Line data Source code
1 : //! Functionality for finding and purging garbage, as in "garbage collection".
2 : //!
3 : //! Garbage means S3 objects which are either not referenced by any metadata,
4 : //! or are referenced by a control plane tenant/timeline in a deleted state.
5 :
6 : use std::{
7 : collections::{HashMap, HashSet},
8 : sync::Arc,
9 : time::Duration,
10 : };
11 :
12 : use anyhow::Context;
13 : use futures_util::TryStreamExt;
14 : use pageserver_api::shard::TenantShardId;
15 : use remote_storage::{GenericRemoteStorage, ListingMode, ListingObject, RemotePath};
16 : use serde::{Deserialize, Serialize};
17 : use tokio_stream::StreamExt;
18 : use tokio_util::sync::CancellationToken;
19 : use utils::{backoff, id::TenantId};
20 :
21 : use crate::{
22 : cloud_admin_api::{CloudAdminApiClient, MaybeDeleted, ProjectData},
23 : init_remote, list_objects_with_retries,
24 : metadata_stream::{stream_tenant_timelines, stream_tenants},
25 : BucketConfig, ConsoleConfig, NodeKind, TenantShardTimelineId, TraversingDepth, MAX_RETRIES,
26 : };
27 :
28 0 : #[derive(Serialize, Deserialize, Debug)]
29 : enum GarbageReason {
30 : DeletedInConsole,
31 : MissingInConsole,
32 :
33 : // The remaining data relates to a known deletion issue, and we're sure that purging this
34 : // will not delete any real data, for example https://github.com/neondatabase/neon/pull/7928 where
35 : // there is nothing in a tenant path apart from a heatmap file.
36 : KnownBug,
37 : }
38 :
39 0 : #[derive(Serialize, Deserialize, Debug)]
40 : enum GarbageEntity {
41 : Tenant(TenantShardId),
42 : Timeline(TenantShardTimelineId),
43 : }
44 :
45 0 : #[derive(Serialize, Deserialize, Debug)]
46 : struct GarbageItem {
47 : entity: GarbageEntity,
48 : reason: GarbageReason,
49 : }
50 :
51 0 : #[derive(Serialize, Deserialize, Debug)]
52 : pub struct GarbageList {
53 : /// Remember what NodeKind we were finding garbage for, so that we can
54 : /// purge the list without re-stating it.
55 : node_kind: NodeKind,
56 :
57 : /// Embed the identity of the bucket, so that we do not risk executing
58 : /// the wrong list against the wrong bucket, and so that the user does not have
59 : /// to re-state the bucket details when purging.
60 : bucket_config: BucketConfig,
61 :
62 : items: Vec<GarbageItem>,
63 :
64 : /// Advisory information to enable consumers to do a validation that if we
65 : /// see garbage, we saw some active tenants too. This protects against classes of bugs
66 : /// in the scrubber that might otherwise generate a "deleted all" result.
67 : active_tenant_count: usize,
68 : active_timeline_count: usize,
69 : }
70 :
71 : impl GarbageList {
72 0 : fn new(node_kind: NodeKind, bucket_config: BucketConfig) -> Self {
73 0 : Self {
74 0 : items: Vec::new(),
75 0 : active_tenant_count: 0,
76 0 : active_timeline_count: 0,
77 0 : node_kind,
78 0 : bucket_config,
79 0 : }
80 0 : }
81 :
82 : /// If an entity has been identified as requiring purge due to a known bug, e.g.
83 : /// a particular type of object left behind after an incomplete deletion.
84 0 : fn append_buggy(&mut self, entity: GarbageEntity) {
85 0 : self.items.push(GarbageItem {
86 0 : entity,
87 0 : reason: GarbageReason::KnownBug,
88 0 : });
89 0 : }
90 :
91 : /// Return true if appended, false if not. False means the result was not garbage.
