Line data Source code
1 : //! This module acts as a switchboard to access different repositories managed by this
2 : //! page server.
3 :
4 : use camino::{Utf8DirEntry, Utf8Path, Utf8PathBuf};
5 : use futures::StreamExt;
6 : use itertools::Itertools;
7 : use pageserver_api::key::Key;
8 : use pageserver_api::models::LocationConfigMode;
9 : use pageserver_api::shard::{
10 : ShardCount, ShardIdentity, ShardIndex, ShardNumber, ShardStripeSize, TenantShardId,
11 : };
12 : use pageserver_api::upcall_api::ReAttachResponseTenant;
13 : use rand::{distributions::Alphanumeric, Rng};
14 : use remote_storage::TimeoutOrCancel;
15 : use std::borrow::Cow;
16 : use std::cmp::Ordering;
17 : use std::collections::{BTreeMap, HashMap, HashSet};
18 : use std::ops::Deref;
19 : use std::sync::Arc;
20 : use std::time::Duration;
21 : use sysinfo::SystemExt;
22 : use tokio::fs;
23 :
24 : use anyhow::Context;
25 : use once_cell::sync::Lazy;
26 : use tokio::task::JoinSet;
27 : use tokio_util::sync::CancellationToken;
28 : use tracing::*;
29 :
30 : use utils::{backoff, completion, crashsafe};
31 :
32 : use crate::config::PageServerConf;
33 : use crate::context::{DownloadBehavior, RequestContext};
34 : use crate::controller_upcall_client::{
35 : ControlPlaneGenerationsApi, ControllerUpcallClient, RetryForeverError,
36 : };
37 : use crate::deletion_queue::DeletionQueueClient;
38 : use crate::http::routes::ACTIVE_TENANT_TIMEOUT;
39 : use crate::metrics::{TENANT, TENANT_MANAGER as METRICS};
40 : use crate::task_mgr::{TaskKind, BACKGROUND_RUNTIME};
41 : use crate::tenant::config::{
42 : AttachedLocationConfig, AttachmentMode, LocationConf, LocationMode, SecondaryLocationConfig,
43 : };
44 : use crate::tenant::span::debug_assert_current_span_has_tenant_id;
45 : use crate::tenant::storage_layer::inmemory_layer;
46 : use crate::tenant::timeline::ShutdownMode;
47 : use crate::tenant::{AttachedTenantConf, GcError, LoadConfigError, SpawnMode, Tenant, TenantState};
48 : use crate::virtual_file::MaybeFatalIo;
49 : use crate::{InitializationOrder, TEMP_FILE_SUFFIX};
50 :
51 : use utils::crashsafe::path_with_suffix_extension;
52 : use utils::fs_ext::PathExt;
53 : use utils::generation::Generation;
54 : use utils::id::{TenantId, TimelineId};
55 :
56 : use super::remote_timeline_client::remote_tenant_path;
57 : use super::secondary::SecondaryTenant;
58 : use super::timeline::detach_ancestor::{self, PreparedTimelineDetach};
59 : use super::{GlobalShutDown, TenantSharedResources};
60 :
61 : /// For a tenant that appears in TenantsMap, it may either be
62 : /// - `Attached`: has a full Tenant object, is elegible to service
63 : /// reads and ingest WAL.
64 : /// - `Secondary`: is only keeping a local cache warm.
65 : ///
66 : /// Secondary is a totally distinct state rather than being a mode of a `Tenant`, because
67 : /// that way we avoid having to carefully switch a tenant's ingestion etc on and off during
68 : /// its lifetime, and we can preserve some important safety invariants like `Tenant` always
69 : /// having a properly acquired generation (Secondary doesn't need a generation)
70 : #[derive(Clone)]
71 : pub(crate) enum TenantSlot {
72 : Attached(Arc<Tenant>),
73 : Secondary(Arc<SecondaryTenant>),
74 : /// In this state, other administrative operations acting on the TenantId should
75 : /// block, or return a retry indicator equivalent to HTTP 503.
76 : InProgress(utils::completion::Barrier),
77 : }
78 :
79 : impl std::fmt::Debug for TenantSlot {
80 0 : fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
81 0 : match self {
82 0 : Self::Attached(tenant) => write!(f, "Attached({})", tenant.current_state()),
83 0 : Self::Secondary(_) => write!(f, "Secondary"),
84 0 : Self::InProgress(_) => write!(f, "InProgress"),
85 : }
86 0 : }
87 : }
88 :
89 : impl TenantSlot {
90 : /// Return the `Tenant` in this slot if attached, else None
91 0 : fn get_attached(&self) -> Option<&Arc<Tenant>> {
92 0 : match self {
93 0 : Self::Attached(t) => Some(t),
94 0 : Self::Secondary(_) => None,
95 0 : Self::InProgress(_) => None,
96 : }
97 0 : }
98 : }
99 :
100 : /// The tenants known to the pageserver.
101 : /// The enum variants are used to distinguish the different states that the pageserver can be in.
102 : pub(crate) enum TenantsMap {
103 : /// [`init_tenant_mgr`] is not done yet.
104 : Initializing,
105 : /// [`init_tenant_mgr`] is done, all on-disk tenants have been loaded.
106 : /// New tenants can be added using [`tenant_map_acquire_slot`].
107 : Open(BTreeMap<TenantShardId, TenantSlot>),
108 : /// The pageserver has entered shutdown mode via [`TenantManager::shutdown`].
109 : /// Existing tenants are still accessible, but no new tenants can be created.
110 : ShuttingDown(BTreeMap<TenantShardId, TenantSlot>),
111 : }
112 :
113 : /// When resolving a TenantId to a shard, we may be looking for the 0th
114 : /// shard, or we might be looking for whichever shard holds a particular page.
115 : #[derive(Copy, Clone)]
116 : pub(crate) enum ShardSelector {
117 : /// Only return the 0th shard, if it is present. If a non-0th shard is present,
118 : /// ignore it.
119 : Zero,
120 : /// Pick the shard that holds this key
121 : Page(Key),
122 : /// The shard ID is known: pick the given shard
123 : Known(ShardIndex),
124 : }
125 :
126 : /// A convenience for use with the re_attach ControllerUpcallClient function: rather
127 : /// than the serializable struct, we build this enum that encapsulates
128 : /// the invariant that attached tenants always have generations.
129 : ///
130 : /// This represents the subset of a LocationConfig that we receive during re-attach.
131 : pub(crate) enum TenantStartupMode {
132 : Attached((AttachmentMode, Generation)),
133 : Secondary,
134 : }
135 :
136 : impl TenantStartupMode {
137 : /// Return the generation & mode that should be used when starting
138 : /// this tenant.
139 : ///
140 : /// If this returns None, the re-attach struct is in an invalid state and
141 : /// should be ignored in the response.
142 0 : fn from_reattach_tenant(rart: ReAttachResponseTenant) -> Option<Self> {
143 0 : match (rart.mode, rart.gen) {
144 0 : (LocationConfigMode::Detached, _) => None,
145 0 : (LocationConfigMode::Secondary, _) => Some(Self::Secondary),
146 0 : (LocationConfigMode::AttachedMulti, Some(g)) => {
147 0 : Some(Self::Attached((AttachmentMode::Multi, Generation::new(g))))
148 : }
149 0 : (LocationConfigMode::AttachedSingle, Some(g)) => {
150 0 : Some(Self::Attached((AttachmentMode::Single, Generation::new(g))))
151 : }
152 0 : (LocationConfigMode::AttachedStale, Some(g)) => {
153 0 : Some(Self::Attached((AttachmentMode::Stale, Generation::new(g))))
154 : }
155 : _ => {
156 0 : tracing::warn!(
157 0 : "Received invalid re-attach state for tenant {}: {rart:?}",
158 : rart.id
159 : );
160 0 : None
161 : }
162 : }
163 0 : }
164 : }
165 :
166 : /// Result type for looking up a TenantId to a specific shard
167 : pub(crate) enum ShardResolveResult {
168 : NotFound,
169 : Found(Arc<Tenant>),
170 : // Wait for this barrrier, then query again
171 : InProgress(utils::completion::Barrier),
172 : }
173 :
174 : impl TenantsMap {
175 : /// Convenience function for typical usage, where we want to get a `Tenant` object, for
176 : /// working with attached tenants. If the TenantId is in the map but in Secondary state,
177 : /// None is returned.
178 0 : pub(crate) fn get(&self, tenant_shard_id: &TenantShardId) -> Option<&Arc<Tenant>> {
179 0 : match self {
180 0 : TenantsMap::Initializing => None,
181 0 : TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => {
182 0 : m.get(tenant_shard_id).and_then(|slot| slot.get_attached())
183 : }
184 : }
185 0 : }
186 :
187 : #[cfg(all(debug_assertions, not(test)))]
188 0 : pub(crate) fn len(&self) -> usize {
189 0 : match self {
190 0 : TenantsMap::Initializing => 0,
191 0 : TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m.len(),
192 : }
193 0 : }
194 : }
195 :
196 : /// Precursor to deletion of a tenant dir: we do a fast rename to a tmp path, and then
197 : /// the slower actual deletion in the background.
198 : ///
199 : /// This is "safe" in that that it won't leave behind a partially deleted directory
200 : /// at the original path, because we rename with TEMP_FILE_SUFFIX before starting deleting
201 : /// the contents.
202 : ///
203 : /// This is pageserver-specific, as it relies on future processes after a crash to check
204 : /// for TEMP_FILE_SUFFIX when loading things.
205 0 : async fn safe_rename_tenant_dir(path: impl AsRef<Utf8Path>) -> std::io::Result<Utf8PathBuf> {
206 0 : let parent = path
207 0 : .as_ref()
208 0 : .parent()
209 0 : // It is invalid to call this function with a relative path. Tenant directories
210 0 : // should always have a parent.
211 0 : .ok_or(std::io::Error::new(
212 0 : std::io::ErrorKind::InvalidInput,
213 0 : "Path must be absolute",
214 0 : ))?;
215 0 : let rand_suffix = rand::thread_rng()
216 0 : .sample_iter(&Alphanumeric)
217 0 : .take(8)
218 0 : .map(char::from)
219 0 : .collect::<String>()
220 0 : + TEMP_FILE_SUFFIX;
221 0 : let tmp_path = path_with_suffix_extension(&path, &rand_suffix);
222 0 : fs::rename(path.as_ref(), &tmp_path).await?;
223 0 : fs::File::open(parent)
224 0 : .await?
225 0 : .sync_all()
226 0 : .await
227 0 : .maybe_fatal_err("safe_rename_tenant_dir")?;
228 0 : Ok(tmp_path)
229 0 : }
230 :
231 : /// See [`Self::spawn`].
232 : #[derive(Clone, Default)]
233 : pub struct BackgroundPurges(tokio_util::task::TaskTracker);
234 :
235 : impl BackgroundPurges {
236 : /// When we have moved a tenant's content to a temporary directory, we may delete it lazily in
237 : /// the background, and thereby avoid blocking any API requests on this deletion completing.
238 : ///
239 : /// Although we are cleaning up the tenant, this task is not meant to be bound by the lifetime of the tenant in memory.
240 : /// Thus the [`BackgroundPurges`] type to keep track of these tasks.
241 0 : pub fn spawn(&self, tmp_path: Utf8PathBuf) {
242 0 : // because on shutdown we close and wait, we are misusing TaskTracker a bit.
243 0 : //
244 0 : // so first acquire a token, then check if the tracker has been closed. the tracker might get closed
245 0 : // right after, but at least the shutdown will wait for what we are spawning next.
246 0 : let token = self.0.token();
247 0 :
248 0 : if self.0.is_closed() {
249 0 : warn!(
250 : %tmp_path,
251 0 : "trying to spawn background purge during shutdown, ignoring"
252 : );
253 0 : return;
254 0 : }
255 :
256 0 : let span = info_span!(parent: None, "background_purge", %tmp_path);
257 :
258 0 : let task = move || {
259 0 : let _token = token;
260 0 : let _entered = span.entered();
261 0 : if let Err(error) = std::fs::remove_dir_all(tmp_path.as_path()) {
262 : // should we fatal_io_error here?
263 0 : warn!(%error, "failed to purge tenant directory");
264 0 : }
265 0 : };
266 :
267 0 : BACKGROUND_RUNTIME.spawn_blocking(task);
268 0 : }
269 :
270 : /// When this future completes, all background purges have completed.
271 : /// The first poll of the future will already lock out new background purges spawned via [`Self::spawn`].
272 : ///
273 : /// Concurrent calls will coalesce.
274 : ///
275 : /// # Cancellation-Safety
276 : ///
277 : /// If this future is dropped before polled to completion, concurrent and subsequent
278 : /// instances of this future will continue to be correct.
279 0 : #[instrument(skip_all)]
280 : pub async fn shutdown(&self) {
281 : // forbid new tasks (can be called many times)
282 : self.0.close();
283 : self.0.wait().await;
284 : }
285 : }
286 :
287 : static TENANTS: Lazy<std::sync::RwLock<TenantsMap>> =
288 2 : Lazy::new(|| std::sync::RwLock::new(TenantsMap::Initializing));
289 :
290 : /// Responsible for storing and mutating the collection of all tenants
291 : /// that this pageserver has state for.
292 : ///
293 : /// Every Tenant and SecondaryTenant instance lives inside the TenantManager.
294 : ///
295 : /// The most important role of the TenantManager is to prevent conflicts: e.g. trying to attach
296 : /// the same tenant twice concurrently, or trying to configure the same tenant into secondary
297 : /// and attached modes concurrently.
298 : pub struct TenantManager {
299 : conf: &'static PageServerConf,
300 : // TODO: currently this is a &'static pointing to TENANTs. When we finish refactoring
301 : // out of that static variable, the TenantManager can own this.
302 : // See https://github.com/neondatabase/neon/issues/5796
303 : tenants: &'static std::sync::RwLock<TenantsMap>,
304 : resources: TenantSharedResources,
305 :
306 : // Long-running operations that happen outside of a [`Tenant`] lifetime should respect this token.
307 : // This is for edge cases like tenant deletion. In normal cases (within a Tenant lifetime),
308 : // tenants have their own cancellation tokens, which we fire individually in [`Self::shutdown`], or
309 : // when the tenant detaches.
310 : cancel: CancellationToken,
311 :
312 : background_purges: BackgroundPurges,
313 : }
314 :
315 0 : fn emergency_generations(
316 0 : tenant_confs: &HashMap<TenantShardId, Result<LocationConf, LoadConfigError>>,
317 0 : ) -> HashMap<TenantShardId, TenantStartupMode> {
318 0 : tenant_confs
319 0 : .iter()
320 0 : .filter_map(|(tid, lc)| {
321 0 : let lc = match lc {
322 0 : Ok(lc) => lc,
323 0 : Err(_) => return None,
324 : };
325 : Some((
326 0 : *tid,
327 0 : match &lc.mode {
328 0 : LocationMode::Attached(alc) => {
329 0 : TenantStartupMode::Attached((alc.attach_mode, alc.generation))
330 : }
331 0 : LocationMode::Secondary(_) => TenantStartupMode::Secondary,
332 : },
333 : ))
334 0 : })
335 0 : .collect()
336 0 : }
337 :
338 0 : async fn init_load_generations(
339 0 : conf: &'static PageServerConf,
340 0 : tenant_confs: &HashMap<TenantShardId, Result<LocationConf, LoadConfigError>>,
341 0 : resources: &TenantSharedResources,
342 0 : cancel: &CancellationToken,
343 0 : ) -> anyhow::Result<Option<HashMap<TenantShardId, TenantStartupMode>>> {
344 0 : let generations = if conf.control_plane_emergency_mode {
345 0 : error!(
346 0 : "Emergency mode! Tenants will be attached unsafely using their last known generation"
347 : );
348 0 : emergency_generations(tenant_confs)
349 0 : } else if let Some(client) = ControllerUpcallClient::new(conf, cancel) {
350 0 : info!("Calling control plane API to re-attach tenants");
351 : // If we are configured to use the control plane API, then it is the source of truth for what tenants to load.
