LCOV - code coverage report
Current view: top level - safekeeper/src - wal_backup.rs (source / functions) Coverage Total Hit
Test: 20b6afc7b7f34578dcaab2b3acdaecfe91cd8bf1.info Lines: 0.0 % 435 0
Test Date: 2024-11-25 17:48:16 Functions: 0.0 % 61 0

            Line data    Source code
       1              : use anyhow::{Context, Result};
       2              : 
       3              : use camino::{Utf8Path, Utf8PathBuf};
       4              : use futures::stream::FuturesOrdered;
       5              : use futures::StreamExt;
       6              : use tokio::task::JoinHandle;
       7              : use tokio_util::sync::CancellationToken;
       8              : use utils::backoff;
       9              : use utils::id::NodeId;
      10              : 
      11              : use std::cmp::min;
      12              : use std::collections::HashSet;
      13              : use std::num::NonZeroU32;
      14              : use std::pin::Pin;
      15              : use std::time::Duration;
      16              : 
      17              : use postgres_ffi::v14::xlog_utils::XLogSegNoOffsetToRecPtr;
      18              : use postgres_ffi::XLogFileName;
      19              : use postgres_ffi::{XLogSegNo, PG_TLI};
      20              : use remote_storage::{
      21              :     DownloadOpts, GenericRemoteStorage, ListingMode, RemotePath, StorageMetadata,
      22              : };
      23              : use tokio::fs::File;
      24              : 
      25              : use tokio::select;
      26              : use tokio::sync::mpsc::{self, Receiver, Sender};
      27              : use tokio::sync::{watch, OnceCell};
      28              : use tracing::*;
      29              : 
      30              : use utils::{id::TenantTimelineId, lsn::Lsn};
      31              : 
      32              : use crate::metrics::{BACKED_UP_SEGMENTS, BACKUP_ERRORS, WAL_BACKUP_TASKS};
      33              : use crate::timeline::{PeerInfo, WalResidentTimeline};
      34              : use crate::timeline_manager::{Manager, StateSnapshot};
      35              : use crate::{SafeKeeperConf, WAL_BACKUP_RUNTIME};
      36              : 
      37              : const UPLOAD_FAILURE_RETRY_MIN_MS: u64 = 10;
      38              : const UPLOAD_FAILURE_RETRY_MAX_MS: u64 = 5000;
      39              : 
      40              : /// Default buffer size when interfacing with [`tokio::fs::File`].
      41              : const BUFFER_SIZE: usize = 32 * 1024;
      42              : 
      43              : pub struct WalBackupTaskHandle {
      44              :     shutdown_tx: Sender<()>,
      45              :     handle: JoinHandle<()>,
      46              : }
      47              : 
      48              : impl WalBackupTaskHandle {
      49            0 :     pub(crate) async fn join(self) {
      50            0 :         if let Err(e) = self.handle.await {
      51            0 :             error!("WAL backup task panicked: {}", e);
      52            0 :         }
      53            0 :     }
      54              : }
      55              : 
      56              : /// Do we have anything to upload to S3, i.e. should safekeepers run backup activity?
      57            0 : pub(crate) fn is_wal_backup_required(
      58            0 :     wal_seg_size: usize,
      59            0 :     num_computes: usize,
      60            0 :     state: &StateSnapshot,
      61            0 : ) -> bool {
      62            0 :     num_computes > 0 ||
      63              :     // Currently only the whole segment is offloaded, so compare segment numbers.
      64            0 :     (state.commit_lsn.segment_number(wal_seg_size) > state.backup_lsn.segment_number(wal_seg_size))
      65            0 : }
      66              : 
      67              : /// Based on peer information determine which safekeeper should offload; if it
      68              : /// is me, run (per timeline) task, if not yet. OTOH, if it is not me and task
      69              : /// is running, kill it.
