Line data Source code
1 : //! This module manages synchronizing local FS with remote storage.
2 : //!
3 : //! # Overview
4 : //!
5 : //! * [`RemoteTimelineClient`] provides functions related to upload/download of a particular timeline.
6 : //! It contains a queue of pending uploads, and manages the queue, performing uploads in parallel
7 : //! when it's safe to do so.
8 : //!
9 : //! * Stand-alone function, [`list_remote_timelines`], to get list of timelines of a tenant.
10 : //!
11 : //! These functions use the low-level remote storage client, [`remote_storage::RemoteStorage`].
12 : //!
13 : //! # APIs & How To Use Them
14 : //!
15 : //! There is a [RemoteTimelineClient] for each [Timeline][`crate::tenant::Timeline`] in the system,
16 : //! unless the pageserver is configured without remote storage.
17 : //!
18 : //! We allocate the client instance in [Timeline][`crate::tenant::Timeline`], i.e.,
19 : //! either in [`crate::tenant::mgr`] during startup or when creating a new
20 : //! timeline.
21 : //! However, the client does not become ready for use until we've initialized its upload queue:
22 : //!
23 : //! - For timelines that already have some state on the remote storage, we use
24 : //! [`RemoteTimelineClient::init_upload_queue`] .
25 : //! - For newly created timelines, we use
26 : //! [`RemoteTimelineClient::init_upload_queue_for_empty_remote`].
27 : //!
28 : //! The former takes the remote's [`IndexPart`] as an argument, possibly retrieved
29 : //! using [`list_remote_timelines`]. We'll elaborate on [`IndexPart`] in the next section.
30 : //!
31 : //! Whenever we've created/updated/deleted a file in a timeline directory, we schedule
32 : //! the corresponding remote operation with the timeline's [`RemoteTimelineClient`]:
33 : //!
34 : //! - [`RemoteTimelineClient::schedule_layer_file_upload`] when we've created a new layer file.
35 : //! - [`RemoteTimelineClient::schedule_index_upload_for_metadata_update`] when we've updated the timeline metadata file.
36 : //! - [`RemoteTimelineClient::schedule_index_upload_for_file_changes`] to upload an updated index file, after we've scheduled file uploads
37 : //! - [`RemoteTimelineClient::schedule_layer_file_deletion`] when we've deleted one or more layer files.
38 : //!
39 : //! Internally, these functions create [`UploadOp`]s and put them in a queue.
40 : //!
41 : //! There are also APIs for downloading files.
42 : //! These are not part of the aforementioned queuing and will not be discussed
43 : //! further here, except in the section covering tenant attach.
44 : //!
45 : //! # Remote Storage Structure & [`IndexPart`] Index File
46 : //!
47 : //! The "directory structure" in the remote storage mirrors the local directory structure, with paths
48 : //! like `tenants/<tenant_id>/timelines/<timeline_id>/<layer filename>`.
49 : //! Yet instead of keeping the `metadata` file remotely, we wrap it with more
50 : //! data in an "index file" aka [`IndexPart`], containing the list of **all** remote
51 : //! files for a given timeline.
52 : //! If a file is not referenced from [`IndexPart`], it's not part of the remote storage state.
53 : //!
54 : //! Having the `IndexPart` also avoids expensive and slow `S3 list` commands.
55 : //!
56 : //! # Consistency
57 : //!
58 : //! To have a consistent remote structure, it's important that uploads and
59 : //! deletions are performed in the right order. For example, the index file
60 : //! contains a list of layer files, so it must not be uploaded until all the
61 : //! layer files that are in its list have been successfully uploaded.
62 : //!
63 : //! The contract between client and its user is that the user is responsible of
64 : //! scheduling operations in an order that keeps the remote consistent as
65 : //! described above.
66 : //! From the user's perspective, the operations are executed sequentially.
67 : //! Internally, the client knows which operations can be performed in parallel,
68 : //! and which operations act like a "barrier" that require preceding operations
69 : //! to finish. The calling code just needs to call the schedule-functions in the
70 : //! correct order, and the client will parallelize the operations in a way that
71 : //! is safe.
72 : //!
73 : //! The caller should be careful with deletion, though. They should not delete
74 : //! local files that have been scheduled for upload but not yet finished uploading.
75 : //! Otherwise the upload will fail. To wait for an upload to finish, use
76 : //! the 'wait_completion' function (more on that later.)
77 : //!
78 : //! All of this relies on the following invariants:
79 : //!
80 : //! - We rely on read-after write consistency in the remote storage.
81 : //! - Layer files are immutable
82 : //!
83 : //! NB: Pageserver assumes that it has exclusive write access to the tenant in remote
84 : //! storage. Different tenants can be attached to different pageservers, but if the
85 : //! same tenant is attached to two pageservers at the same time, they will overwrite
86 : //! each other's index file updates, and confusion will ensue. There's no interlock or
87 : //! mechanism to detect that in the pageserver, we rely on the control plane to ensure
88 : //! that that doesn't happen.
89 : //!
90 : //! ## Implementation Note
91 : //!
92 : //! The *actual* remote state lags behind the *desired* remote state while
93 : //! there are in-flight operations.
94 : //! We keep track of the desired remote state in [`UploadQueueInitialized::dirty`].
95 : //! It is initialized based on the [`IndexPart`] that was passed during init
96 : //! and updated with every `schedule_*` function call.
97 : //! All this is necessary necessary to compute the future [`IndexPart`]s
98 : //! when scheduling an operation while other operations that also affect the
99 : //! remote [`IndexPart`] are in flight.
100 : //!
101 : //! # Retries & Error Handling
102 : //!
103 : //! The client retries operations indefinitely, using exponential back-off.
104 : //! There is no way to force a retry, i.e., interrupt the back-off.
105 : //! This could be built easily.
106 : //!
107 : //! # Cancellation
108 : //!
109 : //! The operations execute as plain [`task_mgr`] tasks, scoped to
110 : //! the client's tenant and timeline.
111 : //! Dropping the client will drop queued operations but not executing operations.
112 : //! These will complete unless the `task_mgr` tasks are cancelled using `task_mgr`
113 : //! APIs, e.g., during pageserver shutdown, timeline delete, or tenant detach.
114 : //!
115 : //! # Completion
116 : //!
117 : //! Once an operation has completed, we update [`UploadQueueInitialized::clean`] immediately,
118 : //! and submit a request through the DeletionQueue to update
119 : //! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
120 : //! validated that our generation is not stale. It is this visible value
121 : //! that is advertized to safekeepers as a signal that that they can
122 : //! delete the WAL up to that LSN.
123 : //!
124 : //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
125 : //! for all pending operations to complete. It does not prevent more
126 : //! operations from getting scheduled.
127 : //!
128 : //! # Crash Consistency
129 : //!
130 : //! We do not persist the upload queue state.
131 : //! If we drop the client, or crash, all unfinished operations are lost.
132 : //!
133 : //! To recover, the following steps need to be taken:
134 : //! - Retrieve the current remote [`IndexPart`]. This gives us a
135 : //! consistent remote state, assuming the user scheduled the operations in
136 : //! the correct order.
137 : //! - Initiate upload queue with that [`IndexPart`].
138 : //! - Reschedule all lost operations by comparing the local filesystem state
139 : //! and remote state as per [`IndexPart`]. This is done in
140 : //! [`Tenant::timeline_init_and_sync`].
141 : //!
142 : //! Note that if we crash during file deletion between the index update
143 : //! that removes the file from the list of files, and deleting the remote file,
144 : //! the file is leaked in the remote storage. Similarly, if a new file is created
145 : //! and uploaded, but the pageserver dies permanently before updating the
146 : //! remote index file, the new file is leaked in remote storage. We accept and
147 : //! tolerate that for now.
148 : //! Note further that we cannot easily fix this by scheduling deletes for every
149 : //! file that is present only on the remote, because we cannot distinguish the
150 : //! following two cases:
151 : //! - (1) We had the file locally, deleted it locally, scheduled a remote delete,
152 : //! but crashed before it finished remotely.
153 : //! - (2) We never had the file locally because we haven't on-demand downloaded
154 : //! it yet.
155 : //!
156 : //! # Downloads
157 : //!
158 : //! In addition to the upload queue, [`RemoteTimelineClient`] has functions for
159 : //! downloading files from the remote storage. Downloads are performed immediately
160 : //! against the `RemoteStorage`, independently of the upload queue.
161 : //!
162 : //! When we attach a tenant, we perform the following steps:
163 : //! - create `Tenant` object in `TenantState::Attaching` state
164 : //! - List timelines that are present in remote storage, and for each:
165 : //! - download their remote [`IndexPart`]s
166 : //! - create `Timeline` struct and a `RemoteTimelineClient`
167 : //! - initialize the client's upload queue with its `IndexPart`
168 : //! - schedule uploads for layers that are only present locally.
169 : //! - After the above is done for each timeline, open the tenant for business by
170 : //! transitioning it from `TenantState::Attaching` to `TenantState::Active` state.
171 : //! This starts the timelines' WAL-receivers and the tenant's GC & Compaction loops.
172 : //!
173 : //! # Operating Without Remote Storage
174 : //!
175 : //! If no remote storage configuration is provided, the [`RemoteTimelineClient`] is
176 : //! not created and the uploads are skipped.
177 : //!
178 : //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
179 : //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
180 :
181 : pub(crate) mod download;
182 : pub mod index;
183 : pub(crate) mod upload;
184 :
185 : use anyhow::Context;
186 : use camino::Utf8Path;
187 : use chrono::{NaiveDateTime, Utc};
188 :
189 : pub(crate) use download::download_initdb_tar_zst;
190 : use pageserver_api::models::AuxFilePolicy;
191 : use pageserver_api::shard::{ShardIndex, TenantShardId};
192 : use scopeguard::ScopeGuard;
193 : use tokio_util::sync::CancellationToken;
194 : pub(crate) use upload::upload_initdb_dir;
195 : use utils::backoff::{
196 : self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
197 : };
198 : use utils::pausable_failpoint;
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 : pub(crate) 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 : ///
336 379 : pub fn new(
337 379 : remote_storage: GenericRemoteStorage,
338 379 : deletion_queue_client: DeletionQueueClient,
339 379 : conf: &'static PageServerConf,
340 379 : tenant_shard_id: TenantShardId,
341 379 : timeline_id: TimelineId,
342 379 : generation: Generation,
343 379 : ) -> RemoteTimelineClient {
344 379 : RemoteTimelineClient {
345 379 : conf,
346 379 : runtime: if cfg!(test) {
347 : // remote_timeline_client.rs tests rely on current-thread runtime
348 379 : tokio::runtime::Handle::current()
349 : } else {
350 0 : BACKGROUND_RUNTIME.handle().clone()
351 : },
352 379 : tenant_shard_id,
353 379 : timeline_id,
354 379 : generation,
355 379 : storage_impl: remote_storage,
356 379 : deletion_queue_client,
357 379 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
358 379 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
359 379 : &tenant_shard_id,
360 379 : &timeline_id,
361 379 : )),
362 379 : cancel: CancellationToken::new(),
363 379 : }
364 379 : }
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`.
