Line data Source code
1 : //! Functionality for finding and purging garbage, as in "garbage collection". Garbage means
2 : //! S3 objects which are either not referenced by any metadata, or are referenced by a
3 : //! control plane tenant/timeline in a deleted state.
4 :
5 : use std::{
6 : collections::{HashMap, HashSet},
7 : sync::Arc,
8 : time::Duration,
9 : };
10 :
11 : use anyhow::Context;
12 : use futures_util::TryStreamExt;
13 : use pageserver_api::shard::TenantShardId;
14 : use remote_storage::{GenericRemoteStorage, ListingMode, ListingObject, RemotePath};
15 : use serde::{Deserialize, Serialize};
16 : use tokio_stream::StreamExt;
17 : use tokio_util::sync::CancellationToken;
18 : use utils::id::TenantId;
19 :
20 : use crate::{
21 : cloud_admin_api::{CloudAdminApiClient, MaybeDeleted, ProjectData},
22 : init_remote, list_objects_with_retries,
23 : metadata_stream::{stream_tenant_timelines, stream_tenants},
24 : BucketConfig, ConsoleConfig, NodeKind, TenantShardTimelineId, TraversingDepth,
25 : };
26 :
27 0 : #[derive(Serialize, Deserialize, Debug)]
28 : enum GarbageReason {
29 : DeletedInConsole,
30 : MissingInConsole,
31 :
32 : // The remaining data relates to a known deletion issue, and we're sure that purging this
33 : // will not delete any real data, for example https://github.com/neondatabase/neon/pull/7928 where
34 : // there is nothing in a tenant path apart from a heatmap file.
35 : KnownBug,
36 : }
37 :
38 0 : #[derive(Serialize, Deserialize, Debug)]
39 : enum GarbageEntity {
40 : Tenant(TenantShardId),
41 : Timeline(TenantShardTimelineId),
42 : }
43 :
44 0 : #[derive(Serialize, Deserialize, Debug)]
45 : struct GarbageItem {
46 : entity: GarbageEntity,
47 : reason: GarbageReason,
48 : }
49 :
50 0 : #[derive(Serialize, Deserialize, Debug)]
51 : pub struct GarbageList {
52 : /// Remember what NodeKind we were finding garbage for, so that we can
53 : /// purge the list without re-stating it.
54 : node_kind: NodeKind,
55 :
56 : /// Embed the identity of the bucket, so that we do not risk executing
57 : /// the wrong list against the wrong bucket, and so that the user does not have
58 : /// to re-state the bucket details when purging.
59 : bucket_config: BucketConfig,
60 :
61 : items: Vec<GarbageItem>,
62 :
63 : /// Advisory information to enable consumers to do a validation that if we
64 : /// see garbage, we saw some active tenants too. This protects against classes of bugs
65 : /// in the scrubber that might otherwise generate a "deleted all" result.
66 : active_tenant_count: usize,
67 : active_timeline_count: usize,
68 : }
69 :
70 : impl GarbageList {
71 0 : fn new(node_kind: NodeKind, bucket_config: BucketConfig) -> Self {
72 0 : Self {
73 0 : items: Vec::new(),
74 0 : active_tenant_count: 0,
75 0 : active_timeline_count: 0,
76 0 : node_kind,
77 0 : bucket_config,
78 0 : }
79 0 : }
80 :
81 : /// If an entity has been identified as requiring purge due to a known bug, e.g.
82 : /// a particular type of object left behind after an incomplete deletion.
83 0 : fn append_buggy(&mut self, entity: GarbageEntity) {
84 0 : self.items.push(GarbageItem {
85 0 : entity,
86 0 : reason: GarbageReason::KnownBug,
87 0 : });
88 0 : }
89 :
90 : /// Return true if appended, false if not. False means the result was not garbage.