352 0 : match client.re_attach(conf).await {
353 0 : Ok(tenants) => tenants
354 0 : .into_iter()
355 0 : .flat_map(|(id, rart)| {
356 0 : TenantStartupMode::from_reattach_tenant(rart).map(|tsm| (id, tsm))
357 0 : })
358 0 : .collect(),
359 : Err(RetryForeverError::ShuttingDown) => {
360 0 : anyhow::bail!("Shut down while waiting for control plane re-attach response")
361 : }
362 : }
363 : } else {
364 0 : info!("Control plane API not configured, tenant generations are disabled");
365 0 : return Ok(None);
366 : };
367 :
368 : // The deletion queue needs to know about the startup attachment state to decide which (if any) stored
369 : // deletion list entries may still be valid. We provide that by pushing a recovery operation into
370 : // the queue. Sequential processing of te queue ensures that recovery is done before any new tenant deletions
371 : // are processed, even though we don't block on recovery completing here.
372 0 : let attached_tenants = generations
373 0 : .iter()
374 0 : .flat_map(|(id, start_mode)| {
375 0 : match start_mode {
376 0 : TenantStartupMode::Attached((_mode, generation)) => Some(generation),
377 0 : TenantStartupMode::Secondary => None,
378 : }
379 0 : .map(|gen| (*id, *gen))
380 0 : })
381 0 : .collect();
382 0 : resources.deletion_queue_client.recover(attached_tenants)?;
383 :
384 0 : Ok(Some(generations))
385 0 : }
386 :
387 : /// Given a directory discovered in the pageserver's tenants/ directory, attempt
388 : /// to load a tenant config from it.
389 : ///
390 : /// If we cleaned up something expected (like an empty dir or a temp dir), return None.
391 0 : fn load_tenant_config(
392 0 : conf: &'static PageServerConf,
393 0 : tenant_shard_id: TenantShardId,
394 0 : dentry: Utf8DirEntry,
395 0 : ) -> Option<Result<LocationConf, LoadConfigError>> {
396 0 : let tenant_dir_path = dentry.path().to_path_buf();
397 0 : if crate::is_temporary(&tenant_dir_path) {
398 0 : info!("Found temporary tenant directory, removing: {tenant_dir_path}");
399 : // No need to use safe_remove_tenant_dir_all because this is already
400 : // a temporary path
401 0 : std::fs::remove_dir_all(&tenant_dir_path).fatal_err("delete temporary tenant dir");
402 0 : return None;
403 0 : }
404 0 :
405 0 : // This case happens if we crash during attachment before writing a config into the dir
406 0 : let is_empty = tenant_dir_path
407 0 : .is_empty_dir()
408 0 : .fatal_err("Checking for empty tenant dir");
409 0 : if is_empty {
410 0 : info!("removing empty tenant directory {tenant_dir_path:?}");
411 0 : std::fs::remove_dir(&tenant_dir_path).fatal_err("delete empty tenant dir");
412 0 : return None;
413 0 : }
414 0 :
415 0 : Some(Tenant::load_tenant_config(conf, &tenant_shard_id))
416 0 : }
417 :
418 : /// Initial stage of load: walk the local tenants directory, clean up any temp files,
419 : /// and load configurations for the tenants we found.
420 : ///
421 : /// Do this in parallel, because we expect 10k+ tenants, so serial execution can take
422 : /// seconds even on reasonably fast drives.
423 0 : async fn init_load_tenant_configs(
424 0 : conf: &'static PageServerConf,
425 0 : ) -> HashMap<TenantShardId, Result<LocationConf, LoadConfigError>> {
426 0 : let tenants_dir = conf.tenants_path();
427 :
428 0 : let dentries = tokio::task::spawn_blocking(move || -> Vec<Utf8DirEntry> {
429 0 : let context = format!("read tenants dir {tenants_dir}");
430 0 : let dir_entries = tenants_dir.read_dir_utf8().fatal_err(&context);
431 0 :
432 0 : dir_entries
433 0 : .collect::<Result<Vec<_>, std::io::Error>>()
434 0 : .fatal_err(&context)
435 0 : })
436 0 : .await
437 0 : .expect("Config load task panicked");
438 0 :
439 0 : let mut configs = HashMap::new();
440 0 :
441 0 : let mut join_set = JoinSet::new();
442 0 : for dentry in dentries {
443 0 : let tenant_shard_id = match dentry.file_name().parse::<TenantShardId>() {
444 0 : Ok(id) => id,
445 : Err(_) => {
446 0 : warn!(
447 0 : "Invalid tenant path (garbage in our repo directory?): '{}'",
448 0 : dentry.file_name()
449 : );
450 0 : continue;
451 : }
452 : };
453 :
454 0 : join_set.spawn_blocking(move || {
455 0 : (
456 0 : tenant_shard_id,
457 0 : load_tenant_config(conf, tenant_shard_id, dentry),
458 0 : )
459 0 : });
460 0 : }
461 :
462 0 : while let Some(r) = join_set.join_next().await {
463 0 : let (tenant_shard_id, tenant_config) = r.expect("Panic in config load task");
464 0 : if let Some(tenant_config) = tenant_config {
465 0 : configs.insert(tenant_shard_id, tenant_config);
466 0 : }
467 : }
468 :
469 0 : configs
470 0 : }
471 :
472 0 : #[derive(Debug, thiserror::Error)]
473 : pub(crate) enum DeleteTenantError {
474 : #[error("Tenant map slot error {0}")]
475 : SlotError(#[from] TenantSlotError),
476 :
477 : #[error("Cancelled")]
478 : Cancelled,
479 :
480 : #[error(transparent)]
481 : Other(#[from] anyhow::Error),
482 : }
483 :
484 : /// Initialize repositories with locally available timelines.
485 : /// Timelines that are only partially available locally (remote storage has more data than this pageserver)
486 : /// are scheduled for download and added to the tenant once download is completed.
487 0 : #[instrument(skip_all)]
488 : pub async fn init_tenant_mgr(
489 : conf: &'static PageServerConf,
490 : background_purges: BackgroundPurges,
491 : resources: TenantSharedResources,
492 : init_order: InitializationOrder,
493 : cancel: CancellationToken,
494 : ) -> anyhow::Result<TenantManager> {
495 : let mut tenants = BTreeMap::new();
496 :
497 : let ctx = RequestContext::todo_child(TaskKind::Startup, DownloadBehavior::Warn);
498 :
499 : // Initialize dynamic limits that depend on system resources
500 : let system_memory =
501 : sysinfo::System::new_with_specifics(sysinfo::RefreshKind::new().with_memory())
502 : .total_memory();
503 : let max_ephemeral_layer_bytes =
504 : conf.ephemeral_bytes_per_memory_kb as u64 * (system_memory / 1024);
505 : tracing::info!("Initialized ephemeral layer size limit to {max_ephemeral_layer_bytes}, for {system_memory} bytes of memory");
506 : inmemory_layer::GLOBAL_RESOURCES.max_dirty_bytes.store(
507 : max_ephemeral_layer_bytes,
508 : std::sync::atomic::Ordering::Relaxed,
509 : );
510 :
511 : // Scan local filesystem for attached tenants
512 : let tenant_configs = init_load_tenant_configs(conf).await;
513 :
514 : // Determine which tenants are to be secondary or attached, and in which generation
515 : let tenant_modes = init_load_generations(conf, &tenant_configs, &resources, &cancel).await?;
516 :
517 : tracing::info!(
518 : "Attaching {} tenants at startup, warming up {} at a time",
519 : tenant_configs.len(),
520 : conf.concurrent_tenant_warmup.initial_permits()
521 : );
522 : TENANT.startup_scheduled.inc_by(tenant_configs.len() as u64);
523 :
524 : // Accumulate futures for writing tenant configs, so that we can execute in parallel
525 : let mut config_write_futs = Vec::new();
526 :
527 : // Update the location configs according to the re-attach response and persist them to disk
528 : tracing::info!("Updating {} location configs", tenant_configs.len());
529 : for (tenant_shard_id, location_conf) in tenant_configs {
530 : let tenant_dir_path = conf.tenant_path(&tenant_shard_id);
531 :
532 : let mut location_conf = match location_conf {
533 : Ok(l) => l,
534 : Err(e) => {
535 : // This should only happen in the case of a serialization bug or critical local I/O error: we cannot load this tenant
536 : error!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Failed to load tenant config, failed to {e:#}");
537 : continue;
538 : }
539 : };
540 :
541 : // FIXME: if we were attached, and get demoted to secondary on re-attach, we
542 : // don't have a place to get a config.
543 : // (https://github.com/neondatabase/neon/issues/5377)
544 : const DEFAULT_SECONDARY_CONF: SecondaryLocationConfig =
545 : SecondaryLocationConfig { warm: true };
546 :
547 : if let Some(tenant_modes) = &tenant_modes {
548 : // We have a generation map: treat it as the authority for whether
549 : // this tenant is really attached.
550 : match tenant_modes.get(&tenant_shard_id) {
551 : None => {
552 : info!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Detaching tenant, control plane omitted it in re-attach response");
553 :
554 : match safe_rename_tenant_dir(&tenant_dir_path).await {
555 : Ok(tmp_path) => {
556 : background_purges.spawn(tmp_path);
557 : }
558 : Err(e) => {
559 : error!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
560 : "Failed to move detached tenant directory '{tenant_dir_path}': {e:?}");
561 : }
562 : };
563 :
564 : // We deleted local content: move on to next tenant, don't try and spawn this one.
565 : continue;
566 : }
567 : Some(TenantStartupMode::Secondary) => {
568 : if !matches!(location_conf.mode, LocationMode::Secondary(_)) {
569 : location_conf.mode = LocationMode::Secondary(DEFAULT_SECONDARY_CONF);
570 : }
571 : }
572 : Some(TenantStartupMode::Attached((attach_mode, generation))) => {
573 : let old_gen_higher = match &location_conf.mode {
574 : LocationMode::Attached(AttachedLocationConfig {
575 : generation: old_generation,
576 : attach_mode: _attach_mode,
577 : }) => {
578 : if old_generation > generation {
579 : Some(old_generation)
580 : } else {
581 : None
582 : }
583 : }
584 : _ => None,
585 : };
586 : if let Some(old_generation) = old_gen_higher {
587 : tracing::error!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
588 : "Control plane gave decreasing generation ({generation:?}) in re-attach response for tenant that was attached in generation {:?}, demoting to secondary",
589 : old_generation
590 : );
591 :
592 : // We cannot safely attach this tenant given a bogus generation number, but let's avoid throwing away
593 : // local disk content: demote to secondary rather than detaching.
594 : location_conf.mode = LocationMode::Secondary(DEFAULT_SECONDARY_CONF);
595 : } else {
596 : location_conf.attach_in_generation(*attach_mode, *generation);
597 : }
598 : }
599 : }
600 : } else {
601 : // Legacy mode: no generation information, any tenant present
602 : // on local disk may activate
603 : info!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Starting tenant in legacy mode, no generation",);
604 : };
605 :
606 : // Presence of a generation number implies attachment: attach the tenant
607 : // if it wasn't already, and apply the generation number.
608 0 : config_write_futs.push(async move {
609 0 : let r = Tenant::persist_tenant_config(conf, &tenant_shard_id, &location_conf).await;
610 0 : (tenant_shard_id, location_conf, r)
611 0 : });
612 : }
613 :
614 : // Execute config writes with concurrency, to avoid bottlenecking on local FS write latency
615 : tracing::info!(
616 : "Writing {} location config files...",
617 : config_write_futs.len()
618 : );
619 : let config_write_results = futures::stream::iter(config_write_futs)
620 : .buffer_unordered(16)
621 : .collect::<Vec<_>>()
622 : .await;
623 :
624 : tracing::info!(
625 : "Spawning {} tenant shard locations...",
626 : config_write_results.len()
627 : );
628 : // For those shards that have live configurations, construct `Tenant` or `SecondaryTenant` objects and start them running
629 : for (tenant_shard_id, location_conf, config_write_result) in config_write_results {
630 : // Writing a config to local disk is foundational to startup up tenants: panic if we can't.
631 : config_write_result.fatal_err("write tenant shard config file");
632 :
633 : let tenant_dir_path = conf.tenant_path(&tenant_shard_id);
634 : let shard_identity = location_conf.shard;
635 : let slot = match location_conf.mode {
636 : LocationMode::Attached(attached_conf) => TenantSlot::Attached(
637 : tenant_spawn(
638 : conf,
639 : tenant_shard_id,
640 : &tenant_dir_path,
641 : resources.clone(),
642 : AttachedTenantConf::new(location_conf.tenant_conf, attached_conf),
643 : shard_identity,
644 : Some(init_order.clone()),
645 : SpawnMode::Lazy,
646 : &ctx,
647 : )
648 : .expect("global shutdown during init_tenant_mgr cannot happen"),
649 : ),
650 : LocationMode::Secondary(secondary_conf) => {
651 : info!(
652 : tenant_id = %tenant_shard_id.tenant_id,
653 : shard_id = %tenant_shard_id.shard_slug(),
654 : "Starting secondary tenant"
655 : );
656 : TenantSlot::Secondary(SecondaryTenant::new(
657 : tenant_shard_id,
658 : shard_identity,
659 : location_conf.tenant_conf,
660 : &secondary_conf,
661 : ))
662 : }
663 : };
664 :
665 : METRICS.slot_inserted(&slot);
666 : tenants.insert(tenant_shard_id, slot);
667 : }
668 :
669 : info!("Processed {} local tenants at startup", tenants.len());
670 :
671 : let mut tenants_map = TENANTS.write().unwrap();
672 : assert!(matches!(&*tenants_map, &TenantsMap::Initializing));
673 :
674 : *tenants_map = TenantsMap::Open(tenants);
675 :
676 : Ok(TenantManager {
677 : conf,
678 : tenants: &TENANTS,
679 : resources,
680 : cancel: CancellationToken::new(),
681 : background_purges,
682 : })
683 : }
684 :
685 : /// Wrapper for Tenant::spawn that checks invariants before running
686 : #[allow(clippy::too_many_arguments)]
687 0 : fn tenant_spawn(
688 0 : conf: &'static PageServerConf,
689 0 : tenant_shard_id: TenantShardId,
690 0 : tenant_path: &Utf8Path,
691 0 : resources: TenantSharedResources,
692 0 : location_conf: AttachedTenantConf,
693 0 : shard_identity: ShardIdentity,
694 0 : init_order: Option<InitializationOrder>,
695 0 : mode: SpawnMode,
696 0 : ctx: &RequestContext,
697 0 : ) -> Result<Arc<Tenant>, GlobalShutDown> {
698 0 : // All these conditions should have been satisfied by our caller: the tenant dir exists, is a well formed
699 0 : // path, and contains a configuration file. Assertions that do synchronous I/O are limited to debug mode
700 0 : // to avoid impacting prod runtime performance.
701 0 : assert!(!crate::is_temporary(tenant_path));
702 0 : debug_assert!(tenant_path.is_dir());
703 0 : debug_assert!(conf
704 0 : .tenant_location_config_path(&tenant_shard_id)
705 0 : .try_exists()
706 0 : .unwrap());
707 :
708 0 : Tenant::spawn(
709 0 : conf,
710 0 : tenant_shard_id,
711 0 : resources,
712 0 : location_conf,
713 0 : shard_identity,
714 0 : init_order,
715 0 : mode,
716 0 : ctx,
717 0 : )
718 0 : }
719 :
720 2 : async fn shutdown_all_tenants0(tenants: &std::sync::RwLock<TenantsMap>) {
721 2 : let mut join_set = JoinSet::new();
722 0 :
723 0 : #[cfg(all(debug_assertions, not(test)))]
724 0 : {
725 0 : // Check that our metrics properly tracked the size of the tenants map. This is a convenient location to check,
726 0 : // as it happens implicitly at the end of tests etc.
727 0 : let m = tenants.read().unwrap();
728 0 : debug_assert_eq!(METRICS.slots_total(), m.len() as u64);
729 : }
730 :
731 : // Atomically, 1. create the shutdown tasks and 2. prevent creation of new tenants.