92 0 : fn maybe_append<T>(&mut self, entity: GarbageEntity, result: Option<T>) -> bool
93 0 : where
94 0 : T: MaybeDeleted,
95 0 : {
96 0 : match result {
97 0 : Some(result_item) if result_item.is_deleted() => {
98 0 : self.items.push(GarbageItem {
99 0 : entity,
100 0 : reason: GarbageReason::DeletedInConsole,
101 0 : });
102 0 : true
103 : }
104 0 : Some(_) => false,
105 : None => {
106 0 : self.items.push(GarbageItem {
107 0 : entity,
108 0 : reason: GarbageReason::MissingInConsole,
109 0 : });
110 0 : true
111 : }
112 : }
113 0 : }
114 : }
115 :
116 0 : pub async fn find_garbage(
117 0 : bucket_config: BucketConfig,
118 0 : console_config: ConsoleConfig,
119 0 : depth: TraversingDepth,
120 0 : node_kind: NodeKind,
121 0 : output_path: String,
122 0 : ) -> anyhow::Result<()> {
123 0 : let garbage = find_garbage_inner(bucket_config, console_config, depth, node_kind).await?;
124 0 : let serialized = serde_json::to_vec_pretty(&garbage)?;
125 :
126 0 : tokio::fs::write(&output_path, &serialized).await?;
127 :
128 0 : tracing::info!("Wrote garbage report to {output_path}");
129 :
130 0 : Ok(())
131 0 : }
132 :
133 : // How many concurrent S3 operations to issue (approximately): this is the concurrency
134 : // for things like listing the timelines within tenant prefixes.
135 : const S3_CONCURRENCY: usize = 32;
136 :
137 : // How many concurrent API requests to make to the console API.
138 : //
139 : // Be careful increasing this; roughly we shouldn't have more than ~100 rps. It
140 : // would be better to implement real rsp limiter.
141 : const CONSOLE_CONCURRENCY: usize = 16;
142 :
143 : struct ConsoleCache {
144 : /// Set of tenants found in the control plane API
145 : projects: HashMap<TenantId, ProjectData>,
146 : /// Set of tenants for which the control plane API returned 404
147 : not_found: HashSet<TenantId>,
148 : }
149 :
150 0 : async fn find_garbage_inner(
151 0 : bucket_config: BucketConfig,
152 0 : console_config: ConsoleConfig,
153 0 : depth: TraversingDepth,
154 0 : node_kind: NodeKind,
155 0 : ) -> anyhow::Result<GarbageList> {
156 : // Construct clients for S3 and for Console API
157 0 : let (remote_client, target) = init_remote(bucket_config.clone(), node_kind).await?;
158 0 : let cloud_admin_api_client = Arc::new(CloudAdminApiClient::new(console_config));
159 :
160 : // Build a set of console-known tenants, for quickly eliminating known-active tenants without having
161 : // to issue O(N) console API requests.
162 0 : let console_projects: HashMap<TenantId, ProjectData> = cloud_admin_api_client
163 0 : .list_projects()
164 0 : .await?
165 0 : .into_iter()
166 0 : .map(|t| (t.tenant, t))
167 0 : .collect();
168 0 : tracing::info!(
169 0 : "Loaded {} console projects tenant IDs",
170 0 : console_projects.len()
171 : );
172 :
173 : // Because many tenant shards may look up the same TenantId, we maintain a cache.
