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 : //!
67 : //! From the user's perspective, the operations are executed sequentially.
68 : //! Internally, the client knows which operations can be performed in parallel,
69 : //! and which operations act like a "barrier" that require preceding operations
70 : //! to finish. The calling code just needs to call the schedule-functions in the
71 : //! correct order, and the client will parallelize the operations in a way that
72 : //! is safe. For more details, see `UploadOp::can_bypass`.
73 : //!
74 : //! All of this relies on the following invariants:
75 : //!
76 : //! - We rely on read-after write consistency in the remote storage.
77 : //! - Layer files are immutable.
78 : //!
79 : //! NB: Pageserver assumes that it has exclusive write access to the tenant in remote
80 : //! storage. Different tenants can be attached to different pageservers, but if the
81 : //! same tenant is attached to two pageservers at the same time, they will overwrite
82 : //! each other's index file updates, and confusion will ensue. There's no interlock or
83 : //! mechanism to detect that in the pageserver, we rely on the control plane to ensure
84 : //! that that doesn't happen.
85 : //!
86 : //! ## Implementation Note
87 : //!
88 : //! The *actual* remote state lags behind the *desired* remote state while
89 : //! there are in-flight operations.
90 : //! We keep track of the desired remote state in [`UploadQueueInitialized::dirty`].
91 : //! It is initialized based on the [`IndexPart`] that was passed during init
92 : //! and updated with every `schedule_*` function call.
93 : //! All this is necessary necessary to compute the future [`IndexPart`]s
94 : //! when scheduling an operation while other operations that also affect the
95 : //! remote [`IndexPart`] are in flight.
96 : //!
97 : //! # Retries & Error Handling
98 : //!
99 : //! The client retries operations indefinitely, using exponential back-off.
100 : //! There is no way to force a retry, i.e., interrupt the back-off.
101 : //! This could be built easily.
102 : //!
103 : //! # Cancellation
104 : //!
105 : //! The operations execute as plain [`task_mgr`] tasks, scoped to
106 : //! the client's tenant and timeline.
107 : //! Dropping the client will drop queued operations but not executing operations.
108 : //! These will complete unless the `task_mgr` tasks are cancelled using `task_mgr`
109 : //! APIs, e.g., during pageserver shutdown, timeline delete, or tenant detach.
110 : //!
111 : //! # Completion
112 : //!
113 : //! Once an operation has completed, we update [`UploadQueueInitialized::clean`] immediately,
114 : //! and submit a request through the DeletionQueue to update
115 : //! [`UploadQueueInitialized::visible_remote_consistent_lsn`] after it has
116 : //! validated that our generation is not stale. It is this visible value
117 : //! that is advertized to safekeepers as a signal that that they can
118 : //! delete the WAL up to that LSN.
119 : //!
120 : //! The [`RemoteTimelineClient::wait_completion`] method can be used to wait
121 : //! for all pending operations to complete. It does not prevent more
122 : //! operations from getting scheduled.
123 : //!
124 : //! # Crash Consistency
125 : //!
126 : //! We do not persist the upload queue state.
127 : //! If we drop the client, or crash, all unfinished operations are lost.
128 : //!
129 : //! To recover, the following steps need to be taken:
130 : //! - Retrieve the current remote [`IndexPart`]. This gives us a
131 : //! consistent remote state, assuming the user scheduled the operations in
132 : //! the correct order.
133 : //! - Initiate upload queue with that [`IndexPart`].
134 : //! - Reschedule all lost operations by comparing the local filesystem state
135 : //! and remote state as per [`IndexPart`]. This is done in
136 : //! [`Tenant::timeline_init_and_sync`].
137 : //!
138 : //! Note that if we crash during file deletion between the index update
139 : //! that removes the file from the list of files, and deleting the remote file,
140 : //! the file is leaked in the remote storage. Similarly, if a new file is created
141 : //! and uploaded, but the pageserver dies permanently before updating the
142 : //! remote index file, the new file is leaked in remote storage. We accept and
143 : //! tolerate that for now.
144 : //! Note further that we cannot easily fix this by scheduling deletes for every
145 : //! file that is present only on the remote, because we cannot distinguish the
146 : //! following two cases:
147 : //! - (1) We had the file locally, deleted it locally, scheduled a remote delete,
148 : //! but crashed before it finished remotely.
149 : //! - (2) We never had the file locally because we haven't on-demand downloaded
150 : //! it yet.
151 : //!
152 : //! # Downloads
153 : //!
154 : //! In addition to the upload queue, [`RemoteTimelineClient`] has functions for
155 : //! downloading files from the remote storage. Downloads are performed immediately
156 : //! against the `RemoteStorage`, independently of the upload queue.
157 : //!
158 : //! When we attach a tenant, we perform the following steps:
159 : //! - create `Tenant` object in `TenantState::Attaching` state
160 : //! - List timelines that are present in remote storage, and for each:
161 : //! - download their remote [`IndexPart`]s
162 : //! - create `Timeline` struct and a `RemoteTimelineClient`
163 : //! - initialize the client's upload queue with its `IndexPart`
164 : //! - schedule uploads for layers that are only present locally.
165 : //! - After the above is done for each timeline, open the tenant for business by
166 : //! transitioning it from `TenantState::Attaching` to `TenantState::Active` state.
167 : //! This starts the timelines' WAL-receivers and the tenant's GC & Compaction loops.
168 : //!
169 : //! # Operating Without Remote Storage
170 : //!
171 : //! If no remote storage configuration is provided, the [`RemoteTimelineClient`] is
172 : //! not created and the uploads are skipped.
173 : //!
174 : //! [`Tenant::timeline_init_and_sync`]: super::Tenant::timeline_init_and_sync
175 : //! [`Timeline::load_layer_map`]: super::Timeline::load_layer_map
176 :
177 : pub(crate) mod download;
178 : pub mod index;
179 : pub mod manifest;
180 : pub(crate) mod upload;
181 :
182 : use anyhow::Context;
183 : use camino::Utf8Path;
184 : use chrono::{NaiveDateTime, Utc};
185 :
186 : pub(crate) use download::download_initdb_tar_zst;
187 : use pageserver_api::models::TimelineArchivalState;
188 : use pageserver_api::shard::{ShardIndex, TenantShardId};
189 : use regex::Regex;
190 : use scopeguard::ScopeGuard;
191 : use tokio_util::sync::CancellationToken;
192 : use utils::backoff::{
193 : self, exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS,
194 : };
195 : use utils::pausable_failpoint;
196 : use utils::shard::ShardNumber;
197 :
198 : use std::collections::{HashMap, HashSet, VecDeque};
199 : use std::sync::atomic::{AtomicU32, Ordering};
200 : use std::sync::{Arc, Mutex, OnceLock};
201 : use std::time::Duration;
202 :
203 : use remote_storage::{
204 : DownloadError, GenericRemoteStorage, ListingMode, RemotePath, TimeoutOrCancel,
205 : };
206 : use std::ops::DerefMut;
207 : use tracing::{debug, error, info, instrument, warn};
208 : use tracing::{info_span, Instrument};
209 : use utils::lsn::Lsn;
210 :
211 : use crate::context::RequestContext;
212 : use crate::deletion_queue::{DeletionQueueClient, DeletionQueueError};
213 : use crate::metrics::{
214 : MeasureRemoteOp, RemoteOpFileKind, RemoteOpKind, RemoteTimelineClientMetrics,
215 : RemoteTimelineClientMetricsCallTrackSize, REMOTE_ONDEMAND_DOWNLOADED_BYTES,
216 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
217 : };
218 : use crate::task_mgr::shutdown_token;
219 : use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
220 : use crate::tenant::remote_timeline_client::download::download_retry;
221 : use crate::tenant::storage_layer::AsLayerDesc;
222 : use crate::tenant::upload_queue::{Delete, OpType, UploadQueueStoppedDeletable};
223 : use crate::tenant::TIMELINES_SEGMENT_NAME;
224 : use crate::{
225 : config::PageServerConf,
226 : task_mgr,
227 : task_mgr::TaskKind,
228 : task_mgr::BACKGROUND_RUNTIME,
229 : tenant::metadata::TimelineMetadata,
230 : tenant::upload_queue::{
231 : UploadOp, UploadQueue, UploadQueueInitialized, UploadQueueStopped, UploadTask,
232 : },
233 : TENANT_HEATMAP_BASENAME,
234 : };
235 :
236 : use utils::id::{TenantId, TimelineId};
237 :
238 : use self::index::IndexPart;
239 :
240 : use super::config::AttachedLocationConfig;
241 : use super::metadata::MetadataUpdate;
242 : use super::storage_layer::{Layer, LayerName, ResidentLayer};
243 : use super::timeline::import_pgdata;
244 : use super::upload_queue::{NotInitialized, SetDeletedFlagProgress};
245 : use super::{DeleteTimelineError, Generation};
246 :
247 : pub(crate) use download::{
248 : download_index_part, download_tenant_manifest, is_temp_download_file,
249 : list_remote_tenant_shards, list_remote_timelines,
250 : };
251 : pub(crate) use index::LayerFileMetadata;
252 : pub(crate) use upload::upload_initdb_dir;
253 :
254 : // Occasional network issues and such can cause remote operations to fail, and
255 : // that's expected. If a download fails, we log it at info-level, and retry.
256 : // But after FAILED_DOWNLOAD_WARN_THRESHOLD retries, we start to log it at WARN
257 : // level instead, as repeated failures can mean a more serious problem. If it
258 : // fails more than FAILED_DOWNLOAD_RETRIES times, we give up
259 : pub(crate) const FAILED_DOWNLOAD_WARN_THRESHOLD: u32 = 3;
260 : pub(crate) const FAILED_REMOTE_OP_RETRIES: u32 = 10;
261 :
262 : // Similarly log failed uploads and deletions at WARN level, after this many
263 : // retries. Uploads and deletions are retried forever, though.
264 : pub(crate) const FAILED_UPLOAD_WARN_THRESHOLD: u32 = 3;
265 :
266 : pub(crate) const INITDB_PATH: &str = "initdb.tar.zst";
267 :
268 : pub(crate) const INITDB_PRESERVED_PATH: &str = "initdb-preserved.tar.zst";
269 :
270 : /// Default buffer size when interfacing with [`tokio::fs::File`].
271 : pub(crate) const BUFFER_SIZE: usize = 32 * 1024;
272 :
273 : /// Doing non-essential flushes of deletion queue is subject to this timeout, after
274 : /// which we warn and skip.
275 : const DELETION_QUEUE_FLUSH_TIMEOUT: Duration = Duration::from_secs(10);
276 :
277 : pub enum MaybeDeletedIndexPart {
278 : IndexPart(IndexPart),
279 : Deleted(IndexPart),
280 : }
281 :
282 : #[derive(Debug, thiserror::Error)]
283 : pub enum PersistIndexPartWithDeletedFlagError {
284 : #[error("another task is already setting the deleted_flag, started at {0:?}")]
285 : AlreadyInProgress(NaiveDateTime),
286 : #[error("the deleted_flag was already set, value is {0:?}")]
287 : AlreadyDeleted(NaiveDateTime),
288 : #[error(transparent)]
289 : Other(#[from] anyhow::Error),
290 : }
291 :
292 : #[derive(Debug, thiserror::Error)]
293 : pub enum WaitCompletionError {
294 : #[error(transparent)]
295 : NotInitialized(NotInitialized),
296 : #[error("wait_completion aborted because upload queue was stopped")]
297 : UploadQueueShutDownOrStopped,
298 : }
299 :
300 : #[derive(Debug, thiserror::Error)]
301 : #[error("Upload queue either in unexpected state or hasn't downloaded manifest yet")]
302 : pub struct UploadQueueNotReadyError;
303 :
304 : #[derive(Debug, thiserror::Error)]
305 : pub enum ShutdownIfArchivedError {
306 : #[error(transparent)]
307 : NotInitialized(NotInitialized),
308 : #[error("timeline is not archived")]
309 : NotArchived,
310 : }
311 :
312 : /// Behavioral modes that enable seamless live migration.
313 : ///
314 : /// See docs/rfcs/028-pageserver-migration.md to understand how these fit in.
315 : struct RemoteTimelineClientConfig {
316 : /// If this is false, then update to remote_consistent_lsn are dropped rather
317 : /// than being submitted to DeletionQueue for validation. This behavior is
318 : /// used when a tenant attachment is known to have a stale generation number,
319 : /// such that validation attempts will always fail. This is not necessary
320 : /// for correctness, but avoids spamming error statistics with failed validations
321 : /// when doing migrations of tenants.
322 : process_remote_consistent_lsn_updates: bool,
323 :
324 : /// If this is true, then object deletions are held in a buffer in RemoteTimelineClient
325 : /// rather than being submitted to the DeletionQueue. This behavior is used when a tenant
326 : /// is known to be multi-attached, in order to avoid disrupting other attached tenants
327 : /// whose generations' metadata refers to the deleted objects.
328 : block_deletions: bool,
329 : }
330 :
331 : /// RemoteTimelineClientConfig's state is entirely driven by LocationConf, but we do
332 : /// not carry the entire LocationConf structure: it's much more than we need. The From
333 : /// impl extracts the subset of the LocationConf that is interesting to RemoteTimelineClient.
334 : impl From<&AttachedLocationConfig> for RemoteTimelineClientConfig {
335 916 : fn from(lc: &AttachedLocationConfig) -> Self {
336 916 : Self {
337 916 : block_deletions: !lc.may_delete_layers_hint(),
338 916 : process_remote_consistent_lsn_updates: lc.may_upload_layers_hint(),
339 916 : }
340 916 : }
341 : }
342 :
343 : /// A client for accessing a timeline's data in remote storage.
344 : ///
345 : /// This takes care of managing the number of connections, and balancing them
346 : /// across tenants. This also handles retries of failed uploads.
347 : ///
348 : /// Upload and delete requests are ordered so that before a deletion is
349 : /// performed, we wait for all preceding uploads to finish. This ensures sure
350 : /// that if you perform a compaction operation that reshuffles data in layer
351 : /// files, we don't have a transient state where the old files have already been
352 : /// deleted, but new files have not yet been uploaded.
353 : ///
354 : /// Similarly, this enforces an order between index-file uploads, and layer
355 : /// uploads. Before an index-file upload is performed, all preceding layer
356 : /// uploads must be finished.
357 : ///
358 : /// This also maintains a list of remote files, and automatically includes that
359 : /// in the index part file, whenever timeline metadata is uploaded.
360 : ///
361 : /// Downloads are not queued, they are performed immediately.
362 : pub(crate) struct RemoteTimelineClient {
363 : conf: &'static PageServerConf,
364 :
365 : runtime: tokio::runtime::Handle,
366 :
367 : tenant_shard_id: TenantShardId,
368 : timeline_id: TimelineId,
369 : generation: Generation,
370 :
371 : upload_queue: Mutex<UploadQueue>,
372 :
373 : pub(crate) metrics: Arc<RemoteTimelineClientMetrics>,
374 :
375 : storage_impl: GenericRemoteStorage,
376 :
377 : deletion_queue_client: DeletionQueueClient,
378 :
379 : /// Subset of tenant configuration used to control upload behaviors during migrations
380 : config: std::sync::RwLock<RemoteTimelineClientConfig>,
381 :
382 : cancel: CancellationToken,
383 : }
384 :
385 : impl Drop for RemoteTimelineClient {
386 40 : fn drop(&mut self) {
387 40 : debug!("dropping RemoteTimelineClient");
388 40 : }
389 : }
390 :
391 : impl RemoteTimelineClient {
392 : ///
393 : /// Create a remote storage client for given timeline
394 : ///
395 : /// Note: the caller must initialize the upload queue before any uploads can be scheduled,
396 : /// by calling init_upload_queue.
397 : ///
398 896 : pub(crate) fn new(
399 896 : remote_storage: GenericRemoteStorage,
400 896 : deletion_queue_client: DeletionQueueClient,
401 896 : conf: &'static PageServerConf,
402 896 : tenant_shard_id: TenantShardId,
403 896 : timeline_id: TimelineId,
404 896 : generation: Generation,
405 896 : location_conf: &AttachedLocationConfig,
406 896 : ) -> RemoteTimelineClient {
407 896 : RemoteTimelineClient {
408 896 : conf,
409 896 : runtime: if cfg!(test) {
410 : // remote_timeline_client.rs tests rely on current-thread runtime
411 896 : tokio::runtime::Handle::current()
412 : } else {
413 0 : BACKGROUND_RUNTIME.handle().clone()
414 : },
415 896 : tenant_shard_id,
416 896 : timeline_id,
417 896 : generation,
418 896 : storage_impl: remote_storage,
419 896 : deletion_queue_client,
420 896 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
421 896 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
422 896 : &tenant_shard_id,
423 896 : &timeline_id,
424 896 : )),
425 896 : config: std::sync::RwLock::new(RemoteTimelineClientConfig::from(location_conf)),
426 896 : cancel: CancellationToken::new(),
427 896 : }
428 896 : }
429 :
430 : /// Initialize the upload queue for a remote storage that already received
431 : /// an index file upload, i.e., it's not empty.