382 373 : pub fn init_upload_queue_for_empty_remote(
383 373 : &self,
384 373 : local_metadata: &TimelineMetadata,
385 373 : ) -> anyhow::Result<()> {
386 373 : let mut upload_queue = self.upload_queue.lock().unwrap();
387 373 : upload_queue.initialize_empty_remote(local_metadata)?;
388 373 : self.update_remote_physical_size_gauge(None);
389 373 : info!("initialized upload queue as empty");
390 373 : Ok(())
391 373 : }
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 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
418 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
419 0 : match &mut *self.upload_queue.lock().unwrap() {
420 0 : UploadQueue::Uninitialized => None,
421 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
422 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
423 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
424 0 : .upload_queue_for_deletion
425 0 : .get_last_remote_consistent_lsn_projected(),
426 : }
427 0 : }
428 :
429 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
430 0 : match &mut *self.upload_queue.lock().unwrap() {
431 0 : UploadQueue::Uninitialized => None,
432 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
433 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
434 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
435 0 : q.upload_queue_for_deletion
436 0 : .get_last_remote_consistent_lsn_visible(),
437 0 : ),
438 : }
439 0 : }
440 :
441 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
442 : /// client is currently initialized.
443 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
444 0 : self.upload_queue
445 0 : .lock()
446 0 : .unwrap()
447 0 : .initialized_mut()
448 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
449 0 : .unwrap_or(false)
450 0 : }
451 :
452 1724 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
453 1724 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
454 1351 : current_remote_index_part
455 1351 : .layer_metadata
456 1351 : .values()
457 17314 : .map(|ilmd| ilmd.file_size)
458 1351 : .sum()
459 : } else {
460 373 : 0
461 : };
462 1724 : self.metrics.remote_physical_size_gauge.set(size);
463 1724 : }
464 :
465 2 : pub fn get_remote_physical_size(&self) -> u64 {
466 2 : self.metrics.remote_physical_size_gauge.get()
467 2 : }
468 :
469 : //
470 : // Download operations.
471 : //
472 : // These don't use the per-timeline queue. They do use the global semaphore in
473 : // S3Bucket, to limit the total number of concurrent operations, though.
474 : //
475 :
476 : /// Download index file
477 20 : pub async fn download_index_file(
478 20 : &self,
479 20 : cancel: &CancellationToken,
480 20 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
481 20 : let _unfinished_gauge_guard = self.metrics.call_begin(
482 20 : &RemoteOpFileKind::Index,
483 20 : &RemoteOpKind::Download,
484 20 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
485 20 : reason: "no need for a downloads gauge",
486 20 : },
487 20 : );
488 :
489 20 : let (index_part, _index_generation) = download::download_index_part(
490 20 : &self.storage_impl,
491 20 : &self.tenant_shard_id,
492 20 : &self.timeline_id,
493 20 : self.generation,
494 20 : cancel,
495 20 : )
496 20 : .measure_remote_op(
497 20 : RemoteOpFileKind::Index,
498 20 : RemoteOpKind::Download,
499 20 : Arc::clone(&self.metrics),
500 20 : )
501 89 : .await?;
502 :
503 20 : if index_part.deleted_at.is_some() {
504 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
505 : } else {
506 20 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
507 : }
508 20 : }
509 :
510 : /// Download a (layer) file from `path`, into local filesystem.
511 : ///
512 : /// 'layer_metadata' is the metadata from the remote index file.
513 : ///
514 : /// On success, returns the size of the downloaded file.
515 6 : pub async fn download_layer_file(
516 6 : &self,
517 6 : layer_file_name: &LayerName,
518 6 : layer_metadata: &LayerFileMetadata,
519 6 : local_path: &Utf8Path,
520 6 : cancel: &CancellationToken,
521 6 : ctx: &RequestContext,
522 6 : ) -> anyhow::Result<u64> {
523 6 : let downloaded_size = {
524 6 : let _unfinished_gauge_guard = self.metrics.call_begin(
525 6 : &RemoteOpFileKind::Layer,
526 6 : &RemoteOpKind::Download,
527 6 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
528 6 : reason: "no need for a downloads gauge",
529 6 : },
530 6 : );
531 6 : download::download_layer_file(
532 6 : self.conf,
533 6 : &self.storage_impl,
534 6 : self.tenant_shard_id,
535 6 : self.timeline_id,
536 6 : layer_file_name,
537 6 : layer_metadata,
538 6 : local_path,
539 6 : cancel,
540 6 : ctx,
541 6 : )
542 6 : .measure_remote_op(
543 6 : RemoteOpFileKind::Layer,
544 6 : RemoteOpKind::Download,
545 6 : Arc::clone(&self.metrics),
546 6 : )
547 81 : .await?
548 : };
549 :
550 6 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
551 6 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
552 6 :
553 6 : Ok(downloaded_size)
554 6 : }
555 :
556 : //
557 : // Upload operations.
558 : //
559 :
560 : /// Launch an index-file upload operation in the background, with
561 : /// fully updated metadata.
562 : ///
563 : /// This should only be used to upload initial metadata to remote storage.
564 : ///
565 : /// The upload will be added to the queue immediately, but it
566 : /// won't be performed until all previously scheduled layer file
567 : /// upload operations have completed successfully. This is to
568 : /// ensure that when the index file claims that layers X, Y and Z
569 : /// exist in remote storage, they really do. To wait for the upload
570 : /// to complete, use `wait_completion`.
571 : ///
572 : /// If there were any changes to the list of files, i.e. if any
573 : /// layer file uploads were scheduled, since the last index file
574 : /// upload, those will be included too.
575 226 : pub fn schedule_index_upload_for_full_metadata_update(
576 226 : self: &Arc<Self>,
577 226 : metadata: &TimelineMetadata,
578 226 : ) -> anyhow::Result<()> {
579 226 : let mut guard = self.upload_queue.lock().unwrap();
580 226 : let upload_queue = guard.initialized_mut()?;
581 :
582 : // As documented in the struct definition, it's ok for latest_metadata to be
583 : // ahead of what's _actually_ on the remote during index upload.
584 226 : upload_queue.dirty.metadata = metadata.clone();
585 226 :
586 226 : self.schedule_index_upload(upload_queue)?;
587 :
588 226 : Ok(())
589 226 : }
590 :
591 : /// Launch an index-file upload operation in the background, with only parts of the metadata
592 : /// updated.
593 : ///
594 : /// This is the regular way of updating metadata on layer flushes or Gc.
595 : ///
596 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
597 : /// `index_part.json`, while being more clear on what values update regularly.
598 1114 : pub(crate) fn schedule_index_upload_for_metadata_update(
599 1114 : self: &Arc<Self>,
600 1114 : update: &MetadataUpdate,
601 1114 : ) -> anyhow::Result<()> {
602 1114 : let mut guard = self.upload_queue.lock().unwrap();
603 1114 : let upload_queue = guard.initialized_mut()?;
604 :
605 1114 : upload_queue.dirty.metadata.apply(update);
606 1114 :
607 1114 : self.schedule_index_upload(upload_queue)?;
608 :
609 1114 : Ok(())
610 1114 : }
611 :
612 : /// Launch an index-file upload operation in the background, with only aux_file_policy flag updated.
613 16 : pub(crate) fn schedule_index_upload_for_aux_file_policy_update(
614 16 : self: &Arc<Self>,
615 16 : last_aux_file_policy: Option<AuxFilePolicy>,
616 16 : ) -> anyhow::Result<()> {
617 16 : let mut guard = self.upload_queue.lock().unwrap();
618 16 : let upload_queue = guard.initialized_mut()?;
619 16 : upload_queue.dirty.last_aux_file_policy = last_aux_file_policy;
620 16 : self.schedule_index_upload(upload_queue)?;
621 16 : Ok(())
622 16 : }
623 : ///
624 : /// Launch an index-file upload operation in the background, if necessary.
625 : ///
626 : /// Use this function to schedule the update of the index file after
627 : /// scheduling file uploads or deletions. If no file uploads or deletions
628 : /// have been scheduled since the last index file upload, this does
629 : /// nothing.
630 : ///
631 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
632 : /// the upload to the upload queue and returns quickly.
633 370 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> anyhow::Result<()> {
634 370 : let mut guard = self.upload_queue.lock().unwrap();
635 370 : let upload_queue = guard.initialized_mut()?;
636 :
637 370 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
638 14 : self.schedule_index_upload(upload_queue)?;
639 356 : }
640 :
641 370 : Ok(())
642 370 : }
643 :
644 : /// Launch an index-file upload operation in the background (internal function)
645 1406 : fn schedule_index_upload(
646 1406 : self: &Arc<Self>,
647 1406 : upload_queue: &mut UploadQueueInitialized,
648 1406 : ) -> anyhow::Result<()> {
649 1406 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
650 1406 : // fix up the duplicated field
651 1406 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
652 1406 :
653 1406 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
654 1406 : // look like a retryable error
655 1406 : let void = std::io::sink();
656 1406 : serde_json::to_writer(void, &upload_queue.dirty).context("serialize index_part.json")?;
657 :
658 1406 : let index_part = &upload_queue.dirty;
659 1406 :
660 1406 : info!(
661 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
662 0 : index_part.layer_metadata.len(),
663 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
664 : );
665 :
666 1406 : let op = UploadOp::UploadMetadata {
667 1406 : uploaded: Box::new(index_part.clone()),
668 1406 : };
669 1406 : self.metric_begin(&op);
670 1406 : upload_queue.queued_operations.push_back(op);
671 1406 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
672 1406 :
673 1406 : // Launch the task immediately, if possible
674 1406 : self.launch_queued_tasks(upload_queue);
675 1406 : Ok(())
676 1406 : }
677 :
678 0 : pub(crate) async fn schedule_reparenting_and_wait(
679 0 : self: &Arc<Self>,
680 0 : new_parent: &TimelineId,
681 0 : ) -> anyhow::Result<()> {
682 : // FIXME: because of how Timeline::schedule_uploads works when called from layer flushing
683 : // and reads the in-memory part we cannot do the detaching like this
684 0 : let receiver = {
685 0 : let mut guard = self.upload_queue.lock().unwrap();
686 0 : let upload_queue = guard.initialized_mut()?;
687 :
688 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
689 0 : return Err(anyhow::anyhow!(
690 0 : "cannot reparent without a current ancestor"
691 0 : ));
692 : };
693 :
694 0 : upload_queue.dirty.metadata.reparent(new_parent);
695 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
696 0 :
697 0 : self.schedule_index_upload(upload_queue)?;
698 :
699 0 : self.schedule_barrier0(upload_queue)
700 0 : };
701 0 :
702 0 : Self::wait_completion0(receiver).await
703 0 : }
704 :
705 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
706 : /// detaching from ancestor and waits for it to complete.
