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_generic, list_objects_with_retries_generic,
23 : metadata_stream::{stream_tenant_timelines_generic, stream_tenants_generic},
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_generic(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_generic(&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 =
241 0 : stream_tenant_timelines_generic(&remote_client, &target, tenant_shard_id)
242 0 : .await?
243 0 : .collect::<Vec<_>>()
244 0 : .await;
245 0 : if timelines.is_empty() {
246 : // No timelines, but a heatmap: the deletion bug where we deleted everything but heatmaps
247 0 : let tenant_objects = list_objects_with_retries_generic(
248 0 : &remote_client,
249 0 : ListingMode::WithDelimiter,
250 0 : &target.tenant_root(&tenant_shard_id),
251 0 : )
252 0 : .await?;
253 0 : let object = tenant_objects.keys.first().unwrap();
254 0 : if object.key.get_path().as_str().ends_with("heatmap-v1.json") {
255 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and is only a heatmap (known historic deletion bug)");
256 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
257 0 : continue;
258 : } else {
259 0 : tracing::info!("Tenant {tenant_shard_id} is missing in console and contains one object: {}", object.key);
260 : }
261 : } else {
262 : // A console-unknown tenant with timelines: check if these timelines only contain initdb.tar.zst, from the initial
263 : // rollout of WAL DR in which we never deleted these.
264 0 : let mut any_non_initdb = false;
265 :
266 0 : for timeline_r in timelines {
267 0 : let timeline = timeline_r?;
268 0 : let timeline_objects = list_objects_with_retries_generic(
269 0 : &remote_client,
270 0 : ListingMode::WithDelimiter,
271 0 : &target.timeline_root(&timeline),
272 0 : )
273 0 : .await?;
274 0 : if !timeline_objects.prefixes.is_empty() {
275 0 : // Sub-paths? Unexpected
276 0 : any_non_initdb = true;
277 0 : } else {
278 0 : let object = timeline_objects.keys.first().unwrap();
279 0 : if object.key.get_path().as_str().ends_with("initdb.tar.zst") {
280 0 : tracing::info!("Timeline {timeline} contains only initdb.tar.zst");
281 0 : } else {
282 0 : any_non_initdb = true;
283 0 : }
284 : }
285 : }
286 :
287 0 : if any_non_initdb {
288 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains timelines, one or more of which are more than just initdb");
289 : } else {
290 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains only timelines that only contain initdb");
291 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
292 0 : continue;
293 : }
294 : }
295 0 : }
296 :
297 0 : if garbage.maybe_append(GarbageEntity::Tenant(tenant_shard_id), console_result) {
298 0 : tracing::debug!("Tenant {tenant_shard_id} is garbage");
299 : } else {
300 0 : tracing::debug!("Tenant {tenant_shard_id} is active");
301 0 : active_tenants.push(tenant_shard_id);
302 0 : garbage.active_tenant_count = active_tenants.len();
303 : }
304 :
305 0 : counter += 1;
306 0 : if counter % 1000 == 0 {
307 0 : tracing::info!(
308 0 : "Progress: {counter} tenants checked, {} active, {} garbage",
309 0 : active_tenants.len(),
310 0 : garbage.items.len()
311 : );
312 0 : }
313 : }
314 :
315 0 : tracing::info!(
316 0 : "Found {}/{} garbage tenants",
317 0 : garbage.items.len(),
318 0 : garbage.items.len() + active_tenants.len()
319 : );
320 :
321 : // If we are only checking tenant-deep, we are done. Otherwise we must
322 : // proceed to check the individual timelines of the active tenants.