91 0 : fn maybe_append<T>(&mut self, entity: GarbageEntity, result: Option<T>) -> bool
92 0 : where
93 0 : T: MaybeDeleted,
94 0 : {
95 0 : match result {
96 0 : Some(result_item) if result_item.is_deleted() => {
97 0 : self.items.push(GarbageItem {
98 0 : entity,
99 0 : reason: GarbageReason::DeletedInConsole,
100 0 : });
101 0 : true
102 : }
103 0 : Some(_) => false,
104 : None => {
105 0 : self.items.push(GarbageItem {
106 0 : entity,
107 0 : reason: GarbageReason::MissingInConsole,
108 0 : });
109 0 : true
110 : }
111 : }
112 0 : }
113 : }
114 :
115 0 : pub async fn find_garbage(
116 0 : bucket_config: BucketConfig,
117 0 : console_config: ConsoleConfig,
118 0 : depth: TraversingDepth,
119 0 : node_kind: NodeKind,
120 0 : output_path: String,
121 0 : ) -> anyhow::Result<()> {
122 0 : let garbage = find_garbage_inner(bucket_config, console_config, depth, node_kind).await?;
123 0 : let serialized = serde_json::to_vec_pretty(&garbage)?;
124 :
125 0 : tokio::fs::write(&output_path, &serialized).await?;
126 :
127 0 : tracing::info!("Wrote garbage report to {output_path}");
128 :
129 0 : Ok(())
130 0 : }
131 :
132 : // How many concurrent S3 operations to issue (approximately): this is the concurrency
133 : // for things like listing the timelines within tenant prefixes.
134 : const S3_CONCURRENCY: usize = 32;
135 :
136 : // How many concurrent API requests to make to the console API.
137 : //
138 : // Be careful increasing this; roughly we shouldn't have more than ~100 rps. It
139 : // would be better to implement real rsp limiter.
140 : const CONSOLE_CONCURRENCY: usize = 16;
141 :
142 : struct ConsoleCache {
143 : /// Set of tenants found in the control plane API
144 : projects: HashMap<TenantId, ProjectData>,
145 : /// Set of tenants for which the control plane API returned 404
146 : not_found: HashSet<TenantId>,
147 : }
148 :
149 0 : async fn find_garbage_inner(
150 0 : bucket_config: BucketConfig,
151 0 : console_config: ConsoleConfig,
152 0 : depth: TraversingDepth,
153 0 : node_kind: NodeKind,
154 0 : ) -> anyhow::Result<GarbageList> {
155 : // Construct clients for S3 and for Console API
156 0 : let (remote_client, target) = init_remote(bucket_config.clone(), node_kind).await?;
157 0 : let cloud_admin_api_client = Arc::new(CloudAdminApiClient::new(console_config));
158 :
159 : // Build a set of console-known tenants, for quickly eliminating known-active tenants without having
160 : // to issue O(N) console API requests.
161 0 : let console_projects: HashMap<TenantId, ProjectData> = cloud_admin_api_client
162 0 : // FIXME: we can't just assume that all console's region ids are aws-<something>. This hack
163 0 : // will go away when we are talking to Control Plane APIs, which are per-region.
164 0 : .list_projects(format!("aws-{}", bucket_config.region))
165 0 : .await?
166 0 : .into_iter()
167 0 : .map(|t| (t.tenant, t))
168 0 : .collect();
169 0 : tracing::info!(
170 0 : "Loaded {} console projects tenant IDs",
171 0 : console_projects.len()
172 : );
173 :
174 : // Because many tenant shards may look up the same TenantId, we maintain a cache.
175 0 : let console_cache = Arc::new(std::sync::Mutex::new(ConsoleCache {
176 0 : projects: console_projects,
177 0 : not_found: HashSet::new(),
178 0 : }));
179 0 :
180 0 : // Enumerate Tenants in S3, and check if each one exists in Console
181 0 : tracing::info!("Finding all tenants in bucket {}...", bucket_config.bucket);
182 0 : let tenants = stream_tenants(&remote_client, &target);
183 0 : let tenants_checked = tenants.map_ok(|t| {
184 0 : let api_client = cloud_admin_api_client.clone();
185 0 : let console_cache = console_cache.clone();
186 0 : async move {
187 : // Check cache before issuing API call
188 0 : let project_data = {
189 0 : let cache = console_cache.lock().unwrap();
190 0 : let result = cache.projects.get(&t.tenant_id).cloned();
191 0 : if result.is_none() && cache.not_found.contains(&t.tenant_id) {
192 0 : return Ok((t, None));
193 0 : }
194 0 : result
195 0 : };
196 0 :
197 0 : match project_data {
198 0 : Some(project_data) => Ok((t, Some(project_data.clone()))),
199 : None => {
200 0 : let project_data = api_client
201 0 : .find_tenant_project(t.tenant_id)
202 0 : .await
203 0 : .map_err(|e| anyhow::anyhow!(e));
204 0 :
205 0 : // Populate cache with result of API call
206 0 : {
207 0 : let mut cache = console_cache.lock().unwrap();
208 0 : if let Ok(Some(project_data)) = &project_data {
209 0 : cache.projects.insert(t.tenant_id, project_data.clone());
210 0 : } else if let Ok(None) = &project_data {
211 0 : cache.not_found.insert(t.tenant_id);
212 0 : }
213 : }
214 :
215 0 : project_data.map(|r| (t, r))
216 : }
217 : }
218 0 : }
219 0 : });
220 0 : let mut tenants_checked =
221 0 : std::pin::pin!(tenants_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
222 0 :
223 0 : // Process the results of Tenant checks. If a Tenant is garbage, it goes into
224 0 : // the `GarbageList`. Else it goes into `active_tenants` for more detailed timeline
225 0 : // checks if they are enabled by the `depth` parameter.
