Line data Source code
1 : mod downloader;
2 : pub mod heatmap;
3 : mod heatmap_uploader;
4 : mod scheduler;
5 :
6 : use std::{sync::Arc, time::SystemTime};
7 :
8 : use crate::{
9 : context::RequestContext,
10 : disk_usage_eviction_task::DiskUsageEvictionInfo,
11 : metrics::SECONDARY_HEATMAP_TOTAL_SIZE,
12 : task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
13 : };
14 :
15 : use self::{
16 : downloader::{downloader_task, SecondaryDetail},
17 : heatmap_uploader::heatmap_uploader_task,
18 : };
19 :
20 : use super::{
21 : config::{SecondaryLocationConfig, TenantConfOpt},
22 : mgr::TenantManager,
23 : span::debug_assert_current_span_has_tenant_id,
24 : storage_layer::LayerName,
25 : };
26 :
27 : use crate::metrics::SECONDARY_RESIDENT_PHYSICAL_SIZE;
28 : use metrics::UIntGauge;
29 : use pageserver_api::{
30 : models,
31 : shard::{ShardIdentity, TenantShardId},
32 : };
33 : use remote_storage::GenericRemoteStorage;
34 :
35 : use tokio::task::JoinHandle;
36 : use tokio_util::sync::CancellationToken;
37 : use tracing::instrument;
38 : use utils::{completion::Barrier, id::TimelineId, sync::gate::Gate};
39 :
40 : enum DownloadCommand {
41 : Download(TenantShardId),
42 : }
43 : enum UploadCommand {
44 : Upload(TenantShardId),
45 : }
46 :
47 : impl UploadCommand {
48 0 : fn get_tenant_shard_id(&self) -> &TenantShardId {
49 0 : match self {
50 0 : Self::Upload(id) => id,
51 0 : }
52 0 : }
53 : }
54 :
55 : impl DownloadCommand {
56 0 : fn get_tenant_shard_id(&self) -> &TenantShardId {
57 0 : match self {
58 0 : Self::Download(id) => id,
59 0 : }
60 0 : }
61 : }
62 :
63 : struct CommandRequest<T> {
64 : payload: T,
65 : response_tx: tokio::sync::oneshot::Sender<CommandResponse>,
66 : }
67 :
68 : struct CommandResponse {
69 : result: anyhow::Result<()>,
70 : }
71 :
72 : // Whereas [`Tenant`] represents an attached tenant, this type represents the work
73 : // we do for secondary tenant locations: where we are not serving clients or
74 : // ingesting WAL, but we are maintaining a warm cache of layer files.
75 : //
76 : // This type is all about the _download_ path for secondary mode. The upload path
77 : // runs separately (see [`heatmap_uploader`]) while a regular attached `Tenant` exists.
78 : //
79 : // This structure coordinates TenantManager and SecondaryDownloader,
80 : // so that the downloader can indicate which tenants it is currently
81 : // operating on, and the manager can indicate when a particular
82 : // secondary tenant should cancel any work in flight.
83 : #[derive(Debug)]
84 : pub(crate) struct SecondaryTenant {
85 : /// Carrying a tenant shard ID simplifies callers such as the downloader
86 : /// which need to organize many of these objects by ID.
87 : tenant_shard_id: TenantShardId,
88 :
89 : /// Cancellation token indicates to SecondaryDownloader that it should stop doing
90 : /// any work for this tenant at the next opportunity.
91 : pub(crate) cancel: CancellationToken,
92 :
93 : pub(crate) gate: Gate,
94 :
95 : // Secondary mode does not need the full shard identity or the TenantConfOpt. However,
96 : // storing these enables us to report our full LocationConf, enabling convenient reconciliation
97 : // by the control plane (see [`Self::get_location_conf`])
98 : shard_identity: ShardIdentity,
99 : tenant_conf: std::sync::Mutex<TenantConfOpt>,
100 :
101 : // Internal state used by the Downloader.