      70            0 : pub(crate) async fn update_task(mgr: &mut Manager, need_backup: bool, state: &StateSnapshot) {
      71            0 :     let (offloader, election_dbg_str) =
      72            0 :         determine_offloader(&state.peers, state.backup_lsn, mgr.tli.ttid, &mgr.conf);
      73            0 :     let elected_me = Some(mgr.conf.my_id) == offloader;
      74              : 
      75            0 :     let should_task_run = need_backup && elected_me;
      76              : 
      77              :     // start or stop the task
      78            0 :     if should_task_run != (mgr.backup_task.is_some()) {
      79            0 :         if should_task_run {
      80            0 :             info!("elected for backup: {}", election_dbg_str);
      81              : 
      82            0 :             let (shutdown_tx, shutdown_rx) = mpsc::channel(1);
      83              : 
      84            0 :             let Ok(resident) = mgr.wal_resident_timeline() else {
      85            0 :                 info!("Timeline shut down");
      86            0 :                 return;
      87              :             };
      88              : 
      89            0 :             let async_task = backup_task_main(resident, mgr.conf.backup_parallel_jobs, shutdown_rx);
      90              : 
      91            0 :             let handle = if mgr.conf.current_thread_runtime {
      92            0 :                 tokio::spawn(async_task)
      93              :             } else {
      94            0 :                 WAL_BACKUP_RUNTIME.spawn(async_task)
      95              :             };
      96              : 
      97            0 :             mgr.backup_task = Some(WalBackupTaskHandle {
      98            0 :                 shutdown_tx,
      99            0 :                 handle,
     100            0 :             });
     101              :         } else {
     102            0 :             if !need_backup {
     103              :                 // don't need backup at all
     104            0 :                 info!("stepping down from backup, need_backup={}", need_backup);
     105              :             } else {
     106              :                 // someone else has been elected
     107            0 :                 info!("stepping down from backup: {}", election_dbg_str);
     108              :             }
     109            0 :             shut_down_task(&mut mgr.backup_task).await;
     110              :         }
     111            0 :     }
     112            0 : }
     113              : 
     114            0 : async fn shut_down_task(entry: &mut Option<WalBackupTaskHandle>) {
     115            0 :     if let Some(wb_handle) = entry.take() {
     116              :         // Tell the task to shutdown. Error means task exited earlier, that's ok.
     117            0 :         let _ = wb_handle.shutdown_tx.send(()).await;
     118              :         // Await the task itself. TODO: restart panicked tasks earlier.
     119            0 :         wb_handle.join().await;
     120            0 :     }
     121            0 : }
     122              : 
     123              : /// The goal is to ensure that normally only one safekeepers offloads. However,
     124              : /// it is fine (and inevitable, as s3 doesn't provide CAS) that for some short
     125              : /// time we have several ones as they PUT the same files. Also,
     126              : /// - frequently changing the offloader would be bad;
     127              : /// - electing seriously lagging safekeeper is undesirable;
     128              : ///
     129              : /// So we deterministically choose among the reasonably caught up candidates.
     130              : /// TODO: take into account failed attempts to deal with hypothetical situation
     131              : /// where s3 is unreachable only for some sks.
     132            0 : fn determine_offloader(
     133            0 :     alive_peers: &[PeerInfo],
     134            0 :     wal_backup_lsn: Lsn,
     135            0 :     ttid: TenantTimelineId,
     136            0 :     conf: &SafeKeeperConf,
     137            0 : ) -> (Option<NodeId>, String) {
     138            0 :     // TODO: remove this once we fill newly joined safekeepers since backup_lsn.
     139            0 :     let capable_peers = alive_peers
     140            0 :         .iter()
     141            0 :         .filter(|p| p.local_start_lsn <= wal_backup_lsn);
     142            0 :     match capable_peers.clone().map(|p| p.commit_lsn).max() {
     143            0 :         None => (None, "no connected peers to elect from".to_string()),
     144            0 :         Some(max_commit_lsn) => {
     145            0 :             let threshold = max_commit_lsn
     146            0 :                 .checked_sub(conf.max_offloader_lag_bytes)
     147            0 :                 .unwrap_or(Lsn(0));
     148            0 :             let mut caughtup_peers = capable_peers
     149            0 :                 .clone()
     150            0 :                 .filter(|p| p.commit_lsn >= threshold)
     151            0 :                 .collect::<Vec<_>>();
     152            0 :             caughtup_peers.sort_by(|p1, p2| p1.sk_id.cmp(&p2.sk_id));
     153            0 : 
     154            0 :             // To distribute the load, shift by timeline_id.