707 : ///
708 : /// This is used with `Timeline::detach_ancestor` functionality.
709 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
710 0 : self: &Arc<Self>,
711 0 : layers: &[Layer],
712 0 : adopted: (TimelineId, Lsn),
713 0 : ) -> anyhow::Result<()> {
714 0 : let barrier = {
715 0 : let mut guard = self.upload_queue.lock().unwrap();
716 0 : let upload_queue = guard.initialized_mut()?;
717 :
718 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
719 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
720 :
721 0 : for layer in layers {
722 0 : upload_queue
723 0 : .dirty
724 0 : .layer_metadata
725 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
726 0 : }
727 :
728 0 : self.schedule_index_upload(upload_queue)?;
729 :
730 0 : let barrier = self.schedule_barrier0(upload_queue);
731 0 : self.launch_queued_tasks(upload_queue);
732 0 : barrier
733 0 : };
734 0 :
735 0 : Self::wait_completion0(barrier).await
736 0 : }
737 :
738 : /// Launch an upload operation in the background; the file is added to be included in next
739 : /// `index_part.json` upload.
740 1156 : pub(crate) fn schedule_layer_file_upload(
741 1156 : self: &Arc<Self>,
742 1156 : layer: ResidentLayer,
743 1156 : ) -> anyhow::Result<()> {
744 1156 : let mut guard = self.upload_queue.lock().unwrap();
745 1156 : let upload_queue = guard.initialized_mut()?;
746 :
747 1156 : self.schedule_layer_file_upload0(upload_queue, layer);
748 1156 : self.launch_queued_tasks(upload_queue);
749 1156 : Ok(())
750 1156 : }
751 :
752 1464 : fn schedule_layer_file_upload0(
753 1464 : self: &Arc<Self>,
754 1464 : upload_queue: &mut UploadQueueInitialized,
755 1464 : layer: ResidentLayer,
756 1464 : ) {
757 1464 : let metadata = layer.metadata();
758 1464 :
759 1464 : upload_queue
760 1464 : .dirty
761 1464 : .layer_metadata
762 1464 : .insert(layer.layer_desc().layer_name(), metadata.clone());
763 1464 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
764 1464 :
765 1464 : info!(
766 : gen=?metadata.generation,
767 : shard=?metadata.shard,
768 0 : "scheduled layer file upload {layer}",
769 : );
770 :
771 1464 : let op = UploadOp::UploadLayer(layer, metadata);
772 1464 : self.metric_begin(&op);
773 1464 : upload_queue.queued_operations.push_back(op);
774 1464 : }
775 :
776 : /// Launch a delete operation in the background.
777 : ///
778 : /// The operation does not modify local filesystem state.
779 : ///
780 : /// Note: This schedules an index file upload before the deletions. The
781 : /// deletion won't actually be performed, until all previously scheduled
782 : /// upload operations, and the index file upload, have completed
783 : /// successfully.
784 8 : pub fn schedule_layer_file_deletion(
785 8 : self: &Arc<Self>,
786 8 : names: &[LayerName],
787 8 : ) -> anyhow::Result<()> {
788 8 : let mut guard = self.upload_queue.lock().unwrap();
789 8 : let upload_queue = guard.initialized_mut()?;
790 :
791 8 : let with_metadata = self
792 8 : .schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned())?;
793 :
794 8 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
795 8 :
796 8 : // Launch the tasks immediately, if possible
797 8 : self.launch_queued_tasks(upload_queue);
798 8 : Ok(())
799 8 : }
800 :
801 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
802 : /// layer files, leaving them dangling.
803 : ///
804 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
805 : /// is invoked on them.
806 8 : pub(crate) fn schedule_gc_update(self: &Arc<Self>, gc_layers: &[Layer]) -> anyhow::Result<()> {
807 8 : let mut guard = self.upload_queue.lock().unwrap();
808 8 : let upload_queue = guard.initialized_mut()?;
809 :
810 : // just forget the return value; after uploading the next index_part.json, we can consider
811 : // the layer files as "dangling". this is fine, at worst case we create work for the
812 : // scrubber.
813 :
814 10 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
815 8 :
816 8 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
817 :
818 8 : self.launch_queued_tasks(upload_queue);
819 8 :
820 8 : Ok(())
821 8 : }
822 :
823 : /// Update the remote index file, removing the to-be-deleted files from the index,
824 : /// allowing scheduling of actual deletions later.
825 44 : fn schedule_unlinking_of_layers_from_index_part0<I>(
826 44 : self: &Arc<Self>,
827 44 : upload_queue: &mut UploadQueueInitialized,
828 44 : names: I,
829 44 : ) -> anyhow::Result<Vec<(LayerName, LayerFileMetadata)>>
830 44 : where
831 44 : I: IntoIterator<Item = LayerName>,
832 44 : {
833 44 : // Decorate our list of names with each name's metadata, dropping
834 44 : // names that are unexpectedly missing from our metadata. This metadata
835 44 : // is later used when physically deleting layers, to construct key paths.
836 44 : let with_metadata: Vec<_> = names
837 44 : .into_iter()
838 414 : .filter_map(|name| {
839 414 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
840 :
841 414 : if let Some(meta) = meta {
842 412 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
843 412 : Some((name, meta))
844 : } else {
845 : // This can only happen if we forgot to to schedule the file upload
846 : // before scheduling the delete. Log it because it is a rare/strange
847 : // situation, and in case something is misbehaving, we'd like to know which
848 : // layers experienced this.
849 2 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
850 2 : None
851 : }
852 414 : })
853 44 : .collect();
854 :
855 : #[cfg(feature = "testing")]
856 456 : for (name, metadata) in &with_metadata {
857 412 : let gen = metadata.generation;
858 412 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
859 0 : if unexpected == gen {
860 0 : tracing::error!("{name} was unlinked twice with same generation");
861 : } else {
862 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
863 : }
864 412 : }
865 : }
866 :
867 : // after unlinking files from the upload_queue.latest_files we must always schedule an
868 : // index_part update, because that needs to be uploaded before we can actually delete the
869 : // files.
870 44 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
871 36 : self.schedule_index_upload(upload_queue)?;
872 8 : }
873 :
874 44 : Ok(with_metadata)
875 44 : }
876 :
877 : /// Schedules deletion for layer files which have previously been unlinked from the
878 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
879 425 : pub(crate) fn schedule_deletion_of_unlinked(
880 425 : self: &Arc<Self>,
881 425 : layers: Vec<(LayerName, LayerFileMetadata)>,
882 425 : ) -> anyhow::Result<()> {
883 425 : let mut guard = self.upload_queue.lock().unwrap();
884 425 : let upload_queue = guard.initialized_mut()?;
885 :
886 425 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
887 425 : self.launch_queued_tasks(upload_queue);
888 425 : Ok(())
889 425 : }
890 :
891 433 : fn schedule_deletion_of_unlinked0(
892 433 : self: &Arc<Self>,
893 433 : upload_queue: &mut UploadQueueInitialized,
894 433 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
895 433 : ) {
896 433 : // Filter out any layers which were not created by this tenant shard. These are
897 433 : // layers that originate from some ancestor shard after a split, and may still
898 433 : // be referenced by other shards. We are free to delete them locally and remove
899 433 : // them from our index (and would have already done so when we reach this point
900 433 : // in the code), but we may not delete them remotely.
901 433 : with_metadata.retain(|(name, meta)| {
902 427 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
903 427 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
904 427 : if !retain {
905 0 : tracing::debug!(
906 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
907 : meta.shard
908 : );
909 427 : }
910 427 : retain
911 433 : });
912 :
913 860 : for (name, meta) in &with_metadata {
914 427 : info!(
915 0 : "scheduling deletion of layer {}{} (shard {})",
916 0 : name,
917 0 : meta.generation.get_suffix(),
918 : meta.shard
919 : );
920 : }
921 :
922 : #[cfg(feature = "testing")]
923 860 : for (name, meta) in &with_metadata {
924 427 : let gen = meta.generation;
925 427 : match upload_queue.dangling_files.remove(name) {
926 411 : Some(same) if same == gen => { /* expected */ }
927 0 : Some(other) => {
928 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
929 : }
930 : None => {
931 16 : tracing::error!("{name} was unlinked but was not dangling");
932 : }
933 : }
934 : }
935 :
936 : // schedule the actual deletions
937 433 : if with_metadata.is_empty() {
938 : // avoid scheduling the op & bumping the metric
939 6 : return;
940 427 : }
941 427 : let op = UploadOp::Delete(Delete {
942 427 : layers: with_metadata,
943 427 : });
944 427 : self.metric_begin(&op);
945 427 : upload_queue.queued_operations.push_back(op);
946 433 : }
947 :
948 : /// Schedules a compaction update to the remote `index_part.json`.
949 : ///
950 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
951 28 : pub(crate) fn schedule_compaction_update(
952 28 : self: &Arc<Self>,
953 28 : compacted_from: &[Layer],
954 28 : compacted_to: &[ResidentLayer],
955 28 : ) -> anyhow::Result<()> {
956 28 : let mut guard = self.upload_queue.lock().unwrap();
957 28 : let upload_queue = guard.initialized_mut()?;
958 :
959 336 : for layer in compacted_to {
960 308 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
961 308 : }
962 :
963 402 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
964 28 :
965 28 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names)?;
966 28 : self.launch_queued_tasks(upload_queue);
967 28 :
968 28 : Ok(())
969 28 : }
970 :
971 : /// Wait for all previously scheduled uploads/deletions to complete
972 148 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
973 148 : let receiver = {
974 148 : let mut guard = self.upload_queue.lock().unwrap();
975 148 : let upload_queue = guard.initialized_mut()?;
976 148 : self.schedule_barrier0(upload_queue)
977 148 : };
978 148 :
979 148 : Self::wait_completion0(receiver).await
980 148 : }
981 :
982 148 : async fn wait_completion0(
983 148 : mut receiver: tokio::sync::watch::Receiver<()>,
984 148 : ) -> anyhow::Result<()> {
985 148 : if receiver.changed().await.is_err() {
986 0 : anyhow::bail!("wait_completion aborted because upload queue was stopped");
987 148 : }
988 148 :
989 148 : Ok(())
990 148 : }
991 :
992 6 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
993 6 : let mut guard = self.upload_queue.lock().unwrap();
994 6 : let upload_queue = guard.initialized_mut()?;
995 6 : self.schedule_barrier0(upload_queue);
996 6 : Ok(())
997 6 : }
998 :
999 154 : fn schedule_barrier0(
1000 154 : self: &Arc<Self>,
1001 154 : upload_queue: &mut UploadQueueInitialized,
1002 154 : ) -> tokio::sync::watch::Receiver<()> {
1003 154 : let (sender, receiver) = tokio::sync::watch::channel(());
1004 154 : let barrier_op = UploadOp::Barrier(sender);
1005 154 :
1006 154 : upload_queue.queued_operations.push_back(barrier_op);
1007 154 : // Don't count this kind of operation!
