1985 :
1986 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
1987 2 : let metadata = TimelineMetadata::new(
1988 2 : disk_consistent_lsn,
1989 2 : None,
1990 2 : None,
1991 2 : Lsn(0),
1992 2 : Lsn(0),
1993 2 : Lsn(0),
1994 2 : // Any version will do
1995 2 : // but it should be consistent with the one in the tests
1996 2 : crate::DEFAULT_PG_VERSION,
1997 2 : );
1998 2 :
1999 2 : // go through serialize + deserialize to fix the header, including checksum
2000 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2001 2 : }
2002 :
2003 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2004 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2005 2 : avec.sort();
2006 2 :
2007 2 : let mut bvec = b.to_vec();
2008 2 : bvec.sort_unstable();
2009 2 :
2010 2 : assert_eq!(avec, bvec);
2011 2 : }
2012 :
2013 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2014 4 : let mut expected: Vec<String> = expected
2015 4 : .iter()
2016 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2017 4 : .collect();
2018 4 : expected.sort();
2019 4 :
2020 4 : let mut found: Vec<String> = Vec::new();
2021 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2022 16 : let entry_name = entry.file_name();
2023 16 : let fname = entry_name.to_str().unwrap();
2024 16 : found.push(String::from(fname));
2025 16 : }
2026 4 : found.sort();
2027 4 :
2028 4 : assert_eq!(found, expected);
2029 4 : }
2030 :
2031 : struct TestSetup {
2032 : harness: TenantHarness,
2033 : tenant: Arc<Tenant>,
2034 : timeline: Arc<Timeline>,
2035 : tenant_ctx: RequestContext,
2036 : }
2037 :
2038 : impl TestSetup {
2039 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2040 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2041 8 : let harness = TenantHarness::create(test_name)?;
2042 32 : let (tenant, ctx) = harness.load().await;
2043 :
2044 8 : let timeline = tenant
2045 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2046 22 : .await?;
2047 :
2048 8 : Ok(Self {
2049 8 : harness,
2050 8 : tenant,
2051 8 : timeline,
2052 8 : tenant_ctx: ctx,
2053 8 : })
2054 8 : }
2055 :
2056 : /// Construct a RemoteTimelineClient in an arbitrary generation
2057 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2058 10 : Arc::new(RemoteTimelineClient {
2059 10 : conf: self.harness.conf,
2060 10 : runtime: tokio::runtime::Handle::current(),
2061 10 : tenant_shard_id: self.harness.tenant_shard_id,
2062 10 : timeline_id: TIMELINE_ID,
2063 10 : generation,
2064 10 : storage_impl: self.harness.remote_storage.clone(),
2065 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2066 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2067 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2068 10 : &self.harness.tenant_shard_id,
2069 10 : &TIMELINE_ID,
2070 10 : )),
2071 10 : cancel: CancellationToken::new(),
2072 10 : })
2073 10 : }
2074 :
2075 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2076 : /// and timeline_id are present.
2077 6 : fn span(&self) -> tracing::Span {
2078 6 : tracing::info_span!(
2079 : "test",
2080 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2081 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2082 : timeline_id = %TIMELINE_ID
2083 : )
2084 6 : }
2085 : }
2086 :
2087 : // Test scheduling
2088 : #[tokio::test]
2089 2 : async fn upload_scheduling() {
2090 2 : // Test outline:
2091 2 : //
2092 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2093 2 : // Schedule upload of index. Check that it is queued
2094 2 : // let the layer file uploads finish. Check that the index-upload is now started
2095 2 : // let the index-upload finish.
2096 2 : //
2097 2 : // Download back the index.json. Check that the list of files is correct
2098 2 : //
2099 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2100 2 : // let upload finish. Check that deletion is now started
2101 2 : // Schedule another deletion. Check that it's launched immediately.