226 0 : let mut garbage = GarbageList::new(node_kind, bucket_config);
227 0 : let mut active_tenants: Vec<TenantShardId> = vec![];
228 0 : let mut counter = 0;
229 0 : while let Some(result) = tenants_checked.next().await {
230 0 : let (tenant_shard_id, console_result) = result?;
231 :
232 : // Paranoia check
233 0 : if let Some(project) = &console_result {
234 0 : assert!(project.tenant == tenant_shard_id.tenant_id);
235 0 : }
236 :
237 : // Special case: If it's missing in console, check for known bugs that would enable us to conclusively
238 : // identify it as purge-able anyway
239 0 : if console_result.is_none() {
240 0 : let timelines = stream_tenant_timelines(&remote_client, &target, tenant_shard_id)
241 0 : .await?
242 0 : .collect::<Vec<_>>()
243 0 : .await;
244 0 : if timelines.is_empty() {
245 : // No timelines, but a heatmap: the deletion bug where we deleted everything but heatmaps
246 0 : let tenant_objects = list_objects_with_retries(
247 0 : &remote_client,
248 0 : ListingMode::WithDelimiter,
249 0 : &target.tenant_root(&tenant_shard_id),
250 0 : )
251 0 : .await?;
252 0 : let object = tenant_objects.keys.first().unwrap();
253 0 : if object.key.get_path().as_str().ends_with("heatmap-v1.json") {
254 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and is only a heatmap (known historic deletion bug)");
255 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
256 0 : continue;
257 : } else {
258 0 : tracing::info!("Tenant {tenant_shard_id} is missing in console and contains one object: {}", object.key);
259 : }
260 : } else {
261 : // A console-unknown tenant with timelines: check if these timelines only contain initdb.tar.zst, from the initial
262 : // rollout of WAL DR in which we never deleted these.
263 0 : let mut any_non_initdb = false;
264 :
265 0 : for timeline_r in timelines {
266 0 : let timeline = timeline_r?;
267 0 : let timeline_objects = list_objects_with_retries(
268 0 : &remote_client,
269 0 : ListingMode::WithDelimiter,
270 0 : &target.timeline_root(&timeline),
271 0 : )
272 0 : .await?;
273 0 : if !timeline_objects.prefixes.is_empty() {
274 0 : // Sub-paths? Unexpected
275 0 : any_non_initdb = true;
276 0 : } else {
277 0 : let object = timeline_objects.keys.first().unwrap();
278 0 : if object.key.get_path().as_str().ends_with("initdb.tar.zst") {
279 0 : tracing::info!("Timeline {timeline} contains only initdb.tar.zst");
280 0 : } else {
281 0 : any_non_initdb = true;
282 0 : }
283 : }
284 : }
285 :
286 0 : if any_non_initdb {
287 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains timelines, one or more of which are more than just initdb");
288 : } else {
289 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains only timelines that only contain initdb");
290 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
291 0 : continue;
292 : }
293 : }
294 0 : }
295 :
296 0 : if garbage.maybe_append(GarbageEntity::Tenant(tenant_shard_id), console_result) {
297 0 : tracing::debug!("Tenant {tenant_shard_id} is garbage");
298 : } else {
299 0 : tracing::debug!("Tenant {tenant_shard_id} is active");
300 0 : active_tenants.push(tenant_shard_id);
301 0 : garbage.active_tenant_count = active_tenants.len();
302 : }
303 :
304 0 : counter += 1;
305 0 : if counter % 1000 == 0 {
306 0 : tracing::info!(
307 0 : "Progress: {counter} tenants checked, {} active, {} garbage",
308 0 : active_tenants.len(),
309 0 : garbage.items.len()
310 : );
311 0 : }
312 : }
313 :
314 0 : tracing::info!(
315 0 : "Found {}/{} garbage tenants",
316 0 : garbage.items.len(),
317 0 : garbage.items.len() + active_tenants.len()
318 : );
319 :
320 : // If we are only checking tenant-deep, we are done. Otherwise we must
321 : // proceed to check the individual timelines of the active tenants.