     155            0 :             let offloader = caughtup_peers
     156            0 :                 [(u128::from(ttid.timeline_id) % caughtup_peers.len() as u128) as usize]
     157            0 :                 .sk_id;
     158            0 : 
     159            0 :             let mut capable_peers_dbg = capable_peers
     160            0 :                 .map(|p| (p.sk_id, p.commit_lsn))
     161            0 :                 .collect::<Vec<_>>();
     162            0 :             capable_peers_dbg.sort_by(|p1, p2| p1.0.cmp(&p2.0));
     163            0 :             (
     164            0 :                 Some(offloader),
     165            0 :                 format!(
     166            0 :                     "elected {} among {:?} peers, with {} of them being caughtup",
     167            0 :                     offloader,
     168            0 :                     capable_peers_dbg,
     169            0 :                     caughtup_peers.len()
     170            0 :                 ),
     171            0 :             )
     172              :         }
     173              :     }
     174            0 : }
     175              : 
     176              : static REMOTE_STORAGE: OnceCell<Option<GenericRemoteStorage>> = OnceCell::const_new();
     177              : 
     178              : // Storage must be configured and initialized when this is called.
     179            0 : fn get_configured_remote_storage() -> &'static GenericRemoteStorage {
     180            0 :     REMOTE_STORAGE
     181            0 :         .get()
     182            0 :         .expect("failed to get remote storage")
     183            0 :         .as_ref()
     184            0 :         .unwrap()
     185            0 : }
     186              : 
     187            0 : pub async fn init_remote_storage(conf: &SafeKeeperConf) {
     188            0 :     // TODO: refactor REMOTE_STORAGE to avoid using global variables, and provide
     189            0 :     // dependencies to all tasks instead.
     190            0 :     REMOTE_STORAGE
     191            0 :         .get_or_init(|| async {
     192            0 :             if let Some(conf) = conf.remote_storage.as_ref() {
     193              :                 Some(
     194            0 :                     GenericRemoteStorage::from_config(conf)
     195            0 :                         .await
     196            0 :                         .expect("failed to create remote storage"),
     197              :                 )
     198              :             } else {
     199            0 :                 None
     200              :             }
     201            0 :         })
     202            0 :         .await;
     203            0 : }
     204              : 
     205              : struct WalBackupTask {
     206              :     timeline: WalResidentTimeline,
     207              :     timeline_dir: Utf8PathBuf,
     208              :     wal_seg_size: usize,
     209              :     parallel_jobs: usize,
     210              :     commit_lsn_watch_rx: watch::Receiver<Lsn>,
     211              : }
     212              : 
     213              : /// Offload single timeline.
     214            0 : #[instrument(name = "wal_backup", skip_all, fields(ttid = %tli.ttid))]
     215              : async fn backup_task_main(
     216              :     tli: WalResidentTimeline,
     217              :     parallel_jobs: usize,
     218              :     mut shutdown_rx: Receiver<()>,
     219              : ) {
     220              :     let _guard = WAL_BACKUP_TASKS.guard();
     221              :     info!("started");
     222              : 
     223              :     let cancel = tli.tli.cancel.clone();
     224              :     let mut wb = WalBackupTask {
     225              :         wal_seg_size: tli.get_wal_seg_size().await,
     226              :         commit_lsn_watch_rx: tli.get_commit_lsn_watch_rx(),
     227              :         timeline_dir: tli.get_timeline_dir(),
     228              :         timeline: tli,
     229              :         parallel_jobs,
     230              :     };
     231              : 
     232              :     // task is spinned up only when wal_seg_size already initialized
     233              :     assert!(wb.wal_seg_size > 0);
     234              : 
     235              :     let mut canceled = false;
     236              :     select! {
     237              :         _ = wb.run() => {}
     238              :         _ = shutdown_rx.recv() => {
     239              :             canceled = true;
     240              :         },
     241              :         _ = cancel.cancelled() => {
     242              :             canceled = true;
     243              :         }
     244              :     }
     245              :     info!("task {}", if canceled { "canceled" } else { "terminated" });
     246              : }
     247              : 
     248              : impl WalBackupTask {
     249              :     /// This function must be called from a select! that also respects self.timeline's
     250              :     /// cancellation token.  This is done in [`backup_task_main`].