732 2 : let (total_in_progress, total_attached) = {
733 2 : let mut m = tenants.write().unwrap();
734 2 : match &mut *m {
735 : TenantsMap::Initializing => {
736 0 : *m = TenantsMap::ShuttingDown(BTreeMap::default());
737 0 : info!("tenants map is empty");
738 0 : return;
739 : }
740 2 : TenantsMap::Open(tenants) => {
741 2 : let mut shutdown_state = BTreeMap::new();
742 2 : let mut total_in_progress = 0;
743 2 : let mut total_attached = 0;
744 :
745 2 : for (tenant_shard_id, v) in std::mem::take(tenants).into_iter() {
746 2 : match v {
747 0 : TenantSlot::Attached(t) => {
748 0 : shutdown_state.insert(tenant_shard_id, TenantSlot::Attached(t.clone()));
749 0 : join_set.spawn(
750 0 : async move {
751 0 : let res = {
752 0 : let (_guard, shutdown_progress) = completion::channel();
753 0 : t.shutdown(shutdown_progress, ShutdownMode::FreezeAndFlush).await
754 : };
755 :
756 0 : if let Err(other_progress) = res {
757 : // join the another shutdown in progress
758 0 : other_progress.wait().await;
759 0 : }
760 :
761 : // we cannot afford per tenant logging here, because if s3 is degraded, we are
762 : // going to log too many lines
763 0 : debug!("tenant successfully stopped");
764 0 : }
765 0 : .instrument(info_span!("shutdown", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug())),
766 : );
767 :
768 0 : total_attached += 1;
769 : }
770 0 : TenantSlot::Secondary(state) => {
771 0 : // We don't need to wait for this individually per-tenant: the
772 0 : // downloader task will be waited on eventually, this cancel
773 0 : // is just to encourage it to drop out if it is doing work
774 0 : // for this tenant right now.
775 0 : state.cancel.cancel();
776 0 :
777 0 : shutdown_state.insert(tenant_shard_id, TenantSlot::Secondary(state));
778 0 : }
779 2 : TenantSlot::InProgress(notify) => {
780 2 : // InProgress tenants are not visible in TenantsMap::ShuttingDown: we will
781 2 : // wait for their notifications to fire in this function.
782 2 : join_set.spawn(async move {
783 2 : notify.wait().await;
784 2 : });
785 2 :
786 2 : total_in_progress += 1;
787 2 : }
788 : }
789 : }
790 2 : *m = TenantsMap::ShuttingDown(shutdown_state);
791 2 : (total_in_progress, total_attached)
792 : }
793 : TenantsMap::ShuttingDown(_) => {
794 0 : error!("already shutting down, this function isn't supposed to be called more than once");
795 0 : return;
796 : }
797 : }
798 : };
799 :
800 2 : let started_at = std::time::Instant::now();
801 2 :
802 2 : info!(
803 0 : "Waiting for {} InProgress tenants and {} Attached tenants to shut down",
804 : total_in_progress, total_attached
805 : );
806 :
807 2 : let total = join_set.len();
808 2 : let mut panicked = 0;
809 2 : let mut buffering = true;
810 : const BUFFER_FOR: std::time::Duration = std::time::Duration::from_millis(500);
811 2 : let mut buffered = std::pin::pin!(tokio::time::sleep(BUFFER_FOR));
812 :
813 6 : while !join_set.is_empty() {
814 4 : tokio::select! {
815 4 : Some(joined) = join_set.join_next() => {
816 0 : match joined {
817 2 : Ok(()) => {},
818 0 : Err(join_error) if join_error.is_cancelled() => {
819 0 : unreachable!("we are not cancelling any of the tasks");
820 : }
821 0 : Err(join_error) if join_error.is_panic() => {
822 0 : // cannot really do anything, as this panic is likely a bug
823 0 : panicked += 1;
824 0 : }
825 0 : Err(join_error) => {
826 0 : warn!("unknown kind of JoinError: {join_error}");
827 : }
828 : }
829 2 : if !buffering {
830 2 : // buffer so that every 500ms since the first update (or starting) we'll log
831 2 : // how far away we are; this is because we will get SIGKILL'd at 10s, and we
832 2 : // are not able to log *then*.
833 2 : buffering = true;
834 2 : buffered.as_mut().reset(tokio::time::Instant::now() + BUFFER_FOR);
835 2 : }
836 : },
837 4 : _ = &mut buffered, if buffering => {
838 2 : buffering = false;
839 2 : info!(remaining = join_set.len(), total, elapsed_ms = started_at.elapsed().as_millis(), "waiting for tenants to shutdown");
840 : }
841 : }
842 : }
843 :
844 2 : if panicked > 0 {
845 0 : warn!(
846 : panicked,
847 0 : total, "observed panicks while shutting down tenants"
848 : );
849 2 : }
850 :
851 : // caller will log how long we took
852 2 : }
853 :
854 0 : #[derive(thiserror::Error, Debug)]
855 : pub(crate) enum UpsertLocationError {
856 : #[error("Bad config request: {0}")]
857 : BadRequest(anyhow::Error),
858 :
859 : #[error("Cannot change config in this state: {0}")]
860 : Unavailable(#[from] TenantMapError),
861 :
862 : #[error("Tenant is already being modified")]
863 : InProgress,
864 :
865 : #[error("Failed to flush: {0}")]
866 : Flush(anyhow::Error),
867 :
868 : /// This error variant is for unexpected situations (soft assertions) where the system is in an unexpected state.
869 : #[error("Internal error: {0}")]
870 : InternalError(anyhow::Error),
871 : }
872 :
873 : impl TenantManager {
874 : /// Convenience function so that anyone with a TenantManager can get at the global configuration, without
875 : /// having to pass it around everywhere as a separate object.
876 0 : pub(crate) fn get_conf(&self) -> &'static PageServerConf {
877 0 : self.conf
878 0 : }
879 :
880 : /// Gets the attached tenant from the in-memory data, erroring if it's absent, in secondary mode, or currently
881 : /// undergoing a state change (i.e. slot is InProgress).
882 : ///
883 : /// The return Tenant is not guaranteed to be active: check its status after obtaing it, or
884 : /// use [`Tenant::wait_to_become_active`] before using it if you will do I/O on it.
885 0 : pub(crate) fn get_attached_tenant_shard(
886 0 : &self,
887 0 : tenant_shard_id: TenantShardId,
888 0 : ) -> Result<Arc<Tenant>, GetTenantError> {
889 0 : let locked = self.tenants.read().unwrap();
890 :
891 0 : let peek_slot = tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Read)?;
892 :
893 0 : match peek_slot {
894 0 : Some(TenantSlot::Attached(tenant)) => Ok(Arc::clone(tenant)),
895 0 : Some(TenantSlot::InProgress(_)) => Err(GetTenantError::NotActive(tenant_shard_id)),
896 : None | Some(TenantSlot::Secondary(_)) => {
897 0 : Err(GetTenantError::NotFound(tenant_shard_id.tenant_id))
898 : }
899 : }
900 0 : }
901 :
902 0 : pub(crate) fn get_secondary_tenant_shard(
903 0 : &self,
904 0 : tenant_shard_id: TenantShardId,
905 0 : ) -> Option<Arc<SecondaryTenant>> {
906 0 : let locked = self.tenants.read().unwrap();
907 0 :
908 0 : let peek_slot = tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Read)
909 0 : .ok()
910 0 : .flatten();
911 :
912 0 : match peek_slot {
913 0 : Some(TenantSlot::Secondary(s)) => Some(s.clone()),
914 0 : _ => None,
915 : }
916 0 : }
917 :
918 : /// Whether the `TenantManager` is responsible for the tenant shard
919 0 : pub(crate) fn manages_tenant_shard(&self, tenant_shard_id: TenantShardId) -> bool {
920 0 : let locked = self.tenants.read().unwrap();
921 0 :
922 0 : let peek_slot = tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Read)
923 0 : .ok()
924 0 : .flatten();
925 0 :
926 0 : peek_slot.is_some()
927 0 : }
928 :
929 0 : #[instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
930 : pub(crate) async fn upsert_location(
931 : &self,
932 : tenant_shard_id: TenantShardId,
933 : new_location_config: LocationConf,
934 : flush: Option<Duration>,
935 : mut spawn_mode: SpawnMode,
936 : ctx: &RequestContext,
937 : ) -> Result<Option<Arc<Tenant>>, UpsertLocationError> {
938 : debug_assert_current_span_has_tenant_id();
939 : info!("configuring tenant location to state {new_location_config:?}");
940 :
941 : enum FastPathModified {
942 : Attached(Arc<Tenant>),
943 : Secondary(Arc<SecondaryTenant>),
944 : }
945 :
946 : // Special case fast-path for updates to existing slots: if our upsert is only updating configuration,
947 : // then we do not need to set the slot to InProgress, we can just call into the
948 : // existng tenant.
949 : let fast_path_taken = {
950 : let locked = self.tenants.read().unwrap();
951 : let peek_slot =
952 : tenant_map_peek_slot(&locked, &tenant_shard_id, TenantSlotPeekMode::Write)?;
953 : match (&new_location_config.mode, peek_slot) {
954 : (LocationMode::Attached(attach_conf), Some(TenantSlot::Attached(tenant))) => {
955 : match attach_conf.generation.cmp(&tenant.generation) {
956 : Ordering::Equal => {
957 : // A transition from Attached to Attached in the same generation, we may
958 : // take our fast path and just provide the updated configuration
959 : // to the tenant.
960 : tenant.set_new_location_config(
961 : AttachedTenantConf::try_from(new_location_config.clone())
962 : .map_err(UpsertLocationError::BadRequest)?,
963 : );
964 :
965 : Some(FastPathModified::Attached(tenant.clone()))
966 : }
967 : Ordering::Less => {
968 : return Err(UpsertLocationError::BadRequest(anyhow::anyhow!(
969 : "Generation {:?} is less than existing {:?}",
970 : attach_conf.generation,
971 : tenant.generation
972 : )));
973 : }
974 : Ordering::Greater => {
975 : // Generation advanced, fall through to general case of replacing `Tenant` object
976 : None
977 : }
978 : }
979 : }
980 : (
981 : LocationMode::Secondary(secondary_conf),
982 : Some(TenantSlot::Secondary(secondary_tenant)),
983 : ) => {
984 : secondary_tenant.set_config(secondary_conf);
985 : secondary_tenant.set_tenant_conf(&new_location_config.tenant_conf);
986 : Some(FastPathModified::Secondary(secondary_tenant.clone()))
987 : }
988 : _ => {
989 : // Not an Attached->Attached transition, fall through to general case
990 : None
991 : }
992 : }
993 : };
994 :
995 : // Fast-path continued: having dropped out of the self.tenants lock, do the async
996 : // phase of writing config and/or waiting for flush, before returning.
997 : match fast_path_taken {
998 : Some(FastPathModified::Attached(tenant)) => {
999 : Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
1000 : .await
1001 : .fatal_err("write tenant shard config");
1002 :
1003 : // Transition to AttachedStale means we may well hold a valid generation
1004 : // still, and have been requested to go stale as part of a migration. If
1005 : // the caller set `flush`, then flush to remote storage.
1006 : if let LocationMode::Attached(AttachedLocationConfig {
1007 : generation: _,
1008 : attach_mode: AttachmentMode::Stale,
1009 : }) = &new_location_config.mode
1010 : {
1011 : if let Some(flush_timeout) = flush {
1012 : match tokio::time::timeout(flush_timeout, tenant.flush_remote()).await {
1013 : Ok(Err(e)) => {
1014 : return Err(UpsertLocationError::Flush(e));
1015 : }
1016 : Ok(Ok(_)) => return Ok(Some(tenant)),
1017 : Err(_) => {
1018 : tracing::warn!(
1019 : timeout_ms = flush_timeout.as_millis(),
1020 : "Timed out waiting for flush to remote storage, proceeding anyway."
1021 : )
1022 : }
1023 : }
1024 : }
1025 : }
1026 :
1027 : return Ok(Some(tenant));
1028 : }
1029 : Some(FastPathModified::Secondary(_secondary_tenant)) => {
1030 : Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
1031 : .await
1032 : .fatal_err("write tenant shard config");
1033 :
1034 : return Ok(None);
1035 : }
1036 : None => {
1037 : // Proceed with the general case procedure, where we will shutdown & remove any existing
1038 : // slot contents and replace with a fresh one
1039 : }
1040 : };
1041 :
1042 : // General case for upserts to TenantsMap, excluding the case above: we will substitute an
1043 : // InProgress value to the slot while we make whatever changes are required. The state for
1044 : // the tenant is inaccessible to the outside world while we are doing this, but that is sensible:
1045 : // the state is ill-defined while we're in transition. Transitions are async, but fast: we do
1046 : // not do significant I/O, and shutdowns should be prompt via cancellation tokens.
1047 : let mut slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)
1048 0 : .map_err(|e| match e {
1049 : TenantSlotError::NotFound(_) => {
1050 0 : unreachable!("Called with mode Any")
1051 : }
1052 0 : TenantSlotError::InProgress => UpsertLocationError::InProgress,
1053 0 : TenantSlotError::MapState(s) => UpsertLocationError::Unavailable(s),
1054 0 : })?;
1055 :
1056 : match slot_guard.get_old_value() {
1057 : Some(TenantSlot::Attached(tenant)) => {
1058 : // The case where we keep a Tenant alive was covered above in the special case
1059 : // for Attached->Attached transitions in the same generation. By this point,
1060 : // if we see an attached tenant we know it will be discarded and should be
1061 : // shut down.
1062 : let (_guard, progress) = utils::completion::channel();
1063 :
1064 : match tenant.get_attach_mode() {
1065 : AttachmentMode::Single | AttachmentMode::Multi => {
1066 : // Before we leave our state as the presumed holder of the latest generation,
1067 : // flush any outstanding deletions to reduce the risk of leaking objects.
1068 : self.resources.deletion_queue_client.flush_advisory()
1069 : }
1070 : AttachmentMode::Stale => {
1071 : // If we're stale there's not point trying to flush deletions
1072 : }
1073 : };
1074 :
1075 : info!("Shutting down attached tenant");
1076 : match tenant.shutdown(progress, ShutdownMode::Hard).await {
1077 : Ok(()) => {}
1078 : Err(barrier) => {
1079 : info!("Shutdown already in progress, waiting for it to complete");
1080 : barrier.wait().await;
1081 : }
1082 : }
1083 : slot_guard.drop_old_value().expect("We just shut it down");
1084 :
1085 : // Edge case: if we were called with SpawnMode::Create, but a Tenant already existed, then
1086 : // the caller thinks they're creating but the tenant already existed. We must switch to
1087 : // Eager mode so that when starting this Tenant we properly probe remote storage for timelines,
1088 : // rather than assuming it to be empty.
1089 : spawn_mode = SpawnMode::Eager;
1090 : }
1091 : Some(TenantSlot::Secondary(state)) => {
1092 : info!("Shutting down secondary tenant");
1093 : state.shutdown().await;
1094 : }
1095 : Some(TenantSlot::InProgress(_)) => {
1096 : // This should never happen: acquire_slot should error out
1097 : // if the contents of a slot were InProgress.
1098 : return Err(UpsertLocationError::InternalError(anyhow::anyhow!(
1099 : "Acquired an InProgress slot, this is a bug."
1100 : )));
1101 : }
1102 : None => {
1103 : // Slot was vacant, nothing needs shutting down.
1104 : }
1105 : }
1106 :
1107 : let tenant_path = self.conf.tenant_path(&tenant_shard_id);
1108 : let timelines_path = self.conf.timelines_path(&tenant_shard_id);
1109 :
1110 : // Directory structure is the same for attached and secondary modes:
1111 : // create it if it doesn't exist. Timeline load/creation expects the
1112 : // timelines/ subdir to already exist.
1113 : //
1114 : // Does not need to be fsync'd because local storage is just a cache.
1115 : tokio::fs::create_dir_all(&timelines_path)
1116 : .await
1117 : .fatal_err("create timelines/ dir");
1118 :
1119 : // Before activating either secondary or attached mode, persist the
1120 : // configuration, so that on restart we will re-attach (or re-start
1121 : // secondary) on the tenant.
