LCOV - code coverage report
Current view: top level - storage_scrubber/src - garbage.rs (source / functions) Coverage Total Hit
Test: 2b0730d767f560e20b6748f57465922aa8bb805e.info Lines: 0.0 % 399 0
Test Date: 2024-09-25 14:04:07 Functions: 0.0 % 84 0

            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::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,
      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 :         // FIXME: we can't just assume that all console's region ids are aws-<something>.  This hack
     164            0 :         // will go away when we are talking to Control Plane APIs, which are per-region.
     165            0 :         .list_projects(format!("aws-{}", bucket_config.region))
     166            0 :         .await?
     167            0 :         .into_iter()
     168            0 :         .map(|t| (t.tenant, t))
     169            0 :         .collect();
     170            0 :     tracing::info!(
     171            0 :         "Loaded {} console projects tenant IDs",
     172            0 :         console_projects.len()
     173              :     );
     174              : 
     175              :     // Because many tenant shards may look up the same TenantId, we maintain a cache.
     176            0 :     let console_cache = Arc::new(std::sync::Mutex::new(ConsoleCache {
     177            0 :         projects: console_projects,
     178            0 :         not_found: HashSet::new(),
     179            0 :     }));
     180            0 : 
     181            0 :     // Enumerate Tenants in S3, and check if each one exists in Console
     182            0 :     tracing::info!("Finding all tenants in bucket {}...", bucket_config.bucket);
     183            0 :     let tenants = stream_tenants(&remote_client, &target);
     184            0 :     let tenants_checked = tenants.map_ok(|t| {
     185            0 :         let api_client = cloud_admin_api_client.clone();
     186            0 :         let console_cache = console_cache.clone();
     187            0 :         async move {
     188              :             // Check cache before issuing API call
     189            0 :             let project_data = {
     190            0 :                 let cache = console_cache.lock().unwrap();
     191            0 :                 let result = cache.projects.get(&t.tenant_id).cloned();
     192            0 :                 if result.is_none() && cache.not_found.contains(&t.tenant_id) {
     193            0 :                     return Ok((t, None));
     194            0 :                 }
     195            0 :                 result
     196            0 :             };
     197            0 : 
     198            0 :             match project_data {
     199            0 :                 Some(project_data) => Ok((t, Some(project_data.clone()))),
     200              :                 None => {
     201            0 :                     let project_data = api_client
     202            0 :                         .find_tenant_project(t.tenant_id)
     203            0 :                         .await
     204            0 :                         .map_err(|e| anyhow::anyhow!(e));
     205            0 : 
     206            0 :                     // Populate cache with result of API call
     207            0 :                     {
     208            0 :                         let mut cache = console_cache.lock().unwrap();
     209            0 :                         if let Ok(Some(project_data)) = &project_data {
     210            0 :                             cache.projects.insert(t.tenant_id, project_data.clone());
     211            0 :                         } else if let Ok(None) = &project_data {
     212            0 :                             cache.not_found.insert(t.tenant_id);
     213            0 :                         }
     214              :                     }
     215              : 
     216            0 :                     project_data.map(|r| (t, r))
     217              :                 }
     218              :             }
     219            0 :         }
     220            0 :     });
     221            0 :     let mut tenants_checked =
     222            0 :         std::pin::pin!(tenants_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
     223            0 : 
     224            0 :     // Process the results of Tenant checks.  If a Tenant is garbage, it goes into
     225            0 :     // the `GarbageList`.  Else it goes into `active_tenants` for more detailed timeline
     226            0 :     // checks if they are enabled by the `depth` parameter.