     251              :     ///
     252              :     /// The future returned by this function is safe to drop at any time because it
     253              :     /// does not write to local disk.
     254            0 :     async fn run(&mut self) {
     255            0 :         let mut backup_lsn = Lsn(0);
     256            0 : 
     257            0 :         let mut retry_attempt = 0u32;
     258              :         // offload loop
     259            0 :         while !self.timeline.cancel.is_cancelled() {
     260            0 :             if retry_attempt == 0 {
     261              :                 // wait for new WAL to arrive
     262            0 :                 if let Err(e) = self.commit_lsn_watch_rx.changed().await {
     263              :                     // should never happen, as we hold Arc to timeline and transmitter's lifetime
     264              :                     // is within Timeline's
     265            0 :                     error!("commit_lsn watch shut down: {:?}", e);
     266            0 :                     return;
     267            0 :                 };
     268              :             } else {
     269              :                 // or just sleep if we errored previously
     270            0 :                 let mut retry_delay = UPLOAD_FAILURE_RETRY_MAX_MS;
     271            0 :                 if let Some(backoff_delay) = UPLOAD_FAILURE_RETRY_MIN_MS.checked_shl(retry_attempt)
     272            0 :                 {
     273            0 :                     retry_delay = min(retry_delay, backoff_delay);
     274            0 :                 }
     275            0 :                 tokio::time::sleep(Duration::from_millis(retry_delay)).await;
     276              :             }
     277              : 
     278            0 :             let commit_lsn = *self.commit_lsn_watch_rx.borrow();
     279            0 : 
     280            0 :             // Note that backup_lsn can be higher than commit_lsn if we
     281            0 :             // don't have much local WAL and others already uploaded
     282            0 :             // segments we don't even have.
     283            0 :             if backup_lsn.segment_number(self.wal_seg_size)
     284            0 :                 >= commit_lsn.segment_number(self.wal_seg_size)
     285              :             {
     286            0 :                 retry_attempt = 0;
     287            0 :                 continue; /* nothing to do, common case as we wake up on every commit_lsn bump */
     288            0 :             }
     289            0 :             // Perhaps peers advanced the position, check shmem value.
     290            0 :             backup_lsn = self.timeline.get_wal_backup_lsn().await;
     291            0 :             if backup_lsn.segment_number(self.wal_seg_size)
     292            0 :                 >= commit_lsn.segment_number(self.wal_seg_size)
     293              :             {
     294            0 :                 retry_attempt = 0;
     295            0 :                 continue;
     296            0 :             }
     297            0 : 
     298            0 :             match backup_lsn_range(
     299            0 :                 &self.timeline,
     300            0 :                 &mut backup_lsn,
     301            0 :                 commit_lsn,
     302            0 :                 self.wal_seg_size,
     303            0 :                 &self.timeline_dir,
     304            0 :                 self.parallel_jobs,
     305            0 :             )
     306            0 :             .await
     307              :             {
     308            0 :                 Ok(()) => {
     309            0 :                     retry_attempt = 0;
     310            0 :                 }
     311            0 :                 Err(e) => {
     312            0 :                     error!(
     313            0 :                         "failed while offloading range {}-{}: {:?}",
     314              :                         backup_lsn, commit_lsn, e
     315              :                     );
     316              : 
     317            0 :                     retry_attempt = retry_attempt.saturating_add(1);
     318              :                 }
     319              :             }
     320              :         }
     321            0 :     }
     322              : }
     323              : 
     324            0 : async fn backup_lsn_range(
     325            0 :     timeline: &WalResidentTimeline,
     326            0 :     backup_lsn: &mut Lsn,
     327            0 :     end_lsn: Lsn,
     328            0 :     wal_seg_size: usize,
     329            0 :     timeline_dir: &Utf8Path,
     330            0 :     parallel_jobs: usize,
     331            0 : ) -> Result<()> {
     332            0 :     if parallel_jobs < 1 {
     333            0 :         anyhow::bail!("parallel_jobs must be >= 1");
     334            0 :     }
     335            0 : 
     336            0 :     let remote_timeline_path = &timeline.remote_path;
     337            0 :     let start_lsn = *backup_lsn;
     338            0 :     let segments = get_segments(start_lsn, end_lsn, wal_seg_size);
     339            0 : 
     340            0 :     // Pool of concurrent upload tasks. We use `FuturesOrdered` to
     341            0 :     // preserve order of uploads, and update `backup_lsn` only after
     342            0 :     // all previous uploads are finished.