174 0 : let console_cache = Arc::new(std::sync::Mutex::new(ConsoleCache {
175 0 : projects: console_projects,
176 0 : not_found: HashSet::new(),
177 0 : }));
178 0 :
179 0 : // Enumerate Tenants in S3, and check if each one exists in Console
180 0 : tracing::info!("Finding all tenants in bucket {}...", bucket_config.bucket);
181 0 : let tenants = stream_tenants(&remote_client, &target);
182 0 : let tenants_checked = tenants.map_ok(|t| {
183 0 : let api_client = cloud_admin_api_client.clone();
184 0 : let console_cache = console_cache.clone();
185 0 : async move {
186 : // Check cache before issuing API call
187 0 : let project_data = {
188 0 : let cache = console_cache.lock().unwrap();
189 0 : let result = cache.projects.get(&t.tenant_id).cloned();
190 0 : if result.is_none() && cache.not_found.contains(&t.tenant_id) {
191 0 : return Ok((t, None));
192 0 : }
193 0 : result
194 0 : };
195 0 :
196 0 : match project_data {
197 0 : Some(project_data) => Ok((t, Some(project_data.clone()))),
198 : None => {
199 0 : let project_data = api_client
200 0 : .find_tenant_project(t.tenant_id)
201 0 : .await
202 0 : .map_err(|e| anyhow::anyhow!(e));
203 0 :
204 0 : // Populate cache with result of API call
205 0 : {
206 0 : let mut cache = console_cache.lock().unwrap();
207 0 : if let Ok(Some(project_data)) = &project_data {
208 0 : cache.projects.insert(t.tenant_id, project_data.clone());
209 0 : } else if let Ok(None) = &project_data {
210 0 : cache.not_found.insert(t.tenant_id);
211 0 : }
212 : }
213 :
214 0 : project_data.map(|r| (t, r))
215 : }
216 : }
217 0 : }
218 0 : });
219 0 : let mut tenants_checked =
220 0 : std::pin::pin!(tenants_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
221 0 :
222 0 : // Process the results of Tenant checks. If a Tenant is garbage, it goes into
223 0 : // the `GarbageList`. Else it goes into `active_tenants` for more detailed timeline
224 0 : // checks if they are enabled by the `depth` parameter.
225 0 : let mut garbage = GarbageList::new(node_kind, bucket_config);
226 0 : let mut active_tenants: Vec<TenantShardId> = vec![];
227 0 : let mut counter = 0;
228 0 : while let Some(result) = tenants_checked.next().await {
229 0 : let (tenant_shard_id, console_result) = result?;
230 :
231 : // Paranoia check
232 0 : if let Some(project) = &console_result {
233 0 : assert!(project.tenant == tenant_shard_id.tenant_id);
234 0 : }
235 :
236 : // Special case: If it's missing in console, check for known bugs that would enable us to conclusively
237 : // identify it as purge-able anyway
238 0 : if console_result.is_none() {
239 0 : let timelines = stream_tenant_timelines(&remote_client, &target, tenant_shard_id)
240 0 : .await?
241 0 : .collect::<Vec<_>>()
242 0 : .await;
243 0 : if timelines.is_empty() {
244 : // No timelines, but a heatmap: the deletion bug where we deleted everything but heatmaps
245 0 : let tenant_objects = list_objects_with_retries(
246 0 : &remote_client,
247 0 : ListingMode::WithDelimiter,
248 0 : &target.tenant_root(&tenant_shard_id),
249 0 : )
250 0 : .await?;
251 0 : if let Some(object) = tenant_objects.keys.first() {
252 0 : if object.key.get_path().as_str().ends_with("heatmap-v1.json") {
253 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and is only a heatmap (known historic deletion bug)");
254 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
255 0 : continue;
256 : } else {
257 0 : tracing::info!("Tenant {tenant_shard_id} is missing in console and contains one object: {}", object.key);
258 : }
259 : } else {
260 0 : tracing::info!("Tenant {tenant_shard_id} is missing in console appears to have been deleted while we ran");
261 : }
262 : } else {
263 : // A console-unknown tenant with timelines: check if these timelines only contain initdb.tar.zst, from the initial
264 : // rollout of WAL DR in which we never deleted these.