     227            0 :     let mut garbage = GarbageList::new(node_kind, bucket_config);
     228            0 :     let mut active_tenants: Vec<TenantShardId> = vec![];
     229            0 :     let mut counter = 0;
     230            0 :     while let Some(result) = tenants_checked.next().await {
     231            0 :         let (tenant_shard_id, console_result) = result?;
     232              : 
     233              :         // Paranoia check
     234            0 :         if let Some(project) = &console_result {
     235            0 :             assert!(project.tenant == tenant_shard_id.tenant_id);
     236            0 :         }
     237              : 
     238              :         // Special case: If it's missing in console, check for known bugs that would enable us to conclusively
     239              :         // identify it as purge-able anyway
     240            0 :         if console_result.is_none() {
     241            0 :             let timelines = stream_tenant_timelines(&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(
     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(
     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 = active_tenants.map_ok(|t| stream_tenant_timelines(&remote_client, &target, *t));
     335            0 :     let timelines = timelines.try_buffer_unordered(S3_CONCURRENCY);
     336            0 :     let timelines = timelines.try_flatten();
     337            0 : 
     338            0 :     // For all timelines within active tenants, call into console API to check their existence
     339            0 :     let timelines_checked = timelines.map_ok(|ttid| {
     340            0 :         let api_client = cloud_admin_api_client.clone();
     341            0 :         async move {
     342            0 :             api_client
     343            0 :                 .find_timeline_branch(ttid.tenant_shard_id.tenant_id, ttid.timeline_id)
     344            0 :                 .await
     345            0 :                 .map_err(|e| anyhow::anyhow!(e))
     346            0 :                 .map(|r| (ttid, r))
     347            0 :         }
     348            0 :     });
     349            0 :     let mut timelines_checked =
     350            0 :         std::pin::pin!(timelines_checked.try_buffer_unordered(CONSOLE_CONCURRENCY));
     351            0 : 
     352            0 :     // Update the GarbageList with any timelines which appear not to exist.
     353            0 :     let mut active_timelines: Vec<TenantShardTimelineId> = vec![];
     354            0 :     while let Some(result) = timelines_checked.next().await {
     355            0 :         let (ttid, console_result) = result?;
     356            0 :         if garbage.maybe_append(GarbageEntity::Timeline(ttid), console_result) {
     357            0 :             tracing::debug!("Timeline {ttid} is garbage");
     358              :         } else {
     359            0 :             tracing::debug!("Timeline {ttid} is active");
     360            0 :             active_timelines.push(ttid);
     361            0 :             garbage.active_timeline_count = active_timelines.len();
     362              :         }
     363              :     }
     364              : 
     365            0 :     let num_garbage_timelines = garbage
     366            0 :         .items
     367            0 :         .iter()
     368            0 :         .filter(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
     369            0 :         .count();
     370            0 :     tracing::info!(
     371            0 :         "Found {}/{} garbage timelines in active tenants",
     372            0 :         num_garbage_timelines,
     373            0 :         active_timelines.len(),
     374              :     );
     375              : 
     376            0 :     Ok(garbage)
     377            0 : }
     378              : 
     379            0 : #[derive(clap::ValueEnum, Debug, Clone)]
     380              : pub enum PurgeMode {
     381              :     /// The safest mode: only delete tenants that were explicitly reported as deleted
     382              :     /// by Console API.
     383              :     DeletedOnly,
     384              : 
     385              :     /// Delete all garbage tenants, including those which are only presumed to be deleted,
     386              :     /// because the Console API could not find them.
     387              :     DeletedAndMissing,
     388              : }
     389              : 
     390              : impl std::fmt::Display for PurgeMode {
     391            0 :     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
     392            0 :         match self {
     393            0 :             PurgeMode::DeletedOnly => write!(f, "deleted-only"),
     394            0 :             PurgeMode::DeletedAndMissing => write!(f, "deleted-and-missing"),
     395              :         }
     396            0 :     }
     397              : }
     398              : 
     399            0 : pub async fn get_tenant_objects(
     400            0 :     s3_client: &GenericRemoteStorage,
     401            0 :     tenant_shard_id: TenantShardId,
     402            0 : ) -> anyhow::Result<Vec<ListingObject>> {
     403            0 :     tracing::debug!("Listing objects in tenant {tenant_shard_id}");
     404            0 :     let tenant_root = super::remote_tenant_path(&tenant_shard_id);
     405              : 
     406              :     // TODO: apply extra validation based on object modification time.  Don't purge
     407              :     // tenants where any timeline's index_part.json has been touched recently.
     408              : 
     409            0 :     let list = s3_client
     410            0 :         .list(
     411            0 :             Some(&tenant_root),
     412            0 :             ListingMode::NoDelimiter,
     413            0 :             None,
     414            0 :             &CancellationToken::new(),
     415            0 :         )
     416            0 :         .await?;
     417            0 :     Ok(list.keys)
     418            0 : }
     419              : 
     420            0 : pub async fn get_timeline_objects(
     421            0 :     s3_client: &GenericRemoteStorage,
     422            0 :     ttid: TenantShardTimelineId,
     423            0 : ) -> anyhow::Result<Vec<ListingObject>> {
     424            0 :     tracing::debug!("Listing objects in timeline {ttid}");
     425            0 :     let timeline_root = super::remote_timeline_path_id(&ttid);
     426              : 
     427            0 :     let list = s3_client
     428            0 :         .list(
     429            0 :             Some(&timeline_root),
     430            0 :             ListingMode::NoDelimiter,
     431            0 :             None,
     432            0 :             &CancellationToken::new(),
     433            0 :         )
     434            0 :         .await?;
     435            0 :     Ok(list.keys)
     436            0 : }
     437              : 
     438              : const MAX_KEYS_PER_DELETE: usize = 1000;
     439              : 
     440              : /// Drain a buffer of keys into DeleteObjects requests
     441              : ///
     442              : /// If `drain` is true, drains keys completely; otherwise stops when <
     443              : /// MAX_KEYS_PER_DELETE keys are left.