     343            0 :     let mut uploads = FuturesOrdered::new();
     344            0 :     let mut iter = segments.iter();
     345              : 
     346              :     loop {
     347            0 :         let added_task = match iter.next() {
     348            0 :             Some(s) => {
     349            0 :                 uploads.push_back(backup_single_segment(s, timeline_dir, remote_timeline_path));
     350            0 :                 true
     351              :             }
     352            0 :             None => false,
     353              :         };
     354              : 
     355              :         // Wait for the next segment to upload if we don't have any more segments,
     356              :         // or if we have too many concurrent uploads.
     357            0 :         if !added_task || uploads.len() >= parallel_jobs {
     358            0 :             let next = uploads.next().await;
     359            0 :             if let Some(res) = next {
     360              :                 // next segment uploaded
     361            0 :                 let segment = res?;
     362            0 :                 let new_backup_lsn = segment.end_lsn;
     363            0 :                 timeline
     364            0 :                     .set_wal_backup_lsn(new_backup_lsn)
     365            0 :                     .await
     366            0 :                     .context("setting wal_backup_lsn")?;
     367            0 :                 *backup_lsn = new_backup_lsn;
     368              :             } else {
     369              :                 // no more segments to upload
     370            0 :                 break;
     371              :             }
     372            0 :         }
     373              :     }
     374              : 
     375            0 :     info!(
     376            0 :         "offloaded segnos {:?} up to {}, previous backup_lsn {}",
     377            0 :         segments.iter().map(|&s| s.seg_no).collect::<Vec<_>>(),
     378              :         end_lsn,
     379              :         start_lsn,
     380              :     );
     381            0 :     Ok(())
     382            0 : }
     383              : 
     384            0 : async fn backup_single_segment(
     385            0 :     seg: &Segment,
     386            0 :     timeline_dir: &Utf8Path,
     387            0 :     remote_timeline_path: &RemotePath,
     388            0 : ) -> Result<Segment> {
     389            0 :     let segment_file_path = seg.file_path(timeline_dir)?;
     390            0 :     let remote_segment_path = seg.remote_path(remote_timeline_path);
     391              : 
     392            0 :     let res = backup_object(&segment_file_path, &remote_segment_path, seg.size()).await;
     393            0 :     if res.is_ok() {
     394            0 :         BACKED_UP_SEGMENTS.inc();
     395            0 :     } else {
     396            0 :         BACKUP_ERRORS.inc();
     397            0 :     }
     398            0 :     res?;
     399            0 :     debug!("Backup of {} done", segment_file_path);
     400              : 
     401            0 :     Ok(*seg)
     402            0 : }
     403              : 
     404              : #[derive(Debug, Copy, Clone)]
     405              : pub struct Segment {
     406              :     seg_no: XLogSegNo,
     407              :     start_lsn: Lsn,
     408              :     end_lsn: Lsn,
     409              : }
     410              : 
     411              : impl Segment {
     412            0 :     pub fn new(seg_no: u64, start_lsn: Lsn, end_lsn: Lsn) -> Self {
     413            0 :         Self {
     414            0 :             seg_no,
     415            0 :             start_lsn,
     416            0 :             end_lsn,
     417            0 :         }
     418            0 :     }
     419              : 
     420            0 :     pub fn object_name(self) -> String {
     421            0 :         XLogFileName(PG_TLI, self.seg_no, self.size())
     422            0 :     }
     423              : 
     424            0 :     pub fn file_path(self, timeline_dir: &Utf8Path) -> Result<Utf8PathBuf> {
     425            0 :         Ok(timeline_dir.join(self.object_name()))
     426            0 :     }
     427              : 
     428            0 :     pub fn remote_path(self, remote_timeline_path: &RemotePath) -> RemotePath {
     429            0 :         remote_timeline_path.join(self.object_name())