2102 2 : // Schedule index upload. Check that it's queued
2103 2 :
2104 13 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2105 2 : let span = test_setup.span();
2106 2 : let _guard = span.enter();
2107 2 :
2108 2 : let TestSetup {
2109 2 : harness,
2110 2 : tenant: _tenant,
2111 2 : timeline,
2112 2 : tenant_ctx: _tenant_ctx,
2113 2 : } = test_setup;
2114 2 :
2115 2 : let client = timeline.remote_client.as_ref().unwrap();
2116 2 :
2117 2 : // Download back the index.json, and check that the list of files is correct
2118 2 : let initial_index_part = match client
2119 2 : .download_index_file(&CancellationToken::new())
2120 8 : .await
2121 2 : .unwrap()
2122 2 : {
2123 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2124 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2125 2 : };
2126 2 : let initial_layers = initial_index_part
2127 2 : .layer_metadata
2128 2 : .keys()
2129 2 : .map(|f| f.to_owned())
2130 2 : .collect::<HashSet<LayerName>>();
2131 2 : let initial_layer = {
2132 2 : assert!(initial_layers.len() == 1);
2133 2 : initial_layers.into_iter().next().unwrap()
2134 2 : };
2135 2 :
2136 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2137 2 :
2138 2 : println!("workdir: {}", harness.conf.workdir);
2139 2 :
2140 2 : let remote_timeline_dir = harness
2141 2 : .remote_fs_dir
2142 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2143 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2144 2 :
2145 2 : let generation = harness.generation;
2146 2 : let shard = harness.shard;
2147 2 :
2148 2 : // Create a couple of dummy files, schedule upload for them
2149 2 :
2150 2 : let layers = [
2151 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2152 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2153 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2154 2 : ]
2155 2 : .into_iter()
2156 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2157 6 :
2158 6 : let local_path = local_layer_path(
2159 6 : harness.conf,
2160 6 : &timeline.tenant_shard_id,
2161 6 : &timeline.timeline_id,
2162 6 : &name,
2163 6 : &generation,
2164 6 : );
2165 6 : std::fs::write(&local_path, &contents).unwrap();
2166 6 :
2167 6 : Layer::for_resident(
2168 6 : harness.conf,
2169 6 : &timeline,
2170 6 : local_path,
2171 6 : name,
2172 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2173 6 : )
2174 6 : }).collect::<Vec<_>>();
2175 2 :
2176 2 : client
2177 2 : .schedule_layer_file_upload(layers[0].clone())
2178 2 : .unwrap();
2179 2 : client
2180 2 : .schedule_layer_file_upload(layers[1].clone())
2181 2 : .unwrap();
2182 2 :
2183 2 : // Check that they are started immediately, not queued
2184 2 : //
2185 2 : // this works because we running within block_on, so any futures are now queued up until
2186 2 : // our next await point.
2187 2 : {
2188 2 : let mut guard = client.upload_queue.lock().unwrap();
2189 2 : let upload_queue = guard.initialized_mut().unwrap();
2190 2 : assert!(upload_queue.queued_operations.is_empty());
2191 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2192 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2193 2 :
2194 2 : // also check that `latest_file_changes` was updated
2195 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2196 2 : }
2197 2 :
2198 2 : // Schedule upload of index. Check that it is queued
2199 2 : let metadata = dummy_metadata(Lsn(0x20));
2200 2 : client
2201 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2202 2 : .unwrap();
2203 2 : {
2204 2 : let mut guard = client.upload_queue.lock().unwrap();
2205 2 : let upload_queue = guard.initialized_mut().unwrap();
2206 2 : assert!(upload_queue.queued_operations.len() == 1);
2207 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2208 2 : }
2209 2 :
2210 2 : // Wait for the uploads to finish
2211 2 : client.wait_completion().await.unwrap();
2212 2 : {
2213 2 : let mut guard = client.upload_queue.lock().unwrap();
2214 2 : let upload_queue = guard.initialized_mut().unwrap();
2215 2 :
2216 2 : assert!(upload_queue.queued_operations.is_empty());
2217 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2218 2 : }
2219 2 :
2220 2 : // Download back the index.json, and check that the list of files is correct
2221 2 : let index_part = match client
2222 2 : .download_index_file(&CancellationToken::new())
2223 6 : .await
2224 2 : .unwrap()
2225 2 : {
2226 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2227 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2228 2 : };
2229 2 :
2230 2 : assert_file_list(
2231 2 : &index_part
2232 2 : .layer_metadata
2233 2 : .keys()
2234 6 : .map(|f| f.to_owned())
2235 2 : .collect(),
2236 2 : &[
2237 2 : &initial_layer.to_string(),
2238 2 : &layers[0].layer_desc().layer_name().to_string(),
2239 2 : &layers[1].layer_desc().layer_name().to_string(),
2240 2 : ],
2241 2 : );
2242 2 : assert_eq!(index_part.metadata, metadata);
2243 2 :
2244 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2245 2 : client
2246 2 : .schedule_layer_file_upload(layers[2].clone())
2247 2 : .unwrap();
2248 2 :
2249 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2250 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2251 2 : // spawn_blocking started by the drop.