432 : /// The given `index_part` must be the one on the remote.
433 12 : pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
434 12 : // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
435 12 : // certainly no point in starting more upload tasks than this.
436 12 : let inprogress_limit = self
437 12 : .conf
438 12 : .remote_storage_config
439 12 : .as_ref()
440 12 : .map_or(0, |r| r.concurrency_limit());
441 12 : let mut upload_queue = self.upload_queue.lock().unwrap();
442 12 : upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
443 12 : self.update_remote_physical_size_gauge(Some(index_part));
444 12 : info!(
445 0 : "initialized upload queue from remote index with {} layer files",
446 0 : index_part.layer_metadata.len()
447 : );
448 12 : Ok(())
449 12 : }
450 :
451 : /// Initialize the upload queue for the case where the remote storage is empty,
452 : /// i.e., it doesn't have an `IndexPart`.
453 884 : pub fn init_upload_queue_for_empty_remote(
454 884 : &self,
455 884 : local_metadata: &TimelineMetadata,
456 884 : ) -> anyhow::Result<()> {
457 884 : // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
458 884 : // certainly no point in starting more upload tasks than this.
459 884 : let inprogress_limit = self
460 884 : .conf
461 884 : .remote_storage_config
462 884 : .as_ref()
463 884 : .map_or(0, |r| r.concurrency_limit());
464 884 : let mut upload_queue = self.upload_queue.lock().unwrap();
465 884 : upload_queue.initialize_empty_remote(local_metadata, inprogress_limit)?;
466 884 : self.update_remote_physical_size_gauge(None);
467 884 : info!("initialized upload queue as empty");
468 884 : Ok(())
469 884 : }
470 :
471 : /// Initialize the queue in stopped state. Used in startup path
472 : /// to continue deletion operation interrupted by pageserver crash or restart.
473 0 : pub fn init_upload_queue_stopped_to_continue_deletion(
474 0 : &self,
475 0 : index_part: &IndexPart,
476 0 : ) -> anyhow::Result<()> {
477 : // FIXME: consider newtype for DeletedIndexPart.
478 0 : let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
479 0 : "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
480 0 : ))?;
481 0 : let inprogress_limit = self
482 0 : .conf
483 0 : .remote_storage_config
484 0 : .as_ref()
485 0 : .map_or(0, |r| r.concurrency_limit());
486 0 :
487 0 : let mut upload_queue = self.upload_queue.lock().unwrap();
488 0 : upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
489 0 : self.update_remote_physical_size_gauge(Some(index_part));
490 0 : self.stop_impl(&mut upload_queue);
491 0 :
492 0 : upload_queue
493 0 : .stopped_mut()
494 0 : .expect("stopped above")
495 0 : .deleted_at = SetDeletedFlagProgress::Successful(deleted_at);
496 0 :
497 0 : Ok(())
498 0 : }
499 :
500 : /// Notify this client of a change to its parent tenant's config, as this may cause us to
501 : /// take action (unblocking deletions when transitioning from AttachedMulti to AttachedSingle)
502 0 : pub(super) fn update_config(&self, location_conf: &AttachedLocationConfig) {
503 0 : let new_conf = RemoteTimelineClientConfig::from(location_conf);
504 0 : let unblocked = !new_conf.block_deletions;
505 0 :
506 0 : // Update config before draining deletions, so that we don't race with more being
507 0 : // inserted. This can result in deletions happening our of order, but that does not
508 0 : // violate any invariants: deletions only need to be ordered relative to upload of the index
509 0 : // that dereferences the deleted objects, and we are not changing that order.
510 0 : *self.config.write().unwrap() = new_conf;
511 0 :
512 0 : if unblocked {
513 : // If we may now delete layers, drain any that were blocked in our old
514 : // configuration state
515 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
516 :
517 0 : if let Ok(queue) = queue_locked.initialized_mut() {
518 0 : let blocked_deletions = std::mem::take(&mut queue.blocked_deletions);
519 0 : for d in blocked_deletions {
520 0 : if let Err(e) = self.deletion_queue_client.push_layers(
521 0 : self.tenant_shard_id,
522 0 : self.timeline_id,
523 0 : self.generation,
524 0 : d.layers,
525 0 : ) {
526 : // This could happen if the pageserver is shut down while a tenant
527 : // is transitioning from a deletion-blocked state: we will leak some
528 : // S3 objects in this case.
529 0 : warn!("Failed to drain blocked deletions: {}", e);
530 0 : break;
531 0 : }
532 : }
533 0 : }
534 0 : }
535 0 : }
536 :
537 : /// Returns `None` if nothing is yet uplodaded, `Some(disk_consistent_lsn)` otherwise.
538 0 : pub fn remote_consistent_lsn_projected(&self) -> Option<Lsn> {
539 0 : match &mut *self.upload_queue.lock().unwrap() {
540 0 : UploadQueue::Uninitialized => None,
541 0 : UploadQueue::Initialized(q) => q.get_last_remote_consistent_lsn_projected(),
542 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
543 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => q
544 0 : .upload_queue_for_deletion
545 0 : .get_last_remote_consistent_lsn_projected(),
546 : }
547 0 : }
548 :
549 0 : pub fn remote_consistent_lsn_visible(&self) -> Option<Lsn> {
550 0 : match &mut *self.upload_queue.lock().unwrap() {
551 0 : UploadQueue::Uninitialized => None,
552 0 : UploadQueue::Initialized(q) => Some(q.get_last_remote_consistent_lsn_visible()),
553 0 : UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => None,
554 0 : UploadQueue::Stopped(UploadQueueStopped::Deletable(q)) => Some(
555 0 : q.upload_queue_for_deletion
556 0 : .get_last_remote_consistent_lsn_visible(),
557 0 : ),
558 : }
559 0 : }
560 :
561 : /// Returns true if this timeline was previously detached at this Lsn and the remote timeline
562 : /// client is currently initialized.
563 0 : pub(crate) fn is_previous_ancestor_lsn(&self, lsn: Lsn) -> bool {
564 0 : self.upload_queue
565 0 : .lock()
566 0 : .unwrap()
567 0 : .initialized_mut()
568 0 : .map(|uq| uq.clean.0.lineage.is_previous_ancestor_lsn(lsn))
569 0 : .unwrap_or(false)
570 0 : }
571 :
572 : /// Returns whether the timeline is archived.
573 : /// Return None if the remote index_part hasn't been downloaded yet.
574 0 : pub(crate) fn is_archived(&self) -> Option<bool> {
575 0 : self.upload_queue
576 0 : .lock()
577 0 : .unwrap()
578 0 : .initialized_mut()
579 0 : .map(|q| q.clean.0.archived_at.is_some())
580 0 : .ok()
581 0 : }
582 :
583 : /// Returns `Ok(Some(timestamp))` if the timeline has been archived, `Ok(None)` if the timeline hasn't been archived.
584 : ///
585 : /// Return Err(_) if the remote index_part hasn't been downloaded yet, or the timeline hasn't been stopped yet.
586 4 : pub(crate) fn archived_at_stopped_queue(
587 4 : &self,
588 4 : ) -> Result<Option<NaiveDateTime>, UploadQueueNotReadyError> {
589 4 : self.upload_queue
590 4 : .lock()
591 4 : .unwrap()
592 4 : .stopped_mut()
593 4 : .map(|q| q.upload_queue_for_deletion.clean.0.archived_at)
594 4 : .map_err(|_| UploadQueueNotReadyError)
595 4 : }
596 :
597 3861 : fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
598 3861 : let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
599 2977 : current_remote_index_part
600 2977 : .layer_metadata
601 2977 : .values()
602 35808 : .map(|ilmd| ilmd.file_size)
603 2977 : .sum()
604 : } else {
605 884 : 0
606 : };
607 3861 : self.metrics.remote_physical_size_gauge.set(size);
608 3861 : }
609 :
610 4 : pub fn get_remote_physical_size(&self) -> u64 {
611 4 : self.metrics.remote_physical_size_gauge.get()
612 4 : }
613 :
614 : //
615 : // Download operations.
616 : //
617 : // These don't use the per-timeline queue. They do use the global semaphore in
618 : // S3Bucket, to limit the total number of concurrent operations, though.
619 : //
620 :
621 : /// Download index file
622 40 : pub async fn download_index_file(
623 40 : &self,
624 40 : cancel: &CancellationToken,
625 40 : ) -> Result<MaybeDeletedIndexPart, DownloadError> {
626 40 : let _unfinished_gauge_guard = self.metrics.call_begin(
627 40 : &RemoteOpFileKind::Index,
628 40 : &RemoteOpKind::Download,
629 40 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
630 40 : reason: "no need for a downloads gauge",
631 40 : },
632 40 : );
633 :
634 40 : let (index_part, index_generation, index_last_modified) = download::download_index_part(
635 40 : &self.storage_impl,
636 40 : &self.tenant_shard_id,
637 40 : &self.timeline_id,
638 40 : self.generation,
639 40 : cancel,
640 40 : )
641 40 : .measure_remote_op(
642 40 : RemoteOpFileKind::Index,
643 40 : RemoteOpKind::Download,
644 40 : Arc::clone(&self.metrics),
645 40 : )
646 40 : .await?;
647 :
648 : // Defense in depth: monotonicity of generation numbers is an important correctness guarantee, so when we see a very
649 : // old index, we do extra checks in case this is the result of backward time-travel of the generation number (e.g.
650 : // in case of a bug in the service that issues generation numbers). Indices are allowed to be old, but we expect that
651 : // when we load an old index we are loading the _latest_ index: if we are asked to load an old index and there is
652 : // also a newer index available, that is surprising.
653 : const INDEX_AGE_CHECKS_THRESHOLD: Duration = Duration::from_secs(14 * 24 * 3600);
654 40 : let index_age = index_last_modified.elapsed().unwrap_or_else(|e| {
655 0 : if e.duration() > Duration::from_secs(5) {
656 : // We only warn if the S3 clock and our local clock are >5s out: because this is a low resolution
657 : // timestamp, it is common to be out by at least 1 second.
658 0 : tracing::warn!("Index has modification time in the future: {e}");
659 0 : }
660 0 : Duration::ZERO
661 40 : });
662 40 : if index_age > INDEX_AGE_CHECKS_THRESHOLD {
663 0 : tracing::info!(
664 : ?index_generation,
665 0 : age = index_age.as_secs_f64(),
666 0 : "Loaded an old index, checking for other indices..."
667 : );
668 :
669 : // Find the highest-generation index
670 0 : let (_latest_index_part, latest_index_generation, latest_index_mtime) =
671 0 : download::download_index_part(
672 0 : &self.storage_impl,
673 0 : &self.tenant_shard_id,
674 0 : &self.timeline_id,
675 0 : Generation::MAX,
676 0 : cancel,
677 0 : )
678 0 : .await?;
679 :
680 0 : if latest_index_generation > index_generation {
681 : // Unexpected! Why are we loading such an old index if a more recent one exists?
682 : // We will refuse to proceed, as there is no reasonable scenario where this should happen, but
683 : // there _is_ a clear bug/corruption scenario where it would happen (controller sets the generation
684 : // backwards).
685 0 : tracing::error!(
686 : ?index_generation,
687 : ?latest_index_generation,
688 : ?latest_index_mtime,
689 0 : "Found a newer index while loading an old one"
690 : );
691 0 : return Err(DownloadError::Fatal(
692 0 : "Index age exceeds threshold and a newer index exists".into(),
693 0 : ));
694 0 : }
695 40 : }
696 :
697 40 : if index_part.deleted_at.is_some() {
698 0 : Ok(MaybeDeletedIndexPart::Deleted(index_part))
699 : } else {
700 40 : Ok(MaybeDeletedIndexPart::IndexPart(index_part))
701 : }
702 40 : }
703 :
704 : /// Download a (layer) file from `path`, into local filesystem.
705 : ///
706 : /// 'layer_metadata' is the metadata from the remote index file.
707 : ///
708 : /// On success, returns the size of the downloaded file.
709 28 : pub async fn download_layer_file(
710 28 : &self,
711 28 : layer_file_name: &LayerName,
712 28 : layer_metadata: &LayerFileMetadata,
713 28 : local_path: &Utf8Path,
714 28 : gate: &utils::sync::gate::Gate,
715 28 : cancel: &CancellationToken,
716 28 : ctx: &RequestContext,
717 28 : ) -> Result<u64, DownloadError> {
718 28 : let downloaded_size = {
719 28 : let _unfinished_gauge_guard = self.metrics.call_begin(
720 28 : &RemoteOpFileKind::Layer,
721 28 : &RemoteOpKind::Download,
722 28 : crate::metrics::RemoteTimelineClientMetricsCallTrackSize::DontTrackSize {
723 28 : reason: "no need for a downloads gauge",
724 28 : },
725 28 : );
726 28 : download::download_layer_file(
727 28 : self.conf,
728 28 : &self.storage_impl,
729 28 : self.tenant_shard_id,
730 28 : self.timeline_id,
731 28 : layer_file_name,
732 28 : layer_metadata,
733 28 : local_path,
734 28 : gate,
735 28 : cancel,
736 28 : ctx,
737 28 : )
738 28 : .measure_remote_op(
739 28 : RemoteOpFileKind::Layer,
740 28 : RemoteOpKind::Download,
741 28 : Arc::clone(&self.metrics),
742 28 : )
743 28 : .await?
744 : };
745 :
746 28 : REMOTE_ONDEMAND_DOWNLOADED_LAYERS.inc();
747 28 : REMOTE_ONDEMAND_DOWNLOADED_BYTES.inc_by(downloaded_size);
748 28 :
749 28 : Ok(downloaded_size)
750 28 : }
751 :
752 : //
753 : // Upload operations.
754 : //
755 :
756 : /// Launch an index-file upload operation in the background, with
757 : /// fully updated metadata.
758 : ///
759 : /// This should only be used to upload initial metadata to remote storage.
760 : ///
761 : /// The upload will be added to the queue immediately, but it
762 : /// won't be performed until all previously scheduled layer file
763 : /// upload operations have completed successfully. This is to
764 : /// ensure that when the index file claims that layers X, Y and Z
765 : /// exist in remote storage, they really do. To wait for the upload
766 : /// to complete, use `wait_completion`.
767 : ///
768 : /// If there were any changes to the list of files, i.e. if any
769 : /// layer file uploads were scheduled, since the last index file
770 : /// upload, those will be included too.
771 460 : pub fn schedule_index_upload_for_full_metadata_update(
772 460 : self: &Arc<Self>,
773 460 : metadata: &TimelineMetadata,
774 460 : ) -> anyhow::Result<()> {
775 460 : let mut guard = self.upload_queue.lock().unwrap();
776 460 : let upload_queue = guard.initialized_mut()?;
777 :
778 : // As documented in the struct definition, it's ok for latest_metadata to be
779 : // ahead of what's _actually_ on the remote during index upload.
780 460 : upload_queue.dirty.metadata = metadata.clone();
781 460 :
782 460 : self.schedule_index_upload(upload_queue);
783 460 :
784 460 : Ok(())
785 460 : }
786 :
787 : /// Launch an index-file upload operation in the background, with only parts of the metadata
788 : /// updated.
789 : ///
790 : /// This is the regular way of updating metadata on layer flushes or Gc.
791 : ///
792 : /// Using this lighter update mechanism allows for reparenting and detaching without changes to
793 : /// `index_part.json`, while being more clear on what values update regularly.
794 2448 : pub(crate) fn schedule_index_upload_for_metadata_update(
795 2448 : self: &Arc<Self>,
796 2448 : update: &MetadataUpdate,
797 2448 : ) -> anyhow::Result<()> {
798 2448 : let mut guard = self.upload_queue.lock().unwrap();
799 2448 : let upload_queue = guard.initialized_mut()?;
800 :
801 2448 : upload_queue.dirty.metadata.apply(update);
802 2448 :
803 2448 : // Defense in depth: if we somehow generated invalid metadata, do not persist it.
804 2448 : upload_queue
805 2448 : .dirty
806 2448 : .validate()
807 2448 : .map_err(|e| anyhow::anyhow!(e))?;
808 :
809 2448 : self.schedule_index_upload(upload_queue);
810 2448 :
811 2448 : Ok(())
812 2448 : }
813 :
814 : /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
815 : ///
816 : /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
817 : /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
818 : /// been in the queue yet.
819 4 : pub(crate) fn schedule_index_upload_for_timeline_archival_state(
820 4 : self: &Arc<Self>,
821 4 : state: TimelineArchivalState,
822 4 : ) -> anyhow::Result<bool> {
823 4 : let mut guard = self.upload_queue.lock().unwrap();
824 4 : let upload_queue = guard.initialized_mut()?;
825 :
826 : /// Returns Some(_) if a change is needed, and Some(true) if it's a
827 : /// change needed to set archived_at.