1122 : Tenant::persist_tenant_config(self.conf, &tenant_shard_id, &new_location_config)
1123 : .await
1124 : .fatal_err("write tenant shard config");
1125 :
1126 : let new_slot = match &new_location_config.mode {
1127 : LocationMode::Secondary(secondary_config) => {
1128 : let shard_identity = new_location_config.shard;
1129 : TenantSlot::Secondary(SecondaryTenant::new(
1130 : tenant_shard_id,
1131 : shard_identity,
1132 : new_location_config.tenant_conf,
1133 : secondary_config,
1134 : ))
1135 : }
1136 : LocationMode::Attached(_attach_config) => {
1137 : let shard_identity = new_location_config.shard;
1138 :
1139 : // Testing hack: if we are configured with no control plane, then drop the generation
1140 : // from upserts. This enables creating generation-less tenants even though neon_local
1141 : // always uses generations when calling the location conf API.
1142 : let attached_conf = if cfg!(feature = "testing") {
1143 : let mut conf = AttachedTenantConf::try_from(new_location_config)
1144 : .map_err(UpsertLocationError::BadRequest)?;
1145 : if self.conf.control_plane_api.is_none() {
1146 : conf.location.generation = Generation::none();
1147 : }
1148 : conf
1149 : } else {
1150 : AttachedTenantConf::try_from(new_location_config)
1151 : .map_err(UpsertLocationError::BadRequest)?
1152 : };
1153 :
1154 : let tenant = tenant_spawn(
1155 : self.conf,
1156 : tenant_shard_id,
1157 : &tenant_path,
1158 : self.resources.clone(),
1159 : attached_conf,
1160 : shard_identity,
1161 : None,
1162 : spawn_mode,
1163 : ctx,
1164 : )
1165 0 : .map_err(|_: GlobalShutDown| {
1166 0 : UpsertLocationError::Unavailable(TenantMapError::ShuttingDown)
1167 0 : })?;
1168 :
1169 : TenantSlot::Attached(tenant)
1170 : }
1171 : };
1172 :
1173 : let attached_tenant = if let TenantSlot::Attached(tenant) = &new_slot {
1174 : Some(tenant.clone())
1175 : } else {
1176 : None
1177 : };
1178 :
1179 : match slot_guard.upsert(new_slot) {
1180 : Err(TenantSlotUpsertError::InternalError(e)) => {
1181 : Err(UpsertLocationError::InternalError(anyhow::anyhow!(e)))
1182 : }
1183 : Err(TenantSlotUpsertError::MapState(e)) => Err(UpsertLocationError::Unavailable(e)),
1184 : Err(TenantSlotUpsertError::ShuttingDown((new_slot, _completion))) => {
1185 : // If we just called tenant_spawn() on a new tenant, and can't insert it into our map, then
1186 : // we must not leak it: this would violate the invariant that after shutdown_all_tenants, all tenants
1187 : // are shutdown.
1188 : //
1189 : // We must shut it down inline here.
1190 : match new_slot {
1191 : TenantSlot::InProgress(_) => {
1192 : // Unreachable because we never insert an InProgress
1193 : unreachable!()
1194 : }
1195 : TenantSlot::Attached(tenant) => {
1196 : let (_guard, progress) = utils::completion::channel();
1197 : info!("Shutting down just-spawned tenant, because tenant manager is shut down");
1198 : match tenant.shutdown(progress, ShutdownMode::Hard).await {
1199 : Ok(()) => {
1200 : info!("Finished shutting down just-spawned tenant");
1201 : }
1202 : Err(barrier) => {
1203 : info!("Shutdown already in progress, waiting for it to complete");
1204 : barrier.wait().await;
1205 : }
1206 : }
1207 : }
1208 : TenantSlot::Secondary(secondary_tenant) => {
1209 : secondary_tenant.shutdown().await;
1210 : }
1211 : }
1212 :
1213 : Err(UpsertLocationError::Unavailable(
1214 : TenantMapError::ShuttingDown,
1215 : ))
1216 : }
1217 : Ok(()) => Ok(attached_tenant),
1218 : }
1219 : }
1220 :
1221 : /// Resetting a tenant is equivalent to detaching it, then attaching it again with the same
1222 : /// LocationConf that was last used to attach it. Optionally, the local file cache may be
1223 : /// dropped before re-attaching.
1224 : ///
1225 : /// This is not part of a tenant's normal lifecycle: it is used for debug/support, in situations
1226 : /// where an issue is identified that would go away with a restart of the tenant.
1227 : ///
1228 : /// This does not have any special "force" shutdown of a tenant: it relies on the tenant's tasks
1229 : /// to respect the cancellation tokens used in normal shutdown().
1230 0 : #[instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %drop_cache))]
1231 : pub(crate) async fn reset_tenant(
1232 : &self,
1233 : tenant_shard_id: TenantShardId,
1234 : drop_cache: bool,
1235 : ctx: &RequestContext,
1236 : ) -> anyhow::Result<()> {
1237 : let mut slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
1238 : let Some(old_slot) = slot_guard.get_old_value() else {
1239 : anyhow::bail!("Tenant not found when trying to reset");
1240 : };
1241 :
1242 : let Some(tenant) = old_slot.get_attached() else {
1243 : slot_guard.revert();
1244 : anyhow::bail!("Tenant is not in attached state");
1245 : };
1246 :
1247 : let (_guard, progress) = utils::completion::channel();
1248 : match tenant.shutdown(progress, ShutdownMode::Hard).await {
1249 : Ok(()) => {
1250 : slot_guard.drop_old_value()?;
1251 : }
1252 : Err(_barrier) => {
1253 : slot_guard.revert();
1254 : anyhow::bail!("Cannot reset Tenant, already shutting down");
1255 : }
1256 : }
1257 :
1258 : let tenant_path = self.conf.tenant_path(&tenant_shard_id);
1259 : let timelines_path = self.conf.timelines_path(&tenant_shard_id);
1260 : let config = Tenant::load_tenant_config(self.conf, &tenant_shard_id)?;
1261 :
1262 : if drop_cache {
1263 : tracing::info!("Dropping local file cache");
1264 :
1265 : match tokio::fs::read_dir(&timelines_path).await {
1266 : Err(e) => {
1267 : tracing::warn!("Failed to list timelines while dropping cache: {}", e);
1268 : }
1269 : Ok(mut entries) => {
1270 : while let Some(entry) = entries.next_entry().await? {
1271 : tokio::fs::remove_dir_all(entry.path()).await?;
1272 : }
1273 : }
1274 : }
1275 : }
1276 :
1277 : let shard_identity = config.shard;
1278 : let tenant = tenant_spawn(
1279 : self.conf,
1280 : tenant_shard_id,
1281 : &tenant_path,
1282 : self.resources.clone(),
1283 : AttachedTenantConf::try_from(config)?,
1284 : shard_identity,
1285 : None,
1286 : SpawnMode::Eager,
1287 : ctx,
1288 : )?;
1289 :
1290 : slot_guard.upsert(TenantSlot::Attached(tenant))?;
1291 :
1292 : Ok(())
1293 : }
1294 :
1295 0 : pub(crate) fn get_attached_active_tenant_shards(&self) -> Vec<Arc<Tenant>> {
1296 0 : let locked = self.tenants.read().unwrap();
1297 0 : match &*locked {
1298 0 : TenantsMap::Initializing => Vec::new(),
1299 0 : TenantsMap::Open(map) | TenantsMap::ShuttingDown(map) => map
1300 0 : .values()
1301 0 : .filter_map(|slot| {
1302 0 : slot.get_attached()
1303 0 : .and_then(|t| if t.is_active() { Some(t.clone()) } else { None })
1304 0 : })
1305 0 : .collect(),
1306 : }
1307 0 : }
1308 : // Do some synchronous work for all tenant slots in Secondary state. The provided
1309 : // callback should be small and fast, as it will be called inside the global
1310 : // TenantsMap lock.
1311 0 : pub(crate) fn foreach_secondary_tenants<F>(&self, mut func: F)
1312 0 : where
1313 0 : // TODO: let the callback return a hint to drop out of the loop early
1314 0 : F: FnMut(&TenantShardId, &Arc<SecondaryTenant>),
1315 0 : {
1316 0 : let locked = self.tenants.read().unwrap();
1317 :
1318 0 : let map = match &*locked {
1319 0 : TenantsMap::Initializing | TenantsMap::ShuttingDown(_) => return,
1320 0 : TenantsMap::Open(m) => m,
1321 : };
1322 :
1323 0 : for (tenant_id, slot) in map {
1324 0 : if let TenantSlot::Secondary(state) = slot {
1325 : // Only expose secondary tenants that are not currently shutting down
1326 0 : if !state.cancel.is_cancelled() {
1327 0 : func(tenant_id, state)
1328 0 : }
1329 0 : }
1330 : }
1331 0 : }
1332 :
1333 : /// Total list of all tenant slots: this includes attached, secondary, and InProgress.
1334 0 : pub(crate) fn list(&self) -> Vec<(TenantShardId, TenantSlot)> {
1335 0 : let locked = self.tenants.read().unwrap();
1336 0 : match &*locked {
1337 0 : TenantsMap::Initializing => Vec::new(),
1338 0 : TenantsMap::Open(map) | TenantsMap::ShuttingDown(map) => {
1339 0 : map.iter().map(|(k, v)| (*k, v.clone())).collect()
1340 : }
1341 : }
1342 0 : }
1343 :
1344 0 : pub(crate) fn get(&self, tenant_shard_id: TenantShardId) -> Option<TenantSlot> {
1345 0 : let locked = self.tenants.read().unwrap();
1346 0 : match &*locked {
1347 0 : TenantsMap::Initializing => None,
1348 0 : TenantsMap::Open(map) | TenantsMap::ShuttingDown(map) => {
1349 0 : map.get(&tenant_shard_id).cloned()
1350 : }
1351 : }
1352 0 : }
1353 :
1354 : /// If a tenant is attached, detach it. Then remove its data from remote storage.
1355 : ///
1356 : /// A tenant is considered deleted once it is gone from remote storage. It is the caller's
1357 : /// responsibility to avoid trying to attach the tenant again or use it any way once deletion
1358 : /// has started: this operation is not atomic, and must be retried until it succeeds.
1359 : ///
1360 : /// As a special case, if an unsharded tenant ID is given for a sharded tenant, it will remove
1361 : /// all tenant shards in remote storage (removing all paths with the tenant prefix). The storage
1362 : /// controller uses this to purge all remote tenant data, including any stale parent shards that
1363 : /// may remain after splits. Ideally, this special case would be handled elsewhere. See:
1364 : /// <https://github.com/neondatabase/neon/pull/9394>.
1365 0 : pub(crate) async fn delete_tenant(
1366 0 : &self,
1367 0 : tenant_shard_id: TenantShardId,
1368 0 : ) -> Result<(), DeleteTenantError> {
1369 0 : super::span::debug_assert_current_span_has_tenant_id();
1370 :
1371 0 : async fn delete_local(
1372 0 : conf: &PageServerConf,
1373 0 : background_purges: &BackgroundPurges,
1374 0 : tenant_shard_id: &TenantShardId,
1375 0 : ) -> anyhow::Result<()> {
1376 0 : let local_tenant_directory = conf.tenant_path(tenant_shard_id);
1377 0 : let tmp_dir = safe_rename_tenant_dir(&local_tenant_directory)
1378 0 : .await
1379 0 : .with_context(|| {
1380 0 : format!("local tenant directory {local_tenant_directory:?} rename")
1381 0 : })?;
1382 0 : background_purges.spawn(tmp_dir);
1383 0 : Ok(())
1384 0 : }
1385 :
1386 0 : let slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
1387 0 : match &slot_guard.old_value {
1388 0 : Some(TenantSlot::Attached(tenant)) => {
1389 0 : // Legacy deletion flow: the tenant remains attached, goes to Stopping state, and
1390 0 : // deletion will be resumed across restarts.
1391 0 : let tenant = tenant.clone();
1392 0 : let (_guard, progress) = utils::completion::channel();
1393 0 : match tenant.shutdown(progress, ShutdownMode::Hard).await {
1394 0 : Ok(()) => {}
1395 0 : Err(barrier) => {
1396 0 : info!("Shutdown already in progress, waiting for it to complete");
1397 0 : barrier.wait().await;
1398 : }
1399 : }
1400 0 : delete_local(self.conf, &self.background_purges, &tenant_shard_id).await?;
1401 : }
1402 0 : Some(TenantSlot::Secondary(secondary_tenant)) => {
1403 0 : secondary_tenant.shutdown().await;
1404 :
1405 0 : delete_local(self.conf, &self.background_purges, &tenant_shard_id).await?;
1406 : }
1407 0 : Some(TenantSlot::InProgress(_)) => unreachable!(),
1408 0 : None => {}
1409 : };
1410 :
1411 : // Fall through: local state for this tenant is no longer present, proceed with remote delete.
1412 : // - We use a retry wrapper here so that common transient S3 errors (e.g. 503, 429) do not result
1413 : // in 500 responses to delete requests.
1414 : // - We keep the `SlotGuard` during this I/O, so that if a concurrent delete request comes in, it will
1415 : // 503/retry, rather than kicking off a wasteful concurrent deletion.
1416 : // NB: this also deletes partial prefixes, i.e. a <tenant_id> path will delete all
1417 : // <tenant_id>_<shard_id>/* objects. See method comment for why.
1418 0 : backoff::retry(
1419 0 : || async move {
1420 0 : self.resources
1421 0 : .remote_storage
1422 0 : .delete_prefix(&remote_tenant_path(&tenant_shard_id), &self.cancel)
1423 0 : .await
1424 0 : },
1425 0 : |_| false, // backoff::retry handles cancellation
1426 0 : 1,
1427 0 : 3,
1428 0 : &format!("delete_tenant[tenant_shard_id={tenant_shard_id}]"),
1429 0 : &self.cancel,
1430 0 : )
1431 0 : .await
1432 0 : .unwrap_or(Err(TimeoutOrCancel::Cancel.into()))
1433 0 : .map_err(|err| {
1434 0 : if TimeoutOrCancel::caused_by_cancel(&err) {
1435 0 : return DeleteTenantError::Cancelled;
1436 0 : }
1437 0 : DeleteTenantError::Other(err)
1438 0 : })
1439 0 : }
1440 :
1441 0 : #[instrument(skip_all, fields(tenant_id=%tenant.get_tenant_shard_id().tenant_id, shard_id=%tenant.get_tenant_shard_id().shard_slug(), new_shard_count=%new_shard_count.literal()))]
1442 : pub(crate) async fn shard_split(
1443 : &self,
1444 : tenant: Arc<Tenant>,
1445 : new_shard_count: ShardCount,
1446 : new_stripe_size: Option<ShardStripeSize>,
1447 : ctx: &RequestContext,
1448 : ) -> anyhow::Result<Vec<TenantShardId>> {
1449 : let tenant_shard_id = *tenant.get_tenant_shard_id();
1450 : let r = self
1451 : .do_shard_split(tenant, new_shard_count, new_stripe_size, ctx)
1452 : .await;
1453 : if r.is_err() {
1454 : // Shard splitting might have left the original shard in a partially shut down state (it
1455 : // stops the shard's remote timeline client). Reset it to ensure we leave things in
1456 : // a working state.
1457 : if self.get(tenant_shard_id).is_some() {
1458 : tracing::warn!("Resetting after shard split failure");
1459 : if let Err(e) = self.reset_tenant(tenant_shard_id, false, ctx).await {
1460 : // Log this error because our return value will still be the original error, not this one. This is
1461 : // a severe error: if this happens, we might be leaving behind a tenant that is not fully functional
1462 : // (e.g. has uploads disabled). We can't do anything else: if reset fails then shutting the tenant down or
1463 : // setting it broken probably won't help either.