323 0 : if depth == TraversingDepth::Tenant {
324 0 : return Ok(garbage);
325 0 : }
326 0 :
327 0 : tracing::info!(
328 0 : "Checking timelines for {} active tenants",
329 0 : active_tenants.len(),
330 : );
331 :
332 : // Construct a stream of all timelines within active tenants
333 0 : let active_tenants = tokio_stream::iter(active_tenants.iter().map(Ok));
334 0 : let timelines =
335 0 : active_tenants.map_ok(|t| stream_tenant_timelines_generic(&remote_client, &target, *t));
336 0 : let timelines = timelines.try_buffer_unordered(S3_CONCURRENCY);
337 0 : let timelines = timelines.try_flatten();
338 0 :
339 0 : // For all timelines within active tenants, call into console API to check their existence
340 0 : let timelines_checked = timelines.map_ok(|ttid| {
341 0 : let api_client = cloud_admin_api_client.clone();
342 0 : async move {
343 0 : api_client
344 0 : .find_timeline_branch(ttid.tenant_shard_id.tenant_id, ttid.timeline_id)
345 0 : .await
346 0 : .map_err(|e| anyhow::anyhow!(e))
347 0 : .map(|r| (ttid, r))
348 0 : }
349 0 : });
350 0 : let mut timelines_checked =
351 0 : std::pin::pin!(timelines_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
352 0 :
353 0 : // Update the GarbageList with any timelines which appear not to exist.
354 0 : let mut active_timelines: Vec<TenantShardTimelineId> = vec![];
355 0 : while let Some(result) = timelines_checked.next().await {
356 0 : let (ttid, console_result) = result?;
357 0 : if garbage.maybe_append(GarbageEntity::Timeline(ttid), console_result) {
358 0 : tracing::debug!("Timeline {ttid} is garbage");
359 : } else {
360 0 : tracing::debug!("Timeline {ttid} is active");
361 0 : active_timelines.push(ttid);
362 0 : garbage.active_timeline_count = active_timelines.len();
363 : }
364 : }
365 :
366 0 : let num_garbage_timelines = garbage
367 0 : .items
368 0 : .iter()
369 0 : .filter(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
370 0 : .count();
371 0 : tracing::info!(
372 0 : "Found {}/{} garbage timelines in active tenants",
373 0 : num_garbage_timelines,
374 0 : active_timelines.len(),
375 : );
376 :
377 0 : Ok(garbage)
378 0 : }
379 :
380 0 : #[derive(clap::ValueEnum, Debug, Clone)]
381 : pub enum PurgeMode {
382 : /// The safest mode: only delete tenants that were explicitly reported as deleted
383 : /// by Console API.
384 : DeletedOnly,
385 :
386 : /// Delete all garbage tenants, including those which are only presumed to be deleted,
387 : /// because the Console API could not find them.
388 : DeletedAndMissing,
389 : }
390 :
391 : impl std::fmt::Display for PurgeMode {
392 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
393 0 : match self {
394 0 : PurgeMode::DeletedOnly => write!(f, "deleted-only"),
395 0 : PurgeMode::DeletedAndMissing => write!(f, "deleted-and-missing"),
396 : }
397 0 : }
398 : }
399 :
400 0 : pub async fn get_tenant_objects(
401 0 : s3_client: &GenericRemoteStorage,
402 0 : tenant_shard_id: TenantShardId,
403 0 : ) -> anyhow::Result<Vec<ListingObject>> {
404 0 : tracing::debug!("Listing objects in tenant {tenant_shard_id}");
405 0 : let tenant_root = super::remote_tenant_path(&tenant_shard_id);
406 :
407 : // TODO: apply extra validation based on object modification time. Don't purge
408 : // tenants where any timeline's index_part.json has been touched recently.
409 :
410 0 : let list = s3_client
411 0 : .list(
412 0 : Some(&tenant_root),
413 0 : ListingMode::NoDelimiter,
414 0 : None,
415 0 : &CancellationToken::new(),
416 0 : )
417 0 : .await?;
418 0 : Ok(list.keys)
419 0 : }
420 :
421 0 : pub async fn get_timeline_objects(
422 0 : s3_client: &GenericRemoteStorage,
423 0 : ttid: TenantShardTimelineId,
424 0 : ) -> anyhow::Result<Vec<ListingObject>> {
425 0 : tracing::debug!("Listing objects in timeline {ttid}");
426 0 : let timeline_root = super::remote_timeline_path_id(&ttid);
427 :
428 0 : let list = s3_client
429 0 : .list(
430 0 : Some(&timeline_root),
431 0 : ListingMode::NoDelimiter,
432 0 : None,
433 0 : &CancellationToken::new(),
434 0 : )
435 0 : .await?;
436 0 : Ok(list.keys)
437 0 : }
438 :
439 : const MAX_KEYS_PER_DELETE: usize = 1000;
440 :
441 : /// Drain a buffer of keys into DeleteObjects requests
442 : ///
443 : /// If `drain` is true, drains keys completely; otherwise stops when <
444 : /// MAX_KEYS_PER_DELETE keys are left.