828 8 : fn need_change(
829 8 : archived_at: &Option<NaiveDateTime>,
830 8 : state: TimelineArchivalState,
831 8 : ) -> Option<bool> {
832 8 : match (archived_at, state) {
833 : (Some(_), TimelineArchivalState::Archived)
834 : | (None, TimelineArchivalState::Unarchived) => {
835 : // Nothing to do
836 0 : tracing::info!("intended state matches present state");
837 0 : None
838 : }
839 8 : (None, TimelineArchivalState::Archived) => Some(true),
840 0 : (Some(_), TimelineArchivalState::Unarchived) => Some(false),
841 : }
842 8 : }
843 4 : let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
844 :
845 4 : if let Some(archived_at_set) = need_upload_scheduled {
846 4 : let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
847 4 : upload_queue.dirty.archived_at = intended_archived_at;
848 4 : self.schedule_index_upload(upload_queue);
849 4 : }
850 :
851 4 : let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
852 4 : Ok(need_wait)
853 4 : }
854 :
855 : /// Shuts the timeline client down, but only if the timeline is archived.
856 : ///
857 : /// This function and [`Self::schedule_index_upload_for_timeline_archival_state`] use the
858 : /// same lock to prevent races between unarchival and offloading: unarchival requires the
859 : /// upload queue to be initialized, and leaves behind an upload queue where either dirty
860 : /// or clean has archived_at of `None`. offloading leaves behind an uninitialized upload
861 : /// queue.
862 4 : pub(crate) async fn shutdown_if_archived(
863 4 : self: &Arc<Self>,
864 4 : ) -> Result<(), ShutdownIfArchivedError> {
865 4 : {
866 4 : let mut guard = self.upload_queue.lock().unwrap();
867 4 : let upload_queue = guard
868 4 : .initialized_mut()
869 4 : .map_err(ShutdownIfArchivedError::NotInitialized)?;
870 :
871 4 : match (
872 4 : upload_queue.dirty.archived_at.is_none(),
873 4 : upload_queue.clean.0.archived_at.is_none(),
874 4 : ) {
875 : // The expected case: the timeline is archived and we don't want to unarchive
876 4 : (false, false) => {}
877 : (true, false) => {
878 0 : tracing::info!("can't shut down timeline: timeline slated for unarchival");
879 0 : return Err(ShutdownIfArchivedError::NotArchived);
880 : }
881 0 : (dirty_archived, true) => {
882 0 : tracing::info!(%dirty_archived, "can't shut down timeline: timeline not archived in remote storage");
883 0 : return Err(ShutdownIfArchivedError::NotArchived);
884 : }
885 : }
886 :
887 : // Set the shutting_down flag while the guard from the archival check is held.
888 : // This prevents a race with unarchival, as initialized_mut will not return
889 : // an upload queue from this point.
890 : // Also launch the queued tasks like shutdown() does.
891 4 : if !upload_queue.shutting_down {
892 4 : upload_queue.shutting_down = true;
893 4 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
894 4 : // this operation is not counted similar to Barrier
895 4 : self.launch_queued_tasks(upload_queue);
896 4 : }
897 : }
898 :
899 4 : self.shutdown().await;
900 :
901 4 : Ok(())
902 4 : }
903 :
904 : /// Launch an index-file upload operation in the background, setting `import_pgdata` field.
905 0 : pub(crate) fn schedule_index_upload_for_import_pgdata_state_update(
906 0 : self: &Arc<Self>,
907 0 : state: Option<import_pgdata::index_part_format::Root>,
908 0 : ) -> anyhow::Result<()> {
909 0 : let mut guard = self.upload_queue.lock().unwrap();
910 0 : let upload_queue = guard.initialized_mut()?;
911 0 : upload_queue.dirty.import_pgdata = state;
912 0 : self.schedule_index_upload(upload_queue);
913 0 : Ok(())
914 0 : }
915 :
916 : ///
917 : /// Launch an index-file upload operation in the background, if necessary.
918 : ///
919 : /// Use this function to schedule the update of the index file after
920 : /// scheduling file uploads or deletions. If no file uploads or deletions
921 : /// have been scheduled since the last index file upload, this does
922 : /// nothing.
923 : ///
924 : /// Like schedule_index_upload_for_metadata_update(), this merely adds
925 : /// the upload to the upload queue and returns quickly.
926 740 : pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
927 740 : let mut guard = self.upload_queue.lock().unwrap();
928 740 : let upload_queue = guard.initialized_mut()?;
929 :
930 740 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
931 28 : self.schedule_index_upload(upload_queue);
932 712 : }
933 :
934 740 : Ok(())
935 740 : }
936 :
937 : /// Launch an index-file upload operation in the background (internal function)
938 3084 : fn schedule_index_upload(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
939 3084 : let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
940 3084 : // fix up the duplicated field
941 3084 : upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
942 3084 :
943 3084 : // make sure it serializes before doing it in perform_upload_task so that it doesn't
944 3084 : // look like a retryable error
945 3084 : let void = std::io::sink();
946 3084 : serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
947 3084 :
948 3084 : let index_part = &upload_queue.dirty;
949 3084 :
950 3084 : info!(
951 0 : "scheduling metadata upload up to consistent LSN {disk_consistent_lsn} with {} files ({} changed)",
952 0 : index_part.layer_metadata.len(),
953 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
954 : );
955 :
956 3084 : let op = UploadOp::UploadMetadata {
957 3084 : uploaded: Box::new(index_part.clone()),
958 3084 : };
959 3084 : self.metric_begin(&op);
960 3084 : upload_queue.queued_operations.push_back(op);
961 3084 : upload_queue.latest_files_changes_since_metadata_upload_scheduled = 0;
962 3084 :
963 3084 : // Launch the task immediately, if possible
964 3084 : self.launch_queued_tasks(upload_queue);
965 3084 : }
966 :
967 : /// Reparent this timeline to a new parent.
968 : ///
969 : /// A retryable step of timeline ancestor detach.
970 0 : pub(crate) async fn schedule_reparenting_and_wait(
971 0 : self: &Arc<Self>,
972 0 : new_parent: &TimelineId,
973 0 : ) -> anyhow::Result<()> {
974 0 : let receiver = {
975 0 : let mut guard = self.upload_queue.lock().unwrap();
976 0 : let upload_queue = guard.initialized_mut()?;
977 :
978 0 : let Some(prev) = upload_queue.dirty.metadata.ancestor_timeline() else {
979 0 : return Err(anyhow::anyhow!(
980 0 : "cannot reparent without a current ancestor"
981 0 : ));
982 : };
983 :
984 0 : let uploaded = &upload_queue.clean.0.metadata;
985 0 :
986 0 : if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
987 : // nothing to do
988 0 : None
989 : } else {
990 0 : upload_queue.dirty.metadata.reparent(new_parent);
991 0 : upload_queue.dirty.lineage.record_previous_ancestor(&prev);
992 0 :
993 0 : self.schedule_index_upload(upload_queue);
994 0 :
995 0 : Some(self.schedule_barrier0(upload_queue))
996 : }
997 : };
998 :
999 0 : if let Some(receiver) = receiver {
1000 0 : Self::wait_completion0(receiver).await?;
1001 0 : }
1002 0 : Ok(())
1003 0 : }
1004 :
1005 : /// Schedules uploading a new version of `index_part.json` with the given layers added,
1006 : /// detaching from ancestor and waits for it to complete.
1007 : ///
1008 : /// This is used with `Timeline::detach_ancestor` functionality.
1009 0 : pub(crate) async fn schedule_adding_existing_layers_to_index_detach_and_wait(
1010 0 : self: &Arc<Self>,
1011 0 : layers: &[Layer],
1012 0 : adopted: (TimelineId, Lsn),
1013 0 : ) -> anyhow::Result<()> {
1014 0 : let barrier = {
1015 0 : let mut guard = self.upload_queue.lock().unwrap();
1016 0 : let upload_queue = guard.initialized_mut()?;
1017 :
1018 0 : if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
1019 0 : None
1020 : } else {
1021 0 : upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
1022 0 : upload_queue.dirty.lineage.record_detaching(&adopted);
1023 :
1024 0 : for layer in layers {
1025 0 : let prev = upload_queue
1026 0 : .dirty
1027 0 : .layer_metadata
1028 0 : .insert(layer.layer_desc().layer_name(), layer.metadata());
1029 0 : assert!(prev.is_none(), "copied layer existed already {layer}");
1030 : }
1031 :
1032 0 : self.schedule_index_upload(upload_queue);
1033 0 :
1034 0 : Some(self.schedule_barrier0(upload_queue))
1035 : }
1036 : };
1037 :
1038 0 : if let Some(barrier) = barrier {
1039 0 : Self::wait_completion0(barrier).await?;
1040 0 : }
1041 0 : Ok(())
1042 0 : }
1043 :
1044 : /// Adds a gc blocking reason for this timeline if one does not exist already.
1045 : ///
1046 : /// A retryable step of timeline detach ancestor.
1047 : ///
1048 : /// Returns a future which waits until the completion of the upload.
1049 0 : pub(crate) fn schedule_insert_gc_block_reason(
1050 0 : self: &Arc<Self>,
1051 0 : reason: index::GcBlockingReason,
1052 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
1053 0 : {
1054 0 : let maybe_barrier = {
1055 0 : let mut guard = self.upload_queue.lock().unwrap();
1056 0 : let upload_queue = guard.initialized_mut()?;
1057 :
1058 0 : if let index::GcBlockingReason::DetachAncestor = reason {
1059 0 : if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
1060 0 : drop(guard);
1061 0 : panic!("cannot start detach ancestor if there is nothing to detach from");
1062 0 : }
1063 0 : }
1064 :
1065 0 : let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
1066 :
1067 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
1068 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
1069 0 :
1070 0 : match (current, uploaded) {
1071 0 : (x, y) if wanted(x) && wanted(y) => None,
1072 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
1073 : // Usual case: !wanted(x) && !wanted(y)
1074 : //
1075 : // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
1076 : // turn on and off some reason.
1077 0 : (x, y) => {
1078 0 : if !wanted(x) && wanted(y) {
1079 : // this could be avoided by having external in-memory synchronization, like
1080 : // timeline detach ancestor
1081 0 : warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
1082 0 : }
1083 :
1084 : // at this point, the metadata must always show that there is a parent
1085 0 : upload_queue.dirty.gc_blocking = current
1086 0 : .map(|x| x.with_reason(reason))
1087 0 : .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
1088 0 : self.schedule_index_upload(upload_queue);
1089 0 : Some(self.schedule_barrier0(upload_queue))
1090 : }
1091 : }
1092 : };
1093 :
1094 0 : Ok(async move {
1095 0 : if let Some(barrier) = maybe_barrier {
1096 0 : Self::wait_completion0(barrier).await?;
1097 0 : }
1098 0 : Ok(())
1099 0 : })
1100 0 : }
1101 :
1102 : /// Removes a gc blocking reason for this timeline if one exists.
1103 : ///
1104 : /// A retryable step of timeline detach ancestor.
1105 : ///
1106 : /// Returns a future which waits until the completion of the upload.
1107 0 : pub(crate) fn schedule_remove_gc_block_reason(
1108 0 : self: &Arc<Self>,
1109 0 : reason: index::GcBlockingReason,
1110 0 : ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
1111 0 : {
1112 0 : let maybe_barrier = {
1113 0 : let mut guard = self.upload_queue.lock().unwrap();
1114 0 : let upload_queue = guard.initialized_mut()?;
1115 :
1116 0 : if let index::GcBlockingReason::DetachAncestor = reason {
1117 0 : if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
1118 0 : drop(guard);
1119 0 : panic!("cannot complete timeline_ancestor_detach while not detached");
1120 0 : }
1121 0 : }
1122 :
1123 0 : let wanted = |x: Option<&index::GcBlocking>| {
1124 0 : x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
1125 0 : };
1126 :
1127 0 : let current = upload_queue.dirty.gc_blocking.as_ref();
1128 0 : let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
1129 0 :
1130 0 : match (current, uploaded) {
1131 0 : (x, y) if wanted(x) && wanted(y) => None,
1132 0 : (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
1133 0 : (x, y) => {
1134 0 : if !wanted(x) && wanted(y) {
1135 0 : warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
1136 0 : }
1137 :
1138 0 : upload_queue.dirty.gc_blocking =
1139 0 : current.as_ref().and_then(|x| x.without_reason(reason));
1140 0 : assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
1141 0 : self.schedule_index_upload(upload_queue);
1142 0 : Some(self.schedule_barrier0(upload_queue))
1143 : }
1144 : }
1145 : };
1146 :
1147 0 : Ok(async move {
1148 0 : if let Some(barrier) = maybe_barrier {
1149 0 : Self::wait_completion0(barrier).await?;
1150 0 : }
1151 0 : Ok(())
1152 0 : })
1153 0 : }
1154 :
1155 : /// Launch an upload operation in the background; the file is added to be included in next
1156 : /// `index_part.json` upload.
1157 2744 : pub(crate) fn schedule_layer_file_upload(
1158 2744 : self: &Arc<Self>,
1159 2744 : layer: ResidentLayer,
1160 2744 : ) -> Result<(), NotInitialized> {
1161 2744 : let mut guard = self.upload_queue.lock().unwrap();
1162 2744 : let upload_queue = guard.initialized_mut()?;
1163 :
1164 2744 : self.schedule_layer_file_upload0(upload_queue, layer);
1165 2744 : self.launch_queued_tasks(upload_queue);
1166 2744 : Ok(())
1167 2744 : }
1168 :
1169 3468 : fn schedule_layer_file_upload0(
1170 3468 : self: &Arc<Self>,
1171 3468 : upload_queue: &mut UploadQueueInitialized,
1172 3468 : layer: ResidentLayer,
1173 3468 : ) {
1174 3468 : let metadata = layer.metadata();
1175 3468 :
1176 3468 : upload_queue
1177 3468 : .dirty
1178 3468 : .layer_metadata
1179 3468 : .insert(layer.layer_desc().layer_name(), metadata.clone());
1180 3468 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1181 3468 :
1182 3468 : info!(
1183 : gen=?metadata.generation,
1184 : shard=?metadata.shard,
1185 0 : "scheduled layer file upload {layer}",
1186 : );
1187 :
1188 3468 : let op = UploadOp::UploadLayer(layer, metadata, None);
1189 3468 : self.metric_begin(&op);
1190 3468 : upload_queue.queued_operations.push_back(op);
1191 3468 : }
1192 :
1193 : /// Launch a delete operation in the background.
1194 : ///
1195 : /// The operation does not modify local filesystem state.
1196 : ///
1197 : /// Note: This schedules an index file upload before the deletions. The
1198 : /// deletion won't actually be performed, until all previously scheduled
1199 : /// upload operations, and the index file upload, have completed
1200 : /// successfully.
1201 16 : pub fn schedule_layer_file_deletion(
1202 16 : self: &Arc<Self>,
1203 16 : names: &[LayerName],
1204 16 : ) -> anyhow::Result<()> {
1205 16 : let mut guard = self.upload_queue.lock().unwrap();
1206 16 : let upload_queue = guard.initialized_mut()?;
1207 :
1208 16 : let with_metadata =
1209 16 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names.iter().cloned());
1210 16 :
1211 16 : self.schedule_deletion_of_unlinked0(upload_queue, with_metadata);
1212 16 :
1213 16 : // Launch the tasks immediately, if possible
1214 16 : self.launch_queued_tasks(upload_queue);
1215 16 : Ok(())
1216 16 : }
1217 :
1218 : /// Unlinks the layer files from `index_part.json` but does not yet schedule deletion for the
1219 : /// layer files, leaving them dangling.
1220 : ///
1221 : /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
1222 : /// is invoked on them.
1223 8 : pub(crate) fn schedule_gc_update(
1224 8 : self: &Arc<Self>,
1225 8 : gc_layers: &[Layer],
1226 8 : ) -> Result<(), NotInitialized> {
1227 8 : let mut guard = self.upload_queue.lock().unwrap();
1228 8 : let upload_queue = guard.initialized_mut()?;
1229 :
1230 : // just forget the return value; after uploading the next index_part.json, we can consider
1231 : // the layer files as "dangling". this is fine, at worst case we create work for the
1232 : // scrubber.
1233 :
1234 8 : let names = gc_layers.iter().map(|x| x.layer_desc().layer_name());
1235 8 :
1236 8 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
1237 8 :
1238 8 : self.launch_queued_tasks(upload_queue);
1239 8 :
1240 8 : Ok(())
1241 8 : }
1242 :
1243 : /// Update the remote index file, removing the to-be-deleted files from the index,
1244 : /// allowing scheduling of actual deletions later.