     444              : /// `num_deleted` returns number of deleted keys.
     445            0 : async fn do_delete(
     446            0 :     remote_client: &GenericRemoteStorage,
     447            0 :     keys: &mut Vec<ListingObject>,
     448            0 :     dry_run: bool,
     449            0 :     drain: bool,
     450            0 :     progress_tracker: &mut DeletionProgressTracker,
     451            0 : ) -> anyhow::Result<()> {
     452            0 :     let cancel = CancellationToken::new();
     453            0 :     while (!keys.is_empty() && drain) || (keys.len() >= MAX_KEYS_PER_DELETE) {
     454            0 :         let request_keys =
     455            0 :             keys.split_off(keys.len() - (std::cmp::min(MAX_KEYS_PER_DELETE, keys.len())));
     456            0 : 
     457            0 :         let request_keys: Vec<RemotePath> = request_keys.into_iter().map(|o| o.key).collect();
     458            0 : 
     459            0 :         let num_deleted = request_keys.len();
     460            0 :         if dry_run {
     461            0 :             tracing::info!("Dry-run deletion of objects: ");
     462            0 :             for k in request_keys {
     463            0 :                 tracing::info!("  {k:?}");
     464              :             }
     465              :         } else {
     466            0 :             remote_client
     467            0 :                 .delete_objects(&request_keys, &cancel)
     468            0 :                 .await
     469            0 :                 .context("deletetion request")?;
     470            0 :             progress_tracker.register(num_deleted);
     471              :         }
     472              :     }
     473              : 
     474            0 :     Ok(())
     475            0 : }
     476              : 
     477              : /// Simple tracker reporting each 10k deleted keys.
     478              : #[derive(Default)]
     479              : struct DeletionProgressTracker {
     480              :     num_deleted: usize,
     481              :     last_reported_num_deleted: usize,
     482              : }
     483              : 
     484              : impl DeletionProgressTracker {
     485            0 :     fn register(&mut self, n: usize) {
     486            0 :         self.num_deleted += n;
     487            0 :         if self.num_deleted - self.last_reported_num_deleted > 10000 {
     488            0 :             tracing::info!("progress: deleted {} keys", self.num_deleted);
     489            0 :             self.last_reported_num_deleted = self.num_deleted;
     490            0 :         }
     491            0 :     }
     492              : }
     493              : 
     494            0 : pub async fn purge_garbage(
     495            0 :     input_path: String,
     496            0 :     mode: PurgeMode,
     497            0 :     min_age: Duration,
     498            0 :     dry_run: bool,
     499            0 : ) -> anyhow::Result<()> {
     500            0 :     let list_bytes = tokio::fs::read(&input_path).await?;
     501            0 :     let garbage_list = serde_json::from_slice::<GarbageList>(&list_bytes)?;
     502            0 :     tracing::info!(
     503            0 :         "Loaded {} items in garbage list from {}",
     504            0 :         garbage_list.items.len(),
     505              :         input_path
     506              :     );
     507              : 
     508            0 :     let (remote_client, _target) =
     509            0 :         init_remote(garbage_list.bucket_config.clone(), garbage_list.node_kind).await?;
     510              : 
     511            0 :     assert_eq!(
     512            0 :         &garbage_list.bucket_config.bucket,
     513            0 :         remote_client.bucket_name().unwrap()
     514            0 :     );
     515              : 
     516              :     // Sanity checks on the incoming list
     517            0 :     if garbage_list.active_tenant_count == 0 {
     518            0 :         anyhow::bail!("Refusing to purge a garbage list that reports 0 active tenants");
     519            0 :     }
     520            0 :     if garbage_list
     521            0 :         .items
     522            0 :         .iter()
     523            0 :         .any(|g| matches!(g.entity, GarbageEntity::Timeline(_)))
     524            0 :         && garbage_list.active_timeline_count == 0
     525              :     {
     526            0 :         anyhow::bail!("Refusing to purge a garbage list containing garbage timelines that reports 0 active timelines");
     527            0 :     }
     528            0 : 
     529            0 :     let filtered_items = garbage_list
     530            0 :         .items
     531            0 :         .iter()
     532            0 :         .filter(|i| match (&mode, &i.reason) {
     533            0 :             (PurgeMode::DeletedAndMissing, _) => true,
     534            0 :             (PurgeMode::DeletedOnly, GarbageReason::DeletedInConsole) => true,
     535            0 :             (PurgeMode::DeletedOnly, GarbageReason::KnownBug) => true,