1008 154 :
1009 154 : // Launch the task immediately, if possible
1010 154 : self.launch_queued_tasks(upload_queue);
1011 154 :
1012 154 : receiver
1013 154 : }
1014 :
1015 : /// Wait for all previously scheduled operations to complete, and then stop.
1016 : ///
1017 : /// Not cancellation safe
1018 6 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1019 6 : // On cancellation the queue is left in ackward state of refusing new operations but
1020 6 : // proper stop is yet to be called. On cancel the original or some later task must call
1021 6 : // `stop` or `shutdown`.
1022 6 : let sg = scopeguard::guard((), |_| {
1023 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1024 6 : });
1025 :
1026 6 : let fut = {
1027 6 : let mut guard = self.upload_queue.lock().unwrap();
1028 6 : let upload_queue = match &mut *guard {
1029 0 : UploadQueue::Stopped(_) => return,
1030 : UploadQueue::Uninitialized => {
1031 : // transition into Stopped state
1032 0 : self.stop_impl(&mut guard);
1033 0 : return;
1034 : }
1035 6 : UploadQueue::Initialized(ref mut init) => init,
1036 6 : };
1037 6 :
1038 6 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1039 6 : // just don't add more of these as they would never complete.
1040 6 : //
1041 6 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1042 6 : // in every place we would not have to jump through this hoop, and this method could be
1043 6 : // made cancellable.
1044 6 : if !upload_queue.shutting_down {
1045 6 : upload_queue.shutting_down = true;
1046 6 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1047 6 : // this operation is not counted similar to Barrier
1048 6 :
1049 6 : self.launch_queued_tasks(upload_queue);
1050 6 : }
1051 :
1052 6 : upload_queue.shutdown_ready.clone().acquire_owned()
1053 : };
1054 :
1055 6 : let res = fut.await;
1056 :
1057 6 : scopeguard::ScopeGuard::into_inner(sg);
1058 6 :
1059 6 : match res {
1060 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1061 6 : Err(_closed) => {
1062 6 : // expected
1063 6 : }
1064 6 : }
1065 6 :
1066 6 : self.stop();
1067 6 : }
1068 :
1069 : /// Set the deleted_at field in the remote index file.
1070 : ///
1071 : /// This fails if the upload queue has not been `stop()`ed.
1072 : ///
1073 : /// The caller is responsible for calling `stop()` AND for waiting
1074 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1075 : /// Check method [`RemoteTimelineClient::stop`] for details.
1076 0 : #[instrument(skip_all)]
1077 : pub(crate) async fn persist_index_part_with_deleted_flag(
1078 : self: &Arc<Self>,
1079 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1080 : let index_part_with_deleted_at = {
1081 : let mut locked = self.upload_queue.lock().unwrap();
1082 :
1083 : // We must be in stopped state because otherwise
1084 : // we can have inprogress index part upload that can overwrite the file
1085 : // with missing is_deleted flag that we going to set below
1086 : let stopped = locked.stopped_mut()?;
1087 :
1088 : match stopped.deleted_at {
1089 : SetDeletedFlagProgress::NotRunning => (), // proceed
1090 : SetDeletedFlagProgress::InProgress(at) => {
1091 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1092 : }
1093 : SetDeletedFlagProgress::Successful(at) => {
1094 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1095 : }
1096 : };
1097 : let deleted_at = Utc::now().naive_utc();
1098 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1099 :
1100 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1101 : index_part.deleted_at = Some(deleted_at);
1102 : index_part
1103 : };
1104 :
1105 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1106 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1107 0 : let stopped = locked
1108 0 : .stopped_mut()
1109 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1110 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1111 0 : });
1112 :
1113 : pausable_failpoint!("persist_deleted_index_part");
1114 :
1115 : backoff::retry(
1116 0 : || {
1117 0 : upload::upload_index_part(
1118 0 : &self.storage_impl,
1119 0 : &self.tenant_shard_id,
1120 0 : &self.timeline_id,
1121 0 : self.generation,
1122 0 : &index_part_with_deleted_at,
1123 0 : &self.cancel,
1124 0 : )
1125 0 : },
1126 0 : |_e| false,
1127 : 1,
1128 : // have just a couple of attempts
1129 : // when executed as part of timeline deletion this happens in context of api call
1130 : // when executed as part of tenant deletion this happens in the background
1131 : 2,
1132 : "persist_index_part_with_deleted_flag",
1133 : &self.cancel,
1134 : )
1135 : .await
1136 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1137 0 : .and_then(|x| x)?;
1138 :
1139 : // all good, disarm the guard and mark as success
1140 : ScopeGuard::into_inner(undo_deleted_at);
1141 : {
1142 : let mut locked = self.upload_queue.lock().unwrap();
1143 :
1144 : let stopped = locked
1145 : .stopped_mut()
1146 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1147 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1148 : index_part_with_deleted_at
1149 : .deleted_at
1150 : .expect("we set it above"),
1151 : );
1152 : }
1153 :
1154 : Ok(())
1155 : }
1156 :
1157 0 : pub(crate) fn is_deleting(&self) -> bool {
1158 0 : let mut locked = self.upload_queue.lock().unwrap();
1159 0 : locked.stopped_mut().is_ok()
1160 0 : }
1161 :
1162 0 : pub(crate) async fn preserve_initdb_archive(
1163 0 : self: &Arc<Self>,
1164 0 : tenant_id: &TenantId,
1165 0 : timeline_id: &TimelineId,
1166 0 : cancel: &CancellationToken,
1167 0 : ) -> anyhow::Result<()> {
1168 0 : backoff::retry(
1169 0 : || async {
1170 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1171 0 : .await
1172 0 : },
1173 0 : TimeoutOrCancel::caused_by_cancel,
1174 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1175 0 : FAILED_REMOTE_OP_RETRIES,
1176 0 : "preserve_initdb_tar_zst",
1177 0 : &cancel.clone(),
1178 0 : )
1179 0 : .await
1180 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1181 0 : .and_then(|x| x)
1182 0 : .context("backing up initdb archive")?;
1183 0 : Ok(())
1184 0 : }
1185 :
1186 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1187 : ///
1188 : /// This is not normally needed.
1189 0 : pub(crate) async fn upload_layer_file(
1190 0 : self: &Arc<Self>,
1191 0 : uploaded: &ResidentLayer,
1192 0 : cancel: &CancellationToken,
1193 0 : ) -> anyhow::Result<()> {
1194 0 : let remote_path = remote_layer_path(
1195 0 : &self.tenant_shard_id.tenant_id,
1196 0 : &self.timeline_id,
1197 0 : self.tenant_shard_id.to_index(),
1198 0 : &uploaded.layer_desc().layer_name(),
1199 0 : uploaded.metadata().generation,
1200 0 : );
1201 0 :
1202 0 : backoff::retry(
1203 0 : || async {
1204 0 : upload::upload_timeline_layer(
1205 0 : &self.storage_impl,
1206 0 : uploaded.local_path(),
1207 0 : &remote_path,
1208 0 : uploaded.metadata().file_size,
1209 0 : cancel,
1210 0 : )
1211 0 : .await
1212 0 : },
1213 0 : TimeoutOrCancel::caused_by_cancel,
1214 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1215 0 : FAILED_REMOTE_OP_RETRIES,
1216 0 : "upload a layer without adding it to latest files",
1217 0 : cancel,
1218 0 : )
1219 0 : .await
1220 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1221 0 : .and_then(|x| x)
1222 0 : .context("upload a layer without adding it to latest files")
1223 0 : }
1224 :
1225 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1226 : /// not added to be part of a future `index_part.json` upload.
1227 0 : pub(crate) async fn copy_timeline_layer(
1228 0 : self: &Arc<Self>,
1229 0 : adopted: &Layer,
1230 0 : adopted_as: &Layer,
1231 0 : cancel: &CancellationToken,
1232 0 : ) -> anyhow::Result<()> {
1233 0 : let source_remote_path = remote_layer_path(
1234 0 : &self.tenant_shard_id.tenant_id,
1235 0 : &adopted
1236 0 : .get_timeline_id()
1237 0 : .expect("Source timeline should be alive"),
1238 0 : self.tenant_shard_id.to_index(),
1239 0 : &adopted.layer_desc().layer_name(),
1240 0 : adopted.metadata().generation,
1241 0 : );
1242 0 :
1243 0 : let target_remote_path = remote_layer_path(
1244 0 : &self.tenant_shard_id.tenant_id,
1245 0 : &self.timeline_id,
1246 0 : self.tenant_shard_id.to_index(),
1247 0 : &adopted_as.layer_desc().layer_name(),
1248 0 : adopted_as.metadata().generation,
1249 0 : );
1250 0 :
1251 0 : backoff::retry(
1252 0 : || async {
1253 0 : upload::copy_timeline_layer(
1254 0 : &self.storage_impl,
1255 0 : &source_remote_path,
1256 0 : &target_remote_path,
1257 0 : cancel,
1258 0 : )
1259 0 : .await
1260 0 : },
1261 0 : TimeoutOrCancel::caused_by_cancel,
1262 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1263 0 : FAILED_REMOTE_OP_RETRIES,
1264 0 : "copy timeline layer",
1265 0 : cancel,
1266 0 : )
1267 0 : .await
1268 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1269 0 : .and_then(|x| x)
1270 0 : .context("remote copy timeline layer")
1271 0 : }
1272 :
1273 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1274 0 : match tokio::time::timeout(
1275 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1276 0 : self.deletion_queue_client.flush_immediate(),
1277 0 : )
1278 0 : .await
1279 : {
1280 0 : Ok(result) => result,
1281 0 : Err(_timeout) => {
1282 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1283 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1284 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1285 0 : tracing::warn!(
1286 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1287 : );
1288 0 : Ok(())
1289 : }
1290 : }
1291 0 : }
1292 :
1293 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1294 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1295 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1296 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> anyhow::Result<()> {
1297 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1298 :
1299 0 : let layers: Vec<RemotePath> = {
1300 0 : let mut locked = self.upload_queue.lock().unwrap();
1301 0 : let stopped = locked.stopped_mut()?;
1302 :
1303 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1304 0 : anyhow::bail!("deleted_at is not set")
1305 0 : }
1306 0 :
1307 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1308 :
1309 0 : stopped
1310 0 : .upload_queue_for_deletion
1311 0 : .dirty
1312 0 : .layer_metadata
1313 0 : .drain()
1314 0 : .map(|(file_name, meta)| {
1315 0 : remote_layer_path(
1316 0 : &self.tenant_shard_id.tenant_id,
1317 0 : &self.timeline_id,
1318 0 : meta.shard,
1319 0 : &file_name,
1320 0 : meta.generation,
1321 0 : )
1322 0 : })
1323 0 : .collect()
1324 0 : };
1325 0 :
1326 0 : let layer_deletion_count = layers.len();
1327 0 : self.deletion_queue_client.push_immediate(layers).await?;
1328 :
1329 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1330 : // inexistant objects are not considered errors.