445 : /// `num_deleted` returns number of deleted keys.
446 0 : async fn do_delete(
447 0 : remote_client: &GenericRemoteStorage,
448 0 : keys: &mut Vec<ListingObject>,
449 0 : dry_run: bool,
450 0 : drain: bool,
451 0 : progress_tracker: &mut DeletionProgressTracker,
452 0 : ) -> anyhow::Result<()> {
453 0 : let cancel = CancellationToken::new();
454 0 : while (!keys.is_empty() && drain) || (keys.len() >= MAX_KEYS_PER_DELETE) {
455 0 : let request_keys =
456 0 : keys.split_off(keys.len() - (std::cmp::min(MAX_KEYS_PER_DELETE, keys.len())));
457 0 :
458 0 : let request_keys: Vec<RemotePath> = request_keys.into_iter().map(|o| o.key).collect();
459 0 :
460 0 : let num_deleted = request_keys.len();
461 0 : if dry_run {
462 0 : tracing::info!("Dry-run deletion of objects: ");
463 0 : for k in request_keys {
464 0 : tracing::info!(" {k:?}");
465 : }
466 : } else {
467 0 : remote_client
468 0 : .delete_objects(&request_keys, &cancel)
469 0 : .await
470 0 : .context("deletetion request")?;
471 0 : progress_tracker.register(num_deleted);
472 : }
473 : }
474 :
475 0 : Ok(())
476 0 : }
477 :
478 : /// Simple tracker reporting each 10k deleted keys.
479 : #[derive(Default)]
480 : struct DeletionProgressTracker {
481 : num_deleted: usize,
482 : last_reported_num_deleted: usize,
483 : }
484 :
485 : impl DeletionProgressTracker {
486 0 : fn register(&mut self, n: usize) {
487 0 : self.num_deleted += n;
488 0 : if self.num_deleted - self.last_reported_num_deleted > 10000 {
489 0 : tracing::info!("progress: deleted {} keys", self.num_deleted);
490 0 : self.last_reported_num_deleted = self.num_deleted;
491 0 : }
492 0 : }
493 : }
494 :
495 0 : pub async fn purge_garbage(
496 0 : input_path: String,
497 0 : mode: PurgeMode,
498 0 : min_age: Duration,
499 0 : dry_run: bool,
500 0 : ) -> anyhow::Result<()> {
501 0 : let list_bytes = tokio::fs::read(&input_path).await?;
502 0 : let garbage_list = serde_json::from_slice::<GarbageList>(&list_bytes)?;
503 0 : tracing::info!(
504 0 : "Loaded {} items in garbage list from {}",
505 0 : garbage_list.items.len(),
506 : input_path
507 : );
508 :
509 0 : let (remote_client, _target) =
510 0 : init_remote_generic(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
511 :
512 0 : assert_eq!(
513 0 : &garbage_list.bucket_config.bucket,
514 0 : remote_client.bucket_name().unwrap()
515 0 : );
516 :
517 : // Sanity checks on the incoming list
518 0 : if garbage_list.active_tenant_count == 0 {
519 0 : anyhow::bail!("Refusing to purge a garbage list that reports 0 active tenants");
520 0 : }
521 0 : if garbage_list
522 0 : .items
523 0 : .iter()
524 0 : .any(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
525 0 : && garbage_list.active_timeline_count == 0
526 : {
527 0 : anyhow::bail!("Refusing to purge a garbage list containing garbage timelines that reports 0 active timelines");
528 0 : }
529 0 :
530 0 : let filtered_items = garbage_list
531 0 : .items
532 0 : .iter()
533 0 : .filter(|i| match (&mode, &i.reason) {
534 0 : (PurgeMode::DeletedAndMissing, _) => true,