1245 176 : fn schedule_unlinking_of_layers_from_index_part0<I>(
1246 176 : self: &Arc<Self>,
1247 176 : upload_queue: &mut UploadQueueInitialized,
1248 176 : names: I,
1249 176 : ) -> Vec<(LayerName, LayerFileMetadata)>
1250 176 : where
1251 176 : I: IntoIterator<Item = LayerName>,
1252 176 : {
1253 176 : // Decorate our list of names with each name's metadata, dropping
1254 176 : // names that are unexpectedly missing from our metadata. This metadata
1255 176 : // is later used when physically deleting layers, to construct key paths.
1256 176 : let with_metadata: Vec<_> = names
1257 176 : .into_iter()
1258 1024 : .filter_map(|name| {
1259 1024 : let meta = upload_queue.dirty.layer_metadata.remove(&name);
1260 :
1261 1024 : if let Some(meta) = meta {
1262 1024 : upload_queue.latest_files_changes_since_metadata_upload_scheduled += 1;
1263 1024 : Some((name, meta))
1264 : } else {
1265 : // This can only happen if we forgot to to schedule the file upload
1266 : // before scheduling the delete. Log it because it is a rare/strange
1267 : // situation, and in case something is misbehaving, we'd like to know which
1268 : // layers experienced this.
1269 0 : info!("Deleting layer {name} not found in latest_files list, never uploaded?");
1270 0 : None
1271 : }
1272 1024 : })
1273 176 : .collect();
1274 :
1275 : #[cfg(feature = "testing")]
1276 1200 : for (name, metadata) in &with_metadata {
1277 1024 : let gen = metadata.generation;
1278 1024 : if let Some(unexpected) = upload_queue.dangling_files.insert(name.to_owned(), gen) {
1279 0 : if unexpected == gen {
1280 0 : tracing::error!("{name} was unlinked twice with same generation");
1281 : } else {
1282 0 : tracing::error!("{name} was unlinked twice with different generations {gen:?} and {unexpected:?}");
1283 : }
1284 1024 : }
1285 : }
1286 :
1287 : // after unlinking files from the upload_queue.latest_files we must always schedule an
1288 : // index_part update, because that needs to be uploaded before we can actually delete the
1289 : // files.
1290 176 : if upload_queue.latest_files_changes_since_metadata_upload_scheduled > 0 {
1291 144 : self.schedule_index_upload(upload_queue);
1292 144 : }
1293 :
1294 176 : with_metadata
1295 176 : }
1296 :
1297 : /// Schedules deletion for layer files which have previously been unlinked from the
1298 : /// `index_part.json` with [`Self::schedule_gc_update`] or [`Self::schedule_compaction_update`].
1299 1003 : pub(crate) fn schedule_deletion_of_unlinked(
1300 1003 : self: &Arc<Self>,
1301 1003 : layers: Vec<(LayerName, LayerFileMetadata)>,
1302 1003 : ) -> anyhow::Result<()> {
1303 1003 : let mut guard = self.upload_queue.lock().unwrap();
1304 1003 : let upload_queue = guard.initialized_mut()?;
1305 :
1306 1003 : self.schedule_deletion_of_unlinked0(upload_queue, layers);
1307 1003 : self.launch_queued_tasks(upload_queue);
1308 1003 : Ok(())
1309 1003 : }
1310 :
1311 1015 : fn schedule_deletion_of_unlinked0(
1312 1015 : self: &Arc<Self>,
1313 1015 : upload_queue: &mut UploadQueueInitialized,
1314 1015 : mut with_metadata: Vec<(LayerName, LayerFileMetadata)>,
1315 1015 : ) {
1316 1015 : // Filter out any layers which were not created by this tenant shard. These are
1317 1015 : // layers that originate from some ancestor shard after a split, and may still
1318 1015 : // be referenced by other shards. We are free to delete them locally and remove
1319 1015 : // them from our index (and would have already done so when we reach this point
1320 1015 : // in the code), but we may not delete them remotely.
1321 1015 : with_metadata.retain(|(name, meta)| {
1322 1003 : let retain = meta.shard.shard_number == self.tenant_shard_id.shard_number
1323 1003 : && meta.shard.shard_count == self.tenant_shard_id.shard_count;
1324 1003 : if !retain {
1325 0 : tracing::debug!(
1326 0 : "Skipping deletion of ancestor-shard layer {name}, from shard {}",
1327 : meta.shard
1328 : );
1329 1003 : }
1330 1003 : retain
1331 1015 : });
1332 :
1333 2018 : for (name, meta) in &with_metadata {
1334 1003 : info!(
1335 0 : "scheduling deletion of layer {}{} (shard {})",
1336 0 : name,
1337 0 : meta.generation.get_suffix(),
1338 : meta.shard
1339 : );
1340 : }
1341 :
1342 : #[cfg(feature = "testing")]
1343 2018 : for (name, meta) in &with_metadata {
1344 1003 : let gen = meta.generation;
1345 1003 : match upload_queue.dangling_files.remove(name) {
1346 995 : Some(same) if same == gen => { /* expected */ }
1347 0 : Some(other) => {
1348 0 : tracing::error!("{name} was unlinked with {other:?} but deleted with {gen:?}");
1349 : }
1350 : None => {
1351 8 : tracing::error!("{name} was unlinked but was not dangling");
1352 : }
1353 : }
1354 : }
1355 :
1356 : // schedule the actual deletions
1357 1015 : if with_metadata.is_empty() {
1358 : // avoid scheduling the op & bumping the metric
1359 12 : return;
1360 1003 : }
1361 1003 : let op = UploadOp::Delete(Delete {
1362 1003 : layers: with_metadata,
1363 1003 : });
1364 1003 : self.metric_begin(&op);
1365 1003 : upload_queue.queued_operations.push_back(op);
1366 1015 : }
1367 :
1368 : /// Schedules a compaction update to the remote `index_part.json`.
1369 : ///
1370 : /// `compacted_from` represent the L0 names which have been `compacted_to` L1 layers.
1371 152 : pub(crate) fn schedule_compaction_update(
1372 152 : self: &Arc<Self>,
1373 152 : compacted_from: &[Layer],
1374 152 : compacted_to: &[ResidentLayer],
1375 152 : ) -> Result<(), NotInitialized> {
1376 152 : let mut guard = self.upload_queue.lock().unwrap();
1377 152 : let upload_queue = guard.initialized_mut()?;
1378 :
1379 876 : for layer in compacted_to {
1380 724 : self.schedule_layer_file_upload0(upload_queue, layer.clone());
1381 724 : }
1382 :
1383 1012 : let names = compacted_from.iter().map(|x| x.layer_desc().layer_name());
1384 152 :
1385 152 : self.schedule_unlinking_of_layers_from_index_part0(upload_queue, names);
1386 152 : self.launch_queued_tasks(upload_queue);
1387 152 :
1388 152 : Ok(())
1389 152 : }
1390 :
1391 : /// Wait for all previously scheduled uploads/deletions to complete
1392 2780 : pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
1393 2780 : let receiver = {
1394 2780 : let mut guard = self.upload_queue.lock().unwrap();
1395 2780 : let upload_queue = guard
1396 2780 : .initialized_mut()
1397 2780 : .map_err(WaitCompletionError::NotInitialized)?;
1398 2780 : self.schedule_barrier0(upload_queue)
1399 2780 : };
1400 2780 :
1401 2780 : Self::wait_completion0(receiver).await
1402 2780 : }
1403 :
1404 2780 : async fn wait_completion0(
1405 2780 : mut receiver: tokio::sync::watch::Receiver<()>,
1406 2780 : ) -> Result<(), WaitCompletionError> {
1407 2780 : if receiver.changed().await.is_err() {
1408 0 : return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
1409 2780 : }
1410 2780 :
1411 2780 : Ok(())
1412 2780 : }
1413 :
1414 12 : pub(crate) fn schedule_barrier(self: &Arc<Self>) -> anyhow::Result<()> {
1415 12 : let mut guard = self.upload_queue.lock().unwrap();
1416 12 : let upload_queue = guard.initialized_mut()?;
1417 12 : self.schedule_barrier0(upload_queue);
1418 12 : Ok(())
1419 12 : }
1420 :
1421 2792 : fn schedule_barrier0(
1422 2792 : self: &Arc<Self>,
1423 2792 : upload_queue: &mut UploadQueueInitialized,
1424 2792 : ) -> tokio::sync::watch::Receiver<()> {
1425 2792 : let (sender, receiver) = tokio::sync::watch::channel(());
1426 2792 : let barrier_op = UploadOp::Barrier(sender);
1427 2792 :
1428 2792 : upload_queue.queued_operations.push_back(barrier_op);
1429 2792 : // Don't count this kind of operation!
1430 2792 :
1431 2792 : // Launch the task immediately, if possible
1432 2792 : self.launch_queued_tasks(upload_queue);
1433 2792 :
1434 2792 : receiver
1435 2792 : }
1436 :
1437 : /// Wait for all previously scheduled operations to complete, and then stop.
1438 : ///
1439 : /// Not cancellation safe
1440 20 : pub(crate) async fn shutdown(self: &Arc<Self>) {
1441 20 : // On cancellation the queue is left in ackward state of refusing new operations but
1442 20 : // proper stop is yet to be called. On cancel the original or some later task must call
1443 20 : // `stop` or `shutdown`.
1444 20 : let sg = scopeguard::guard((), |_| {
1445 0 : tracing::error!("RemoteTimelineClient::shutdown was cancelled; this should not happen, do not make this into an allowed_error")
1446 20 : });
1447 :
1448 16 : let fut = {
1449 20 : let mut guard = self.upload_queue.lock().unwrap();
1450 20 : let upload_queue = match &mut *guard {
1451 : UploadQueue::Stopped(_) => {
1452 4 : scopeguard::ScopeGuard::into_inner(sg);
1453 4 : return;
1454 : }
1455 : UploadQueue::Uninitialized => {
1456 : // transition into Stopped state
1457 0 : self.stop_impl(&mut guard);
1458 0 : scopeguard::ScopeGuard::into_inner(sg);
1459 0 : return;
1460 : }
1461 16 : UploadQueue::Initialized(ref mut init) => init,
1462 16 : };
1463 16 :
1464 16 : // if the queue is already stuck due to a shutdown operation which was cancelled, then
1465 16 : // just don't add more of these as they would never complete.
1466 16 : //
1467 16 : // TODO: if launch_queued_tasks were to be refactored to accept a &mut UploadQueue
1468 16 : // in every place we would not have to jump through this hoop, and this method could be
1469 16 : // made cancellable.
1470 16 : if !upload_queue.shutting_down {
1471 12 : upload_queue.shutting_down = true;
1472 12 : upload_queue.queued_operations.push_back(UploadOp::Shutdown);
1473 12 : // this operation is not counted similar to Barrier
1474 12 :
1475 12 : self.launch_queued_tasks(upload_queue);
1476 12 : }
1477 :
1478 16 : upload_queue.shutdown_ready.clone().acquire_owned()
1479 : };
1480 :
1481 16 : let res = fut.await;
1482 :
1483 16 : scopeguard::ScopeGuard::into_inner(sg);
1484 16 :
1485 16 : match res {
1486 0 : Ok(_permit) => unreachable!("shutdown_ready should not have been added permits"),
1487 16 : Err(_closed) => {
1488 16 : // expected
1489 16 : }
1490 16 : }
1491 16 :
1492 16 : self.stop();
1493 20 : }
1494 :
1495 : /// Set the deleted_at field in the remote index file.
1496 : ///
1497 : /// This fails if the upload queue has not been `stop()`ed.
1498 : ///
1499 : /// The caller is responsible for calling `stop()` AND for waiting
1500 : /// for any ongoing upload tasks to finish after `stop()` has succeeded.
1501 : /// Check method [`RemoteTimelineClient::stop`] for details.
1502 : #[instrument(skip_all)]
1503 : pub(crate) async fn persist_index_part_with_deleted_flag(
1504 : self: &Arc<Self>,
1505 : ) -> Result<(), PersistIndexPartWithDeletedFlagError> {
1506 : let index_part_with_deleted_at = {
1507 : let mut locked = self.upload_queue.lock().unwrap();
1508 :
1509 : // We must be in stopped state because otherwise
1510 : // we can have inprogress index part upload that can overwrite the file
1511 : // with missing is_deleted flag that we going to set below
1512 : let stopped = locked.stopped_mut()?;
1513 :
1514 : match stopped.deleted_at {
1515 : SetDeletedFlagProgress::NotRunning => (), // proceed
1516 : SetDeletedFlagProgress::InProgress(at) => {
1517 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyInProgress(at));
1518 : }
1519 : SetDeletedFlagProgress::Successful(at) => {
1520 : return Err(PersistIndexPartWithDeletedFlagError::AlreadyDeleted(at));
1521 : }
1522 : };
1523 : let deleted_at = Utc::now().naive_utc();
1524 : stopped.deleted_at = SetDeletedFlagProgress::InProgress(deleted_at);
1525 :
1526 : let mut index_part = stopped.upload_queue_for_deletion.dirty.clone();
1527 : index_part.deleted_at = Some(deleted_at);
1528 : index_part
1529 : };
1530 :
1531 0 : let undo_deleted_at = scopeguard::guard(Arc::clone(self), |self_clone| {
1532 0 : let mut locked = self_clone.upload_queue.lock().unwrap();
1533 0 : let stopped = locked
1534 0 : .stopped_mut()
1535 0 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1536 0 : stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
1537 0 : });
1538 :
1539 : pausable_failpoint!("persist_deleted_index_part");
1540 :
1541 : backoff::retry(
1542 0 : || {
1543 0 : upload::upload_index_part(
1544 0 : &self.storage_impl,
1545 0 : &self.tenant_shard_id,
1546 0 : &self.timeline_id,
1547 0 : self.generation,
1548 0 : &index_part_with_deleted_at,
1549 0 : &self.cancel,
1550 0 : )
1551 0 : },
1552 0 : |_e| false,
1553 : 1,
1554 : // have just a couple of attempts
1555 : // when executed as part of timeline deletion this happens in context of api call
1556 : // when executed as part of tenant deletion this happens in the background
1557 : 2,
1558 : "persist_index_part_with_deleted_flag",
1559 : &self.cancel,
1560 : )
1561 : .await
1562 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1563 0 : .and_then(|x| x)?;
1564 :
1565 : // all good, disarm the guard and mark as success
1566 : ScopeGuard::into_inner(undo_deleted_at);
1567 : {
1568 : let mut locked = self.upload_queue.lock().unwrap();
1569 :
1570 : let stopped = locked
1571 : .stopped_mut()
1572 : .expect("there's no way out of Stopping, and we checked it's Stopping above");
1573 : stopped.deleted_at = SetDeletedFlagProgress::Successful(
1574 : index_part_with_deleted_at
1575 : .deleted_at
1576 : .expect("we set it above"),
1577 : );
1578 : }
1579 :
1580 : Ok(())
1581 : }
1582 :
1583 0 : pub(crate) fn is_deleting(&self) -> bool {
1584 0 : let mut locked = self.upload_queue.lock().unwrap();
1585 0 : locked.stopped_mut().is_ok()
1586 0 : }
1587 :
1588 0 : pub(crate) async fn preserve_initdb_archive(
1589 0 : self: &Arc<Self>,
1590 0 : tenant_id: &TenantId,
1591 0 : timeline_id: &TimelineId,
1592 0 : cancel: &CancellationToken,
1593 0 : ) -> anyhow::Result<()> {
1594 0 : backoff::retry(
1595 0 : || async {
1596 0 : upload::preserve_initdb_archive(&self.storage_impl, tenant_id, timeline_id, cancel)
1597 0 : .await
1598 0 : },
1599 0 : TimeoutOrCancel::caused_by_cancel,
1600 0 : FAILED_DOWNLOAD_WARN_THRESHOLD,
1601 0 : FAILED_REMOTE_OP_RETRIES,
1602 0 : "preserve_initdb_tar_zst",
1603 0 : &cancel.clone(),
1604 0 : )
1605 0 : .await
1606 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1607 0 : .and_then(|x| x)
1608 0 : .context("backing up initdb archive")?;
1609 0 : Ok(())
1610 0 : }
1611 :
1612 : /// Uploads the given layer **without** adding it to be part of a future `index_part.json` upload.
1613 : ///
1614 : /// This is not normally needed.