1464 : tracing::error!("Failed to reset: {e}");
1465 : }
1466 : }
1467 : }
1468 :
1469 : r
1470 : }
1471 :
1472 0 : pub(crate) async fn do_shard_split(
1473 0 : &self,
1474 0 : tenant: Arc<Tenant>,
1475 0 : new_shard_count: ShardCount,
1476 0 : new_stripe_size: Option<ShardStripeSize>,
1477 0 : ctx: &RequestContext,
1478 0 : ) -> anyhow::Result<Vec<TenantShardId>> {
1479 0 : let tenant_shard_id = *tenant.get_tenant_shard_id();
1480 0 :
1481 0 : // Validate the incoming request
1482 0 : if new_shard_count.count() <= tenant_shard_id.shard_count.count() {
1483 0 : anyhow::bail!("Requested shard count is not an increase");
1484 0 : }
1485 0 : let expansion_factor = new_shard_count.count() / tenant_shard_id.shard_count.count();
1486 0 : if !expansion_factor.is_power_of_two() {
1487 0 : anyhow::bail!("Requested split is not a power of two");
1488 0 : }
1489 :
1490 0 : if let Some(new_stripe_size) = new_stripe_size {
1491 0 : if tenant.get_shard_stripe_size() != new_stripe_size
1492 0 : && tenant_shard_id.shard_count.count() > 1
1493 : {
1494 : // This tenant already has multiple shards, it is illegal to try and change its stripe size
1495 0 : anyhow::bail!(
1496 0 : "Shard stripe size may not be modified once tenant has multiple shards"
1497 0 : );
1498 0 : }
1499 0 : }
1500 :
1501 : // Plan: identify what the new child shards will be
1502 0 : let child_shards = tenant_shard_id.split(new_shard_count);
1503 0 : tracing::info!(
1504 0 : "Shard {} splits into: {}",
1505 0 : tenant_shard_id.to_index(),
1506 0 : child_shards
1507 0 : .iter()
1508 0 : .map(|id| format!("{}", id.to_index()))
1509 0 : .join(",")
1510 : );
1511 :
1512 0 : fail::fail_point!("shard-split-pre-prepare", |_| Err(anyhow::anyhow!(
1513 0 : "failpoint"
1514 0 : )));
1515 :
1516 0 : let parent_shard_identity = tenant.shard_identity;
1517 0 : let parent_tenant_conf = tenant.get_tenant_conf();
1518 0 : let parent_generation = tenant.generation;
1519 :
1520 : // Phase 1: Write out child shards' remote index files, in the parent tenant's current generation
1521 0 : if let Err(e) = tenant.split_prepare(&child_shards).await {
1522 : // If [`Tenant::split_prepare`] fails, we must reload the tenant, because it might
1523 : // have been left in a partially-shut-down state.
1524 0 : tracing::warn!("Failed to prepare for split: {e}, reloading Tenant before returning");
1525 0 : return Err(e);
1526 0 : }
1527 0 :
1528 0 : fail::fail_point!("shard-split-post-prepare", |_| Err(anyhow::anyhow!(
1529 0 : "failpoint"
1530 0 : )));
1531 :
1532 0 : self.resources.deletion_queue_client.flush_advisory();
1533 0 :
1534 0 : // Phase 2: Put the parent shard to InProgress and grab a reference to the parent Tenant
1535 0 : drop(tenant);
1536 0 : let mut parent_slot_guard =
1537 0 : tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
1538 0 : let parent = match parent_slot_guard.get_old_value() {
1539 0 : Some(TenantSlot::Attached(t)) => t,
1540 0 : Some(TenantSlot::Secondary(_)) => anyhow::bail!("Tenant location in secondary mode"),
1541 : Some(TenantSlot::InProgress(_)) => {
1542 : // tenant_map_acquire_slot never returns InProgress, if a slot was InProgress
1543 : // it would return an error.
1544 0 : unreachable!()
1545 : }
1546 : None => {
1547 : // We don't actually need the parent shard to still be attached to do our work, but it's
1548 : // a weird enough situation that the caller probably didn't want us to continue working
1549 : // if they had detached the tenant they requested the split on.
1550 0 : anyhow::bail!("Detached parent shard in the middle of split!")
1551 : }
1552 : };
1553 0 : fail::fail_point!("shard-split-pre-hardlink", |_| Err(anyhow::anyhow!(
1554 0 : "failpoint"
1555 0 : )));
1556 : // Optimization: hardlink layers from the parent into the children, so that they don't have to
1557 : // re-download & duplicate the data referenced in their initial IndexPart
1558 0 : self.shard_split_hardlink(parent, child_shards.clone())
1559 0 : .await?;
1560 0 : fail::fail_point!("shard-split-post-hardlink", |_| Err(anyhow::anyhow!(
1561 0 : "failpoint"
1562 0 : )));
1563 :
1564 : // Take a snapshot of where the parent's WAL ingest had got to: we will wait for
1565 : // child shards to reach this point.
1566 0 : let mut target_lsns = HashMap::new();
1567 0 : for timeline in parent.timelines.lock().unwrap().clone().values() {
1568 0 : target_lsns.insert(timeline.timeline_id, timeline.get_last_record_lsn());
1569 0 : }
1570 :
1571 : // TODO: we should have the parent shard stop its WAL ingest here, it's a waste of resources
1572 : // and could slow down the children trying to catch up.
1573 :
1574 : // Phase 3: Spawn the child shards
1575 0 : for child_shard in &child_shards {
1576 0 : let mut child_shard_identity = parent_shard_identity;
1577 0 : if let Some(new_stripe_size) = new_stripe_size {
1578 0 : child_shard_identity.stripe_size = new_stripe_size;
1579 0 : }
1580 0 : child_shard_identity.count = child_shard.shard_count;
1581 0 : child_shard_identity.number = child_shard.shard_number;
1582 0 :
1583 0 : let child_location_conf = LocationConf {
1584 0 : mode: LocationMode::Attached(AttachedLocationConfig {
1585 0 : generation: parent_generation,
1586 0 : attach_mode: AttachmentMode::Single,
1587 0 : }),
1588 0 : shard: child_shard_identity,
1589 0 : tenant_conf: parent_tenant_conf.clone(),
1590 0 : };
1591 0 :
1592 0 : self.upsert_location(
1593 0 : *child_shard,
1594 0 : child_location_conf,
1595 0 : None,
1596 0 : SpawnMode::Eager,
1597 0 : ctx,
1598 0 : )
1599 0 : .await?;
1600 : }
1601 :
1602 0 : fail::fail_point!("shard-split-post-child-conf", |_| Err(anyhow::anyhow!(
1603 0 : "failpoint"
1604 0 : )));
1605 :
1606 : // Phase 4: wait for child chards WAL ingest to catch up to target LSN
1607 0 : for child_shard_id in &child_shards {
1608 0 : let child_shard_id = *child_shard_id;
1609 0 : let child_shard = {
1610 0 : let locked = self.tenants.read().unwrap();
1611 0 : let peek_slot =
1612 0 : tenant_map_peek_slot(&locked, &child_shard_id, TenantSlotPeekMode::Read)?;
1613 0 : peek_slot.and_then(|s| s.get_attached()).cloned()
1614 : };
1615 0 : if let Some(t) = child_shard {
1616 : // Wait for the child shard to become active: this should be very quick because it only
1617 : // has to download the index_part that we just uploaded when creating it.
1618 0 : if let Err(e) = t.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await {
1619 : // This is not fatal: we have durably created the child shard. It just makes the
1620 : // split operation less seamless for clients, as we will may detach the parent
1621 : // shard before the child shards are fully ready to serve requests.
1622 0 : tracing::warn!("Failed to wait for shard {child_shard_id} to activate: {e}");
1623 0 : continue;
1624 0 : }
1625 0 :
1626 0 : let timelines = t.timelines.lock().unwrap().clone();
1627 0 : for timeline in timelines.values() {
1628 0 : let Some(target_lsn) = target_lsns.get(&timeline.timeline_id) else {
1629 0 : continue;
1630 : };
1631 :
1632 0 : tracing::info!(
1633 0 : "Waiting for child shard {}/{} to reach target lsn {}...",
1634 0 : child_shard_id,
1635 0 : timeline.timeline_id,
1636 : target_lsn
1637 : );
1638 :
1639 0 : fail::fail_point!("shard-split-lsn-wait", |_| Err(anyhow::anyhow!(
1640 0 : "failpoint"
1641 0 : )));
1642 0 : if let Err(e) = timeline
1643 0 : .wait_lsn(
1644 0 : *target_lsn,
1645 0 : crate::tenant::timeline::WaitLsnWaiter::Tenant,
1646 0 : ctx,
1647 0 : )
1648 0 : .await
1649 : {
1650 : // Failure here might mean shutdown, in any case this part is an optimization
1651 : // and we shouldn't hold up the split operation.
1652 0 : tracing::warn!(
1653 0 : "Failed to wait for timeline {} to reach lsn {target_lsn}: {e}",
1654 0 : timeline.timeline_id
1655 : );
1656 : } else {
1657 0 : tracing::info!(
1658 0 : "Child shard {}/{} reached target lsn {}",
1659 0 : child_shard_id,
1660 0 : timeline.timeline_id,
1661 : target_lsn
1662 : );
1663 : }
1664 : }
1665 0 : }
1666 : }
1667 :
1668 : // Phase 5: Shut down the parent shard, and erase it from disk
1669 0 : let (_guard, progress) = completion::channel();
1670 0 : match parent.shutdown(progress, ShutdownMode::Hard).await {
1671 0 : Ok(()) => {}
1672 0 : Err(other) => {
1673 0 : other.wait().await;
1674 : }
1675 : }
1676 0 : let local_tenant_directory = self.conf.tenant_path(&tenant_shard_id);
1677 0 : let tmp_path = safe_rename_tenant_dir(&local_tenant_directory)
1678 0 : .await
1679 0 : .with_context(|| format!("local tenant directory {local_tenant_directory:?} rename"))?;
1680 0 : self.background_purges.spawn(tmp_path);
1681 0 :
1682 0 : fail::fail_point!("shard-split-pre-finish", |_| Err(anyhow::anyhow!(
1683 0 : "failpoint"
1684 0 : )));
1685 :
1686 0 : parent_slot_guard.drop_old_value()?;
1687 :
1688 : // Phase 6: Release the InProgress on the parent shard
1689 0 : drop(parent_slot_guard);
1690 0 :
1691 0 : Ok(child_shards)
1692 0 : }
1693 :
1694 : /// Part of [`Self::shard_split`]: hard link parent shard layers into child shards, as an optimization
1695 : /// to avoid the children downloading them again.
1696 : ///
1697 : /// For each resident layer in the parent shard, we will hard link it into all of the child shards.
1698 0 : async fn shard_split_hardlink(
1699 0 : &self,
1700 0 : parent_shard: &Tenant,
1701 0 : child_shards: Vec<TenantShardId>,
1702 0 : ) -> anyhow::Result<()> {
1703 0 : debug_assert_current_span_has_tenant_id();
1704 0 :
1705 0 : let parent_path = self.conf.tenant_path(parent_shard.get_tenant_shard_id());
1706 0 : let (parent_timelines, parent_layers) = {
1707 0 : let mut parent_layers = Vec::new();
1708 0 : let timelines = parent_shard.timelines.lock().unwrap().clone();
1709 0 : let parent_timelines = timelines.keys().cloned().collect::<Vec<_>>();
1710 0 : for timeline in timelines.values() {
1711 0 : tracing::info!(timeline_id=%timeline.timeline_id, "Loading list of layers to hardlink");
1712 0 : let layers = timeline.layers.read().await;
1713 :
1714 0 : for layer in layers.likely_resident_layers() {
1715 0 : let relative_path = layer
1716 0 : .local_path()
1717 0 : .strip_prefix(&parent_path)
1718 0 : .context("Removing prefix from parent layer path")?;
1719 0 : parent_layers.push(relative_path.to_owned());
1720 : }
1721 : }
1722 :
1723 0 : if parent_layers.is_empty() {
1724 0 : tracing::info!("Ancestor shard has no resident layer to hard link");
1725 0 : }
1726 :
1727 0 : (parent_timelines, parent_layers)
1728 0 : };
1729 0 :
1730 0 : let mut child_prefixes = Vec::new();
1731 0 : let mut create_dirs = Vec::new();
1732 :
1733 0 : for child in child_shards {
1734 0 : let child_prefix = self.conf.tenant_path(&child);
1735 0 : create_dirs.push(child_prefix.clone());
1736 0 : create_dirs.extend(
1737 0 : parent_timelines
1738 0 : .iter()
1739 0 : .map(|t| self.conf.timeline_path(&child, t)),
1740 0 : );
1741 0 :
1742 0 : child_prefixes.push(child_prefix);
1743 0 : }
1744 :
1745 : // Since we will do a large number of small filesystem metadata operations, batch them into
1746 : // spawn_blocking calls rather than doing each one as a tokio::fs round-trip.
1747 0 : let span = tracing::Span::current();
1748 0 : let jh = tokio::task::spawn_blocking(move || -> anyhow::Result<usize> {
1749 0 : // Run this synchronous code in the same log context as the outer function that spawned it.
1750 0 : let _span = span.enter();
1751 0 :
1752 0 : tracing::info!("Creating {} directories", create_dirs.len());
1753 0 : for dir in &create_dirs {
1754 0 : if let Err(e) = std::fs::create_dir_all(dir) {
1755 : // Ignore AlreadyExists errors, drop out on all other errors
1756 0 : match e.kind() {
1757 0 : std::io::ErrorKind::AlreadyExists => {}
1758 : _ => {
1759 0 : return Err(anyhow::anyhow!(e).context(format!("Creating {dir}")));
1760 : }
1761 : }
1762 0 : }
1763 : }
1764 :
1765 0 : for child_prefix in child_prefixes {
1766 0 : tracing::info!(
1767 0 : "Hard-linking {} parent layers into child path {}",
1768 0 : parent_layers.len(),
1769 : child_prefix
1770 : );
1771 0 : for relative_layer in &parent_layers {
1772 0 : let parent_path = parent_path.join(relative_layer);
1773 0 : let child_path = child_prefix.join(relative_layer);
1774 0 : if let Err(e) = std::fs::hard_link(&parent_path, &child_path) {
1775 0 : match e.kind() {
1776 0 : std::io::ErrorKind::AlreadyExists => {}
1777 : std::io::ErrorKind::NotFound => {
1778 0 : tracing::info!(
1779 0 : "Layer {} not found during hard-linking, evicted during split?",
1780 : relative_layer
1781 : );
1782 : }
1783 : _ => {
1784 0 : return Err(anyhow::anyhow!(e).context(format!(
1785 0 : "Hard linking {relative_layer} into {child_prefix}"
1786 0 : )))
1787 : }
1788 : }
1789 0 : }
1790 : }
1791 : }
1792 :
1793 : // Durability is not required for correctness, but if we crashed during split and
1794 : // then came restarted with empty timeline dirs, it would be very inefficient to
1795 : // re-populate from remote storage.
1796 0 : tracing::info!("fsyncing {} directories", create_dirs.len());
1797 0 : for dir in create_dirs {
1798 0 : if let Err(e) = crashsafe::fsync(&dir) {
1799 : // Something removed a newly created timeline dir out from underneath us? Extremely
1800 : // unexpected, but not worth panic'ing over as this whole function is just an
1801 : // optimization.
1802 0 : tracing::warn!("Failed to fsync directory {dir}: {e}")
1803 0 : }
1804 : }
1805 :
1806 0 : Ok(parent_layers.len())
1807 0 : });
1808 0 :
1809 0 : match jh.await {
1810 0 : Ok(Ok(layer_count)) => {
1811 0 : tracing::info!(count = layer_count, "Hard linked layers into child shards");
1812 : }
1813 0 : Ok(Err(e)) => {
1814 0 : // This is an optimization, so we tolerate failure.
1815 0 : tracing::warn!("Error hard-linking layers, proceeding anyway: {e}")
1816 : }
1817 0 : Err(e) => {
1818 0 : // This is something totally unexpected like a panic, so bail out.
1819 0 : anyhow::bail!("Error joining hard linking task: {e}");
1820 : }
1821 : }
1822 :
1823 0 : Ok(())
1824 0 : }
1825 :
1826 : ///
1827 : /// Shut down all tenants. This runs as part of pageserver shutdown.