1331 0 : let initdb_path =
1332 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1333 0 : self.deletion_queue_client
1334 0 : .push_immediate(vec![initdb_path])
1335 0 : .await?;
1336 :
1337 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1338 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1339 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1340 0 :
1341 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1342 0 : // taking the burden of listing all the layers that we already know we should delete.
1343 0 : self.flush_deletion_queue().await?;
1344 :
1345 0 : let cancel = shutdown_token();
1346 :
1347 0 : let remaining = download_retry(
1348 0 : || async {
1349 0 : self.storage_impl
1350 0 : .list(
1351 0 : Some(&timeline_storage_path),
1352 0 : ListingMode::NoDelimiter,
1353 0 : None,
1354 0 : &cancel,
1355 0 : )
1356 0 : .await
1357 0 : },
1358 0 : "list remaining files",
1359 0 : &cancel,
1360 0 : )
1361 0 : .await
1362 0 : .context("list files remaining files")?
1363 : .keys;
1364 :
1365 : // We will delete the current index_part object last, since it acts as a deletion
1366 : // marker via its deleted_at attribute
1367 0 : let latest_index = remaining
1368 0 : .iter()
1369 0 : .filter(|p| {
1370 0 : p.object_name()
1371 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1372 0 : .unwrap_or(false)
1373 0 : })
1374 0 : .filter_map(|path| parse_remote_index_path(path.clone()).map(|gen| (path, gen)))
1375 0 : .max_by_key(|i| i.1)
1376 0 : .map(|i| i.0.clone())
1377 0 : .unwrap_or(
1378 0 : // No generation-suffixed indices, assume we are dealing with
1379 0 : // a legacy index.
1380 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1381 0 : );
1382 0 :
1383 0 : let remaining_layers: Vec<RemotePath> = remaining
1384 0 : .into_iter()
1385 0 : .filter(|p| {
1386 0 : if p == &latest_index {
1387 0 : return false;
1388 0 : }
1389 0 : if p.object_name() == Some(INITDB_PRESERVED_PATH) {
1390 0 : return false;
1391 0 : }
1392 0 : true
1393 0 : })
1394 0 : .inspect(|path| {
1395 0 : if let Some(name) = path.object_name() {
1396 0 : info!(%name, "deleting a file not referenced from index_part.json");
1397 : } else {
1398 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1399 : }
1400 0 : })
1401 0 : .collect();
1402 0 :
1403 0 : let not_referenced_count = remaining_layers.len();
1404 0 : if !remaining_layers.is_empty() {
1405 0 : self.deletion_queue_client
1406 0 : .push_immediate(remaining_layers)
1407 0 : .await?;
1408 0 : }
1409 :
1410 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1411 0 : Err(anyhow::anyhow!(
1412 0 : "failpoint: timeline-delete-before-index-delete"
1413 0 : ))?
1414 0 : });
1415 :
1416 0 : debug!("enqueuing index part deletion");
1417 0 : self.deletion_queue_client
1418 0 : .push_immediate([latest_index].to_vec())
1419 0 : .await?;
1420 :
1421 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1422 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1423 0 : self.flush_deletion_queue().await?;
1424 :
1425 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1426 0 : Err(anyhow::anyhow!(
1427 0 : "failpoint: timeline-delete-after-index-delete"
1428 0 : ))?
1429 0 : });
1430 :
1431 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1432 :
1433 0 : Ok(())
1434 0 : }
1435 :
1436 : ///
1437 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1438 : /// the ordering constraints.
1439 : ///
1440 : /// The caller needs to already hold the `upload_queue` lock.
1441 5922 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1442 9126 : while let Some(next_op) = upload_queue.queued_operations.front() {
1443 : // Can we run this task now?
1444 5280 : let can_run_now = match next_op {
1445 : UploadOp::UploadLayer(..) => {
1446 : // Can always be scheduled.
1447 1461 : true
1448 : }
1449 : UploadOp::UploadMetadata { .. } => {
1450 : // These can only be performed after all the preceding operations
1451 : // have finished.
1452 3268 : upload_queue.inprogress_tasks.is_empty()
1453 : }
1454 : UploadOp::Delete(_) => {
1455 : // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
1456 245 : upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
1457 : }
1458 :
1459 : UploadOp::Barrier(_) | UploadOp::Shutdown => {
1460 306 : upload_queue.inprogress_tasks.is_empty()
1461 : }
1462 : };
1463 :
1464 : // If we cannot launch this task, don't look any further.
1465 : //
1466 : // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
1467 : // them now, but we don't try to do that currently. For example, if the frontmost task
1468 : // is an index-file upload that cannot proceed until preceding uploads have finished, we
1469 : // could still start layer uploads that were scheduled later.
1470 5280 : if !can_run_now {
1471 2070 : break;
1472 3210 : }
1473 3210 :
1474 3210 : if let UploadOp::Shutdown = next_op {
1475 : // leave the op in the queue but do not start more tasks; it will be dropped when
1476 : // the stop is called.
1477 6 : upload_queue.shutdown_ready.close();
1478 6 : break;
1479 3204 : }
1480 3204 :
1481 3204 : // We can launch this task. Remove it from the queue first.
1482 3204 : let next_op = upload_queue.queued_operations.pop_front().unwrap();
1483 3204 :
1484 3204 : debug!("starting op: {}", next_op);
1485 :
1486 : // Update the counters
1487 3204 : match next_op {
1488 1461 : UploadOp::UploadLayer(_, _) => {
1489 1461 : upload_queue.num_inprogress_layer_uploads += 1;
1490 1461 : }
1491 1375 : UploadOp::UploadMetadata { .. } => {
1492 1375 : upload_queue.num_inprogress_metadata_uploads += 1;
1493 1375 : }
1494 214 : UploadOp::Delete(_) => {
1495 214 : upload_queue.num_inprogress_deletions += 1;
1496 214 : }
1497 154 : UploadOp::Barrier(sender) => {
1498 154 : sender.send_replace(());
1499 154 : continue;
1500 : }
1501 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1502 : };
1503 :
1504 : // Assign unique ID to this task
1505 3050 : upload_queue.task_counter += 1;
1506 3050 : let upload_task_id = upload_queue.task_counter;
1507 3050 :
1508 3050 : // Add it to the in-progress map
1509 3050 : let task = Arc::new(UploadTask {
1510 3050 : task_id: upload_task_id,
1511 3050 : op: next_op,
1512 3050 : retries: AtomicU32::new(0),
1513 3050 : });
1514 3050 : upload_queue
1515 3050 : .inprogress_tasks
1516 3050 : .insert(task.task_id, Arc::clone(&task));
1517 3050 :
1518 3050 : // Spawn task to perform the task
1519 3050 : let self_rc = Arc::clone(self);
1520 3050 : let tenant_shard_id = self.tenant_shard_id;
1521 3050 : let timeline_id = self.timeline_id;
1522 3050 : task_mgr::spawn(
1523 3050 : &self.runtime,
1524 3050 : TaskKind::RemoteUploadTask,
1525 3050 : Some(self.tenant_shard_id),
1526 3050 : Some(self.timeline_id),
1527 3050 : "remote upload",
1528 : false,
1529 2944 : async move {
1530 36331 : self_rc.perform_upload_task(task).await;
1531 2731 : Ok(())
1532 2731 : }
1533 3050 : .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)),
1534 : );
1535 :
1536 : // Loop back to process next task
1537 : }
1538 5922 : }
1539 :
1540 : ///
1541 : /// Perform an upload task.
1542 : ///
1543 : /// The task is in the `inprogress_tasks` list. This function will try to
1544 : /// execute it, retrying forever. On successful completion, the task is
1545 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1546 : /// queue that were waiting by the completion are launched.
1547 : ///
1548 : /// The task can be shut down, however. That leads to stopping the whole
1549 : /// queue.
1550 : ///
1551 2944 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1552 2944 : let cancel = shutdown_token();
1553 : // Loop to retry until it completes.
1554 2944 : loop {
1555 2944 : // If we're requested to shut down, close up shop and exit.
1556 2944 : //
1557 2944 : // Note: We only check for the shutdown requests between retries, so
1558 2944 : // if a shutdown request arrives while we're busy uploading, in the
1559 2944 : // upload::upload:*() call below, we will wait not exit until it has
1560 2944 : // finished. We probably could cancel the upload by simply dropping
1561 2944 : // the Future, but we're not 100% sure if the remote storage library
1562 2944 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1563 2944 : // upload finishes or times out soon enough.
1564 2944 : if cancel.is_cancelled() {
1565 0 : info!("upload task cancelled by shutdown request");
1566 0 : self.stop();
1567 0 : return;
1568 2944 : }
1569 :
1570 2944 : let upload_result: anyhow::Result<()> = match &task.op {
1571 1376 : UploadOp::UploadLayer(ref layer, ref layer_metadata) => {
1572 1376 : let local_path = layer.local_path();
1573 1376 :
1574 1376 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
1575 1376 : // the metadata in the upload should always match our current generation.