1615 0 : pub(crate) async fn upload_layer_file(
1616 0 : self: &Arc<Self>,
1617 0 : uploaded: &ResidentLayer,
1618 0 : cancel: &CancellationToken,
1619 0 : ) -> anyhow::Result<()> {
1620 0 : let remote_path = remote_layer_path(
1621 0 : &self.tenant_shard_id.tenant_id,
1622 0 : &self.timeline_id,
1623 0 : uploaded.metadata().shard,
1624 0 : &uploaded.layer_desc().layer_name(),
1625 0 : uploaded.metadata().generation,
1626 0 : );
1627 0 :
1628 0 : backoff::retry(
1629 0 : || async {
1630 0 : upload::upload_timeline_layer(
1631 0 : &self.storage_impl,
1632 0 : uploaded.local_path(),
1633 0 : &remote_path,
1634 0 : uploaded.metadata().file_size,
1635 0 : cancel,
1636 0 : )
1637 0 : .await
1638 0 : },
1639 0 : TimeoutOrCancel::caused_by_cancel,
1640 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1641 0 : FAILED_REMOTE_OP_RETRIES,
1642 0 : "upload a layer without adding it to latest files",
1643 0 : cancel,
1644 0 : )
1645 0 : .await
1646 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1647 0 : .and_then(|x| x)
1648 0 : .context("upload a layer without adding it to latest files")
1649 0 : }
1650 :
1651 : /// Copies the `adopted` remote existing layer to the remote path of `adopted_as`. The layer is
1652 : /// not added to be part of a future `index_part.json` upload.
1653 0 : pub(crate) async fn copy_timeline_layer(
1654 0 : self: &Arc<Self>,
1655 0 : adopted: &Layer,
1656 0 : adopted_as: &Layer,
1657 0 : cancel: &CancellationToken,
1658 0 : ) -> anyhow::Result<()> {
1659 0 : let source_remote_path = remote_layer_path(
1660 0 : &self.tenant_shard_id.tenant_id,
1661 0 : &adopted
1662 0 : .get_timeline_id()
1663 0 : .expect("Source timeline should be alive"),
1664 0 : adopted.metadata().shard,
1665 0 : &adopted.layer_desc().layer_name(),
1666 0 : adopted.metadata().generation,
1667 0 : );
1668 0 :
1669 0 : let target_remote_path = remote_layer_path(
1670 0 : &self.tenant_shard_id.tenant_id,
1671 0 : &self.timeline_id,
1672 0 : adopted_as.metadata().shard,
1673 0 : &adopted_as.layer_desc().layer_name(),
1674 0 : adopted_as.metadata().generation,
1675 0 : );
1676 0 :
1677 0 : backoff::retry(
1678 0 : || async {
1679 0 : upload::copy_timeline_layer(
1680 0 : &self.storage_impl,
1681 0 : &source_remote_path,
1682 0 : &target_remote_path,
1683 0 : cancel,
1684 0 : )
1685 0 : .await
1686 0 : },
1687 0 : TimeoutOrCancel::caused_by_cancel,
1688 0 : FAILED_UPLOAD_WARN_THRESHOLD,
1689 0 : FAILED_REMOTE_OP_RETRIES,
1690 0 : "copy timeline layer",
1691 0 : cancel,
1692 0 : )
1693 0 : .await
1694 0 : .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
1695 0 : .and_then(|x| x)
1696 0 : .context("remote copy timeline layer")
1697 0 : }
1698 :
1699 0 : async fn flush_deletion_queue(&self) -> Result<(), DeletionQueueError> {
1700 0 : match tokio::time::timeout(
1701 0 : DELETION_QUEUE_FLUSH_TIMEOUT,
1702 0 : self.deletion_queue_client.flush_immediate(),
1703 0 : )
1704 0 : .await
1705 : {
1706 0 : Ok(result) => result,
1707 0 : Err(_timeout) => {
1708 0 : // Flushing remote deletions is not mandatory: we flush here to make the system easier to test, and
1709 0 : // to ensure that _usually_ objects are really gone after a DELETE is acked. However, in case of deletion
1710 0 : // queue issues (https://github.com/neondatabase/neon/issues/6440), we don't want to wait indefinitely here.
1711 0 : tracing::warn!(
1712 0 : "Timed out waiting for deletion queue flush, acking deletion anyway"
1713 : );
1714 0 : Ok(())
1715 : }
1716 : }
1717 0 : }
1718 :
1719 : /// Prerequisites: UploadQueue should be in stopped state and deleted_at should be successfuly set.
1720 : /// The function deletes layer files one by one, then lists the prefix to see if we leaked something
1721 : /// deletes leaked files if any and proceeds with deletion of index file at the end.
1722 0 : pub(crate) async fn delete_all(self: &Arc<Self>) -> Result<(), DeleteTimelineError> {
1723 0 : debug_assert_current_span_has_tenant_and_timeline_id();
1724 :
1725 0 : let layers: Vec<RemotePath> = {
1726 0 : let mut locked = self.upload_queue.lock().unwrap();
1727 0 : let stopped = locked.stopped_mut().map_err(DeleteTimelineError::Other)?;
1728 :
1729 0 : if !matches!(stopped.deleted_at, SetDeletedFlagProgress::Successful(_)) {
1730 0 : return Err(DeleteTimelineError::Other(anyhow::anyhow!(
1731 0 : "deleted_at is not set"
1732 0 : )));
1733 0 : }
1734 0 :
1735 0 : debug_assert!(stopped.upload_queue_for_deletion.no_pending_work());
1736 :
1737 0 : stopped
1738 0 : .upload_queue_for_deletion
1739 0 : .dirty
1740 0 : .layer_metadata
1741 0 : .drain()
1742 0 : .filter(|(_file_name, meta)| {
1743 0 : // Filter out layers that belonged to an ancestor shard. Since we are deleting the whole timeline from
1744 0 : // all shards anyway, we _could_ delete these, but
1745 0 : // - it creates a potential race if other shards are still
1746 0 : // using the layers while this shard deletes them.
1747 0 : // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
1748 0 : // these timelines are present but corrupt (their index exists but some layers don't)
1749 0 : //
1750 0 : // These layers will eventually be cleaned up by the scrubber when it does physical GC.
1751 0 : meta.shard.shard_number == self.tenant_shard_id.shard_number
1752 0 : && meta.shard.shard_count == self.tenant_shard_id.shard_count
1753 0 : })
1754 0 : .map(|(file_name, meta)| {
1755 0 : remote_layer_path(
1756 0 : &self.tenant_shard_id.tenant_id,
1757 0 : &self.timeline_id,
1758 0 : meta.shard,
1759 0 : &file_name,
1760 0 : meta.generation,
1761 0 : )
1762 0 : })
1763 0 : .collect()
1764 0 : };
1765 0 :
1766 0 : let layer_deletion_count = layers.len();
1767 0 : self.deletion_queue_client
1768 0 : .push_immediate(layers)
1769 0 : .await
1770 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1771 :
1772 : // Delete the initdb.tar.zst, which is not always present, but deletion attempts of
1773 : // inexistant objects are not considered errors.
1774 0 : let initdb_path =
1775 0 : remote_initdb_archive_path(&self.tenant_shard_id.tenant_id, &self.timeline_id);
1776 0 : self.deletion_queue_client
1777 0 : .push_immediate(vec![initdb_path])
1778 0 : .await
1779 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1780 :
1781 : // Do not delete index part yet, it is needed for possible retry. If we remove it first
1782 : // and retry will arrive to different pageserver there wont be any traces of it on remote storage
1783 0 : let timeline_storage_path = remote_timeline_path(&self.tenant_shard_id, &self.timeline_id);
1784 0 :
1785 0 : // Execute all pending deletions, so that when we proceed to do a listing below, we aren't
1786 0 : // taking the burden of listing all the layers that we already know we should delete.
1787 0 : self.flush_deletion_queue()
1788 0 : .await
1789 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1790 :
1791 0 : let cancel = shutdown_token();
1792 :
1793 0 : let remaining = download_retry(
1794 0 : || async {
1795 0 : self.storage_impl
1796 0 : .list(
1797 0 : Some(&timeline_storage_path),
1798 0 : ListingMode::NoDelimiter,
1799 0 : None,
1800 0 : &cancel,
1801 0 : )
1802 0 : .await
1803 0 : },
1804 0 : "list remaining files",
1805 0 : &cancel,
1806 0 : )
1807 0 : .await
1808 0 : .context("list files remaining files")?
1809 : .keys;
1810 :
1811 : // We will delete the current index_part object last, since it acts as a deletion
1812 : // marker via its deleted_at attribute
1813 0 : let latest_index = remaining
1814 0 : .iter()
1815 0 : .filter(|o| {
1816 0 : o.key
1817 0 : .object_name()
1818 0 : .map(|n| n.starts_with(IndexPart::FILE_NAME))
1819 0 : .unwrap_or(false)
1820 0 : })
1821 0 : .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
1822 0 : .max_by_key(|i| i.1)
1823 0 : .map(|i| i.0.clone())
1824 0 : .unwrap_or(
1825 0 : // No generation-suffixed indices, assume we are dealing with
1826 0 : // a legacy index.
1827 0 : remote_index_path(&self.tenant_shard_id, &self.timeline_id, Generation::none()),
1828 0 : );
1829 0 :
1830 0 : let remaining_layers: Vec<RemotePath> = remaining
1831 0 : .into_iter()
1832 0 : .filter_map(|o| {
1833 0 : if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
1834 0 : None
1835 : } else {
1836 0 : Some(o.key)
1837 : }
1838 0 : })
1839 0 : .inspect(|path| {
1840 0 : if let Some(name) = path.object_name() {
1841 0 : info!(%name, "deleting a file not referenced from index_part.json");
1842 : } else {
1843 0 : warn!(%path, "deleting a nameless or non-utf8 object not referenced from index_part.json");
1844 : }
1845 0 : })
1846 0 : .collect();
1847 0 :
1848 0 : let not_referenced_count = remaining_layers.len();
1849 0 : if !remaining_layers.is_empty() {
1850 0 : self.deletion_queue_client
1851 0 : .push_immediate(remaining_layers)
1852 0 : .await
1853 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1854 0 : }
1855 :
1856 0 : fail::fail_point!("timeline-delete-before-index-delete", |_| {
1857 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1858 0 : "failpoint: timeline-delete-before-index-delete"
1859 0 : )))?
1860 0 : });
1861 :
1862 0 : debug!("enqueuing index part deletion");
1863 0 : self.deletion_queue_client
1864 0 : .push_immediate([latest_index].to_vec())
1865 0 : .await
1866 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1867 :
1868 : // Timeline deletion is rare and we have probably emitted a reasonably number of objects: wait
1869 : // for a flush to a persistent deletion list so that we may be sure deletion will occur.
1870 0 : self.flush_deletion_queue()
1871 0 : .await
1872 0 : .map_err(|_| DeleteTimelineError::Cancelled)?;
1873 :
1874 0 : fail::fail_point!("timeline-delete-after-index-delete", |_| {
1875 0 : Err(DeleteTimelineError::Other(anyhow::anyhow!(
1876 0 : "failpoint: timeline-delete-after-index-delete"
1877 0 : )))?
1878 0 : });
1879 :
1880 0 : info!(prefix=%timeline_storage_path, referenced=layer_deletion_count, not_referenced=%not_referenced_count, "done deleting in timeline prefix, including index_part.json");
1881 :
1882 0 : Ok(())
1883 0 : }
1884 :
1885 : /// Pick next tasks from the queue, and start as many of them as possible without violating
1886 : /// the ordering constraints.
1887 : ///
1888 : /// TODO: consider limiting the number of in-progress tasks, beyond what remote_storage does.
1889 : /// This can launch an unbounded number of queued tasks. `UploadQueue::next_ready()` also has
1890 : /// worst-case quadratic cost in the number of tasks, and may struggle beyond 10,000 tasks.
1891 16521 : fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
1892 26475 : while let Some((mut next_op, coalesced_ops)) = upload_queue.next_ready() {
1893 9954 : debug!("starting op: {next_op}");
1894 :
1895 : // Prepare upload.
1896 9954 : match &mut next_op {
1897 3468 : UploadOp::UploadLayer(layer, meta, mode) => {
1898 3468 : if upload_queue
1899 3468 : .recently_deleted
1900 3468 : .remove(&(layer.layer_desc().layer_name().clone(), meta.generation))
1901 0 : {
1902 0 : *mode = Some(OpType::FlushDeletion);
1903 0 : } else {
1904 3468 : *mode = Some(OpType::MayReorder)
1905 : }
1906 : }
1907 3007 : UploadOp::UploadMetadata { .. } => {}
1908 687 : UploadOp::Delete(Delete { layers }) => {
1909 1374 : for (name, meta) in layers {
1910 687 : upload_queue
1911 687 : .recently_deleted
1912 687 : .insert((name.clone(), meta.generation));
1913 687 : }
1914 : }
1915 2792 : UploadOp::Barrier(sender) => {
1916 2792 : sender.send_replace(());
1917 2792 : continue;
1918 : }
1919 0 : UploadOp::Shutdown => unreachable!("shutdown is intentionally never popped off"),
1920 : };
1921 :
1922 : // Assign unique ID to this task
1923 7162 : upload_queue.task_counter += 1;
1924 7162 : let upload_task_id = upload_queue.task_counter;
1925 7162 :
1926 7162 : // Add it to the in-progress map
1927 7162 : let task = Arc::new(UploadTask {
1928 7162 : task_id: upload_task_id,
1929 7162 : op: next_op,
1930 7162 : coalesced_ops,
1931 7162 : retries: AtomicU32::new(0),
1932 7162 : });
1933 7162 : upload_queue
1934 7162 : .inprogress_tasks
1935 7162 : .insert(task.task_id, Arc::clone(&task));
1936 7162 :
1937 7162 : // Spawn task to perform the task
1938 7162 : let self_rc = Arc::clone(self);
1939 7162 : let tenant_shard_id = self.tenant_shard_id;
1940 7162 : let timeline_id = self.timeline_id;
1941 7162 : task_mgr::spawn(
1942 7162 : &self.runtime,
1943 7162 : TaskKind::RemoteUploadTask,
1944 7162 : self.tenant_shard_id,
1945 7162 : Some(self.timeline_id),
1946 7162 : "remote upload",
1947 6993 : async move {
1948 6993 : self_rc.perform_upload_task(task).await;
1949 6710 : Ok(())
1950 6710 : }
1951 7162 : .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)),
1952 : );
1953 :
1954 : // Loop back to process next task
1955 : }
1956 16521 : }
1957 :
1958 : ///
1959 : /// Perform an upload task.
1960 : ///
1961 : /// The task is in the `inprogress_tasks` list. This function will try to
1962 : /// execute it, retrying forever. On successful completion, the task is
1963 : /// removed it from the `inprogress_tasks` list, and any next task(s) in the
1964 : /// queue that were waiting by the completion are launched.
1965 : ///
1966 : /// The task can be shut down, however. That leads to stopping the whole
1967 : /// queue.
1968 : ///
1969 6993 : async fn perform_upload_task(self: &Arc<Self>, task: Arc<UploadTask>) {
1970 6993 : let cancel = shutdown_token();
1971 : // Loop to retry until it completes.
1972 : loop {
1973 : // If we're requested to shut down, close up shop and exit.
1974 : //
1975 : // Note: We only check for the shutdown requests between retries, so
1976 : // if a shutdown request arrives while we're busy uploading, in the
1977 : // upload::upload:*() call below, we will wait not exit until it has
1978 : // finished. We probably could cancel the upload by simply dropping
1979 : // the Future, but we're not 100% sure if the remote storage library
1980 : // is cancellation safe, so we don't dare to do that. Hopefully, the
1981 : // upload finishes or times out soon enough.
1982 6993 : if cancel.is_cancelled() {
1983 0 : info!("upload task cancelled by shutdown request");
1984 0 : self.stop();
1985 0 : return;
1986 6993 : }
1987 6993 :
1988 6993 : // Assert that we don't modify a layer that's referenced by the current index.
1989 6993 : if cfg!(debug_assertions) {
1990 6993 : let modified = match &task.op {
1991 3347 : UploadOp::UploadLayer(layer, layer_metadata, _) => {
1992 3347 : vec![(layer.layer_desc().layer_name(), layer_metadata)]
1993 : }
1994 661 : UploadOp::Delete(delete) => {
1995 661 : delete.layers.iter().map(|(n, m)| (n.clone(), m)).collect()
1996 : }
1997 : // These don't modify layers.
1998 2985 : UploadOp::UploadMetadata { .. } => Vec::new(),
1999 0 : UploadOp::Barrier(_) => Vec::new(),
2000 0 : UploadOp::Shutdown => Vec::new(),
2001 : };
2002 6993 : if let Ok(queue) = self.upload_queue.lock().unwrap().initialized_mut() {
2003 10993 : for (ref name, metadata) in modified {
2004 4008 : debug_assert!(
2005 4008 : !queue.clean.0.references(name, metadata),
2006 8 : "layer {name} modified while referenced by index",
2007 : );
2008 : }
2009 0 : }
2010 0 : }
2011 :
2012 6985 : let upload_result: anyhow::Result<()> = match &task.op {
2013 3347 : UploadOp::UploadLayer(ref layer, ref layer_metadata, mode) => {
2014 : // TODO: check if this mechanism can be removed now that can_bypass() performs
2015 : // conflict checks during scheduling.