1828 : ///
1829 : /// NB: We leave the tenants in the map, so that they remain accessible through
1830 : /// the management API until we shut it down. If we removed the shut-down tenants
1831 : /// from the tenants map, the management API would return 404 for these tenants,
1832 : /// because TenantsMap::get() now returns `None`.
1833 : /// That could be easily misinterpreted by control plane, the consumer of the
1834 : /// management API. For example, it could attach the tenant on a different pageserver.
1835 : /// We would then be in split-brain once this pageserver restarts.
1836 0 : #[instrument(skip_all)]
1837 : pub(crate) async fn shutdown(&self) {
1838 : self.cancel.cancel();
1839 :
1840 : shutdown_all_tenants0(self.tenants).await
1841 : }
1842 :
1843 0 : pub(crate) async fn detach_tenant(
1844 0 : &self,
1845 0 : conf: &'static PageServerConf,
1846 0 : tenant_shard_id: TenantShardId,
1847 0 : deletion_queue_client: &DeletionQueueClient,
1848 0 : ) -> Result<(), TenantStateError> {
1849 0 : let tmp_path = self
1850 0 : .detach_tenant0(conf, tenant_shard_id, deletion_queue_client)
1851 0 : .await?;
1852 0 : self.background_purges.spawn(tmp_path);
1853 0 :
1854 0 : Ok(())
1855 0 : }
1856 :
1857 0 : async fn detach_tenant0(
1858 0 : &self,
1859 0 : conf: &'static PageServerConf,
1860 0 : tenant_shard_id: TenantShardId,
1861 0 : deletion_queue_client: &DeletionQueueClient,
1862 0 : ) -> Result<Utf8PathBuf, TenantStateError> {
1863 0 : let tenant_dir_rename_operation = |tenant_id_to_clean: TenantShardId| async move {
1864 0 : let local_tenant_directory = conf.tenant_path(&tenant_id_to_clean);
1865 0 : safe_rename_tenant_dir(&local_tenant_directory)
1866 0 : .await
1867 0 : .with_context(|| {
1868 0 : format!("local tenant directory {local_tenant_directory:?} rename")
1869 0 : })
1870 0 : };
1871 :
1872 0 : let removal_result = remove_tenant_from_memory(
1873 0 : self.tenants,
1874 0 : tenant_shard_id,
1875 0 : tenant_dir_rename_operation(tenant_shard_id),
1876 0 : )
1877 0 : .await;
1878 :
1879 : // Flush pending deletions, so that they have a good chance of passing validation
1880 : // before this tenant is potentially re-attached elsewhere.
1881 0 : deletion_queue_client.flush_advisory();
1882 0 :
1883 0 : removal_result
1884 0 : }
1885 :
1886 0 : pub(crate) fn list_tenants(
1887 0 : &self,
1888 0 : ) -> Result<Vec<(TenantShardId, TenantState, Generation)>, TenantMapListError> {
1889 0 : let tenants = self.tenants.read().unwrap();
1890 0 : let m = match &*tenants {
1891 0 : TenantsMap::Initializing => return Err(TenantMapListError::Initializing),
1892 0 : TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m,
1893 0 : };
1894 0 : Ok(m.iter()
1895 0 : .filter_map(|(id, tenant)| match tenant {
1896 0 : TenantSlot::Attached(tenant) => {
1897 0 : Some((*id, tenant.current_state(), tenant.generation()))
1898 : }
1899 0 : TenantSlot::Secondary(_) => None,
1900 0 : TenantSlot::InProgress(_) => None,
1901 0 : })
1902 0 : .collect())
1903 0 : }
1904 :
1905 : /// Completes an earlier prepared timeline detach ancestor.
1906 0 : pub(crate) async fn complete_detaching_timeline_ancestor(
1907 0 : &self,
1908 0 : tenant_shard_id: TenantShardId,
1909 0 : timeline_id: TimelineId,
1910 0 : prepared: PreparedTimelineDetach,
1911 0 : mut attempt: detach_ancestor::Attempt,
1912 0 : ctx: &RequestContext,
1913 0 : ) -> Result<HashSet<TimelineId>, detach_ancestor::Error> {
1914 : use detach_ancestor::Error;
1915 :
1916 0 : let slot_guard =
1917 0 : tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::MustExist).map_err(
1918 0 : |e| {
1919 : use TenantSlotError::*;
1920 :
1921 0 : match e {
1922 0 : MapState(TenantMapError::ShuttingDown) => Error::ShuttingDown,
1923 0 : NotFound(_) | InProgress | MapState(_) => Error::DetachReparent(e.into()),
1924 : }
1925 0 : },
1926 0 : )?;
1927 :
1928 0 : let tenant = {
1929 0 : let old_slot = slot_guard
1930 0 : .get_old_value()
1931 0 : .as_ref()
1932 0 : .expect("requested MustExist");
1933 :
1934 0 : let Some(tenant) = old_slot.get_attached() else {
1935 0 : return Err(Error::DetachReparent(anyhow::anyhow!(
1936 0 : "Tenant is not in attached state"
1937 0 : )));
1938 : };
1939 :
1940 0 : if !tenant.is_active() {
1941 0 : return Err(Error::DetachReparent(anyhow::anyhow!(
1942 0 : "Tenant is not active"
1943 0 : )));
1944 0 : }
1945 0 :
1946 0 : tenant.clone()
1947 : };
1948 :
1949 0 : let timeline = tenant
1950 0 : .get_timeline(timeline_id, true)
1951 0 : .map_err(Error::NotFound)?;
1952 :
1953 0 : let resp = timeline
1954 0 : .detach_from_ancestor_and_reparent(&tenant, prepared, ctx)
1955 0 : .await?;
1956 :
1957 0 : let mut slot_guard = slot_guard;
1958 :
1959 0 : let tenant = if resp.reset_tenant_required() {
1960 0 : attempt.before_reset_tenant();
1961 0 :
1962 0 : let (_guard, progress) = utils::completion::channel();
1963 0 : match tenant.shutdown(progress, ShutdownMode::Flush).await {
1964 0 : Ok(()) => {
1965 0 : slot_guard.drop_old_value().expect("it was just shutdown");
1966 0 : }
1967 0 : Err(_barrier) => {
1968 0 : slot_guard.revert();
1969 0 : // this really should not happen, at all, unless a shutdown without acquiring
1970 0 : // tenant slot was already going? regardless, on restart the attempt tracking
1971 0 : // will reset to retryable.
1972 0 : return Err(Error::ShuttingDown);
1973 : }
1974 : }
1975 :
1976 0 : let tenant_path = self.conf.tenant_path(&tenant_shard_id);
1977 0 : let config = Tenant::load_tenant_config(self.conf, &tenant_shard_id)
1978 0 : .map_err(|e| Error::DetachReparent(e.into()))?;
1979 :
1980 0 : let shard_identity = config.shard;
1981 0 : let tenant = tenant_spawn(
1982 0 : self.conf,
1983 0 : tenant_shard_id,
1984 0 : &tenant_path,
1985 0 : self.resources.clone(),
1986 0 : AttachedTenantConf::try_from(config).map_err(Error::DetachReparent)?,
1987 0 : shard_identity,
1988 0 : None,
1989 0 : SpawnMode::Eager,
1990 0 : ctx,
1991 0 : )
1992 0 : .map_err(|_| Error::ShuttingDown)?;
1993 :
1994 : {
1995 0 : let mut g = tenant.ongoing_timeline_detach.lock().unwrap();
1996 0 : assert!(
1997 0 : g.is_none(),
1998 0 : "there cannot be any new timeline detach ancestor on newly created tenant"
1999 : );
2000 0 : *g = Some((attempt.timeline_id, attempt.new_barrier()));
2001 0 : }
2002 0 :
2003 0 : // if we bail out here, we will not allow a new attempt, which should be fine.
2004 0 : // pageserver should be shutting down regardless? tenant_reset would help, unless it
2005 0 : // runs into the same problem.
2006 0 : slot_guard
2007 0 : .upsert(TenantSlot::Attached(tenant.clone()))
2008 0 : .map_err(|e| match e {
2009 0 : TenantSlotUpsertError::ShuttingDown(_) => Error::ShuttingDown,
2010 0 : other => Error::DetachReparent(other.into()),
2011 0 : })?;
2012 0 : tenant
2013 : } else {
2014 0 : tracing::info!("skipping tenant_reset as no changes made required it");
2015 0 : tenant
2016 : };
2017 :
2018 0 : if let Some(reparented) = resp.completed() {
2019 : // finally ask the restarted tenant to complete the detach
2020 : //
2021 : // rationale for 9999s: we don't really have a timetable here; if retried, the caller
2022 : // will get an 503.
2023 0 : tenant
2024 0 : .wait_to_become_active(std::time::Duration::from_secs(9999))
2025 0 : .await
2026 0 : .map_err(|e| {
2027 : use pageserver_api::models::TenantState;
2028 : use GetActiveTenantError::{Cancelled, WillNotBecomeActive};
2029 0 : match e {
2030 : Cancelled | WillNotBecomeActive(TenantState::Stopping { .. }) => {
2031 0 : Error::ShuttingDown
2032 : }
2033 0 : other => Error::Complete(other.into()),
2034 : }
2035 0 : })?;
2036 :
2037 0 : utils::pausable_failpoint!(
2038 0 : "timeline-detach-ancestor::after_activating_before_finding-pausable"
2039 0 : );
2040 :
2041 0 : let timeline = tenant
2042 0 : .get_timeline(attempt.timeline_id, true)
2043 0 : .map_err(Error::NotFound)?;
2044 :
2045 0 : timeline
2046 0 : .complete_detaching_timeline_ancestor(&tenant, attempt, ctx)
2047 0 : .await
2048 0 : .map(|()| reparented)
2049 : } else {
2050 : // at least the latest versions have now been downloaded and refreshed; be ready to
2051 : // retry another time.
2052 0 : Err(Error::FailedToReparentAll)
2053 : }
2054 0 : }
2055 :
2056 : /// A page service client sends a TenantId, and to look up the correct Tenant we must
2057 : /// resolve this to a fully qualified TenantShardId.
2058 : ///
2059 : /// During shard splits: we shall see parent shards in InProgress state and skip them, and
2060 : /// instead match on child shards which should appear in Attached state. Very early in a shard
2061 : /// split, or in other cases where a shard is InProgress, we will return our own InProgress result
2062 : /// to instruct the caller to wait for that to finish before querying again.
2063 0 : pub(crate) fn resolve_attached_shard(
2064 0 : &self,
2065 0 : tenant_id: &TenantId,
2066 0 : selector: ShardSelector,
2067 0 : ) -> ShardResolveResult {
2068 0 : let tenants = self.tenants.read().unwrap();
2069 0 : let mut want_shard = None;
2070 0 : let mut any_in_progress = None;
2071 0 :
2072 0 : match &*tenants {
2073 0 : TenantsMap::Initializing => ShardResolveResult::NotFound,
2074 0 : TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => {
2075 0 : for slot in m.range(TenantShardId::tenant_range(*tenant_id)) {
2076 : // Ignore all slots that don't contain an attached tenant
2077 0 : let tenant = match &slot.1 {
2078 0 : TenantSlot::Attached(t) => t,
2079 0 : TenantSlot::InProgress(barrier) => {
2080 0 : // We might still find a usable shard, but in case we don't, remember that
2081 0 : // we saw at least one InProgress slot, so that we can distinguish this case
2082 0 : // from a simple NotFound in our return value.
2083 0 : any_in_progress = Some(barrier.clone());
2084 0 : continue;
2085 : }
2086 0 : _ => continue,
2087 : };
2088 :
2089 0 : match selector {
2090 0 : ShardSelector::Zero if slot.0.shard_number == ShardNumber(0) => {
2091 0 : return ShardResolveResult::Found(tenant.clone())
2092 : }
2093 0 : ShardSelector::Page(key) => {
2094 0 : // First slot we see for this tenant, calculate the expected shard number
2095 0 : // for the key: we will use this for checking if this and subsequent
2096 0 : // slots contain the key, rather than recalculating the hash each time.
2097 0 : if want_shard.is_none() {
2098 0 : want_shard = Some(tenant.shard_identity.get_shard_number(&key));
2099 0 : }
2100 :
2101 0 : if Some(tenant.shard_identity.number) == want_shard {
2102 0 : return ShardResolveResult::Found(tenant.clone());
2103 0 : }
2104 : }
2105 0 : ShardSelector::Known(shard)
2106 0 : if tenant.shard_identity.shard_index() == shard =>
2107 0 : {
2108 0 : return ShardResolveResult::Found(tenant.clone());
2109 : }
2110 0 : _ => continue,
2111 : }
2112 : }
2113 :
2114 : // Fall through: we didn't find a slot that was in Attached state & matched our selector. If
2115 : // we found one or more InProgress slot, indicate to caller that they should retry later. Otherwise
2116 : // this requested shard simply isn't found.
2117 0 : if let Some(barrier) = any_in_progress {
2118 0 : ShardResolveResult::InProgress(barrier)
2119 : } else {
2120 0 : ShardResolveResult::NotFound
2121 : }
2122 : }
2123 : }
2124 0 : }
2125 :
2126 : /// Calculate the tenant shards' contributions to this pageserver's utilization metrics. The
2127 : /// returned values are:
2128 : /// - the number of bytes of local disk space this pageserver's shards are requesting, i.e.
2129 : /// how much space they would use if not impacted by disk usage eviction.
2130 : /// - the number of tenant shards currently on this pageserver, including attached
2131 : /// and secondary.
2132 : ///
2133 : /// This function is quite expensive: callers are expected to cache the result and
2134 : /// limit how often they call it.
2135 0 : pub(crate) fn calculate_utilization(&self) -> Result<(u64, u32), TenantMapListError> {
2136 0 : let tenants = self.tenants.read().unwrap();
2137 0 : let m = match &*tenants {
2138 0 : TenantsMap::Initializing => return Err(TenantMapListError::Initializing),
2139 0 : TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m,
2140 0 : };
2141 0 : let shard_count = m.len();
2142 0 : let mut wanted_bytes = 0;
2143 :
2144 0 : for tenant_slot in m.values() {
2145 0 : match tenant_slot {
2146 0 : TenantSlot::InProgress(_barrier) => {
2147 0 : // While a slot is being changed, we can't know how much storage it wants. This
2148 0 : // means this function's output can fluctuate if a lot of changes are going on
2149 0 : // (such as transitions from secondary to attached).
2150 0 : //
2151 0 : // We could wait for the barrier and retry, but it's important that the utilization
2152 0 : // API is responsive, and the data quality impact is not very significant.
2153 0 : continue;
2154 : }
2155 0 : TenantSlot::Attached(tenant) => {
2156 0 : wanted_bytes += tenant.local_storage_wanted();
2157 0 : }
2158 0 : TenantSlot::Secondary(secondary) => {
2159 0 : let progress = secondary.progress.lock().unwrap();
2160 0 : wanted_bytes += if progress.heatmap_mtime.is_some() {
2161 : // If we have heatmap info, then we will 'want' the sum
2162 : // of the size of layers in the heatmap: this is how much space
2163 : // we would use if not doing any eviction.
2164 0 : progress.bytes_total
2165 : } else {
2166 : // In the absence of heatmap info, assume that the secondary location simply
2167 : // needs as much space as it is currently using.
2168 0 : secondary.resident_size_metric.get()
2169 : }
2170 : }
2171 : }
2172 : }
2173 :
2174 0 : Ok((wanted_bytes, shard_count as u32))
2175 0 : }
2176 :
2177 0 : #[instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), %timeline_id))]
2178 : pub(crate) async fn immediate_gc(
2179 : &self,
2180 : tenant_shard_id: TenantShardId,
2181 : timeline_id: TimelineId,
2182 : gc_req: TimelineGcRequest,
2183 : cancel: CancellationToken,
2184 : ctx: &RequestContext,
2185 : ) -> Result<GcResult, ApiError> {
2186 : let tenant = {
2187 : let guard = self.tenants.read().unwrap();
2188 : guard
2189 : .get(&tenant_shard_id)
2190 : .cloned()
2191 0 : .with_context(|| format!("tenant {tenant_shard_id}"))
2192 0 : .map_err(|e| ApiError::NotFound(e.into()))?