1576 1376 : assert_eq!(layer_metadata.generation, self.generation);
1577 :
1578 1376 : let remote_path = remote_layer_path(
1579 1376 : &self.tenant_shard_id.tenant_id,
1580 1376 : &self.timeline_id,
1581 1376 : layer_metadata.shard,
1582 1376 : &layer.layer_desc().layer_name(),
1583 1376 : layer_metadata.generation,
1584 1376 : );
1585 1376 :
1586 1376 : upload::upload_timeline_layer(
1587 1376 : &self.storage_impl,
1588 1376 : local_path,
1589 1376 : &remote_path,
1590 1376 : layer_metadata.file_size,
1591 1376 : &self.cancel,
1592 1376 : )
1593 1376 : .measure_remote_op(
1594 1376 : RemoteOpFileKind::Layer,
1595 1376 : RemoteOpKind::Upload,
1596 1376 : Arc::clone(&self.metrics),
1597 1376 : )
1598 30995 : .await
1599 : }
1600 1359 : UploadOp::UploadMetadata { ref uploaded } => {
1601 1359 : let res = upload::upload_index_part(
1602 1359 : &self.storage_impl,
1603 1359 : &self.tenant_shard_id,
1604 1359 : &self.timeline_id,
1605 1359 : self.generation,
1606 1359 : uploaded,
1607 1359 : &self.cancel,
1608 1359 : )
1609 1359 : .measure_remote_op(
1610 1359 : RemoteOpFileKind::Index,
1611 1359 : RemoteOpKind::Upload,
1612 1359 : Arc::clone(&self.metrics),
1613 1359 : )
1614 5141 : .await;
1615 1345 : if res.is_ok() {
1616 1345 : self.update_remote_physical_size_gauge(Some(uploaded));
1617 1345 : let mention_having_future_layers = if cfg!(feature = "testing") {
1618 1345 : uploaded
1619 1345 : .layer_metadata
1620 1345 : .keys()
1621 17292 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
1622 : } else {
1623 0 : false
1624 : };
1625 1345 : if mention_having_future_layers {
1626 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
1627 4 : tracing::info!(
1628 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
1629 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
1630 : );
1631 1341 : }
1632 0 : }
1633 1345 : res
1634 : }
1635 209 : UploadOp::Delete(delete) => {
1636 : pausable_failpoint!("before-delete-layer-pausable");
1637 207 : self.deletion_queue_client
1638 207 : .push_layers(
1639 207 : self.tenant_shard_id,
1640 207 : self.timeline_id,
1641 207 : self.generation,
1642 207 : delete.layers.clone(),
1643 207 : )
1644 0 : .await
1645 207 : .map_err(|e| anyhow::anyhow!(e))
1646 : }
1647 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
1648 : // unreachable. Barrier operations are handled synchronously in
1649 : // launch_queued_tasks
1650 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
1651 0 : break;
1652 : }
1653 : };
1654 :
1655 0 : match upload_result {
1656 : Ok(()) => {
1657 2731 : break;
1658 : }
1659 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
1660 0 : // loop around to do the proper stopping
1661 0 : continue;
1662 : }
1663 0 : Err(e) => {
1664 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
1665 0 :
1666 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
1667 0 : // or other external reasons. Such issues are relatively regular, so log them
1668 0 : // at info level at first, and only WARN if the operation fails repeatedly.
1669 0 : //
1670 0 : // (See similar logic for downloads in `download::download_retry`)
1671 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
1672 0 : info!(
1673 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
1674 0 : task.op, retries, e
1675 : );
1676 : } else {
1677 0 : warn!(
1678 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
1679 0 : task.op, retries, e
1680 : );
1681 : }
1682 :
1683 : // sleep until it's time to retry, or we're cancelled
1684 0 : exponential_backoff(
1685 0 : retries,
1686 0 : DEFAULT_BASE_BACKOFF_SECONDS,
1687 0 : DEFAULT_MAX_BACKOFF_SECONDS,
1688 0 : &cancel,
1689 0 : )
1690 0 : .await;
1691 : }
1692 : }
1693 : }
1694 :
1695 2731 : let retries = task.retries.load(Ordering::SeqCst);
1696 2731 : if retries > 0 {
1697 0 : info!(
1698 0 : "remote task {} completed successfully after {} retries",
1699 0 : task.op, retries
1700 : );
1701 : } else {
1702 2731 : debug!("remote task {} completed successfully", task.op);
1703 : }
1704 :
1705 : // The task has completed successfully. Remove it from the in-progress list.
1706 2731 : let lsn_update = {
1707 2731 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
1708 2731 : let upload_queue = match upload_queue_guard.deref_mut() {
1709 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
1710 0 : UploadQueue::Stopped(_stopped) => {
1711 0 : None
1712 : },
1713 2731 : UploadQueue::Initialized(qi) => { Some(qi) }
1714 : };
1715 :
1716 2731 : let upload_queue = match upload_queue {
1717 2731 : Some(upload_queue) => upload_queue,
1718 : None => {
1719 0 : info!("another concurrent task already stopped the queue");
1720 0 : return;
1721 : }
1722 : };
1723 :
1724 2731 : upload_queue.inprogress_tasks.remove(&task.task_id);
1725 :
1726 2731 : let lsn_update = match task.op {
1727 : UploadOp::UploadLayer(_, _) => {
1728 1179 : upload_queue.num_inprogress_layer_uploads -= 1;
1729 1179 : None
1730 : }
1731 1345 : UploadOp::UploadMetadata { ref uploaded } => {
1732 1345 : upload_queue.num_inprogress_metadata_uploads -= 1;
1733 1345 :
1734 1345 : // the task id is reused as a monotonicity check for storing the "clean"
1735 1345 : // IndexPart.
1736 1345 : let last_updater = upload_queue.clean.1;
1737 1345 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
1738 1345 : let monotone = is_later || last_updater.is_none();
1739 :
1740 1345 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
1741 :
1742 : // not taking ownership is wasteful
1743 1345 : upload_queue.clean.0.clone_from(uploaded);
1744 1345 : upload_queue.clean.1 = Some(task.task_id);
1745 1345 :
1746 1345 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
1747 1345 :
1748 1345 : if self.generation.is_none() {
1749 : // Legacy mode: skip validating generation
1750 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
1751 0 : None
1752 : } else {
1753 1345 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
1754 : }
1755 : }
1756 : UploadOp::Delete(_) => {
1757 207 : upload_queue.num_inprogress_deletions -= 1;
1758 207 : None
1759 : }
1760 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
1761 : };
1762 :
1763 : // Launch any queued tasks that were unblocked by this one.
1764 2731 : self.launch_queued_tasks(upload_queue);
1765 2731 : lsn_update
1766 : };
1767 :
1768 2731 : if let Some((lsn, slot)) = lsn_update {
1769 : // Updates to the remote_consistent_lsn we advertise to pageservers
1770 : // are all routed through the DeletionQueue, to enforce important
1771 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
1772 1345 : self.deletion_queue_client
1773 1345 : .update_remote_consistent_lsn(
1774 1345 : self.tenant_shard_id,
1775 1345 : self.timeline_id,
1776 1345 : self.generation,
1777 1345 : lsn,
1778 1345 : slot,
1779 1345 : )
1780 0 : .await;
1781 1386 : }
1782 :
1783 2731 : self.metric_end(&task.op);
1784 2731 : }
1785 :
1786 6034 : fn metric_impl(
1787 6034 : &self,
1788 6034 : op: &UploadOp,
1789 6034 : ) -> Option<(
1790 6034 : RemoteOpFileKind,
1791 6034 : RemoteOpKind,
1792 6034 : RemoteTimelineClientMetricsCallTrackSize,
1793 6034 : )> {
1794 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
1795 6034 : let res = match op {
1796 2643 : UploadOp::UploadLayer(_, m) => (
1797 2643 : RemoteOpFileKind::Layer,
1798 2643 : RemoteOpKind::Upload,
1799 2643 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
1800 2643 : ),
1801 2751 : UploadOp::UploadMetadata { .. } => (
1802 2751 : RemoteOpFileKind::Index,
1803 2751 : RemoteOpKind::Upload,
1804 2751 : DontTrackSize {
1805 2751 : reason: "metadata uploads are tiny",
1806 2751 : },
1807 2751 : ),
1808 634 : UploadOp::Delete(_delete) => (
1809 634 : RemoteOpFileKind::Layer,
1810 634 : RemoteOpKind::Delete,
1811 634 : DontTrackSize {
1812 634 : reason: "should we track deletes? positive or negative sign?",
1813 634 : },
1814 634 : ),
1815 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
1816 : // we do not account these
1817 6 : return None;
1818 : }
1819 : };
1820 6028 : Some(res)
1821 6034 : }
1822 :
1823 3297 : fn metric_begin(&self, op: &UploadOp) {
1824 3297 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1825 3297 : Some(x) => x,
1826 0 : None => return,
1827 : };
1828 3297 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
1829 3297 : guard.will_decrement_manually(); // in metric_end(), see right below
1830 3297 : }
1831 :
1832 2737 : fn metric_end(&self, op: &UploadOp) {
1833 2737 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
1834 2731 : Some(x) => x,
1835 6 : None => return,
1836 : };
1837 2731 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
1838 2737 : }
1839 :
1840 : /// Close the upload queue for new operations and cancel queued operations.
1841 : ///
1842 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
1843 : ///
1844 : /// In-progress operations will still be running after this function returns.
1845 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
1846 : /// to wait for them to complete, after calling this function.
1847 14 : pub(crate) fn stop(&self) {
1848 14 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
1849 14 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
1850 14 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
1851 14 : let mut guard = self.upload_queue.lock().unwrap();
1852 14 : self.stop_impl(&mut guard);
1853 14 : }
1854 :
1855 14 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
1856 14 : match &mut **guard {
1857 : UploadQueue::Uninitialized => {
1858 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
1859 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
1860 : }
1861 : UploadQueue::Stopped(_) => {
1862 : // nothing to do
1863 6 : info!("another concurrent task already shut down the queue");
1864 : }
1865 8 : UploadQueue::Initialized(initialized) => {
1866 8 : info!("shutting down upload queue");
1867 :
1868 : // Replace the queue with the Stopped state, taking ownership of the old
1869 : // Initialized queue. We will do some checks on it, and then drop it.
1870 8 : let qi = {
1871 : // Here we preserve working version of the upload queue for possible use during deletions.
1872 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
1873 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
1874 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
1875 8 : let upload_queue_for_deletion = UploadQueueInitialized {
1876 8 : task_counter: 0,
1877 8 : dirty: initialized.dirty.clone(),
1878 8 : clean: initialized.clean.clone(),
1879 8 : latest_files_changes_since_metadata_upload_scheduled: 0,
1880 8 : visible_remote_consistent_lsn: initialized
1881 8 : .visible_remote_consistent_lsn
1882 8 : .clone(),
1883 8 : num_inprogress_layer_uploads: 0,
1884 8 : num_inprogress_metadata_uploads: 0,
1885 8 : num_inprogress_deletions: 0,
1886 8 : inprogress_tasks: HashMap::default(),
1887 8 : queued_operations: VecDeque::default(),
1888 8 : #[cfg(feature = "testing")]
1889 8 : dangling_files: HashMap::default(),
1890 8 : shutting_down: false,
1891 8 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
1892 8 : };
1893 8 :
1894 8 : let upload_queue = std::mem::replace(
1895 8 : &mut **guard,
1896 8 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
1897 8 : UploadQueueStoppedDeletable {
1898 8 : upload_queue_for_deletion,
1899 8 : deleted_at: SetDeletedFlagProgress::NotRunning,
1900 8 : },
1901 8 : )),
1902 8 : );
1903 8 : if let UploadQueue::Initialized(qi) = upload_queue {
1904 8 : qi
1905 : } else {
1906 0 : unreachable!("we checked in the match above that it is Initialized");
1907 : }
1908 : };
1909 :
1910 : // consistency check
1911 8 : assert_eq!(
1912 8 : qi.num_inprogress_layer_uploads
1913 8 : + qi.num_inprogress_metadata_uploads
1914 8 : + qi.num_inprogress_deletions,
1915 8 : qi.inprogress_tasks.len()
1916 8 : );
1917 :
1918 : // We don't need to do anything here for in-progress tasks. They will finish
1919 : // on their own, decrement the unfinished-task counter themselves, and observe
1920 : // that the queue is Stopped.