2016 3347 : if let Some(OpType::FlushDeletion) = mode {
2017 0 : if self.config.read().unwrap().block_deletions {
2018 : // Of course, this is not efficient... but usually the queue should be empty.
2019 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2020 0 : let mut detected = false;
2021 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2022 0 : for list in queue.blocked_deletions.iter_mut() {
2023 0 : list.layers.retain(|(name, meta)| {
2024 0 : if name == &layer.layer_desc().layer_name()
2025 0 : && meta.generation == layer_metadata.generation
2026 : {
2027 0 : detected = true;
2028 0 : // remove the layer from deletion queue
2029 0 : false
2030 : } else {
2031 : // keep the layer
2032 0 : true
2033 : }
2034 0 : });
2035 0 : }
2036 0 : }
2037 0 : if detected {
2038 0 : info!(
2039 0 : "cancelled blocked deletion of layer {} at gen {:?}",
2040 0 : layer.layer_desc().layer_name(),
2041 : layer_metadata.generation
2042 : );
2043 0 : }
2044 : } else {
2045 : // TODO: we did not guarantee that upload task starts after deletion task, so there could be possibly race conditions
2046 : // that we still get the layer deleted. But this only happens if someone creates a layer immediately after it's deleted,
2047 : // which is not possible in the current system.
2048 0 : info!(
2049 0 : "waiting for deletion queue flush to complete before uploading layer {} at gen {:?}",
2050 0 : layer.layer_desc().layer_name(),
2051 : layer_metadata.generation
2052 : );
2053 : {
2054 : // We are going to flush, we can clean up the recently deleted list.
2055 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2056 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2057 0 : queue.recently_deleted.clear();
2058 0 : }
2059 : }
2060 0 : if let Err(e) = self.deletion_queue_client.flush_execute().await {
2061 0 : warn!(
2062 0 : "failed to flush the deletion queue before uploading layer {} at gen {:?}, still proceeding to upload: {e:#} ",
2063 0 : layer.layer_desc().layer_name(),
2064 : layer_metadata.generation
2065 : );
2066 : } else {
2067 0 : info!(
2068 0 : "done flushing deletion queue before uploading layer {} at gen {:?}",
2069 0 : layer.layer_desc().layer_name(),
2070 : layer_metadata.generation
2071 : );
2072 : }
2073 : }
2074 3347 : }
2075 3347 : let local_path = layer.local_path();
2076 3347 :
2077 3347 : // We should only be uploading layers created by this `Tenant`'s lifetime, so
2078 3347 : // the metadata in the upload should always match our current generation.
2079 3347 : assert_eq!(layer_metadata.generation, self.generation);
2080 :
2081 3347 : let remote_path = remote_layer_path(
2082 3347 : &self.tenant_shard_id.tenant_id,
2083 3347 : &self.timeline_id,
2084 3347 : layer_metadata.shard,
2085 3347 : &layer.layer_desc().layer_name(),
2086 3347 : layer_metadata.generation,
2087 3347 : );
2088 3347 :
2089 3347 : upload::upload_timeline_layer(
2090 3347 : &self.storage_impl,
2091 3347 : local_path,
2092 3347 : &remote_path,
2093 3347 : layer_metadata.file_size,
2094 3347 : &self.cancel,
2095 3347 : )
2096 3347 : .measure_remote_op(
2097 3347 : RemoteOpFileKind::Layer,
2098 3347 : RemoteOpKind::Upload,
2099 3347 : Arc::clone(&self.metrics),
2100 3347 : )
2101 3347 : .await
2102 : }
2103 2985 : UploadOp::UploadMetadata { ref uploaded } => {
2104 2985 : let res = upload::upload_index_part(
2105 2985 : &self.storage_impl,
2106 2985 : &self.tenant_shard_id,
2107 2985 : &self.timeline_id,
2108 2985 : self.generation,
2109 2985 : uploaded,
2110 2985 : &self.cancel,
2111 2985 : )
2112 2985 : .measure_remote_op(
2113 2985 : RemoteOpFileKind::Index,
2114 2985 : RemoteOpKind::Upload,
2115 2985 : Arc::clone(&self.metrics),
2116 2985 : )
2117 2985 : .await;
2118 2965 : if res.is_ok() {
2119 2965 : self.update_remote_physical_size_gauge(Some(uploaded));
2120 2965 : let mention_having_future_layers = if cfg!(feature = "testing") {
2121 2965 : uploaded
2122 2965 : .layer_metadata
2123 2965 : .keys()
2124 35455 : .any(|x| x.is_in_future(uploaded.metadata.disk_consistent_lsn()))
2125 : } else {
2126 0 : false
2127 : };
2128 2965 : if mention_having_future_layers {
2129 : // find rationale near crate::tenant::timeline::init::cleanup_future_layer
2130 116 : tracing::info!(
2131 0 : disk_consistent_lsn = %uploaded.metadata.disk_consistent_lsn(),
2132 0 : "uploaded an index_part.json with future layers -- this is ok! if shutdown now, expect future layer cleanup"
2133 : );
2134 2849 : }
2135 0 : }
2136 2965 : res
2137 : }
2138 653 : UploadOp::Delete(delete) => {
2139 653 : if self.config.read().unwrap().block_deletions {
2140 0 : let mut queue_locked = self.upload_queue.lock().unwrap();
2141 0 : if let Ok(queue) = queue_locked.initialized_mut() {
2142 0 : queue.blocked_deletions.push(delete.clone());
2143 0 : }
2144 0 : Ok(())
2145 : } else {
2146 653 : pausable_failpoint!("before-delete-layer-pausable");
2147 653 : self.deletion_queue_client
2148 653 : .push_layers(
2149 653 : self.tenant_shard_id,
2150 653 : self.timeline_id,
2151 653 : self.generation,
2152 653 : delete.layers.clone(),
2153 653 : )
2154 653 : .map_err(|e| anyhow::anyhow!(e))
2155 : }
2156 : }
2157 0 : unexpected @ UploadOp::Barrier(_) | unexpected @ UploadOp::Shutdown => {
2158 : // unreachable. Barrier operations are handled synchronously in
2159 : // launch_queued_tasks
2160 0 : warn!("unexpected {unexpected:?} operation in perform_upload_task");
2161 0 : break;
2162 : }
2163 : };
2164 :
2165 0 : match upload_result {
2166 : Ok(()) => {
2167 6710 : break;
2168 : }
2169 0 : Err(e) if TimeoutOrCancel::caused_by_cancel(&e) => {
2170 0 : // loop around to do the proper stopping
2171 0 : continue;
2172 : }
2173 0 : Err(e) => {
2174 0 : let retries = task.retries.fetch_add(1, Ordering::SeqCst);
2175 0 :
2176 0 : // Uploads can fail due to rate limits (IAM, S3), spurious network problems,
2177 0 : // or other external reasons. Such issues are relatively regular, so log them
2178 0 : // at info level at first, and only WARN if the operation fails repeatedly.
2179 0 : //
2180 0 : // (See similar logic for downloads in `download::download_retry`)
2181 0 : if retries < FAILED_UPLOAD_WARN_THRESHOLD {
2182 0 : info!(
2183 0 : "failed to perform remote task {}, will retry (attempt {}): {:#}",
2184 0 : task.op, retries, e
2185 : );
2186 : } else {
2187 0 : warn!(
2188 0 : "failed to perform remote task {}, will retry (attempt {}): {:?}",
2189 0 : task.op, retries, e
2190 : );
2191 : }
2192 :
2193 : // sleep until it's time to retry, or we're cancelled
2194 0 : exponential_backoff(
2195 0 : retries,
2196 0 : DEFAULT_BASE_BACKOFF_SECONDS,
2197 0 : DEFAULT_MAX_BACKOFF_SECONDS,
2198 0 : &cancel,
2199 0 : )
2200 0 : .await;
2201 : }
2202 : }
2203 : }
2204 :
2205 6710 : let retries = task.retries.load(Ordering::SeqCst);
2206 6710 : if retries > 0 {
2207 0 : info!(
2208 0 : "remote task {} completed successfully after {} retries",
2209 0 : task.op, retries
2210 : );
2211 : } else {
2212 6710 : debug!("remote task {} completed successfully", task.op);
2213 : }
2214 :
2215 : // The task has completed successfully. Remove it from the in-progress list.
2216 6710 : let lsn_update = {
2217 6710 : let mut upload_queue_guard = self.upload_queue.lock().unwrap();
2218 6710 : let upload_queue = match upload_queue_guard.deref_mut() {
2219 0 : UploadQueue::Uninitialized => panic!("callers are responsible for ensuring this is only called on an initialized queue"),
2220 0 : UploadQueue::Stopped(_stopped) => {
2221 0 : None
2222 : },
2223 6710 : UploadQueue::Initialized(qi) => { Some(qi) }
2224 : };
2225 :
2226 6710 : let upload_queue = match upload_queue {
2227 6710 : Some(upload_queue) => upload_queue,
2228 : None => {
2229 0 : info!("another concurrent task already stopped the queue");
2230 0 : return;
2231 : }
2232 : };
2233 :
2234 6710 : upload_queue.inprogress_tasks.remove(&task.task_id);
2235 :
2236 6710 : let lsn_update = match task.op {
2237 3092 : UploadOp::UploadLayer(_, _, _) => None,
2238 2965 : UploadOp::UploadMetadata { ref uploaded } => {
2239 2965 : // the task id is reused as a monotonicity check for storing the "clean"
2240 2965 : // IndexPart.
2241 2965 : let last_updater = upload_queue.clean.1;
2242 2965 : let is_later = last_updater.is_some_and(|task_id| task_id < task.task_id);
2243 2965 : let monotone = is_later || last_updater.is_none();
2244 :
2245 2965 : assert!(monotone, "no two index uploads should be completing at the same time, prev={last_updater:?}, task.task_id={}", task.task_id);
2246 :
2247 : // not taking ownership is wasteful
2248 2965 : upload_queue.clean.0.clone_from(uploaded);
2249 2965 : upload_queue.clean.1 = Some(task.task_id);
2250 2965 :
2251 2965 : let lsn = upload_queue.clean.0.metadata.disk_consistent_lsn();
2252 2965 : self.metrics
2253 2965 : .projected_remote_consistent_lsn_gauge
2254 2965 : .set(lsn.0);
2255 2965 :
2256 2965 : if self.generation.is_none() {
2257 : // Legacy mode: skip validating generation
2258 0 : upload_queue.visible_remote_consistent_lsn.store(lsn);
2259 0 : None
2260 2965 : } else if self
2261 2965 : .config
2262 2965 : .read()
2263 2965 : .unwrap()
2264 2965 : .process_remote_consistent_lsn_updates
2265 : {
2266 2965 : Some((lsn, upload_queue.visible_remote_consistent_lsn.clone()))
2267 : } else {
2268 : // Our config disables remote_consistent_lsn updates: drop it.
2269 0 : None
2270 : }
2271 : }
2272 653 : UploadOp::Delete(_) => None,
2273 0 : UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
2274 : };
2275 :
2276 : // Launch any queued tasks that were unblocked by this one.
2277 6710 : self.launch_queued_tasks(upload_queue);
2278 6710 : lsn_update
2279 : };
2280 :
2281 6710 : if let Some((lsn, slot)) = lsn_update {
2282 : // Updates to the remote_consistent_lsn we advertise to pageservers
2283 : // are all routed through the DeletionQueue, to enforce important
2284 : // data safety guarantees (see docs/rfcs/025-generation-numbers.md)
2285 2965 : self.deletion_queue_client
2286 2965 : .update_remote_consistent_lsn(
2287 2965 : self.tenant_shard_id,
2288 2965 : self.timeline_id,
2289 2965 : self.generation,
2290 2965 : lsn,
2291 2965 : slot,
2292 2965 : )
2293 2965 : .await;
2294 3745 : }
2295 :
2296 6710 : self.metric_end(&task.op);
2297 6710 : for coalesced_op in &task.coalesced_ops {
2298 5 : self.metric_end(coalesced_op);
2299 5 : }
2300 6710 : }
2301 :
2302 14286 : fn metric_impl(
2303 14286 : &self,
2304 14286 : op: &UploadOp,
2305 14286 : ) -> Option<(
2306 14286 : RemoteOpFileKind,
2307 14286 : RemoteOpKind,
2308 14286 : RemoteTimelineClientMetricsCallTrackSize,
2309 14286 : )> {
2310 : use RemoteTimelineClientMetricsCallTrackSize::DontTrackSize;
2311 14286 : let res = match op {
2312 6560 : UploadOp::UploadLayer(_, m, _) => (
2313 6560 : RemoteOpFileKind::Layer,
2314 6560 : RemoteOpKind::Upload,
2315 6560 : RemoteTimelineClientMetricsCallTrackSize::Bytes(m.file_size),
2316 6560 : ),
2317 6054 : UploadOp::UploadMetadata { .. } => (
2318 6054 : RemoteOpFileKind::Index,
2319 6054 : RemoteOpKind::Upload,
2320 6054 : DontTrackSize {
2321 6054 : reason: "metadata uploads are tiny",
2322 6054 : },
2323 6054 : ),
2324 1656 : UploadOp::Delete(_delete) => (
2325 1656 : RemoteOpFileKind::Layer,
2326 1656 : RemoteOpKind::Delete,
2327 1656 : DontTrackSize {
2328 1656 : reason: "should we track deletes? positive or negative sign?",
2329 1656 : },
2330 1656 : ),
2331 : UploadOp::Barrier(..) | UploadOp::Shutdown => {
2332 : // we do not account these
2333 16 : return None;
2334 : }
2335 : };
2336 14270 : Some(res)
2337 14286 : }
2338 :
2339 7555 : fn metric_begin(&self, op: &UploadOp) {
2340 7555 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2341 7555 : Some(x) => x,
2342 0 : None => return,
2343 : };
2344 7555 : let guard = self.metrics.call_begin(&file_kind, &op_kind, track_bytes);
2345 7555 : guard.will_decrement_manually(); // in metric_end(), see right below
2346 7555 : }
2347 :
2348 6731 : fn metric_end(&self, op: &UploadOp) {
2349 6731 : let (file_kind, op_kind, track_bytes) = match self.metric_impl(op) {
2350 6715 : Some(x) => x,
2351 16 : None => return,
2352 : };
2353 6715 : self.metrics.call_end(&file_kind, &op_kind, track_bytes);
2354 6731 : }
2355 :
2356 : /// Close the upload queue for new operations and cancel queued operations.
2357 : ///
2358 : /// Use [`RemoteTimelineClient::shutdown`] for graceful stop.
2359 : ///
2360 : /// In-progress operations will still be running after this function returns.
2361 : /// Use `task_mgr::shutdown_tasks(Some(TaskKind::RemoteUploadTask), Some(self.tenant_shard_id), Some(timeline_id))`
2362 : /// to wait for them to complete, after calling this function.
2363 36 : pub(crate) fn stop(&self) {
2364 36 : // Whichever *task* for this RemoteTimelineClient grabs the mutex first will transition the queue
2365 36 : // into stopped state, thereby dropping all off the queued *ops* which haven't become *tasks* yet.
2366 36 : // The other *tasks* will come here and observe an already shut down queue and hence simply wrap up their business.
2367 36 : let mut guard = self.upload_queue.lock().unwrap();
2368 36 : self.stop_impl(&mut guard);
2369 36 : }
2370 :
2371 36 : fn stop_impl(&self, guard: &mut std::sync::MutexGuard<UploadQueue>) {
2372 36 : match &mut **guard {
2373 : UploadQueue::Uninitialized => {
2374 0 : info!("UploadQueue is in state Uninitialized, nothing to do");
2375 0 : **guard = UploadQueue::Stopped(UploadQueueStopped::Uninitialized);
2376 : }
2377 : UploadQueue::Stopped(_) => {
2378 : // nothing to do
2379 16 : info!("another concurrent task already shut down the queue");
2380 : }
2381 20 : UploadQueue::Initialized(initialized) => {
2382 20 : info!("shutting down upload queue");
2383 :
2384 : // Replace the queue with the Stopped state, taking ownership of the old
2385 : // Initialized queue. We will do some checks on it, and then drop it.
2386 20 : let qi = {
2387 : // Here we preserve working version of the upload queue for possible use during deletions.