2252 2 : client
2253 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2254 2 : .unwrap();
2255 2 : {
2256 2 : let mut guard = client.upload_queue.lock().unwrap();
2257 2 : let upload_queue = guard.initialized_mut().unwrap();
2258 2 :
2259 2 : // Deletion schedules upload of the index file, and the file deletion itself
2260 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2261 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2262 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2263 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2264 2 : assert_eq!(
2265 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2266 2 : 0
2267 2 : );
2268 2 : }
2269 2 : assert_remote_files(
2270 2 : &[
2271 2 : &initial_layer.to_string(),
2272 2 : &layers[0].layer_desc().layer_name().to_string(),
2273 2 : &layers[1].layer_desc().layer_name().to_string(),
2274 2 : "index_part.json",
2275 2 : ],
2276 2 : &remote_timeline_dir,
2277 2 : generation,
2278 2 : );
2279 2 :
2280 2 : // Finish them
2281 2 : client.wait_completion().await.unwrap();
2282 2 : harness.deletion_queue.pump().await;
2283 2 :
2284 2 : assert_remote_files(
2285 2 : &[
2286 2 : &initial_layer.to_string(),
2287 2 : &layers[1].layer_desc().layer_name().to_string(),
2288 2 : &layers[2].layer_desc().layer_name().to_string(),
2289 2 : "index_part.json",
2290 2 : ],
2291 2 : &remote_timeline_dir,
2292 2 : generation,
2293 2 : );
2294 2 : }
2295 :
2296 : #[tokio::test]
2297 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2298 2 : // Setup
2299 2 :
2300 2 : let TestSetup {
2301 2 : harness,
2302 2 : tenant: _tenant,
2303 2 : timeline,
2304 2 : ..
2305 13 : } = TestSetup::new("metrics").await.unwrap();
2306 2 : let client = timeline.remote_client.as_ref().unwrap();
2307 2 :
2308 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2309 2 : let local_path = local_layer_path(
2310 2 : harness.conf,
2311 2 : &timeline.tenant_shard_id,
2312 2 : &timeline.timeline_id,
2313 2 : &layer_file_name_1,
2314 2 : &harness.generation,
2315 2 : );
2316 2 : let content_1 = dummy_contents("foo");
2317 2 : std::fs::write(&local_path, &content_1).unwrap();
2318 2 :
2319 2 : let layer_file_1 = Layer::for_resident(
2320 2 : harness.conf,
2321 2 : &timeline,
2322 2 : local_path,
2323 2 : layer_file_name_1.clone(),
2324 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2325 2 : );
2326 2 :
2327 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2328 2 : struct BytesStartedFinished {
2329 2 : started: Option<usize>,
2330 2 : finished: Option<usize>,
2331 2 : }
2332 2 : impl std::ops::Add for BytesStartedFinished {
2333 2 : type Output = Self;
2334 4 : fn add(self, rhs: Self) -> Self::Output {
2335 4 : Self {
2336 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2337 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2338 4 : }
2339 4 : }
2340 2 : }
2341 6 : let get_bytes_started_stopped = || {
2342 6 : let started = client
2343 6 : .metrics
2344 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2345 6 : .map(|v| v.try_into().unwrap());
2346 6 : let stopped = client
2347 6 : .metrics
2348 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2349 6 : .map(|v| v.try_into().unwrap());
2350 6 : BytesStartedFinished {
2351 6 : started,
2352 6 : finished: stopped,
2353 6 : }
2354 6 : };
2355 2 :
2356 2 : // Test
2357 2 : tracing::info!("now doing actual test");
2358 2 :
2359 2 : let actual_a = get_bytes_started_stopped();
2360 2 :
2361 2 : client
2362 2 : .schedule_layer_file_upload(layer_file_1.clone())
2363 2 : .unwrap();
2364 2 :
2365 2 : let actual_b = get_bytes_started_stopped();
2366 2 :
2367 2 : client.wait_completion().await.unwrap();
2368 2 :
2369 2 : let actual_c = get_bytes_started_stopped();
2370 2 :
2371 2 : // Validate
2372 2 :
2373 2 : let expected_b = actual_a
2374 2 : + BytesStartedFinished {
2375 2 : started: Some(content_1.len()),