265 0 : let mut any_non_initdb = false;
266 :
267 0 : for timeline_r in timelines {
268 0 : let timeline = timeline_r?;
269 0 : let timeline_objects = list_objects_with_retries(
270 0 : &remote_client,
271 0 : ListingMode::WithDelimiter,
272 0 : &target.timeline_root(&timeline),
273 0 : )
274 0 : .await?;
275 0 : if !timeline_objects.prefixes.is_empty() {
276 0 : // Sub-paths? Unexpected
277 0 : any_non_initdb = true;
278 0 : } else {
279 0 : let object = timeline_objects.keys.first().unwrap();
280 0 : if object.key.get_path().as_str().ends_with("initdb.tar.zst") {
281 0 : tracing::info!("Timeline {timeline} contains only initdb.tar.zst");
282 0 : } else {
283 0 : any_non_initdb = true;
284 0 : }
285 : }
286 : }
287 :
288 0 : if any_non_initdb {
289 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains timelines, one or more of which are more than just initdb");
290 : } else {
291 0 : tracing::info!("Tenant {tenant_shard_id}: is missing in console and contains only timelines that only contain initdb");
292 0 : garbage.append_buggy(GarbageEntity::Tenant(tenant_shard_id));
293 0 : continue;
294 : }
295 : }
296 0 : }
297 :
298 0 : if garbage.maybe_append(GarbageEntity::Tenant(tenant_shard_id), console_result) {
299 0 : tracing::debug!("Tenant {tenant_shard_id} is garbage");
300 : } else {
301 0 : tracing::debug!("Tenant {tenant_shard_id} is active");
302 0 : active_tenants.push(tenant_shard_id);
303 0 : garbage.active_tenant_count = active_tenants.len();
304 : }
305 :
306 0 : counter += 1;
307 0 : if counter % 1000 == 0 {
308 0 : tracing::info!(
309 0 : "Progress: {counter} tenants checked, {} active, {} garbage",
310 0 : active_tenants.len(),
311 0 : garbage.items.len()
312 : );
313 0 : }
314 : }
315 :
316 0 : tracing::info!(
317 0 : "Found {}/{} garbage tenants",
318 0 : garbage.items.len(),
319 0 : garbage.items.len() + active_tenants.len()
320 : );
321 :
322 : // If we are only checking tenant-deep, we are done. Otherwise we must
323 : // proceed to check the individual timelines of the active tenants.
324 0 : if depth == TraversingDepth::Tenant {
325 0 : return Ok(garbage);
326 0 : }
327 0 :
328 0 : tracing::info!(
329 0 : "Checking timelines for {} active tenants",
330 0 : active_tenants.len(),
331 : );
332 :
333 : // Construct a stream of all timelines within active tenants
334 0 : let active_tenants = tokio_stream::iter(active_tenants.iter().map(Ok));
335 0 : let timelines = active_tenants.map_ok(|t| stream_tenant_timelines(&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 0 :
407 0 : // TODO: apply extra validation based on object modification time. Don't purge
408 0 : // tenants where any timeline's index_part.json has been touched recently.
409 0 :
410 0 : let cancel = CancellationToken::new();
411 0 : let list = backoff::retry(
412 0 : || s3_client.list(Some(&tenant_root), ListingMode::NoDelimiter, None, &cancel),
413 0 : |_| false,
414 0 : 3,
415 0 : MAX_RETRIES as u32,
416 0 : "get_tenant_objects",
417 0 : &cancel,
418 0 : )
419 0 : .await
420 0 : .expect("dummy cancellation token")?;
421 0 : Ok(list.keys)
422 0 : }
423 :
424 0 : pub async fn get_timeline_objects(
425 0 : s3_client: &GenericRemoteStorage,
426 0 : ttid: TenantShardTimelineId,
427 0 : ) -> anyhow::Result<Vec<ListingObject>> {
428 0 : tracing::debug!("Listing objects in timeline {ttid}");
429 0 : let timeline_root = super::remote_timeline_path_id(&ttid);
430 0 :
431 0 : let cancel = CancellationToken::new();
432 0 : let list = backoff::retry(
433 0 : || {
434 0 : s3_client.list(
435 0 : Some(&timeline_root),
436 0 : ListingMode::NoDelimiter,
437 0 : None,
438 0 : &cancel,
439 0 : )
440 0 : },
441 0 : |_| false,
442 0 : 3,
443 0 : MAX_RETRIES as u32,
444 0 : "get_timeline_objects",
445 0 : &cancel,
446 0 : )
447 0 : .await
448 0 : .expect("dummy cancellation token")?;
449 :
450 0 : Ok(list.keys)
451 0 : }
452 :
453 : const MAX_KEYS_PER_DELETE: usize = 1000;
454 :
455 : /// Drain a buffer of keys into DeleteObjects requests
456 : ///
457 : /// If `drain` is true, drains keys completely; otherwise stops when <
458 : /// MAX_KEYS_PER_DELETE keys are left.