1921 8 : drop(qi.inprogress_tasks);
1922 :
1923 : // Tear down queued ops
1924 8 : for op in qi.queued_operations.into_iter() {
1925 6 : self.metric_end(&op);
1926 6 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
1927 6 : // which is exactly what we want to happen.
1928 6 : drop(op);
1929 6 : }
1930 : }
1931 : }
1932 14 : }
1933 : }
1934 :
1935 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1936 0 : let path = format!("tenants/{tenant_shard_id}");
1937 0 : RemotePath::from_string(&path).expect("Failed to construct path")
1938 0 : }
1939 :
1940 187 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
1941 187 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
1942 187 : RemotePath::from_string(&path).expect("Failed to construct path")
1943 187 : }
1944 :
1945 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
1946 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
1947 0 : RemotePath::from_string(&path).expect("Failed to construct path")
1948 0 : }
1949 :
1950 30 : pub fn remote_timeline_path(
1951 30 : tenant_shard_id: &TenantShardId,
1952 30 : timeline_id: &TimelineId,
1953 30 : ) -> RemotePath {
1954 30 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
1955 30 : }
1956 :
1957 : /// Note that the shard component of a remote layer path is _not_ always the same
1958 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
1959 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
1960 1394 : pub fn remote_layer_path(
1961 1394 : tenant_id: &TenantId,
1962 1394 : timeline_id: &TimelineId,
1963 1394 : shard: ShardIndex,
1964 1394 : layer_file_name: &LayerName,
1965 1394 : generation: Generation,
1966 1394 : ) -> RemotePath {
1967 1394 : // Generation-aware key format
1968 1394 : let path = format!(
1969 1394 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
1970 1394 : shard.get_suffix(),
1971 1394 : layer_file_name,
1972 1394 : generation.get_suffix()
1973 1394 : );
1974 1394 :
1975 1394 : RemotePath::from_string(&path).expect("Failed to construct path")
1976 1394 : }
1977 :
1978 4 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
1979 4 : RemotePath::from_string(&format!(
1980 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
1981 4 : ))
1982 4 : .expect("Failed to construct path")
1983 4 : }
1984 :
1985 2 : pub fn remote_initdb_preserved_archive_path(
1986 2 : tenant_id: &TenantId,
1987 2 : timeline_id: &TimelineId,
1988 2 : ) -> RemotePath {
1989 2 : RemotePath::from_string(&format!(
1990 2 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
1991 2 : ))
1992 2 : .expect("Failed to construct path")
1993 2 : }
1994 :
1995 1407 : pub fn remote_index_path(
1996 1407 : tenant_shard_id: &TenantShardId,
1997 1407 : timeline_id: &TimelineId,
1998 1407 : generation: Generation,
1999 1407 : ) -> RemotePath {
2000 1407 : RemotePath::from_string(&format!(
2001 1407 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2002 1407 : IndexPart::FILE_NAME,
2003 1407 : generation.get_suffix()
2004 1407 : ))
2005 1407 : .expect("Failed to construct path")
2006 1407 : }
2007 :
2008 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2009 0 : RemotePath::from_string(&format!(
2010 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2011 0 : ))
2012 0 : .expect("Failed to construct path")
2013 0 : }
2014 :
2015 : /// Given the key of an index, parse out the generation part of the name
2016 18 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2017 18 : let file_name = match path.get_path().file_name() {
2018 18 : Some(f) => f,
2019 : None => {
2020 : // Unexpected: we should be seeing index_part.json paths only
2021 0 : tracing::warn!("Malformed index key {}", path);
2022 0 : return None;
2023 : }
2024 : };
2025 :
2026 18 : match file_name.split_once('-') {
2027 12 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2028 6 : None => None,
2029 : }
2030 18 : }
2031 :
2032 : #[cfg(test)]
2033 : mod tests {
2034 : use super::*;
2035 : use crate::{
2036 : context::RequestContext,
2037 : tenant::{
2038 : harness::{TenantHarness, TIMELINE_ID},
2039 : storage_layer::layer::local_layer_path,
2040 : Tenant, Timeline,
2041 : },
2042 : DEFAULT_PG_VERSION,
2043 : };
2044 :
2045 : use std::collections::HashSet;
2046 :
2047 8 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2048 8 : format!("contents for {name}").into()
2049 8 : }
2050 :
2051 2 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2052 2 : let metadata = TimelineMetadata::new(
2053 2 : disk_consistent_lsn,
2054 2 : None,
2055 2 : None,
2056 2 : Lsn(0),
2057 2 : Lsn(0),
2058 2 : Lsn(0),
2059 2 : // Any version will do
2060 2 : // but it should be consistent with the one in the tests
2061 2 : crate::DEFAULT_PG_VERSION,
2062 2 : );
2063 2 :
2064 2 : // go through serialize + deserialize to fix the header, including checksum
2065 2 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2066 2 : }
2067 :
2068 2 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2069 6 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2070 2 : avec.sort();
2071 2 :
2072 2 : let mut bvec = b.to_vec();
2073 2 : bvec.sort_unstable();
2074 2 :
2075 2 : assert_eq!(avec, bvec);
2076 2 : }
2077 :
2078 4 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2079 4 : let mut expected: Vec<String> = expected
2080 4 : .iter()
2081 16 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2082 4 : .collect();
2083 4 : expected.sort();
2084 4 :
2085 4 : let mut found: Vec<String> = Vec::new();
2086 16 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2087 16 : let entry_name = entry.file_name();
2088 16 : let fname = entry_name.to_str().unwrap();
2089 16 : found.push(String::from(fname));
2090 16 : }
2091 4 : found.sort();
2092 4 :
2093 4 : assert_eq!(found, expected);
2094 4 : }
2095 :
2096 : struct TestSetup {
2097 : harness: TenantHarness,
2098 : tenant: Arc<Tenant>,
2099 : timeline: Arc<Timeline>,
2100 : tenant_ctx: RequestContext,
2101 : }
2102 :
2103 : impl TestSetup {
2104 8 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2105 8 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2106 8 : let harness = TenantHarness::create(test_name)?;
2107 32 : let (tenant, ctx) = harness.load().await;
2108 :
2109 8 : let timeline = tenant
2110 8 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2111 23 : .await?;
2112 :
2113 8 : Ok(Self {
2114 8 : harness,
2115 8 : tenant,
2116 8 : timeline,
2117 8 : tenant_ctx: ctx,
2118 8 : })
2119 8 : }
2120 :
2121 : /// Construct a RemoteTimelineClient in an arbitrary generation
2122 10 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2123 10 : Arc::new(RemoteTimelineClient {
2124 10 : conf: self.harness.conf,
2125 10 : runtime: tokio::runtime::Handle::current(),
2126 10 : tenant_shard_id: self.harness.tenant_shard_id,
2127 10 : timeline_id: TIMELINE_ID,
2128 10 : generation,
2129 10 : storage_impl: self.harness.remote_storage.clone(),
2130 10 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2131 10 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2132 10 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2133 10 : &self.harness.tenant_shard_id,
2134 10 : &TIMELINE_ID,
2135 10 : )),
2136 10 : cancel: CancellationToken::new(),
2137 10 : })
2138 10 : }
2139 :
2140 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2141 : /// and timeline_id are present.
2142 6 : fn span(&self) -> tracing::Span {
2143 6 : tracing::info_span!(
2144 : "test",
2145 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2146 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2147 : timeline_id = %TIMELINE_ID
2148 : )
2149 6 : }
2150 : }
2151 :
2152 : // Test scheduling
2153 : #[tokio::test]
2154 2 : async fn upload_scheduling() {
2155 2 : // Test outline:
2156 2 : //
2157 2 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2158 2 : // Schedule upload of index. Check that it is queued
2159 2 : // let the layer file uploads finish. Check that the index-upload is now started
2160 2 : // let the index-upload finish.
2161 2 : //
2162 2 : // Download back the index.json. Check that the list of files is correct
2163 2 : //
2164 2 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2165 2 : // let upload finish. Check that deletion is now started
2166 2 : // Schedule another deletion. Check that it's launched immediately.
2167 2 : // Schedule index upload. Check that it's queued
2168 2 :
2169 14 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2170 2 : let span = test_setup.span();
2171 2 : let _guard = span.enter();
2172 2 :
2173 2 : let TestSetup {
2174 2 : harness,
2175 2 : tenant: _tenant,
2176 2 : timeline,
2177 2 : tenant_ctx: _tenant_ctx,
2178 2 : } = test_setup;
2179 2 :
2180 2 : let client = &timeline.remote_client;
2181 2 :
2182 2 : // Download back the index.json, and check that the list of files is correct
2183 2 : let initial_index_part = match client
2184 2 : .download_index_file(&CancellationToken::new())
2185 8 : .await
2186 2 : .unwrap()
2187 2 : {
2188 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2189 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2190 2 : };
2191 2 : let initial_layers = initial_index_part
2192 2 : .layer_metadata
2193 2 : .keys()
2194 2 : .map(|f| f.to_owned())
2195 2 : .collect::<HashSet<LayerName>>();
2196 2 : let initial_layer = {
2197 2 : assert!(initial_layers.len() == 1);
2198 2 : initial_layers.into_iter().next().unwrap()
2199 2 : };
2200 2 :
2201 2 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2202 2 :
2203 2 : println!("workdir: {}", harness.conf.workdir);
2204 2 :
2205 2 : let remote_timeline_dir = harness
2206 2 : .remote_fs_dir
2207 2 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2208 2 : println!("remote_timeline_dir: {remote_timeline_dir}");
2209 2 :
2210 2 : let generation = harness.generation;
2211 2 : let shard = harness.shard;
2212 2 :
2213 2 : // Create a couple of dummy files, schedule upload for them
2214 2 :
2215 2 : let layers = [
2216 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2217 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2218 2 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2219 2 : ]
2220 2 : .into_iter()
2221 6 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2222 6 :
2223 6 : let local_path = local_layer_path(
2224 6 : harness.conf,
2225 6 : &timeline.tenant_shard_id,
2226 6 : &timeline.timeline_id,
2227 6 : &name,
2228 6 : &generation,
2229 6 : );
2230 6 : std::fs::write(&local_path, &contents).unwrap();
2231 6 :
2232 6 : Layer::for_resident(
2233 6 : harness.conf,
2234 6 : &timeline,
2235 6 : local_path,
2236 6 : name,
2237 6 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2238 6 : )
2239 6 : }).collect::<Vec<_>>();
2240 2 :
2241 2 : client
2242 2 : .schedule_layer_file_upload(layers[0].clone())
2243 2 : .unwrap();
2244 2 : client
2245 2 : .schedule_layer_file_upload(layers[1].clone())
2246 2 : .unwrap();
2247 2 :
2248 2 : // Check that they are started immediately, not queued
2249 2 : //
2250 2 : // this works because we running within block_on, so any futures are now queued up until
2251 2 : // our next await point.