322 0 : if depth == TraversingDepth::Tenant {
323 0 : return Ok(garbage);
324 0 : }
325 0 :
326 0 : tracing::info!(
327 0 : "Checking timelines for {} active tenants",
328 0 : active_tenants.len(),
329 : );
330 :
331 : // Construct a stream of all timelines within active tenants
332 0 : let active_tenants = tokio_stream::iter(active_tenants.iter().map(Ok));
333 0 : let timelines = active_tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, *t));
334 0 : let timelines = timelines.try_buffer_unordered(S3_CONCURRENCY);
335 0 : let timelines = timelines.try_flatten();
336 0 :
337 0 : // For all timelines within active tenants, call into console API to check their existence
338 0 : let timelines_checked = timelines.map_ok(|ttid| {
339 0 : let api_client = cloud_admin_api_client.clone();
340 0 : async move {
341 0 : api_client
342 0 : .find_timeline_branch(ttid.tenant_shard_id.tenant_id, ttid.timeline_id)
343 0 : .await
344 0 : .map_err(|e| anyhow::anyhow!(e))
345 0 : .map(|r| (ttid, r))
346 0 : }
347 0 : });
348 0 : let mut timelines_checked =
349 0 : std::pin::pin!(timelines_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
350 0 :
351 0 : // Update the GarbageList with any timelines which appear not to exist.
352 0 : let mut active_timelines: Vec<TenantShardTimelineId> = vec![];
353 0 : while let Some(result) = timelines_checked.next().await {
354 0 : let (ttid, console_result) = result?;
355 0 : if garbage.maybe_append(GarbageEntity::Timeline(ttid), console_result) {
356 0 : tracing::debug!("Timeline {ttid} is garbage");
357 : } else {
358 0 : tracing::debug!("Timeline {ttid} is active");
359 0 : active_timelines.push(ttid);
360 0 : garbage.active_timeline_count = active_timelines.len();
361 : }
362 : }
363 :
364 0 : let num_garbage_timelines = garbage
365 0 : .items
366 0 : .iter()
367 0 : .filter(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
368 0 : .count();
369 0 : tracing::info!(
370 0 : "Found {}/{} garbage timelines in active tenants",
371 0 : num_garbage_timelines,
372 0 : active_timelines.len(),
373 : );
374 :
375 0 : Ok(garbage)
376 0 : }
377 :
378 0 : #[derive(clap::ValueEnum, Debug, Clone)]
379 : pub enum PurgeMode {
380 : /// The safest mode: only delete tenants that were explicitly reported as deleted
381 : /// by Console API.
382 : DeletedOnly,
383 :
384 : /// Delete all garbage tenants, including those which are only presumed to be deleted,
385 : /// because the Console API could not find them.
386 : DeletedAndMissing,
387 : }
388 :
389 : impl std::fmt::Display for PurgeMode {
390 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
391 0 : match self {
392 0 : PurgeMode::DeletedOnly => write!(f, "deleted-only"),
393 0 : PurgeMode::DeletedAndMissing => write!(f, "deleted-and-missing"),
394 : }
395 0 : }
396 : }
397 :
398 0 : pub async fn get_tenant_objects(
399 0 : s3_client: &GenericRemoteStorage,
400 0 : tenant_shard_id: TenantShardId,
401 0 : ) -> anyhow::Result<Vec<ListingObject>> {
402 0 : tracing::debug!("Listing objects in tenant {tenant_shard_id}");
403 0 : let tenant_root = super::remote_tenant_path(&tenant_shard_id);
404 :
405 : // TODO: apply extra validation based on object modification time. Don't purge
406 : // tenants where any timeline's index_part.json has been touched recently.