2388 : // In-place replace of Initialized to Stopped can be done with the help of https://github.com/Sgeo/take_mut
2389 : // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
2390 : // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
2391 20 : let upload_queue_for_deletion = UploadQueueInitialized {
2392 20 : inprogress_limit: initialized.inprogress_limit,
2393 20 : task_counter: 0,
2394 20 : dirty: initialized.dirty.clone(),
2395 20 : clean: initialized.clean.clone(),
2396 20 : latest_files_changes_since_metadata_upload_scheduled: 0,
2397 20 : visible_remote_consistent_lsn: initialized
2398 20 : .visible_remote_consistent_lsn
2399 20 : .clone(),
2400 20 : inprogress_tasks: HashMap::default(),
2401 20 : queued_operations: VecDeque::default(),
2402 20 : #[cfg(feature = "testing")]
2403 20 : dangling_files: HashMap::default(),
2404 20 : blocked_deletions: Vec::new(),
2405 20 : shutting_down: false,
2406 20 : shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
2407 20 : recently_deleted: HashSet::new(),
2408 20 : };
2409 20 :
2410 20 : let upload_queue = std::mem::replace(
2411 20 : &mut **guard,
2412 20 : UploadQueue::Stopped(UploadQueueStopped::Deletable(
2413 20 : UploadQueueStoppedDeletable {
2414 20 : upload_queue_for_deletion,
2415 20 : deleted_at: SetDeletedFlagProgress::NotRunning,
2416 20 : },
2417 20 : )),
2418 20 : );
2419 20 : if let UploadQueue::Initialized(qi) = upload_queue {
2420 20 : qi
2421 : } else {
2422 0 : unreachable!("we checked in the match above that it is Initialized");
2423 : }
2424 : };
2425 :
2426 : // We don't need to do anything here for in-progress tasks. They will finish
2427 : // on their own, decrement the unfinished-task counter themselves, and observe
2428 : // that the queue is Stopped.
2429 20 : drop(qi.inprogress_tasks);
2430 :
2431 : // Tear down queued ops
2432 20 : for op in qi.queued_operations.into_iter() {
2433 16 : self.metric_end(&op);
2434 16 : // Dropping UploadOp::Barrier() here will make wait_completion() return with an Err()
2435 16 : // which is exactly what we want to happen.
2436 16 : drop(op);
2437 16 : }
2438 : }
2439 : }
2440 36 : }
2441 :
2442 : /// Returns an accessor which will hold the UploadQueue mutex for accessing the upload queue
2443 : /// externally to RemoteTimelineClient.
2444 0 : pub(crate) fn initialized_upload_queue(
2445 0 : &self,
2446 0 : ) -> Result<UploadQueueAccessor<'_>, NotInitialized> {
2447 0 : let mut inner = self.upload_queue.lock().unwrap();
2448 0 : inner.initialized_mut()?;
2449 0 : Ok(UploadQueueAccessor { inner })
2450 0 : }
2451 :
2452 16 : pub(crate) fn no_pending_work(&self) -> bool {
2453 16 : let inner = self.upload_queue.lock().unwrap();
2454 16 : match &*inner {
2455 : UploadQueue::Uninitialized
2456 0 : | UploadQueue::Stopped(UploadQueueStopped::Uninitialized) => true,
2457 16 : UploadQueue::Stopped(UploadQueueStopped::Deletable(x)) => {
2458 16 : x.upload_queue_for_deletion.no_pending_work()
2459 : }
2460 0 : UploadQueue::Initialized(x) => x.no_pending_work(),
2461 : }
2462 16 : }
2463 :
2464 : /// 'foreign' in the sense that it does not belong to this tenant shard. This method
2465 : /// is used during GC for other shards to get the index of shard zero.
2466 0 : pub(crate) async fn download_foreign_index(
2467 0 : &self,
2468 0 : shard_number: ShardNumber,
2469 0 : cancel: &CancellationToken,
2470 0 : ) -> Result<(IndexPart, Generation, std::time::SystemTime), DownloadError> {
2471 0 : let foreign_shard_id = TenantShardId {
2472 0 : shard_number,
2473 0 : shard_count: self.tenant_shard_id.shard_count,
2474 0 : tenant_id: self.tenant_shard_id.tenant_id,
2475 0 : };
2476 0 : download_index_part(
2477 0 : &self.storage_impl,
2478 0 : &foreign_shard_id,
2479 0 : &self.timeline_id,
2480 0 : Generation::MAX,
2481 0 : cancel,
2482 0 : )
2483 0 : .await
2484 0 : }
2485 : }
2486 :
2487 : pub(crate) struct UploadQueueAccessor<'a> {
2488 : inner: std::sync::MutexGuard<'a, UploadQueue>,
2489 : }
2490 :
2491 : impl UploadQueueAccessor<'_> {
2492 0 : pub(crate) fn latest_uploaded_index_part(&self) -> &IndexPart {
2493 0 : match &*self.inner {
2494 0 : UploadQueue::Initialized(x) => &x.clean.0,
2495 : UploadQueue::Uninitialized | UploadQueue::Stopped(_) => {
2496 0 : unreachable!("checked before constructing")
2497 : }
2498 : }
2499 0 : }
2500 : }
2501 :
2502 0 : pub fn remote_tenant_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2503 0 : let path = format!("tenants/{tenant_shard_id}");
2504 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2505 0 : }
2506 :
2507 1336 : pub fn remote_tenant_manifest_path(
2508 1336 : tenant_shard_id: &TenantShardId,
2509 1336 : generation: Generation,
2510 1336 : ) -> RemotePath {
2511 1336 : let path = format!(
2512 1336 : "tenants/{tenant_shard_id}/tenant-manifest{}.json",
2513 1336 : generation.get_suffix()
2514 1336 : );
2515 1336 : RemotePath::from_string(&path).expect("Failed to construct path")
2516 1336 : }
2517 :
2518 : /// Prefix to all generations' manifest objects in a tenant shard
2519 444 : pub fn remote_tenant_manifest_prefix(tenant_shard_id: &TenantShardId) -> RemotePath {
2520 444 : let path = format!("tenants/{tenant_shard_id}/tenant-manifest",);
2521 444 : RemotePath::from_string(&path).expect("Failed to construct path")
2522 444 : }
2523 :
2524 504 : pub fn remote_timelines_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2525 504 : let path = format!("tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}");
2526 504 : RemotePath::from_string(&path).expect("Failed to construct path")
2527 504 : }
2528 :
2529 0 : fn remote_timelines_path_unsharded(tenant_id: &TenantId) -> RemotePath {
2530 0 : let path = format!("tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}");
2531 0 : RemotePath::from_string(&path).expect("Failed to construct path")
2532 0 : }
2533 :
2534 60 : pub fn remote_timeline_path(
2535 60 : tenant_shard_id: &TenantShardId,
2536 60 : timeline_id: &TimelineId,
2537 60 : ) -> RemotePath {
2538 60 : remote_timelines_path(tenant_shard_id).join(Utf8Path::new(&timeline_id.to_string()))
2539 60 : }
2540 :
2541 : /// Obtains the path of the given Layer in the remote
2542 : ///
2543 : /// Note that the shard component of a remote layer path is _not_ always the same
2544 : /// as in the TenantShardId of the caller: tenants may reference layers from a different
2545 : /// ShardIndex. Use the ShardIndex from the layer's metadata.
2546 3973 : pub fn remote_layer_path(
2547 3973 : tenant_id: &TenantId,
2548 3973 : timeline_id: &TimelineId,
2549 3973 : shard: ShardIndex,
2550 3973 : layer_file_name: &LayerName,
2551 3973 : generation: Generation,
2552 3973 : ) -> RemotePath {
2553 3973 : // Generation-aware key format
2554 3973 : let path = format!(
2555 3973 : "tenants/{tenant_id}{0}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{1}{2}",
2556 3973 : shard.get_suffix(),
2557 3973 : layer_file_name,
2558 3973 : generation.get_suffix()
2559 3973 : );
2560 3973 :
2561 3973 : RemotePath::from_string(&path).expect("Failed to construct path")
2562 3973 : }
2563 :
2564 : /// Returns true if a and b have the same layer path within a tenant/timeline. This is essentially
2565 : /// remote_layer_path(a) == remote_layer_path(b) without the string allocations.
2566 : ///
2567 : /// TODO: there should be a variant of LayerName for the physical path that contains information
2568 : /// about the shard and generation, such that this could be replaced by a simple comparison.
2569 2351030 : pub fn is_same_remote_layer_path(
2570 2351030 : aname: &LayerName,
2571 2351030 : ameta: &LayerFileMetadata,
2572 2351030 : bname: &LayerName,
2573 2351030 : bmeta: &LayerFileMetadata,
2574 2351030 : ) -> bool {
2575 2351030 : // NB: don't assert remote_layer_path(a) == remote_layer_path(b); too expensive even for debug.
2576 2351030 : aname == bname && ameta.shard == bmeta.shard && ameta.generation == bmeta.generation
2577 2351030 : }
2578 :
2579 8 : pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
2580 8 : RemotePath::from_string(&format!(
2581 8 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"
2582 8 : ))
2583 8 : .expect("Failed to construct path")
2584 8 : }
2585 :
2586 4 : pub fn remote_initdb_preserved_archive_path(
2587 4 : tenant_id: &TenantId,
2588 4 : timeline_id: &TimelineId,
2589 4 : ) -> RemotePath {
2590 4 : RemotePath::from_string(&format!(
2591 4 : "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PRESERVED_PATH}"
2592 4 : ))
2593 4 : .expect("Failed to construct path")
2594 4 : }
2595 :
2596 3114 : pub fn remote_index_path(
2597 3114 : tenant_shard_id: &TenantShardId,
2598 3114 : timeline_id: &TimelineId,
2599 3114 : generation: Generation,
2600 3114 : ) -> RemotePath {
2601 3114 : RemotePath::from_string(&format!(
2602 3114 : "tenants/{tenant_shard_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{0}{1}",
2603 3114 : IndexPart::FILE_NAME,
2604 3114 : generation.get_suffix()
2605 3114 : ))
2606 3114 : .expect("Failed to construct path")
2607 3114 : }
2608 :
2609 0 : pub(crate) fn remote_heatmap_path(tenant_shard_id: &TenantShardId) -> RemotePath {
2610 0 : RemotePath::from_string(&format!(
2611 0 : "tenants/{tenant_shard_id}/{TENANT_HEATMAP_BASENAME}"
2612 0 : ))
2613 0 : .expect("Failed to construct path")
2614 0 : }
2615 :
2616 : /// Given the key of an index, parse out the generation part of the name
2617 36 : pub fn parse_remote_index_path(path: RemotePath) -> Option<Generation> {
2618 36 : let file_name = match path.get_path().file_name() {
2619 36 : Some(f) => f,
2620 : None => {
2621 : // Unexpected: we should be seeing index_part.json paths only
2622 0 : tracing::warn!("Malformed index key {}", path);
2623 0 : return None;
2624 : }
2625 : };
2626 :
2627 36 : match file_name.split_once('-') {
2628 24 : Some((_, gen_suffix)) => Generation::parse_suffix(gen_suffix),
2629 12 : None => None,
2630 : }
2631 36 : }
2632 :
2633 : /// Given the key of a tenant manifest, parse out the generation number
2634 0 : pub fn parse_remote_tenant_manifest_path(path: RemotePath) -> Option<Generation> {
2635 : static RE: OnceLock<Regex> = OnceLock::new();
2636 0 : let re = RE.get_or_init(|| Regex::new(r".*tenant-manifest-([0-9a-f]{8}).json").unwrap());
2637 0 : re.captures(path.get_path().as_str())
2638 0 : .and_then(|c| c.get(1))
2639 0 : .and_then(|m| Generation::parse_suffix(m.as_str()))
2640 0 : }
2641 :
2642 : #[cfg(test)]
2643 : mod tests {
2644 : use super::*;
2645 : use crate::{
2646 : context::RequestContext,
2647 : tenant::{
2648 : config::AttachmentMode,
2649 : harness::{TenantHarness, TIMELINE_ID},
2650 : storage_layer::layer::local_layer_path,
2651 : Tenant, Timeline,
2652 : },
2653 : DEFAULT_PG_VERSION,
2654 : };
2655 :
2656 : use std::collections::HashSet;
2657 :
2658 16 : pub(super) fn dummy_contents(name: &str) -> Vec<u8> {
2659 16 : format!("contents for {name}").into()
2660 16 : }
2661 :
2662 4 : pub(super) fn dummy_metadata(disk_consistent_lsn: Lsn) -> TimelineMetadata {
2663 4 : let metadata = TimelineMetadata::new(
2664 4 : disk_consistent_lsn,
2665 4 : None,
2666 4 : None,
2667 4 : Lsn(0),
2668 4 : Lsn(0),
2669 4 : Lsn(0),
2670 4 : // Any version will do
2671 4 : // but it should be consistent with the one in the tests
2672 4 : crate::DEFAULT_PG_VERSION,
2673 4 : );
2674 4 :
2675 4 : // go through serialize + deserialize to fix the header, including checksum
2676 4 : TimelineMetadata::from_bytes(&metadata.to_bytes().unwrap()).unwrap()
2677 4 : }
2678 :
2679 4 : fn assert_file_list(a: &HashSet<LayerName>, b: &[&str]) {
2680 12 : let mut avec: Vec<String> = a.iter().map(|x| x.to_string()).collect();
2681 4 : avec.sort();
2682 4 :
2683 4 : let mut bvec = b.to_vec();
2684 4 : bvec.sort_unstable();
2685 4 :
2686 4 : assert_eq!(avec, bvec);
2687 4 : }
2688 :
2689 8 : fn assert_remote_files(expected: &[&str], remote_path: &Utf8Path, generation: Generation) {
2690 8 : let mut expected: Vec<String> = expected
2691 8 : .iter()
2692 32 : .map(|x| format!("{}{}", x, generation.get_suffix()))
2693 8 : .collect();
2694 8 : expected.sort();
2695 8 :
2696 8 : let mut found: Vec<String> = Vec::new();
2697 32 : for entry in std::fs::read_dir(remote_path).unwrap().flatten() {
2698 32 : let entry_name = entry.file_name();
2699 32 : let fname = entry_name.to_str().unwrap();
2700 32 : found.push(String::from(fname));
2701 32 : }
2702 8 : found.sort();
2703 8 :
2704 8 : assert_eq!(found, expected);
2705 8 : }
2706 :
2707 : struct TestSetup {
2708 : harness: TenantHarness,
2709 : tenant: Arc<Tenant>,
2710 : timeline: Arc<Timeline>,
2711 : tenant_ctx: RequestContext,
2712 : }
2713 :
2714 : impl TestSetup {
2715 16 : async fn new(test_name: &str) -> anyhow::Result<Self> {
2716 16 : let test_name = Box::leak(Box::new(format!("remote_timeline_client__{test_name}")));
2717 16 : let harness = TenantHarness::create(test_name).await?;
2718 16 : let (tenant, ctx) = harness.load().await;
2719 :
2720 16 : let timeline = tenant
2721 16 : .create_test_timeline(TIMELINE_ID, Lsn(8), DEFAULT_PG_VERSION, &ctx)
2722 16 : .await?;
2723 :
2724 16 : Ok(Self {
2725 16 : harness,
2726 16 : tenant,
2727 16 : timeline,
2728 16 : tenant_ctx: ctx,
2729 16 : })
2730 16 : }
2731 :
2732 : /// Construct a RemoteTimelineClient in an arbitrary generation
2733 20 : fn build_client(&self, generation: Generation) -> Arc<RemoteTimelineClient> {
2734 20 : let location_conf = AttachedLocationConfig {
2735 20 : generation,
2736 20 : attach_mode: AttachmentMode::Single,
2737 20 : };
2738 20 : Arc::new(RemoteTimelineClient {
2739 20 : conf: self.harness.conf,
2740 20 : runtime: tokio::runtime::Handle::current(),
2741 20 : tenant_shard_id: self.harness.tenant_shard_id,
2742 20 : timeline_id: TIMELINE_ID,
2743 20 : generation,
2744 20 : storage_impl: self.harness.remote_storage.clone(),
2745 20 : deletion_queue_client: self.harness.deletion_queue.new_client(),
2746 20 : upload_queue: Mutex::new(UploadQueue::Uninitialized),
2747 20 : metrics: Arc::new(RemoteTimelineClientMetrics::new(
2748 20 : &self.harness.tenant_shard_id,
2749 20 : &TIMELINE_ID,
2750 20 : )),
2751 20 : config: std::sync::RwLock::new(RemoteTimelineClientConfig::from(&location_conf)),
2752 20 : cancel: CancellationToken::new(),
2753 20 : })
2754 20 : }
2755 :
2756 : /// A tracing::Span that satisfies remote_timeline_client methods that assert tenant_id
2757 : /// and timeline_id are present.
2758 12 : fn span(&self) -> tracing::Span {
2759 12 : tracing::info_span!(
2760 : "test",
2761 : tenant_id = %self.harness.tenant_shard_id.tenant_id,
2762 0 : shard_id = %self.harness.tenant_shard_id.shard_slug(),
2763 : timeline_id = %TIMELINE_ID
2764 : )
2765 12 : }
2766 : }
2767 :
2768 : // Test scheduling
2769 : #[tokio::test]
2770 4 : async fn upload_scheduling() {
2771 4 : // Test outline:
2772 4 : //
2773 4 : // Schedule upload of a bunch of layers. Check that they are started immediately, not queued
2774 4 : // Schedule upload of index. Check that it is queued
2775 4 : // let the layer file uploads finish. Check that the index-upload is now started
2776 4 : // let the index-upload finish.