2252 2 : {
2253 2 : let mut guard = client.upload_queue.lock().unwrap();
2254 2 : let upload_queue = guard.initialized_mut().unwrap();
2255 2 : assert!(upload_queue.queued_operations.is_empty());
2256 2 : assert!(upload_queue.inprogress_tasks.len() == 2);
2257 2 : assert!(upload_queue.num_inprogress_layer_uploads == 2);
2258 2 :
2259 2 : // also check that `latest_file_changes` was updated
2260 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2261 2 : }
2262 2 :
2263 2 : // Schedule upload of index. Check that it is queued
2264 2 : let metadata = dummy_metadata(Lsn(0x20));
2265 2 : client
2266 2 : .schedule_index_upload_for_full_metadata_update(&metadata)
2267 2 : .unwrap();
2268 2 : {
2269 2 : let mut guard = client.upload_queue.lock().unwrap();
2270 2 : let upload_queue = guard.initialized_mut().unwrap();
2271 2 : assert!(upload_queue.queued_operations.len() == 1);
2272 2 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2273 2 : }
2274 2 :
2275 2 : // Wait for the uploads to finish
2276 2 : client.wait_completion().await.unwrap();
2277 2 : {
2278 2 : let mut guard = client.upload_queue.lock().unwrap();
2279 2 : let upload_queue = guard.initialized_mut().unwrap();
2280 2 :
2281 2 : assert!(upload_queue.queued_operations.is_empty());
2282 2 : assert!(upload_queue.inprogress_tasks.is_empty());
2283 2 : }
2284 2 :
2285 2 : // Download back the index.json, and check that the list of files is correct
2286 2 : let index_part = match client
2287 2 : .download_index_file(&CancellationToken::new())
2288 8 : .await
2289 2 : .unwrap()
2290 2 : {
2291 2 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2292 2 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2293 2 : };
2294 2 :
2295 2 : assert_file_list(
2296 2 : &index_part
2297 2 : .layer_metadata
2298 2 : .keys()
2299 6 : .map(|f| f.to_owned())
2300 2 : .collect(),
2301 2 : &[
2302 2 : &initial_layer.to_string(),
2303 2 : &layers[0].layer_desc().layer_name().to_string(),
2304 2 : &layers[1].layer_desc().layer_name().to_string(),
2305 2 : ],
2306 2 : );
2307 2 : assert_eq!(index_part.metadata, metadata);
2308 2 :
2309 2 : // Schedule upload and then a deletion. Check that the deletion is queued
2310 2 : client
2311 2 : .schedule_layer_file_upload(layers[2].clone())
2312 2 : .unwrap();
2313 2 :
2314 2 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2315 2 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2316 2 : // spawn_blocking started by the drop.
2317 2 : client
2318 2 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2319 2 : .unwrap();
2320 2 : {
2321 2 : let mut guard = client.upload_queue.lock().unwrap();
2322 2 : let upload_queue = guard.initialized_mut().unwrap();
2323 2 :
2324 2 : // Deletion schedules upload of the index file, and the file deletion itself
2325 2 : assert_eq!(upload_queue.queued_operations.len(), 2);
2326 2 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2327 2 : assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
2328 2 : assert_eq!(upload_queue.num_inprogress_deletions, 0);
2329 2 : assert_eq!(
2330 2 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2331 2 : 0
2332 2 : );
2333 2 : }
2334 2 : assert_remote_files(
2335 2 : &[
2336 2 : &initial_layer.to_string(),
2337 2 : &layers[0].layer_desc().layer_name().to_string(),
2338 2 : &layers[1].layer_desc().layer_name().to_string(),
2339 2 : "index_part.json",
2340 2 : ],
2341 2 : &remote_timeline_dir,
2342 2 : generation,
2343 2 : );
2344 2 :
2345 2 : // Finish them
2346 2 : client.wait_completion().await.unwrap();
2347 2 : harness.deletion_queue.pump().await;
2348 2 :
2349 2 : assert_remote_files(
2350 2 : &[
2351 2 : &initial_layer.to_string(),
2352 2 : &layers[1].layer_desc().layer_name().to_string(),
2353 2 : &layers[2].layer_desc().layer_name().to_string(),
2354 2 : "index_part.json",
2355 2 : ],
2356 2 : &remote_timeline_dir,
2357 2 : generation,
2358 2 : );
2359 2 : }
2360 :
2361 : #[tokio::test]
2362 2 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2363 2 : // Setup
2364 2 :
2365 2 : let TestSetup {
2366 2 : harness,
2367 2 : tenant: _tenant,
2368 2 : timeline,
2369 2 : ..
2370 14 : } = TestSetup::new("metrics").await.unwrap();
2371 2 : let client = &timeline.remote_client;
2372 2 :
2373 2 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2374 2 : let local_path = local_layer_path(
2375 2 : harness.conf,
2376 2 : &timeline.tenant_shard_id,
2377 2 : &timeline.timeline_id,
2378 2 : &layer_file_name_1,
2379 2 : &harness.generation,
2380 2 : );
2381 2 : let content_1 = dummy_contents("foo");
2382 2 : std::fs::write(&local_path, &content_1).unwrap();
2383 2 :
2384 2 : let layer_file_1 = Layer::for_resident(
2385 2 : harness.conf,
2386 2 : &timeline,
2387 2 : local_path,
2388 2 : layer_file_name_1.clone(),
2389 2 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
2390 2 : );
2391 2 :
2392 2 : #[derive(Debug, PartialEq, Clone, Copy)]
2393 2 : struct BytesStartedFinished {
2394 2 : started: Option<usize>,
2395 2 : finished: Option<usize>,
2396 2 : }
2397 2 : impl std::ops::Add for BytesStartedFinished {
2398 2 : type Output = Self;
2399 4 : fn add(self, rhs: Self) -> Self::Output {
2400 4 : Self {
2401 4 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
2402 4 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
2403 4 : }
2404 4 : }
2405 2 : }
2406 6 : let get_bytes_started_stopped = || {
2407 6 : let started = client
2408 6 : .metrics
2409 6 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2410 6 : .map(|v| v.try_into().unwrap());
2411 6 : let stopped = client
2412 6 : .metrics
2413 6 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
2414 6 : .map(|v| v.try_into().unwrap());
2415 6 : BytesStartedFinished {
2416 6 : started,
2417 6 : finished: stopped,
2418 6 : }
2419 6 : };
2420 2 :
2421 2 : // Test
2422 2 : tracing::info!("now doing actual test");
2423 2 :
2424 2 : let actual_a = get_bytes_started_stopped();
2425 2 :
2426 2 : client
2427 2 : .schedule_layer_file_upload(layer_file_1.clone())
2428 2 : .unwrap();
2429 2 :
2430 2 : let actual_b = get_bytes_started_stopped();
2431 2 :
2432 2 : client.wait_completion().await.unwrap();
2433 2 :
2434 2 : let actual_c = get_bytes_started_stopped();
2435 2 :
2436 2 : // Validate
2437 2 :
2438 2 : let expected_b = actual_a
2439 2 : + BytesStartedFinished {
2440 2 : started: Some(content_1.len()),
2441 2 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
2442 2 : finished: Some(0),
2443 2 : };
2444 2 : assert_eq!(actual_b, expected_b);
2445 2 :
2446 2 : let expected_c = actual_a
2447 2 : + BytesStartedFinished {
2448 2 : started: Some(content_1.len()),
2449 2 : finished: Some(content_1.len()),
2450 2 : };
2451 2 : assert_eq!(actual_c, expected_c);
2452 2 : }
2453 :
2454 12 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
2455 12 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
2456 12 : let example_index_part = IndexPart::example();
2457 12 :
2458 12 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
2459 12 :
2460 12 : let index_path = test_state.harness.remote_fs_dir.join(
2461 12 : remote_index_path(
2462 12 : &test_state.harness.tenant_shard_id,
2463 12 : &TIMELINE_ID,
2464 12 : generation,
2465 12 : )
2466 12 : .get_path(),
2467 12 : );
2468 12 :
2469 12 : std::fs::create_dir_all(index_path.parent().unwrap())
2470 12 : .expect("creating test dir should work");
2471 12 :
2472 12 : eprintln!("Writing {index_path}");
2473 12 : std::fs::write(&index_path, index_part_bytes).unwrap();
2474 12 : example_index_part
2475 12 : }
2476 :
2477 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
2478 : /// index, the IndexPart returned is equal to `expected`
2479 10 : async fn assert_got_index_part(
2480 10 : test_state: &TestSetup,
2481 10 : get_generation: Generation,
2482 10 : expected: &IndexPart,
2483 10 : ) {
2484 10 : let client = test_state.build_client(get_generation);
2485 :
2486 10 : let download_r = client
2487 10 : .download_index_file(&CancellationToken::new())
2488 59 : .await
2489 10 : .expect("download should always succeed");
2490 10 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
2491 10 : match download_r {
2492 10 : MaybeDeletedIndexPart::IndexPart(index_part) => {
2493 10 : assert_eq!(&index_part, expected);
2494 : }
2495 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
2496 : }
2497 10 : }
2498 :
2499 : #[tokio::test]
2500 2 : async fn index_part_download_simple() -> anyhow::Result<()> {
2501 13 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
2502 2 : let span = test_state.span();
2503 2 : let _guard = span.enter();
2504 2 :
2505 2 : // Simple case: we are in generation N, load the index from generation N - 1
2506 2 : let generation_n = 5;
2507 2 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2508 2 :
2509 10 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
2510 2 :
2511 2 : Ok(())
2512 2 : }
2513 :
2514 : #[tokio::test]
2515 2 : async fn index_part_download_ordering() -> anyhow::Result<()> {
2516 2 : let test_state = TestSetup::new("index_part_download_ordering")
2517 14 : .await
2518 2 : .unwrap();
2519 2 :
2520 2 : let span = test_state.span();
2521 2 : let _guard = span.enter();
2522 2 :
2523 2 : // A generation-less IndexPart exists in the bucket, we should find it
2524 2 : let generation_n = 5;
2525 2 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
2526 14 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
2527 2 :
2528 2 : // If a more recent-than-none generation exists, we should prefer to load that
2529 2 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
2530 14 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2531 2 :
2532 2 : // If a more-recent-than-me generation exists, we should ignore it.
2533 2 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
2534 14 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
2535 2 :
2536 2 : // If a directly previous generation exists, _and_ an index exists in my own
2537 2 : // generation, I should prefer my own generation.
2538 2 : let _injected_prev =
2539 2 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
2540 2 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
2541 2 : assert_got_index_part(
2542 2 : &test_state,
2543 2 : Generation::new(generation_n),
2544 2 : &injected_current,
2545 2 : )
2546 7 : .await;
2547 2 :
2548 2 : Ok(())
2549 2 : }
2550 : }
|