2193 : };
2194 :
2195 0 : let gc_horizon = gc_req.gc_horizon.unwrap_or_else(|| tenant.get_gc_horizon());
2196 : // Use tenant's pitr setting
2197 : let pitr = tenant.get_pitr_interval();
2198 :
2199 : tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
2200 :
2201 : // Run in task_mgr to avoid race with tenant_detach operation
2202 : let ctx: RequestContext =
2203 : ctx.detached_child(TaskKind::GarbageCollector, DownloadBehavior::Download);
2204 :
2205 0 : let _gate_guard = tenant.gate.enter().map_err(|_| ApiError::ShuttingDown)?;
2206 :
2207 : fail::fail_point!("immediate_gc_task_pre");
2208 :
2209 : #[allow(unused_mut)]
2210 : let mut result = tenant
2211 : .gc_iteration(Some(timeline_id), gc_horizon, pitr, &cancel, &ctx)
2212 : .await;
2213 : // FIXME: `gc_iteration` can return an error for multiple reasons; we should handle it
2214 : // better once the types support it.
2215 :
2216 : #[cfg(feature = "testing")]
2217 : {
2218 : // we need to synchronize with drop completion for python tests without polling for
2219 : // log messages
2220 : if let Ok(result) = result.as_mut() {
2221 : let mut js = tokio::task::JoinSet::new();
2222 : for layer in std::mem::take(&mut result.doomed_layers) {
2223 : js.spawn(layer.wait_drop());
2224 : }
2225 : tracing::info!(
2226 : total = js.len(),
2227 : "starting to wait for the gc'd layers to be dropped"
2228 : );
2229 : while let Some(res) = js.join_next().await {
2230 : res.expect("wait_drop should not panic");
2231 : }
2232 : }
2233 :
2234 : let timeline = tenant.get_timeline(timeline_id, false).ok();
2235 0 : let rtc = timeline.as_ref().map(|x| &x.remote_client);
2236 :
2237 : if let Some(rtc) = rtc {
2238 : // layer drops schedule actions on remote timeline client to actually do the
2239 : // deletions; don't care about the shutdown error, just exit fast
2240 : drop(rtc.wait_completion().await);
2241 : }
2242 : }
2243 :
2244 0 : result.map_err(|e| match e {
2245 0 : GcError::TenantCancelled | GcError::TimelineCancelled => ApiError::ShuttingDown,
2246 : GcError::TimelineNotFound => {
2247 0 : ApiError::NotFound(anyhow::anyhow!("Timeline not found").into())
2248 : }
2249 0 : other => ApiError::InternalServerError(anyhow::anyhow!(other)),
2250 0 : })
2251 : }
2252 : }
2253 :
2254 0 : #[derive(Debug, thiserror::Error)]
2255 : pub(crate) enum GetTenantError {
2256 : /// NotFound is a TenantId rather than TenantShardId, because this error type is used from
2257 : /// getters that use a TenantId and a ShardSelector, not just getters that target a specific shard.
2258 : #[error("Tenant {0} not found")]
2259 : NotFound(TenantId),
2260 :
2261 : #[error("Tenant {0} is not active")]
2262 : NotActive(TenantShardId),
2263 :
2264 : // Initializing or shutting down: cannot authoritatively say whether we have this tenant
2265 : #[error("Tenant map is not available: {0}")]
2266 : MapState(#[from] TenantMapError),
2267 : }
2268 :
2269 0 : #[derive(thiserror::Error, Debug)]
2270 : pub(crate) enum GetActiveTenantError {
2271 : /// We may time out either while TenantSlot is InProgress, or while the Tenant
2272 : /// is in a non-Active state
2273 : #[error(
2274 : "Timed out waiting {wait_time:?} for tenant active state. Latest state: {latest_state:?}"
2275 : )]
2276 : WaitForActiveTimeout {
2277 : latest_state: Option<TenantState>,
2278 : wait_time: Duration,
2279 : },
2280 :
2281 : /// The TenantSlot is absent, or in secondary mode
2282 : #[error(transparent)]
2283 : NotFound(#[from] GetTenantError),
2284 :
2285 : /// Cancellation token fired while we were waiting
2286 : #[error("cancelled")]
2287 : Cancelled,
2288 :
2289 : /// Tenant exists, but is in a state that cannot become active (e.g. Stopping, Broken)
2290 : #[error("will not become active. Current state: {0}")]
2291 : WillNotBecomeActive(TenantState),
2292 :
2293 : /// Broken is logically a subset of WillNotBecomeActive, but a distinct error is useful as
2294 : /// WillNotBecomeActive is a permitted error under some circumstances, whereas broken should
2295 : /// never happen.
2296 : #[error("Tenant is broken: {0}")]
2297 : Broken(String),
2298 :
2299 : #[error("reconnect to switch tenant id")]
2300 : SwitchedTenant,
2301 : }
2302 :
2303 0 : #[derive(Debug, thiserror::Error)]
2304 : pub(crate) enum DeleteTimelineError {
2305 : #[error("Tenant {0}")]
2306 : Tenant(#[from] GetTenantError),
2307 :
2308 : #[error("Timeline {0}")]
2309 : Timeline(#[from] crate::tenant::DeleteTimelineError),
2310 : }
2311 :
2312 0 : #[derive(Debug, thiserror::Error)]
2313 : pub(crate) enum TenantStateError {
2314 : #[error("Tenant {0} is stopping")]
2315 : IsStopping(TenantShardId),
2316 : #[error(transparent)]
2317 : SlotError(#[from] TenantSlotError),
2318 : #[error(transparent)]
2319 : SlotUpsertError(#[from] TenantSlotUpsertError),
2320 : #[error(transparent)]
2321 : Other(#[from] anyhow::Error),
2322 : }
2323 :
2324 0 : #[derive(Debug, thiserror::Error)]
2325 : pub(crate) enum TenantMapListError {
2326 : #[error("tenant map is still initiailizing")]
2327 : Initializing,
2328 : }
2329 :
2330 0 : #[derive(Debug, thiserror::Error)]
2331 : pub(crate) enum TenantMapInsertError {
2332 : #[error(transparent)]
2333 : SlotError(#[from] TenantSlotError),
2334 : #[error(transparent)]
2335 : SlotUpsertError(#[from] TenantSlotUpsertError),
2336 : #[error(transparent)]
2337 : Other(#[from] anyhow::Error),
2338 : }
2339 :
2340 : /// Superset of TenantMapError: issues that can occur when acquiring a slot
2341 : /// for a particular tenant ID.
2342 0 : #[derive(Debug, thiserror::Error)]
2343 : pub(crate) enum TenantSlotError {
2344 : /// When acquiring a slot with the expectation that the tenant already exists.
2345 : #[error("Tenant {0} not found")]
2346 : NotFound(TenantShardId),
2347 :
2348 : // Tried to read a slot that is currently being mutated by another administrative
2349 : // operation.
2350 : #[error("tenant has a state change in progress, try again later")]
2351 : InProgress,
2352 :
2353 : #[error(transparent)]
2354 : MapState(#[from] TenantMapError),
2355 : }
2356 :
2357 : /// Superset of TenantMapError: issues that can occur when using a SlotGuard
2358 : /// to insert a new value.
2359 0 : #[derive(thiserror::Error)]
2360 : pub(crate) enum TenantSlotUpsertError {
2361 : /// An error where the slot is in an unexpected state, indicating a code bug
2362 : #[error("Internal error updating Tenant")]
2363 : InternalError(Cow<'static, str>),
2364 :
2365 : #[error(transparent)]
2366 : MapState(TenantMapError),
2367 :
2368 : // If we encounter TenantManager shutdown during upsert, we must carry the Completion
2369 : // from the SlotGuard, so that the caller can hold it while they clean up: otherwise
2370 : // TenantManager shutdown might race ahead before we're done cleaning up any Tenant that
2371 : // was protected by the SlotGuard.
2372 : #[error("Shutting down")]
2373 : ShuttingDown((TenantSlot, utils::completion::Completion)),
2374 : }
2375 :
2376 : impl std::fmt::Debug for TenantSlotUpsertError {
2377 0 : fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
2378 0 : match self {
2379 0 : Self::InternalError(reason) => write!(f, "Internal Error {reason}"),
2380 0 : Self::MapState(map_error) => write!(f, "Tenant map state: {map_error:?}"),
2381 0 : Self::ShuttingDown(_completion) => write!(f, "Tenant map shutting down"),
2382 : }
2383 0 : }
2384 : }
2385 :
2386 0 : #[derive(Debug, thiserror::Error)]
2387 : enum TenantSlotDropError {
2388 : /// It is only legal to drop a TenantSlot if its contents are fully shut down
2389 : #[error("Tenant was not shut down")]
2390 : NotShutdown,
2391 : }
2392 :
2393 : /// Errors that can happen any time we are walking the tenant map to try and acquire
2394 : /// the TenantSlot for a particular tenant.
2395 0 : #[derive(Debug, thiserror::Error)]
2396 : pub(crate) enum TenantMapError {
2397 : // Tried to read while initializing
2398 : #[error("tenant map is still initializing")]
2399 : StillInitializing,
2400 :
2401 : // Tried to read while shutting down
2402 : #[error("tenant map is shutting down")]
2403 : ShuttingDown,
2404 : }
2405 :
2406 : /// Guards a particular tenant_id's content in the TenantsMap.
2407 : ///
2408 : /// While this structure exists, the TenantsMap will contain a [`TenantSlot::InProgress`]
2409 : /// for this tenant, which acts as a marker for any operations targeting
2410 : /// this tenant to retry later, or wait for the InProgress state to end.
2411 : ///
2412 : /// This structure enforces the important invariant that we do not have overlapping
2413 : /// tasks that will try use local storage for a the same tenant ID: we enforce that
2414 : /// the previous contents of a slot have been shut down before the slot can be
2415 : /// left empty or used for something else
2416 : ///
2417 : /// Holders of a SlotGuard should explicitly dispose of it, using either `upsert`
2418 : /// to provide a new value, or `revert` to put the slot back into its initial
2419 : /// state. If the SlotGuard is dropped without calling either of these, then
2420 : /// we will leave the slot empty if our `old_value` is already shut down, else
2421 : /// we will replace the slot with `old_value` (equivalent to doing a revert).
2422 : ///
2423 : /// The `old_value` may be dropped before the SlotGuard is dropped, by calling
2424 : /// `drop_old_value`. It is an error to call this without shutting down
2425 : /// the conents of `old_value`.
2426 : pub(crate) struct SlotGuard {
2427 : tenant_shard_id: TenantShardId,
2428 : old_value: Option<TenantSlot>,
2429 : upserted: bool,
2430 :
2431 : /// [`TenantSlot::InProgress`] carries the corresponding Barrier: it will
2432 : /// release any waiters as soon as this SlotGuard is dropped.
2433 : completion: utils::completion::Completion,
2434 : }
2435 :
2436 : impl SlotGuard {
2437 2 : fn new(
2438 2 : tenant_shard_id: TenantShardId,
2439 2 : old_value: Option<TenantSlot>,
2440 2 : completion: utils::completion::Completion,
2441 2 : ) -> Self {
2442 2 : Self {
2443 2 : tenant_shard_id,
2444 2 : old_value,
2445 2 : upserted: false,
2446 2 : completion,
2447 2 : }
2448 2 : }
2449 :
2450 : /// Get any value that was present in the slot before we acquired ownership
2451 : /// of it: in state transitions, this will be the old state.
2452 : ///
2453 : // FIXME: get_ prefix
2454 : // FIXME: this should be .as_ref() -- unsure why no clippy
2455 2 : fn get_old_value(&self) -> &Option<TenantSlot> {
2456 2 : &self.old_value
2457 2 : }
2458 :
2459 : /// Emplace a new value in the slot. This consumes the guard, and after
2460 : /// returning, the slot is no longer protected from concurrent changes.
2461 0 : fn upsert(mut self, new_value: TenantSlot) -> Result<(), TenantSlotUpsertError> {
2462 0 : if !self.old_value_is_shutdown() {
2463 : // This is a bug: callers should never try to drop an old value without
2464 : // shutting it down
2465 0 : return Err(TenantSlotUpsertError::InternalError(
2466 0 : "Old TenantSlot value not shut down".into(),
2467 0 : ));
2468 0 : }
2469 :
2470 0 : let replaced = {
2471 0 : let mut locked = TENANTS.write().unwrap();
2472 0 :
2473 0 : if let TenantSlot::InProgress(_) = new_value {
2474 : // It is never expected to try and upsert InProgress via this path: it should
2475 : // only be written via the tenant_map_acquire_slot path. If we hit this it's a bug.
2476 0 : return Err(TenantSlotUpsertError::InternalError(
2477 0 : "Attempt to upsert an InProgress state".into(),
2478 0 : ));
2479 0 : }
2480 :
2481 0 : let m = match &mut *locked {
2482 : TenantsMap::Initializing => {
2483 0 : return Err(TenantSlotUpsertError::MapState(
2484 0 : TenantMapError::StillInitializing,
2485 0 : ))
2486 : }
2487 : TenantsMap::ShuttingDown(_) => {
2488 0 : return Err(TenantSlotUpsertError::ShuttingDown((
2489 0 : new_value,
2490 0 : self.completion.clone(),
2491 0 : )));
2492 : }
2493 0 : TenantsMap::Open(m) => m,
2494 0 : };
2495 0 :
2496 0 : METRICS.slot_inserted(&new_value);
2497 0 :
2498 0 : let replaced = m.insert(self.tenant_shard_id, new_value);
2499 0 : self.upserted = true;
2500 0 : if let Some(replaced) = replaced.as_ref() {
2501 0 : METRICS.slot_removed(replaced);
2502 0 : }
2503 :
2504 0 : replaced
2505 : };
2506 :
2507 : // Sanity check: on an upsert we should always be replacing an InProgress marker
2508 0 : match replaced {
2509 : Some(TenantSlot::InProgress(_)) => {
2510 : // Expected case: we find our InProgress in the map: nothing should have
2511 : // replaced it because the code that acquires slots will not grant another
2512 : // one for the same TenantId.
2513 0 : Ok(())
2514 : }
2515 : None => {
2516 0 : METRICS.unexpected_errors.inc();
2517 0 : error!(
2518 : tenant_shard_id = %self.tenant_shard_id,
2519 0 : "Missing InProgress marker during tenant upsert, this is a bug."
2520 : );
2521 0 : Err(TenantSlotUpsertError::InternalError(
2522 0 : "Missing InProgress marker during tenant upsert".into(),
2523 0 : ))
2524 : }
2525 0 : Some(slot) => {
2526 0 : METRICS.unexpected_errors.inc();
2527 0 : error!(tenant_shard_id=%self.tenant_shard_id, "Unexpected contents of TenantSlot during upsert, this is a bug. Contents: {:?}", slot);
2528 0 : Err(TenantSlotUpsertError::InternalError(
2529 0 : "Unexpected contents of TenantSlot".into(),
2530 0 : ))
2531 : }
2532 : }
2533 0 : }
2534 :
2535 : /// Replace the InProgress slot with whatever was in the guard when we started
2536 0 : fn revert(mut self) {
2537 0 : if let Some(value) = self.old_value.take() {
2538 0 : match self.upsert(value) {
2539 0 : Err(TenantSlotUpsertError::InternalError(_)) => {
2540 0 : // We already logged the error, nothing else we can do.