2777 4 : //
2778 4 : // Download back the index.json. Check that the list of files is correct
2779 4 : //
2780 4 : // Schedule upload. Schedule deletion. Check that the deletion is queued
2781 4 : // let upload finish. Check that deletion is now started
2782 4 : // Schedule another deletion. Check that it's launched immediately.
2783 4 : // Schedule index upload. Check that it's queued
2784 4 :
2785 4 : let test_setup = TestSetup::new("upload_scheduling").await.unwrap();
2786 4 : let span = test_setup.span();
2787 4 : let _guard = span.enter();
2788 4 :
2789 4 : let TestSetup {
2790 4 : harness,
2791 4 : tenant: _tenant,
2792 4 : timeline,
2793 4 : tenant_ctx: _tenant_ctx,
2794 4 : } = test_setup;
2795 4 :
2796 4 : let client = &timeline.remote_client;
2797 4 :
2798 4 : // Download back the index.json, and check that the list of files is correct
2799 4 : let initial_index_part = match client
2800 4 : .download_index_file(&CancellationToken::new())
2801 4 : .await
2802 4 : .unwrap()
2803 4 : {
2804 4 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2805 4 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2806 4 : };
2807 4 : let initial_layers = initial_index_part
2808 4 : .layer_metadata
2809 4 : .keys()
2810 4 : .map(|f| f.to_owned())
2811 4 : .collect::<HashSet<LayerName>>();
2812 4 : let initial_layer = {
2813 4 : assert!(initial_layers.len() == 1);
2814 4 : initial_layers.into_iter().next().unwrap()
2815 4 : };
2816 4 :
2817 4 : let timeline_path = harness.timeline_path(&TIMELINE_ID);
2818 4 :
2819 4 : println!("workdir: {}", harness.conf.workdir);
2820 4 :
2821 4 : let remote_timeline_dir = harness
2822 4 : .remote_fs_dir
2823 4 : .join(timeline_path.strip_prefix(&harness.conf.workdir).unwrap());
2824 4 : println!("remote_timeline_dir: {remote_timeline_dir}");
2825 4 :
2826 4 : let generation = harness.generation;
2827 4 : let shard = harness.shard;
2828 4 :
2829 4 : // Create a couple of dummy files, schedule upload for them
2830 4 :
2831 4 : let layers = [
2832 4 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), dummy_contents("foo")),
2833 4 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D9-00000000016B5A52".parse().unwrap(), dummy_contents("bar")),
2834 4 : ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59DA-00000000016B5A53".parse().unwrap(), dummy_contents("baz"))
2835 4 : ]
2836 4 : .into_iter()
2837 12 : .map(|(name, contents): (LayerName, Vec<u8>)| {
2838 12 :
2839 12 : let local_path = local_layer_path(
2840 12 : harness.conf,
2841 12 : &timeline.tenant_shard_id,
2842 12 : &timeline.timeline_id,
2843 12 : &name,
2844 12 : &generation,
2845 12 : );
2846 12 : std::fs::write(&local_path, &contents).unwrap();
2847 12 :
2848 12 : Layer::for_resident(
2849 12 : harness.conf,
2850 12 : &timeline,
2851 12 : local_path,
2852 12 : name,
2853 12 : LayerFileMetadata::new(contents.len() as u64, generation, shard),
2854 12 : )
2855 12 : }).collect::<Vec<_>>();
2856 4 :
2857 4 : client
2858 4 : .schedule_layer_file_upload(layers[0].clone())
2859 4 : .unwrap();
2860 4 : client
2861 4 : .schedule_layer_file_upload(layers[1].clone())
2862 4 : .unwrap();
2863 4 :
2864 4 : // Check that they are started immediately, not queued
2865 4 : //
2866 4 : // this works because we running within block_on, so any futures are now queued up until
2867 4 : // our next await point.
2868 4 : {
2869 4 : let mut guard = client.upload_queue.lock().unwrap();
2870 4 : let upload_queue = guard.initialized_mut().unwrap();
2871 4 : assert!(upload_queue.queued_operations.is_empty());
2872 4 : assert_eq!(upload_queue.inprogress_tasks.len(), 2);
2873 4 : assert_eq!(upload_queue.num_inprogress_layer_uploads(), 2);
2874 4 :
2875 4 : // also check that `latest_file_changes` was updated
2876 4 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
2877 4 : }
2878 4 :
2879 4 : // Schedule upload of index. Check that it is queued
2880 4 : let metadata = dummy_metadata(Lsn(0x20));
2881 4 : client
2882 4 : .schedule_index_upload_for_full_metadata_update(&metadata)
2883 4 : .unwrap();
2884 4 : {
2885 4 : let mut guard = client.upload_queue.lock().unwrap();
2886 4 : let upload_queue = guard.initialized_mut().unwrap();
2887 4 : assert!(upload_queue.queued_operations.len() == 1);
2888 4 : assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 0);
2889 4 : }
2890 4 :
2891 4 : // Wait for the uploads to finish
2892 4 : client.wait_completion().await.unwrap();
2893 4 : {
2894 4 : let mut guard = client.upload_queue.lock().unwrap();
2895 4 : let upload_queue = guard.initialized_mut().unwrap();
2896 4 :
2897 4 : assert!(upload_queue.queued_operations.is_empty());
2898 4 : assert!(upload_queue.inprogress_tasks.is_empty());
2899 4 : }
2900 4 :
2901 4 : // Download back the index.json, and check that the list of files is correct
2902 4 : let index_part = match client
2903 4 : .download_index_file(&CancellationToken::new())
2904 4 : .await
2905 4 : .unwrap()
2906 4 : {
2907 4 : MaybeDeletedIndexPart::IndexPart(index_part) => index_part,
2908 4 : MaybeDeletedIndexPart::Deleted(_) => panic!("unexpectedly got deleted index part"),
2909 4 : };
2910 4 :
2911 4 : assert_file_list(
2912 4 : &index_part
2913 4 : .layer_metadata
2914 4 : .keys()
2915 12 : .map(|f| f.to_owned())
2916 4 : .collect(),
2917 4 : &[
2918 4 : &initial_layer.to_string(),
2919 4 : &layers[0].layer_desc().layer_name().to_string(),
2920 4 : &layers[1].layer_desc().layer_name().to_string(),
2921 4 : ],
2922 4 : );
2923 4 : assert_eq!(index_part.metadata, metadata);
2924 4 :
2925 4 : // Schedule upload and then a deletion. Check that the deletion is queued
2926 4 : client
2927 4 : .schedule_layer_file_upload(layers[2].clone())
2928 4 : .unwrap();
2929 4 :
2930 4 : // this is no longer consistent with how deletion works with Layer::drop, but in this test
2931 4 : // keep using schedule_layer_file_deletion because we don't have a way to wait for the
2932 4 : // spawn_blocking started by the drop.
2933 4 : client
2934 4 : .schedule_layer_file_deletion(&[layers[0].layer_desc().layer_name()])
2935 4 : .unwrap();
2936 4 : {
2937 4 : let mut guard = client.upload_queue.lock().unwrap();
2938 4 : let upload_queue = guard.initialized_mut().unwrap();
2939 4 :
2940 4 : // Deletion schedules upload of the index file, and the file deletion itself
2941 4 : assert_eq!(upload_queue.queued_operations.len(), 2);
2942 4 : assert_eq!(upload_queue.inprogress_tasks.len(), 1);
2943 4 : assert_eq!(upload_queue.num_inprogress_layer_uploads(), 1);
2944 4 : assert_eq!(upload_queue.num_inprogress_deletions(), 0);
2945 4 : assert_eq!(
2946 4 : upload_queue.latest_files_changes_since_metadata_upload_scheduled,
2947 4 : 0
2948 4 : );
2949 4 : }
2950 4 : assert_remote_files(
2951 4 : &[
2952 4 : &initial_layer.to_string(),
2953 4 : &layers[0].layer_desc().layer_name().to_string(),
2954 4 : &layers[1].layer_desc().layer_name().to_string(),
2955 4 : "index_part.json",
2956 4 : ],
2957 4 : &remote_timeline_dir,
2958 4 : generation,
2959 4 : );
2960 4 :
2961 4 : // Finish them
2962 4 : client.wait_completion().await.unwrap();
2963 4 : harness.deletion_queue.pump().await;
2964 4 :
2965 4 : assert_remote_files(
2966 4 : &[
2967 4 : &initial_layer.to_string(),
2968 4 : &layers[1].layer_desc().layer_name().to_string(),
2969 4 : &layers[2].layer_desc().layer_name().to_string(),
2970 4 : "index_part.json",
2971 4 : ],
2972 4 : &remote_timeline_dir,
2973 4 : generation,
2974 4 : );
2975 4 : }
2976 :
2977 : #[tokio::test]
2978 4 : async fn bytes_unfinished_gauge_for_layer_file_uploads() {
2979 4 : // Setup
2980 4 :
2981 4 : let TestSetup {
2982 4 : harness,
2983 4 : tenant: _tenant,
2984 4 : timeline,
2985 4 : ..
2986 4 : } = TestSetup::new("metrics").await.unwrap();
2987 4 : let client = &timeline.remote_client;
2988 4 :
2989 4 : let layer_file_name_1: LayerName = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap();
2990 4 : let local_path = local_layer_path(
2991 4 : harness.conf,
2992 4 : &timeline.tenant_shard_id,
2993 4 : &timeline.timeline_id,
2994 4 : &layer_file_name_1,
2995 4 : &harness.generation,
2996 4 : );
2997 4 : let content_1 = dummy_contents("foo");
2998 4 : std::fs::write(&local_path, &content_1).unwrap();
2999 4 :
3000 4 : let layer_file_1 = Layer::for_resident(
3001 4 : harness.conf,
3002 4 : &timeline,
3003 4 : local_path,
3004 4 : layer_file_name_1.clone(),
3005 4 : LayerFileMetadata::new(content_1.len() as u64, harness.generation, harness.shard),
3006 4 : );
3007 4 :
3008 4 : #[derive(Debug, PartialEq, Clone, Copy)]
3009 4 : struct BytesStartedFinished {
3010 4 : started: Option<usize>,
3011 4 : finished: Option<usize>,
3012 4 : }
3013 4 : impl std::ops::Add for BytesStartedFinished {
3014 4 : type Output = Self;
3015 8 : fn add(self, rhs: Self) -> Self::Output {
3016 8 : Self {
3017 8 : started: self.started.map(|v| v + rhs.started.unwrap_or(0)),
3018 8 : finished: self.finished.map(|v| v + rhs.finished.unwrap_or(0)),
3019 8 : }
3020 8 : }
3021 4 : }
3022 12 : let get_bytes_started_stopped = || {
3023 12 : let started = client
3024 12 : .metrics
3025 12 : .get_bytes_started_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
3026 12 : .map(|v| v.try_into().unwrap());
3027 12 : let stopped = client
3028 12 : .metrics
3029 12 : .get_bytes_finished_counter_value(&RemoteOpFileKind::Layer, &RemoteOpKind::Upload)
3030 12 : .map(|v| v.try_into().unwrap());
3031 12 : BytesStartedFinished {
3032 12 : started,
3033 12 : finished: stopped,
3034 12 : }
3035 12 : };
3036 4 :
3037 4 : // Test
3038 4 : tracing::info!("now doing actual test");
3039 4 :
3040 4 : let actual_a = get_bytes_started_stopped();
3041 4 :
3042 4 : client
3043 4 : .schedule_layer_file_upload(layer_file_1.clone())
3044 4 : .unwrap();
3045 4 :
3046 4 : let actual_b = get_bytes_started_stopped();
3047 4 :
3048 4 : client.wait_completion().await.unwrap();
3049 4 :
3050 4 : let actual_c = get_bytes_started_stopped();
3051 4 :
3052 4 : // Validate
3053 4 :
3054 4 : let expected_b = actual_a
3055 4 : + BytesStartedFinished {
3056 4 : started: Some(content_1.len()),
3057 4 : // assert that the _finished metric is created eagerly so that subtractions work on first sample
3058 4 : finished: Some(0),
3059 4 : };
3060 4 : assert_eq!(actual_b, expected_b);
3061 4 :
3062 4 : let expected_c = actual_a
3063 4 : + BytesStartedFinished {
3064 4 : started: Some(content_1.len()),
3065 4 : finished: Some(content_1.len()),
3066 4 : };
3067 4 : assert_eq!(actual_c, expected_c);
3068 4 : }
3069 :
3070 24 : async fn inject_index_part(test_state: &TestSetup, generation: Generation) -> IndexPart {
3071 24 : // An empty IndexPart, just sufficient to ensure deserialization will succeed
3072 24 : let example_index_part = IndexPart::example();
3073 24 :
3074 24 : let index_part_bytes = serde_json::to_vec(&example_index_part).unwrap();
3075 24 :
3076 24 : let index_path = test_state.harness.remote_fs_dir.join(
3077 24 : remote_index_path(
3078 24 : &test_state.harness.tenant_shard_id,
3079 24 : &TIMELINE_ID,
3080 24 : generation,
3081 24 : )
3082 24 : .get_path(),
3083 24 : );
3084 24 :
3085 24 : std::fs::create_dir_all(index_path.parent().unwrap())
3086 24 : .expect("creating test dir should work");
3087 24 :
3088 24 : eprintln!("Writing {index_path}");
3089 24 : std::fs::write(&index_path, index_part_bytes).unwrap();
3090 24 : example_index_part
3091 24 : }
3092 :
3093 : /// Assert that when a RemoteTimelineclient in generation `get_generation` fetches its
3094 : /// index, the IndexPart returned is equal to `expected`
3095 20 : async fn assert_got_index_part(
3096 20 : test_state: &TestSetup,
3097 20 : get_generation: Generation,
3098 20 : expected: &IndexPart,
3099 20 : ) {
3100 20 : let client = test_state.build_client(get_generation);
3101 :
3102 20 : let download_r = client
3103 20 : .download_index_file(&CancellationToken::new())
3104 20 : .await
3105 20 : .expect("download should always succeed");
3106 20 : assert!(matches!(download_r, MaybeDeletedIndexPart::IndexPart(_)));
3107 20 : match download_r {
3108 20 : MaybeDeletedIndexPart::IndexPart(index_part) => {
3109 20 : assert_eq!(&index_part, expected);
3110 : }
3111 0 : MaybeDeletedIndexPart::Deleted(_index_part) => panic!("Test doesn't set deleted_at"),
3112 : }
3113 20 : }
3114 :
3115 : #[tokio::test]
3116 4 : async fn index_part_download_simple() -> anyhow::Result<()> {
3117 4 : let test_state = TestSetup::new("index_part_download_simple").await.unwrap();
3118 4 : let span = test_state.span();
3119 4 : let _guard = span.enter();
3120 4 :
3121 4 : // Simple case: we are in generation N, load the index from generation N - 1
3122 4 : let generation_n = 5;
3123 4 : let injected = inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
3124 4 :
3125 4 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected).await;
3126 4 :
3127 4 : Ok(())
3128 4 : }
3129 :
3130 : #[tokio::test]
3131 4 : async fn index_part_download_ordering() -> anyhow::Result<()> {
3132 4 : let test_state = TestSetup::new("index_part_download_ordering")
3133 4 : .await
3134 4 : .unwrap();
3135 4 :
3136 4 : let span = test_state.span();
3137 4 : let _guard = span.enter();
3138 4 :
3139 4 : // A generation-less IndexPart exists in the bucket, we should find it
3140 4 : let generation_n = 5;
3141 4 : let injected_none = inject_index_part(&test_state, Generation::none()).await;
3142 4 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_none).await;
3143 4 :
3144 4 : // If a more recent-than-none generation exists, we should prefer to load that
3145 4 : let injected_1 = inject_index_part(&test_state, Generation::new(1)).await;
3146 4 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
3147 4 :
3148 4 : // If a more-recent-than-me generation exists, we should ignore it.
3149 4 : let _injected_10 = inject_index_part(&test_state, Generation::new(10)).await;
3150 4 : assert_got_index_part(&test_state, Generation::new(generation_n), &injected_1).await;
3151 4 :
3152 4 : // If a directly previous generation exists, _and_ an index exists in my own
3153 4 : // generation, I should prefer my own generation.
3154 4 : let _injected_prev =
3155 4 : inject_index_part(&test_state, Generation::new(generation_n - 1)).await;
3156 4 : let injected_current = inject_index_part(&test_state, Generation::new(generation_n)).await;
3157 4 : assert_got_index_part(
3158 4 : &test_state,
3159 4 : Generation::new(generation_n),
3160 4 : &injected_current,
3161 4 : )
3162 4 : .await;
3163 4 :
3164 4 : Ok(())
3165 4 : }
3166 : }
|