459 : /// `num_deleted` returns number of deleted keys.
460 0 : async fn do_delete(
461 0 : remote_client: &GenericRemoteStorage,
462 0 : keys: &mut Vec<ListingObject>,
463 0 : dry_run: bool,
464 0 : drain: bool,
465 0 : progress_tracker: &mut DeletionProgressTracker,
466 0 : ) -> anyhow::Result<()> {
467 0 : let cancel = CancellationToken::new();
468 0 : while (!keys.is_empty() && drain) || (keys.len() >= MAX_KEYS_PER_DELETE) {
469 0 : let request_keys =
470 0 : keys.split_off(keys.len() - (std::cmp::min(MAX_KEYS_PER_DELETE, keys.len())));
471 0 :
472 0 : let request_keys: Vec<RemotePath> = request_keys.into_iter().map(|o| o.key).collect();
473 0 :
474 0 : let num_deleted = request_keys.len();
475 0 : if dry_run {
476 0 : tracing::info!("Dry-run deletion of objects: ");
477 0 : for k in request_keys {
478 0 : tracing::info!(" {k:?}");
479 : }
480 : } else {
481 0 : remote_client
482 0 : .delete_objects(&request_keys, &cancel)
483 0 : .await
484 0 : .context("deletetion request")?;
485 0 : progress_tracker.register(num_deleted);
486 : }
487 : }
488 :
489 0 : Ok(())
490 0 : }
491 :
492 : /// Simple tracker reporting each 10k deleted keys.
493 : #[derive(Default)]
494 : struct DeletionProgressTracker {
495 : num_deleted: usize,
496 : last_reported_num_deleted: usize,
497 : }
498 :
499 : impl DeletionProgressTracker {
500 0 : fn register(&mut self, n: usize) {
501 0 : self.num_deleted += n;
502 0 : if self.num_deleted - self.last_reported_num_deleted > 10000 {
503 0 : tracing::info!("progress: deleted {} keys", self.num_deleted);
504 0 : self.last_reported_num_deleted = self.num_deleted;
505 0 : }
506 0 : }
507 : }
508 :
509 0 : pub async fn purge_garbage(
510 0 : input_path: String,
511 0 : mode: PurgeMode,
512 0 : min_age: Duration,
513 0 : dry_run: bool,
514 0 : ) -> anyhow::Result<()> {
515 0 : let list_bytes = tokio::fs::read(&input_path).await?;
516 0 : let garbage_list = serde_json::from_slice::<GarbageList>(&list_bytes)?;
517 0 : tracing::info!(
518 0 : "Loaded {} items in garbage list from {}",
519 0 : garbage_list.items.len(),
520 : input_path
521 : );
522 :
523 0 : let (remote_client, _target) =
524 0 : init_remote(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
525 :
526 0 : assert_eq!(
527 0 : &garbage_list.bucket_config.bucket,
528 0 : remote_client.bucket_name().unwrap()
529 0 : );
530 :
531 : // Sanity checks on the incoming list
532 0 : if garbage_list.active_tenant_count == 0 {
533 0 : anyhow::bail!("Refusing to purge a garbage list that reports 0 active tenants");
534 0 : }
535 0 : if garbage_list
536 0 : .items
537 0 : .iter()
538 0 : .any(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
539 0 : && garbage_list.active_timeline_count == 0
540 : {
541 0 : anyhow::bail!("Refusing to purge a garbage list containing garbage timelines that reports 0 active timelines");
542 0 : }
543 0 :
544 0 : let filtered_items = garbage_list
545 0 : .items
546 0 : .iter()
547 0 : .filter(|i| match (&mode, &i.reason) {
548 0 : (PurgeMode::DeletedAndMissing, _) => true,