407 :
408 0 : let list = s3_client
409 0 : .list(
410 0 : Some(&tenant_root),
411 0 : ListingMode::NoDelimiter,
412 0 : None,
413 0 : &CancellationToken::new(),
414 0 : )
415 0 : .await?;
416 0 : Ok(list.keys)
417 0 : }
418 :
419 0 : pub async fn get_timeline_objects(
420 0 : s3_client: &GenericRemoteStorage,
421 0 : ttid: TenantShardTimelineId,
422 0 : ) -> anyhow::Result<Vec<ListingObject>> {
423 0 : tracing::debug!("Listing objects in timeline {ttid}");
424 0 : let timeline_root = super::remote_timeline_path_id(&ttid);
425 :
426 0 : let list = s3_client
427 0 : .list(
428 0 : Some(&timeline_root),
429 0 : ListingMode::NoDelimiter,
430 0 : None,
431 0 : &CancellationToken::new(),
432 0 : )
433 0 : .await?;
434 0 : Ok(list.keys)
435 0 : }
436 :
437 : const MAX_KEYS_PER_DELETE: usize = 1000;
438 :
439 : /// Drain a buffer of keys into DeleteObjects requests
440 : ///
441 : /// If `drain` is true, drains keys completely; otherwise stops when <
442 : /// MAX_KEYS_PER_DELETE keys are left.
443 : /// `num_deleted` returns number of deleted keys.
444 0 : async fn do_delete(
445 0 : remote_client: &GenericRemoteStorage,
446 0 : keys: &mut Vec<ListingObject>,
447 0 : dry_run: bool,
448 0 : drain: bool,
449 0 : progress_tracker: &mut DeletionProgressTracker,
450 0 : ) -> anyhow::Result<()> {
451 0 : let cancel = CancellationToken::new();
452 0 : while (!keys.is_empty() && drain) || (keys.len() >= MAX_KEYS_PER_DELETE) {
453 0 : let request_keys =
454 0 : keys.split_off(keys.len() - (std::cmp::min(MAX_KEYS_PER_DELETE, keys.len())));
455 0 :
456 0 : let request_keys: Vec<RemotePath> = request_keys.into_iter().map(|o| o.key).collect();
457 0 :
458 0 : let num_deleted = request_keys.len();
459 0 : if dry_run {
460 0 : tracing::info!("Dry-run deletion of objects: ");
461 0 : for k in request_keys {
462 0 : tracing::info!(" {k:?}");
463 : }
464 : } else {
465 0 : remote_client
466 0 : .delete_objects(&request_keys, &cancel)
467 0 : .await
468 0 : .context("deletetion request")?;
469 0 : progress_tracker.register(num_deleted);
470 : }
471 : }
472 :
473 0 : Ok(())
474 0 : }
475 :
476 : /// Simple tracker reporting each 10k deleted keys.
477 : #[derive(Default)]
478 : struct DeletionProgressTracker {
479 : num_deleted: usize,
480 : last_reported_num_deleted: usize,
481 : }
482 :
483 : impl DeletionProgressTracker {
484 0 : fn register(&mut self, n: usize) {
485 0 : self.num_deleted += n;
486 0 : if self.num_deleted - self.last_reported_num_deleted > 10000 {
487 0 : tracing::info!("progress: deleted {} keys", self.num_deleted);
488 0 : self.last_reported_num_deleted = self.num_deleted;
489 0 : }
490 0 : }
491 : }
492 :
493 0 : pub async fn purge_garbage(
494 0 : input_path: String,
495 0 : mode: PurgeMode,
496 0 : min_age: Duration,
497 0 : dry_run: bool,
498 0 : ) -> anyhow::Result<()> {
499 0 : let list_bytes = tokio::fs::read(&input_path).await?;
500 0 : let garbage_list = serde_json::from_slice::<GarbageList>(&list_bytes)?;
501 0 : tracing::info!(
502 0 : "Loaded {} items in garbage list from {}",
503 0 : garbage_list.items.len(),
504 : input_path
505 : );
506 :
507 0 : let (remote_client, _target) =
508 0 : init_remote(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
509 :
510 0 : assert_eq!(
511 0 : &garbage_list.bucket_config.bucket,
512 0 : remote_client.bucket_name().unwrap()
513 0 : );
514 :
515 : // Sanity checks on the incoming list
516 0 : if garbage_list.active_tenant_count == 0 {
517 0 : anyhow::bail!("Refusing to purge a garbage list that reports 0 active tenants");
518 0 : }