535 0 : (PurgeMode::DeletedOnly, GarbageReason::DeletedInConsole) => true,
536 0 : (PurgeMode::DeletedOnly, GarbageReason::KnownBug) => true,
537 0 : (PurgeMode::DeletedOnly, GarbageReason::MissingInConsole) => false,
538 0 : });
539 0 :
540 0 : tracing::info!(
541 0 : "Filtered down to {} garbage items based on mode {}",
542 0 : garbage_list.items.len(),
543 : mode
544 : );
545 :
546 0 : let items = tokio_stream::iter(filtered_items.map(Ok));
547 0 : let get_objects_results = items.map_ok(|i| {
548 0 : let remote_client = remote_client.clone();
549 0 : async move {
550 0 : match i.entity {
551 0 : GarbageEntity::Tenant(tenant_id) => {
552 0 : get_tenant_objects(&remote_client, tenant_id).await
553 : }
554 0 : GarbageEntity::Timeline(ttid) => get_timeline_objects(&remote_client, ttid).await,
555 : }
556 0 : }
557 0 : });
558 0 : let mut get_objects_results =
559 0 : std::pin::pin!(get_objects_results.try_buffer_unordered(S3_CONCURRENCY));
560 0 :
561 0 : let mut objects_to_delete = Vec::new();
562 0 : let mut progress_tracker = DeletionProgressTracker::default();
563 0 : while let Some(result) = get_objects_results.next().await {
564 0 : let mut object_list = result?;
565 :
566 : // Extra safety check: even if a collection of objects is garbage, check max() of modification
567 : // times before purging, so that if we incorrectly marked a live tenant as garbage then we would
568 : // notice that its index has been written recently and would omit deleting it.
569 0 : if object_list.is_empty() {
570 : // Simplify subsequent code by ensuring list always has at least one item
571 : // Usually, this only occurs if there is parallel deletions racing us, as there is no empty prefixes
572 0 : continue;
573 0 : }
574 0 : let max_mtime = object_list.iter().map(|o| o.last_modified).max().unwrap();
575 0 : let age = max_mtime.elapsed();
576 0 : match age {
577 : Err(_) => {
578 0 : tracing::warn!("Bad last_modified time");
579 0 : continue;
580 : }
581 0 : Ok(a) if a < min_age => {
582 0 : // Failed age check. This doesn't mean we did something wrong: a tenant might really be garbage and recently
583 0 : // written, but out of an abundance of caution we still don't purge it.
584 0 : tracing::info!(
585 0 : "Skipping tenant with young objects {}..{}",
586 0 : object_list.first().as_ref().unwrap().key,
587 0 : object_list.last().as_ref().unwrap().key
588 : );
589 0 : continue;
590 : }
591 0 : Ok(_) => {
592 0 : // Passed age check
593 0 : }
594 0 : }
595 0 :
596 0 : objects_to_delete.append(&mut object_list);
597 0 : if objects_to_delete.len() >= MAX_KEYS_PER_DELETE {
598 0 : do_delete(
599 0 : &remote_client,
600 0 : &mut objects_to_delete,
601 0 : dry_run,
602 0 : false,
603 0 : &mut progress_tracker,
604 0 : )
605 0 : .await?;
606 0 : }
607 : }
608 :
609 0 : do_delete(
610 0 : &remote_client,
611 0 : &mut objects_to_delete,
612 0 : dry_run,
613 0 : true,
614 0 : &mut progress_tracker,
615 0 : )
616 0 : .await?;
617 :
618 0 : tracing::info!("{} keys deleted in total", progress_tracker.num_deleted);
619 :
620 0 : Ok(())
621 0 : }
|