     430            0 :     }
     431              : 
     432            0 :     pub fn size(self) -> usize {
     433            0 :         (u64::from(self.end_lsn) - u64::from(self.start_lsn)) as usize
     434            0 :     }
     435              : }
     436              : 
     437            0 : fn get_segments(start: Lsn, end: Lsn, seg_size: usize) -> Vec<Segment> {
     438            0 :     let first_seg = start.segment_number(seg_size);
     439            0 :     let last_seg = end.segment_number(seg_size);
     440            0 : 
     441            0 :     let res: Vec<Segment> = (first_seg..last_seg)
     442            0 :         .map(|s| {
     443            0 :             let start_lsn = XLogSegNoOffsetToRecPtr(s, 0, seg_size);
     444            0 :             let end_lsn = XLogSegNoOffsetToRecPtr(s + 1, 0, seg_size);
     445            0 :             Segment::new(s, Lsn::from(start_lsn), Lsn::from(end_lsn))
     446            0 :         })
     447            0 :         .collect();
     448            0 :     res
     449            0 : }
     450              : 
     451            0 : async fn backup_object(
     452            0 :     source_file: &Utf8Path,
     453            0 :     target_file: &RemotePath,
     454            0 :     size: usize,
     455            0 : ) -> Result<()> {
     456            0 :     let storage = get_configured_remote_storage();
     457              : 
     458            0 :     let file = File::open(&source_file)
     459            0 :         .await
     460            0 :         .with_context(|| format!("Failed to open file {source_file:?} for wal backup"))?;
     461              : 
     462            0 :     let file = tokio_util::io::ReaderStream::with_capacity(file, BUFFER_SIZE);
     463            0 : 
     464            0 :     let cancel = CancellationToken::new();
     465            0 : 
     466            0 :     storage
     467            0 :         .upload_storage_object(file, size, target_file, &cancel)
     468            0 :         .await
     469            0 : }
     470              : 
     471            0 : pub(crate) async fn backup_partial_segment(
     472            0 :     source_file: &Utf8Path,
     473            0 :     target_file: &RemotePath,
     474            0 :     size: usize,
     475            0 : ) -> Result<()> {
     476            0 :     let storage = get_configured_remote_storage();
     477              : 
     478            0 :     let file = File::open(&source_file)
     479            0 :         .await
     480            0 :         .with_context(|| format!("Failed to open file {source_file:?} for wal backup"))?;
     481              : 
     482              :     // limiting the file to read only the first `size` bytes
     483            0 :     let limited_file = tokio::io::AsyncReadExt::take(file, size as u64);
     484            0 : 
     485            0 :     let file = tokio_util::io::ReaderStream::with_capacity(limited_file, BUFFER_SIZE);
     486            0 : 
     487            0 :     let cancel = CancellationToken::new();
     488            0 : 
     489            0 :     storage
     490            0 :         .upload(
     491            0 :             file,
     492            0 :             size,
     493            0 :             target_file,
     494            0 :             Some(StorageMetadata::from([("sk_type", "partial_segment")])),
     495            0 :             &cancel,
     496            0 :         )
     497            0 :         .await
     498            0 : }
     499              : 
     500            0 : pub(crate) async fn copy_partial_segment(
     501            0 :     source: &RemotePath,
     502            0 :     destination: &RemotePath,
     503            0 : ) -> Result<()> {
     504            0 :     let storage = get_configured_remote_storage();
     505            0 :     let cancel = CancellationToken::new();
     506            0 : 
     507            0 :     storage.copy_object(source, destination, &cancel).await
     508            0 : }
     509              : 
     510            0 : pub async fn read_object(
     511            0 :     file_path: &RemotePath,
     512            0 :     offset: u64,
     513            0 : ) -> anyhow::Result<Pin<Box<dyn tokio::io::AsyncRead + Send + Sync>>> {
     514            0 :     let storage = REMOTE_STORAGE
     515            0 :         .get()
     516            0 :         .context("Failed to get remote storage")?