519 0 : if garbage_list
520 0 : .items
521 0 : .iter()
522 0 : .any(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
523 0 : && garbage_list.active_timeline_count == 0
524 : {
525 0 : anyhow::bail!("Refusing to purge a garbage list containing garbage timelines that reports 0 active timelines");
526 0 : }
527 0 :
528 0 : let filtered_items = garbage_list
529 0 : .items
530 0 : .iter()
531 0 : .filter(|i| match (&mode, &i.reason) {
532 0 : (PurgeMode::DeletedAndMissing, _) => true,
533 0 : (PurgeMode::DeletedOnly, GarbageReason::DeletedInConsole) => true,
534 0 : (PurgeMode::DeletedOnly, GarbageReason::KnownBug) => true,
535 0 : (PurgeMode::DeletedOnly, GarbageReason::MissingInConsole) => false,
536 0 : });
537 0 :
538 0 : tracing::info!(
539 0 : "Filtered down to {} garbage items based on mode {}",
540 0 : garbage_list.items.len(),
541 : mode
542 : );
543 :
544 0 : let items = tokio_stream::iter(filtered_items.map(Ok));
545 0 : let get_objects_results = items.map_ok(|i| {
546 0 : let remote_client = remote_client.clone();
547 0 : async move {
548 0 : match i.entity {
549 0 : GarbageEntity::Tenant(tenant_id) => {
550 0 : get_tenant_objects(&remote_client, tenant_id).await
551 : }
552 0 : GarbageEntity::Timeline(ttid) => get_timeline_objects(&remote_client, ttid).await,
553 : }
554 0 : }
555 0 : });
556 0 : let mut get_objects_results =
557 0 : std::pin::pin!(get_objects_results.try_buffer_unordered(S3_CONCURRENCY));
558 0 :
559 0 : let mut objects_to_delete = Vec::new();
560 0 : let mut progress_tracker = DeletionProgressTracker::default();
561 0 : while let Some(result) = get_objects_results.next().await {
562 0 : let mut object_list = result?;
563 :
564 : // Extra safety check: even if a collection of objects is garbage, check max() of modification
565 : // times before purging, so that if we incorrectly marked a live tenant as garbage then we would
566 : // notice that its index has been written recently and would omit deleting it.
567 0 : if object_list.is_empty() {
568 : // Simplify subsequent code by ensuring list always has at least one item
569 : // Usually, this only occurs if there is parallel deletions racing us, as there is no empty prefixes
570 0 : continue;
571 0 : }
572 0 : let max_mtime = object_list.iter().map(|o| o.last_modified).max().unwrap();
573 0 : let age = max_mtime.elapsed();
574 0 : match age {
575 : Err(_) => {
576 0 : tracing::warn!("Bad last_modified time");
577 0 : continue;
578 : }
579 0 : Ok(a) if a < min_age => {
580 0 : // Failed age check. This doesn't mean we did something wrong: a tenant might really be garbage and recently
581 0 : // written, but out of an abundance of caution we still don't purge it.
582 0 : tracing::info!(
583 0 : "Skipping tenant with young objects {}..{}",
584 0 : object_list.first().as_ref().unwrap().key,
585 0 : object_list.last().as_ref().unwrap().key
586 : );
587 0 : continue;
588 : }
589 0 : Ok(_) => {
590 0 : // Passed age check
591 0 : }
592 0 : }
593 0 :
594 0 : objects_to_delete.append(&mut object_list);
595 0 : if objects_to_delete.len() >= MAX_KEYS_PER_DELETE {
596 0 : do_delete(
597 0 : &remote_client,
598 0 : &mut objects_to_delete,
599 0 : dry_run,
600 0 : false,
601 0 : &mut progress_tracker,
602 0 : )
603 0 : .await?;
604 0 : }
605 : }
606 :
607 0 : do_delete(
608 0 : &remote_client,
609 0 : &mut objects_to_delete,
610 0 : dry_run,
611 0 : true,
612 0 : &mut progress_tracker,
613 0 : )
614 0 : .await?;
615 :
616 0 : tracing::info!("{} keys deleted in total", progress_tracker.num_deleted);
617 :
618 0 : Ok(())
619 0 : }
|