2541 0 : }
2542 : Err(
2543 : TenantSlotUpsertError::MapState(_) | TenantSlotUpsertError::ShuttingDown(_),
2544 0 : ) => {
2545 0 : // If the map is shutting down, we need not replace anything
2546 0 : }
2547 0 : Ok(()) => {}
2548 : }
2549 0 : }
2550 0 : }
2551 :
2552 : /// We may never drop our old value until it is cleanly shut down: otherwise we might leave
2553 : /// rogue background tasks that would write to the local tenant directory that this guard
2554 : /// is responsible for protecting
2555 2 : fn old_value_is_shutdown(&self) -> bool {
2556 2 : match self.old_value.as_ref() {
2557 2 : Some(TenantSlot::Attached(tenant)) => tenant.gate.close_complete(),
2558 0 : Some(TenantSlot::Secondary(secondary_tenant)) => secondary_tenant.gate.close_complete(),
2559 : Some(TenantSlot::InProgress(_)) => {
2560 : // A SlotGuard cannot be constructed for a slot that was already InProgress
2561 0 : unreachable!()
2562 : }
2563 0 : None => true,
2564 : }
2565 2 : }
2566 :
2567 : /// The guard holder is done with the old value of the slot: they are obliged to already
2568 : /// shut it down before we reach this point.
2569 2 : fn drop_old_value(&mut self) -> Result<(), TenantSlotDropError> {
2570 2 : if !self.old_value_is_shutdown() {
2571 0 : Err(TenantSlotDropError::NotShutdown)
2572 : } else {
2573 2 : self.old_value.take();
2574 2 : Ok(())
2575 : }
2576 2 : }
2577 : }
2578 :
2579 : impl Drop for SlotGuard {
2580 2 : fn drop(&mut self) {
2581 2 : if self.upserted {
2582 0 : return;
2583 2 : }
2584 2 : // Our old value is already shutdown, or it never existed: it is safe
2585 2 : // for us to fully release the TenantSlot back into an empty state
2586 2 :
2587 2 : let mut locked = TENANTS.write().unwrap();
2588 :
2589 2 : let m = match &mut *locked {
2590 : TenantsMap::Initializing => {
2591 : // There is no map, this should never happen.
2592 2 : return;
2593 : }
2594 : TenantsMap::ShuttingDown(_) => {
2595 : // When we transition to shutdown, InProgress elements are removed
2596 : // from the map, so we do not need to clean up our Inprogress marker.
2597 : // See [`shutdown_all_tenants0`]
2598 0 : return;
2599 : }
2600 0 : TenantsMap::Open(m) => m,
2601 : };
2602 :
2603 : use std::collections::btree_map::Entry;
2604 0 : match m.entry(self.tenant_shard_id) {
2605 0 : Entry::Occupied(mut entry) => {
2606 0 : if !matches!(entry.get(), TenantSlot::InProgress(_)) {
2607 0 : METRICS.unexpected_errors.inc();
2608 0 : error!(tenant_shard_id=%self.tenant_shard_id, "Unexpected contents of TenantSlot during drop, this is a bug. Contents: {:?}", entry.get());
2609 0 : }
2610 :
2611 0 : if self.old_value_is_shutdown() {
2612 0 : METRICS.slot_removed(entry.get());
2613 0 : entry.remove();
2614 0 : } else {
2615 0 : let inserting = self.old_value.take().unwrap();
2616 0 : METRICS.slot_inserted(&inserting);
2617 0 : let replaced = entry.insert(inserting);
2618 0 : METRICS.slot_removed(&replaced);
2619 0 : }
2620 : }
2621 : Entry::Vacant(_) => {
2622 0 : METRICS.unexpected_errors.inc();
2623 0 : error!(
2624 : tenant_shard_id = %self.tenant_shard_id,
2625 0 : "Missing InProgress marker during SlotGuard drop, this is a bug."
2626 : );
2627 : }
2628 : }
2629 2 : }
2630 : }
2631 :
2632 : enum TenantSlotPeekMode {
2633 : /// In Read mode, peek will be permitted to see the slots even if the pageserver is shutting down
2634 : Read,
2635 : /// In Write mode, trying to peek at a slot while the pageserver is shutting down is an error
2636 : Write,
2637 : }
2638 :
2639 0 : fn tenant_map_peek_slot<'a>(
2640 0 : tenants: &'a std::sync::RwLockReadGuard<'a, TenantsMap>,
2641 0 : tenant_shard_id: &TenantShardId,
2642 0 : mode: TenantSlotPeekMode,
2643 0 : ) -> Result<Option<&'a TenantSlot>, TenantMapError> {
2644 0 : match tenants.deref() {
2645 0 : TenantsMap::Initializing => Err(TenantMapError::StillInitializing),
2646 0 : TenantsMap::ShuttingDown(m) => match mode {
2647 : TenantSlotPeekMode::Read => Ok(Some(
2648 : // When reading in ShuttingDown state, we must translate None results
2649 : // into a ShuttingDown error, because absence of a tenant shard ID in the map
2650 : // isn't a reliable indicator of the tenant being gone: it might have been
2651 : // InProgress when shutdown started, and cleaned up from that state such
2652 : // that it's now no longer in the map. Callers will have to wait until
2653 : // we next start up to get a proper answer. This avoids incorrect 404 API responses.
2654 0 : m.get(tenant_shard_id).ok_or(TenantMapError::ShuttingDown)?,
2655 : )),
2656 0 : TenantSlotPeekMode::Write => Err(TenantMapError::ShuttingDown),
2657 : },
2658 0 : TenantsMap::Open(m) => Ok(m.get(tenant_shard_id)),
2659 : }
2660 0 : }
2661 :
2662 : enum TenantSlotAcquireMode {
2663 : /// Acquire the slot irrespective of current state, or whether it already exists
2664 : Any,
2665 : /// Return an error if trying to acquire a slot and it doesn't already exist
2666 : MustExist,
2667 : }
2668 :
2669 0 : fn tenant_map_acquire_slot(
2670 0 : tenant_shard_id: &TenantShardId,
2671 0 : mode: TenantSlotAcquireMode,
2672 0 : ) -> Result<SlotGuard, TenantSlotError> {
2673 0 : tenant_map_acquire_slot_impl(tenant_shard_id, &TENANTS, mode)
2674 0 : }
2675 :
2676 2 : fn tenant_map_acquire_slot_impl(
2677 2 : tenant_shard_id: &TenantShardId,
2678 2 : tenants: &std::sync::RwLock<TenantsMap>,
2679 2 : mode: TenantSlotAcquireMode,
2680 2 : ) -> Result<SlotGuard, TenantSlotError> {
2681 : use TenantSlotAcquireMode::*;
2682 2 : METRICS.tenant_slot_writes.inc();
2683 2 :
2684 2 : let mut locked = tenants.write().unwrap();
2685 2 : let span = tracing::info_span!("acquire_slot", tenant_id=%tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug());
2686 2 : let _guard = span.enter();
2687 :
2688 2 : let m = match &mut *locked {
2689 0 : TenantsMap::Initializing => return Err(TenantMapError::StillInitializing.into()),
2690 0 : TenantsMap::ShuttingDown(_) => return Err(TenantMapError::ShuttingDown.into()),
2691 2 : TenantsMap::Open(m) => m,
2692 : };
2693 :
2694 : use std::collections::btree_map::Entry;
2695 :
2696 2 : let entry = m.entry(*tenant_shard_id);
2697 2 :
2698 2 : match entry {
2699 0 : Entry::Vacant(v) => match mode {
2700 : MustExist => {
2701 0 : tracing::debug!("Vacant && MustExist: return NotFound");
2702 0 : Err(TenantSlotError::NotFound(*tenant_shard_id))
2703 : }
2704 : _ => {
2705 0 : let (completion, barrier) = utils::completion::channel();
2706 0 : let inserting = TenantSlot::InProgress(barrier);
2707 0 : METRICS.slot_inserted(&inserting);
2708 0 : v.insert(inserting);
2709 0 : tracing::debug!("Vacant, inserted InProgress");
2710 0 : Ok(SlotGuard::new(*tenant_shard_id, None, completion))
2711 : }
2712 : },
2713 2 : Entry::Occupied(mut o) => {
2714 2 : // Apply mode-driven checks
2715 2 : match (o.get(), mode) {
2716 : (TenantSlot::InProgress(_), _) => {
2717 0 : tracing::debug!("Occupied, failing for InProgress");
2718 0 : Err(TenantSlotError::InProgress)
2719 : }
2720 : _ => {
2721 : // Happy case: the slot was not in any state that violated our mode
2722 2 : let (completion, barrier) = utils::completion::channel();
2723 2 : let in_progress = TenantSlot::InProgress(barrier);
2724 2 : METRICS.slot_inserted(&in_progress);
2725 2 : let old_value = o.insert(in_progress);
2726 2 : METRICS.slot_removed(&old_value);
2727 2 : tracing::debug!("Occupied, replaced with InProgress");
2728 2 : Ok(SlotGuard::new(
2729 2 : *tenant_shard_id,
2730 2 : Some(old_value),
2731 2 : completion,
2732 2 : ))
2733 : }
2734 : }
2735 : }
2736 : }
2737 2 : }
2738 :
2739 : /// Stops and removes the tenant from memory, if it's not [`TenantState::Stopping`] already, bails otherwise.
2740 : /// Allows to remove other tenant resources manually, via `tenant_cleanup`.
2741 : /// If the cleanup fails, tenant will stay in memory in [`TenantState::Broken`] state, and another removal
2742 : /// operation would be needed to remove it.
2743 2 : async fn remove_tenant_from_memory<V, F>(
2744 2 : tenants: &std::sync::RwLock<TenantsMap>,
2745 2 : tenant_shard_id: TenantShardId,
2746 2 : tenant_cleanup: F,
2747 2 : ) -> Result<V, TenantStateError>
2748 2 : where
2749 2 : F: std::future::Future<Output = anyhow::Result<V>>,
2750 2 : {
2751 2 : let mut slot_guard =
2752 2 : tenant_map_acquire_slot_impl(&tenant_shard_id, tenants, TenantSlotAcquireMode::MustExist)?;
2753 :
2754 : // allow pageserver shutdown to await for our completion
2755 2 : let (_guard, progress) = completion::channel();
2756 :
2757 : // The SlotGuard allows us to manipulate the Tenant object without fear of some
2758 : // concurrent API request doing something else for the same tenant ID.
2759 2 : let attached_tenant = match slot_guard.get_old_value() {
2760 2 : Some(TenantSlot::Attached(tenant)) => {
2761 2 : // whenever we remove a tenant from memory, we don't want to flush and wait for upload
2762 2 : let shutdown_mode = ShutdownMode::Hard;
2763 2 :
2764 2 : // shutdown is sure to transition tenant to stopping, and wait for all tasks to complete, so
2765 2 : // that we can continue safely to cleanup.
2766 2 : match tenant.shutdown(progress, shutdown_mode).await {
2767 2 : Ok(()) => {}
2768 0 : Err(_other) => {
2769 0 : // if pageserver shutdown or other detach/ignore is already ongoing, we don't want to
2770 0 : // wait for it but return an error right away because these are distinct requests.
2771 0 : slot_guard.revert();
2772 0 : return Err(TenantStateError::IsStopping(tenant_shard_id));
2773 : }
2774 : }
2775 2 : Some(tenant)
2776 : }
2777 0 : Some(TenantSlot::Secondary(secondary_state)) => {
2778 0 : tracing::info!("Shutting down in secondary mode");
2779 0 : secondary_state.shutdown().await;
2780 0 : None
2781 : }
2782 : Some(TenantSlot::InProgress(_)) => {
2783 : // Acquiring a slot guarantees its old value was not InProgress
2784 0 : unreachable!();
2785 : }
2786 0 : None => None,
2787 : };
2788 :
2789 2 : match tenant_cleanup
2790 2 : .await
2791 2 : .with_context(|| format!("Failed to run cleanup for tenant {tenant_shard_id}"))
2792 : {
2793 2 : Ok(hook_value) => {
2794 2 : // Success: drop the old TenantSlot::Attached.
2795 2 : slot_guard
2796 2 : .drop_old_value()
2797 2 : .expect("We just called shutdown");
2798 2 :
2799 2 : Ok(hook_value)
2800 : }
2801 0 : Err(e) => {
2802 : // If we had a Tenant, set it to Broken and put it back in the TenantsMap
2803 0 : if let Some(attached_tenant) = attached_tenant {
2804 0 : attached_tenant.set_broken(e.to_string()).await;
2805 0 : }
2806 : // Leave the broken tenant in the map
2807 0 : slot_guard.revert();
2808 0 :
2809 0 : Err(TenantStateError::Other(e))
2810 : }
2811 : }
2812 2 : }
2813 :
2814 : use {
2815 : crate::tenant::gc_result::GcResult, pageserver_api::models::TimelineGcRequest,
2816 : utils::http::error::ApiError,
2817 : };
2818 :
2819 : #[cfg(test)]
2820 : mod tests {
2821 : use std::collections::BTreeMap;
2822 : use std::sync::Arc;
2823 : use tracing::Instrument;
2824 :
2825 : use crate::tenant::mgr::TenantSlot;
2826 :
2827 : use super::{super::harness::TenantHarness, TenantsMap};
2828 :
2829 : #[tokio::test(start_paused = true)]
2830 2 : async fn shutdown_awaits_in_progress_tenant() {
2831 2 : // Test that if an InProgress tenant is in the map during shutdown, the shutdown will gracefully
2832 2 : // wait for it to complete before proceeding.
2833 2 :
2834 2 : let h = TenantHarness::create("shutdown_awaits_in_progress_tenant")
2835 2 : .await
2836 2 : .unwrap();
2837 20 : let (t, _ctx) = h.load().await;
2838 2 :
2839 2 : // harness loads it to active, which is forced and nothing is running on the tenant
2840 2 :
2841 2 : let id = t.tenant_shard_id();
2842 2 :
2843 2 : // tenant harness configures the logging and we cannot escape it
2844 2 : let span = h.span();
2845 2 : let _e = span.enter();
2846 2 :
2847 2 : let tenants = BTreeMap::from([(id, TenantSlot::Attached(t.clone()))]);
2848 2 : let tenants = Arc::new(std::sync::RwLock::new(TenantsMap::Open(tenants)));
2849 2 :
2850 2 : // Invoke remove_tenant_from_memory with a cleanup hook that blocks until we manually
2851 2 : // permit it to proceed: that will stick the tenant in InProgress
2852 2 :
2853 2 : let (until_cleanup_completed, can_complete_cleanup) = utils::completion::channel();
2854 2 : let (until_cleanup_started, cleanup_started) = utils::completion::channel();
2855 2 : let mut remove_tenant_from_memory_task = {
2856 2 : let jh = tokio::spawn({
2857 2 : let tenants = tenants.clone();
2858 2 : async move {
2859 2 : let cleanup = async move {
2860 2 : drop(until_cleanup_started);
2861 2 : can_complete_cleanup.wait().await;
2862 2 : anyhow::Ok(())
2863 2 : };
2864 2 : super::remove_tenant_from_memory(&tenants, id, cleanup).await
2865 2 : }
2866 2 : .instrument(h.span())
2867 2 : });
2868 2 :
2869 2 : // now the long cleanup should be in place, with the stopping state
2870 2 : cleanup_started.wait().await;
2871 2 : jh
2872 2 : };
2873 2 :
2874 2 : let mut shutdown_task = {
2875 2 : let (until_shutdown_started, shutdown_started) = utils::completion::channel();
2876 2 :
2877 2 : let shutdown_task = tokio::spawn(async move {
2878 2 : drop(until_shutdown_started);
2879 4 : super::shutdown_all_tenants0(&tenants).await;
2880 2 : });
2881 2 :
2882 2 : shutdown_started.wait().await;
2883 2 : shutdown_task
2884 2 : };
2885 2 :
2886 2 : let long_time = std::time::Duration::from_secs(15);
2887 2 : tokio::select! {
2888 2 : _ = &mut shutdown_task => unreachable!("shutdown should block on remove_tenant_from_memory completing"),
2889 2 : _ = &mut remove_tenant_from_memory_task => unreachable!("remove_tenant_from_memory_task should not complete until explicitly unblocked"),
2890 2 : _ = tokio::time::sleep(long_time) => {},
2891 2 : }
2892 2 :
2893 2 : drop(until_cleanup_completed);
2894 2 :
2895 2 : // Now that we allow it to proceed, shutdown should complete immediately
2896 2 : remove_tenant_from_memory_task.await.unwrap().unwrap();
2897 2 : shutdown_task.await.unwrap();
2898 2 : }
2899 : }
|