2376 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2377 2 : finished: Some(0),
2378 2 : };
2379 2 : assert_eq!(actual_b, expected_b);
2380 2 :
2381 2 : let expected_c = actual_a
2382 2 : + BytesStartedFinished {
2383 2 : started: Some(content_1.len()),
2384 2 : finished: Some(content_1.len()),
2385 2 : };
2386 2 : assert_eq!(actual_c, expected_c);
2387 2 : }
2388 :
2389 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2390 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2391 12 : let example_index_part = IndexPart::example();
2392 12 :
2393 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2394 12 :
2395 12 : let index_path = test_state.harness.remote_fs_dir.join(
2396 12 : remote_index_path(
2397 12 : &test_state.harness.tenant_shard_id,
2398 12 : &TIMELINE_ID,
2399 12 : generation,
2400 12 : )
2401 12 : .get_path(),
2402 12 : );
2403 12 :
2404 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2405 12 : .expect("creating test dir should work");
2406 12 :
2407 12 : eprintln!("Writing {index_path}");
2408 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2409 12 : example_index_part
2410 12 : }
2411 :
2412 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2413 : /// index, the IndexPart returned is equal to `expected`
2414 10 : async fn assert_got_index_part(
2415 10 : test_state: &TestSetup,
2416 10 : get_generation: Generation,
2417 10 : expected: &IndexPart,
2418 10 : ) {
2419 10 : let client = test_state.build_client(get_generation);
2420 :
2421 10 : let download_r = client
2422 10 : .download_index_file(&CancellationToken::new())
2423 53 : .await
2424 10 : .expect("download should always succeed");
2425 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2426 10 : match download_r {
2427 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2428 10 : assert_eq!(&index_part, expected);
2429 : }
2430 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2431 : }
2432 10 : }
2433 :
2434 : #[tokio::test]
2435 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2436 14 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2437 2 : let span = test_state.span();
2438 2 : let _guard = span.enter();
2439 2 :
2440 2 : // Simple case: we are in generation N, load the index from generation N - 1
2441 2 : let generation_n = 5;
2442 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2443 2 :
2444 10 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2445 2 :
2446 2 : Ok(())
2447 2 : }
2448 :
2449 : #[tokio::test]
2450 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2451 2 : let test_state = TestSetup::new("index_part_download_ordering")
2452 14 : .await
2453 2 : .unwrap();
2454 2 :
2455 2 : let span = test_state.span();
2456 2 : let _guard = span.enter();
2457 2 :
2458 2 : // A generation-less IndexPart exists in the bucket, we should find it
2459 2 : let generation_n = 5;
2460 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2461 14 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2462 2 :
2463 2 : // If a more recent-than-none generation exists, we should prefer to load that
2464 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2465 14 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2466 2 :
2467 2 : // If a more-recent-than-me generation exists, we should ignore it.
2468 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2469 7 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2470 2 :
2471 2 : // If a directly previous generation exists, _and_ an index exists in my own
2472 2 : // generation, I should prefer my own generation.
2473 2 : let _injected_prev =
2474 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2475 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2476 2 : assert_got_index_part(
2477 2 : &test_state,
2478 2 : Generation::new(generation_n),
2479 2 : &injected_current,
2480 2 : )
2481 8 : .await;
2482 2 :
2483 2 : Ok(())
2484 2 : }
2485 : }
|