     536            0 :             (PurgeMode::DeletedOnly, GarbageReason::MissingInConsole) => false,
     537            0 :         });
     538            0 : 
     539            0 :     tracing::info!(
     540            0 :         "Filtered down to {} garbage items based on mode {}",
     541            0 :         garbage_list.items.len(),
     542              :         mode
     543              :     );
     544              : 
     545            0 :     let items = tokio_stream::iter(filtered_items.map(Ok));
     546            0 :     let get_objects_results = items.map_ok(|i| {
     547            0 :         let remote_client = remote_client.clone();
     548            0 :         async move {
     549            0 :             match i.entity {
     550            0 :                 GarbageEntity::Tenant(tenant_id) => {
     551            0 :                     get_tenant_objects(&remote_client, tenant_id).await
     552              :                 }
     553            0 :                 GarbageEntity::Timeline(ttid) => get_timeline_objects(&remote_client, ttid).await,
     554              :             }
     555            0 :         }
     556            0 :     });
     557            0 :     let mut get_objects_results =
     558            0 :         std::pin::pin!(get_objects_results.try_buffer_unordered(S3_CONCURRENCY));
     559            0 : 
     560            0 :     let mut objects_to_delete = Vec::new();
     561            0 :     let mut progress_tracker = DeletionProgressTracker::default();
     562            0 :     while let Some(result) = get_objects_results.next().await {
     563            0 :         let mut object_list = result?;
     564              : 
     565              :         // Extra safety check: even if a collection of objects is garbage, check max() of modification
     566              :         // times before purging, so that if we incorrectly marked a live tenant as garbage then we would
     567              :         // notice that its index has been written recently and would omit deleting it.
     568            0 :         if object_list.is_empty() {
     569              :             // Simplify subsequent code by ensuring list always has at least one item
     570              :             // Usually, this only occurs if there is parallel deletions racing us, as there is no empty prefixes
     571            0 :             continue;
     572            0 :         }
     573            0 :         let max_mtime = object_list.iter().map(|o| o.last_modified).max().unwrap();
     574            0 :         let age = max_mtime.elapsed();
     575            0 :         match age {
     576              :             Err(_) => {
     577            0 :                 tracing::warn!("Bad last_modified time");
     578            0 :                 continue;
     579              :             }
     580            0 :             Ok(a) if a < min_age => {
     581            0 :                 // Failed age check.  This doesn't mean we did something wrong: a tenant might really be garbage and recently
     582            0 :                 // written, but out of an abundance of caution we still don't purge it.
     583            0 :                 tracing::info!(
     584            0 :                     "Skipping tenant with young objects {}..{}",
     585            0 :                     object_list.first().as_ref().unwrap().key,
     586            0 :                     object_list.last().as_ref().unwrap().key
     587              :                 );
     588            0 :                 continue;
     589              :             }
     590            0 :             Ok(_) => {
     591            0 :                 // Passed age check
     592            0 :             }
     593            0 :         }
     594            0 : 
     595            0 :         objects_to_delete.append(&mut object_list);
     596            0 :         if objects_to_delete.len() >= MAX_KEYS_PER_DELETE {
     597            0 :             do_delete(
     598            0 :                 &remote_client,
     599            0 :                 &mut objects_to_delete,
     600            0 :                 dry_run,
     601            0 :                 false,
     602            0 :                 &mut progress_tracker,
     603            0 :             )
     604            0 :             .await?;
     605            0 :         }
     606              :     }
     607              : 
     608            0 :     do_delete(
     609            0 :         &remote_client,
     610            0 :         &mut objects_to_delete,
     611            0 :         dry_run,
     612            0 :         true,
     613            0 :         &mut progress_tracker,
     614            0 :     )
     615            0 :     .await?;
     616              : 
     617            0 :     tracing::info!("{} keys deleted in total", progress_tracker.num_deleted);
     618              : 
     619            0 :     Ok(())
     620            0 : }
        

Generated by: LCOV version 2.1-beta