549 0 : (PurgeMode::DeletedOnly, GarbageReason::DeletedInConsole) => true,
550 0 : (PurgeMode::DeletedOnly, GarbageReason::KnownBug) => true,
551 0 : (PurgeMode::DeletedOnly, GarbageReason::MissingInConsole) => false,
552 0 : });
553 0 :
554 0 : tracing::info!(
555 0 : "Filtered down to {} garbage items based on mode {}",
556 0 : garbage_list.items.len(),
557 : mode
558 : );
559 :
560 0 : let items = tokio_stream::iter(filtered_items.map(Ok));
561 0 : let get_objects_results = items.map_ok(|i| {
562 0 : let remote_client = remote_client.clone();
563 0 : async move {
564 0 : match i.entity {
565 0 : GarbageEntity::Tenant(tenant_id) => {
566 0 : get_tenant_objects(&remote_client, tenant_id).await
567 : }
568 0 : GarbageEntity::Timeline(ttid) => get_timeline_objects(&remote_client, ttid).await,
569 : }
570 0 : }
571 0 : });
572 0 : let mut get_objects_results =
573 0 : std::pin::pin!(get_objects_results.try_buffer_unordered(S3_CONCURRENCY));
574 0 :
575 0 : let mut objects_to_delete = Vec::new();
576 0 : let mut progress_tracker = DeletionProgressTracker::default();
577 0 : while let Some(result) = get_objects_results.next().await {
578 0 : let mut object_list = result?;
579 :
580 : // Extra safety check: even if a collection of objects is garbage, check max() of modification
581 : // times before purging, so that if we incorrectly marked a live tenant as garbage then we would
582 : // notice that its index has been written recently and would omit deleting it.
583 0 : if object_list.is_empty() {
584 : // Simplify subsequent code by ensuring list always has at least one item
585 : // Usually, this only occurs if there is parallel deletions racing us, as there is no empty prefixes
586 0 : continue;
587 0 : }
588 0 : let max_mtime = object_list.iter().map(|o| o.last_modified).max().unwrap();
589 0 : let age = max_mtime.elapsed();
590 0 : match age {
591 : Err(_) => {
592 0 : tracing::warn!("Bad last_modified time");
593 0 : continue;
594 : }
595 0 : Ok(a) if a < min_age => {
596 0 : // Failed age check. This doesn't mean we did something wrong: a tenant might really be garbage and recently
597 0 : // written, but out of an abundance of caution we still don't purge it.
598 0 : tracing::info!(
599 0 : "Skipping tenant with young objects {}..{}",
600 0 : object_list.first().as_ref().unwrap().key,
601 0 : object_list.last().as_ref().unwrap().key
602 : );
603 0 : continue;
604 : }
605 0 : Ok(_) => {
606 0 : // Passed age check
607 0 : }
608 0 : }
609 0 :
610 0 : objects_to_delete.append(&mut object_list);
611 0 : if objects_to_delete.len() >= MAX_KEYS_PER_DELETE {
612 0 : do_delete(
613 0 : &remote_client,
614 0 : &mut objects_to_delete,
615 0 : dry_run,
616 0 : false,
617 0 : &mut progress_tracker,
618 0 : )
619 0 : .await?;
620 0 : }
621 : }
622 :
623 0 : do_delete(
624 0 : &remote_client,
625 0 : &mut objects_to_delete,
626 0 : dry_run,
627 0 : true,
628 0 : &mut progress_tracker,
629 0 : )
630 0 : .await?;
631 :
632 0 : tracing::info!("{} keys deleted in total", progress_tracker.num_deleted);
633 :
634 0 : Ok(())
635 0 : }
|