     517            0 :         .as_ref()
     518            0 :         .context("No remote storage configured")?;
     519              : 
     520            0 :     info!("segment download about to start from remote path {file_path:?} at offset {offset}");
     521              : 
     522            0 :     let cancel = CancellationToken::new();
     523            0 : 
     524            0 :     let opts = DownloadOpts {
     525            0 :         byte_start: std::ops::Bound::Included(offset),
     526            0 :         ..Default::default()
     527            0 :     };
     528            0 :     let download = storage
     529            0 :         .download(file_path, &opts, &cancel)
     530            0 :         .await
     531            0 :         .with_context(|| {
     532            0 :             format!("Failed to open WAL segment download stream for remote path {file_path:?}")
     533            0 :         })?;
     534              : 
     535            0 :     let reader = tokio_util::io::StreamReader::new(download.download_stream);
     536            0 : 
     537            0 :     let reader = tokio::io::BufReader::with_capacity(BUFFER_SIZE, reader);
     538            0 : 
     539            0 :     Ok(Box::pin(reader))
     540            0 : }
     541              : 
     542              : /// Delete WAL files for the given timeline. Remote storage must be configured
     543              : /// when called.
     544            0 : pub async fn delete_timeline(ttid: &TenantTimelineId) -> Result<()> {
     545            0 :     let storage = get_configured_remote_storage();
     546            0 :     let remote_path = remote_timeline_path(ttid)?;
     547              : 
     548              :     // see DEFAULT_MAX_KEYS_PER_LIST_RESPONSE
     549              :     // const Option unwrap is not stable, otherwise it would be const.
     550            0 :     let batch_size: NonZeroU32 = NonZeroU32::new(1000).unwrap();
     551            0 : 
     552            0 :     // A backoff::retry is used here for two reasons:
     553            0 :     // - To provide a backoff rather than busy-polling the API on errors
     554            0 :     // - To absorb transient 429/503 conditions without hitting our error
     555            0 :     //   logging path for issues deleting objects.
     556            0 :     //
     557            0 :     // Note: listing segments might take a long time if there are many of them.
     558            0 :     // We don't currently have http requests timeout cancellation, but if/once
     559            0 :     // we have listing should get streaming interface to make progress.
     560            0 : 
     561            0 :     let cancel = CancellationToken::new(); // not really used
     562            0 :     backoff::retry(
     563            0 :         || async {
     564              :             // Do list-delete in batch_size batches to make progress even if there a lot of files.
     565              :             // Alternatively we could make remote storage list return iterator, but it is more complicated and
     566              :             // I'm not sure deleting while iterating is expected in s3.
     567              :             loop {
     568            0 :                 let files = storage
     569            0 :                     .list(
     570            0 :                         Some(&remote_path),
     571            0 :                         ListingMode::NoDelimiter,
     572            0 :                         Some(batch_size),
     573            0 :                         &cancel,
     574            0 :                     )
     575            0 :                     .await?
     576              :                     .keys
     577            0 :                     .into_iter()
     578            0 :                     .map(|o| o.key)
     579            0 :                     .collect::<Vec<_>>();
     580            0 :                 if files.is_empty() {
     581            0 :                     return Ok(()); // done
     582            0 :                 }
     583            0 :                 // (at least) s3 results are sorted, so can log min/max:
     584            0 :                 // "List results are always returned in UTF-8 binary order."
     585            0 :                 info!(
     586            0 :                     "deleting batch of {} WAL segments [{}-{}]",
     587            0 :                     files.len(),
     588            0 :                     files.first().unwrap().object_name().unwrap_or(""),
     589            0 :                     files.last().unwrap().object_name().unwrap_or("")
     590              :                 );
     591            0 :                 storage.delete_objects(&files, &cancel).await?;
     592              :             }
     593            0 :         },
     594            0 :         // consider TimeoutOrCancel::caused_by_cancel when using cancellation
     595            0 :         |_| false,
     596            0 :         3,
     597            0 :         10,
     598            0 :         "executing WAL segments deletion batch",
     599            0 :         &cancel,
     600            0 :     )
     601            0 :     .await
     602            0 :     .ok_or_else(|| anyhow::anyhow!("canceled"))
     603            0 :     .and_then(|x| x)?;
     604              : 
     605            0 :     Ok(())
     606            0 : }
     607              : 
     608              : /// Used by wal_backup_partial.
     609            0 : pub async fn delete_objects(paths: &[RemotePath]) -> Result<()> {
     610            0 :     let cancel = CancellationToken::new(); // not really used
     611            0 :     let storage = get_configured_remote_storage();
     612            0 :     storage.delete_objects(paths, &cancel).await
     613            0 : }
     614              : 
     615              : /// Copy segments from one timeline to another. Used in copy_timeline.
     616            0 : pub async fn copy_s3_segments(
     617            0 :     wal_seg_size: usize,
     618            0 :     src_ttid: &TenantTimelineId,
     619            0 :     dst_ttid: &TenantTimelineId,
     620            0 :     from_segment: XLogSegNo,
     621            0 :     to_segment: XLogSegNo,
     622            0 : ) -> Result<()> {
     623              :     const SEGMENTS_PROGRESS_REPORT_INTERVAL: u64 = 1024;
     624              : 
     625            0 :     let storage = REMOTE_STORAGE
     626            0 :         .get()
     627            0 :         .expect("failed to get remote storage")
     628            0 :         .as_ref()
     629            0 :         .unwrap();
     630              : 
     631            0 :     let remote_dst_path = remote_timeline_path(dst_ttid)?;
     632              : 
     633            0 :     let cancel = CancellationToken::new();
     634              : 
     635            0 :     let files = storage
     636            0 :         .list(
     637            0 :             Some(&remote_dst_path),
     638            0 :             ListingMode::NoDelimiter,
     639            0 :             None,
     640            0 :             &cancel,
     641            0 :         )
     642            0 :         .await?
     643              :         .keys;
     644              : 
     645            0 :     let uploaded_segments = &files
     646            0 :         .iter()
     647            0 :         .filter_map(|o| o.key.object_name().map(ToOwned::to_owned))
     648            0 :         .collect::<HashSet<_>>();
     649            0 : 
     650            0 :     debug!(
     651            0 :         "these segments have already been uploaded: {:?}",
     652              :         uploaded_segments
     653              :     );
     654              : 
     655            0 :     for segno in from_segment..to_segment {
     656            0 :         if segno % SEGMENTS_PROGRESS_REPORT_INTERVAL == 0 {
     657            0 :             info!("copied all segments from {} until {}", from_segment, segno);
     658            0 :         }
     659              : 
     660            0 :         let segment_name = XLogFileName(PG_TLI, segno, wal_seg_size);
     661            0 :         if uploaded_segments.contains(&segment_name) {
     662            0 :             continue;
     663            0 :         }
     664            0 :         debug!("copying segment {}", segment_name);
     665              : 
     666            0 :         let from = remote_timeline_path(src_ttid)?.join(&segment_name);
     667            0 :         let to = remote_dst_path.join(&segment_name);
     668            0 : 
     669            0 :         storage.copy_object(&from, &to, &cancel).await?;
     670              :     }
     671              : 
     672            0 :     info!(
     673            0 :         "finished copying segments from {} until {}",
     674              :         from_segment, to_segment
     675              :     );
     676            0 :     Ok(())
     677            0 : }
     678              : 
     679              : /// Get S3 (remote_storage) prefix path used for timeline files.
     680            0 : pub fn remote_timeline_path(ttid: &TenantTimelineId) -> Result<RemotePath> {
     681            0 :     RemotePath::new(&Utf8Path::new(&ttid.tenant_id.to_string()).join(ttid.timeline_id.to_string()))
     682            0 : }
        

Generated by: LCOV version 2.1-beta