102 : detail: std::sync::Mutex<SecondaryDetail>,
103 :
104 : // Public state indicating overall progress of downloads relative to the last heatmap seen
105 : pub(crate) progress: std::sync::Mutex<models::SecondaryProgress>,
106 :
107 : // Sum of layer sizes on local disk
108 : pub(super) resident_size_metric: UIntGauge,
109 :
110 : // Sum of layer sizes in the most recently downloaded heatmap
111 : pub(super) heatmap_total_size_metric: UIntGauge,
112 : }
113 :
114 : impl Drop for SecondaryTenant {
115 0 : fn drop(&mut self) {
116 0 : let tenant_id = self.tenant_shard_id.tenant_id.to_string();
117 0 : let shard_id = format!("{}", self.tenant_shard_id.shard_slug());
118 0 : let _ = SECONDARY_RESIDENT_PHYSICAL_SIZE.remove_label_values(&[&tenant_id, &shard_id]);
119 0 : let _ = SECONDARY_HEATMAP_TOTAL_SIZE.remove_label_values(&[&tenant_id, &shard_id]);
120 0 : }
121 : }
122 :
123 : impl SecondaryTenant {
124 0 : pub(crate) fn new(
125 0 : tenant_shard_id: TenantShardId,
126 0 : shard_identity: ShardIdentity,
127 0 : tenant_conf: TenantConfOpt,
128 0 : config: &SecondaryLocationConfig,
129 0 : ) -> Arc<Self> {
130 0 : let tenant_id = tenant_shard_id.tenant_id.to_string();
131 0 : let shard_id = format!("{}", tenant_shard_id.shard_slug());
132 0 : let resident_size_metric = SECONDARY_RESIDENT_PHYSICAL_SIZE
133 0 : .get_metric_with_label_values(&[&tenant_id, &shard_id])
134 0 : .unwrap();
135 0 :
136 0 : let heatmap_total_size_metric = SECONDARY_HEATMAP_TOTAL_SIZE
137 0 : .get_metric_with_label_values(&[&tenant_id, &shard_id])
138 0 : .unwrap();
139 0 :
140 0 : Arc::new(Self {
141 0 : tenant_shard_id,
142 0 : // todo: shall we make this a descendent of the
143 0 : // main cancellation token, or is it sufficient that
144 0 : // on shutdown we walk the tenants and fire their
145 0 : // individual cancellations?
146 0 : cancel: CancellationToken::new(),
147 0 : gate: Gate::default(),
148 0 :
149 0 : shard_identity,
150 0 : tenant_conf: std::sync::Mutex::new(tenant_conf),
151 0 :
152 0 : detail: std::sync::Mutex::new(SecondaryDetail::new(config.clone())),
153 0 :
154 0 : progress: std::sync::Mutex::default(),
155 0 :
156 0 : resident_size_metric,
157 0 : heatmap_total_size_metric,
158 0 : })
159 0 : }
160 :
161 0 : pub(crate) fn tenant_shard_id(&self) -> TenantShardId {
162 0 : self.tenant_shard_id
163 0 : }
164 :
165 0 : pub(crate) async fn shutdown(&self) {
166 0 : self.cancel.cancel();
167 0 :
168 0 : // Wait for any secondary downloader work to complete
169 0 : self.gate.close().await;
170 0 : }
171 :
172 0 : pub(crate) fn set_config(&self, config: &SecondaryLocationConfig) {
173 0 : self.detail.lock().unwrap().config = config.clone();
174 0 : }
175 :
176 0 : pub(crate) fn set_tenant_conf(&self, config: &TenantConfOpt) {
177 0 : *(self.tenant_conf.lock().unwrap()) = config.clone();
178 0 : }
179 :
180 : /// For API access: generate a LocationConfig equivalent to the one that would be used to
181 : /// create a Tenant in the same state. Do not use this in hot paths: it's for relatively
182 : /// rare external API calls, like a reconciliation at startup.
183 0 : pub(crate) fn get_location_conf(&self) -> models::LocationConfig {
184 0 : let conf = self.detail.lock().unwrap().config.clone();
185 0 :
186 0 : let conf = models::LocationConfigSecondary { warm: conf.warm };
187 0 :
188 0 : let tenant_conf = self.tenant_conf.lock().unwrap().clone();
189 0 : models::LocationConfig {
190 0 : mode: models::LocationConfigMode::Secondary,
191 0 : generation: None,
192 0 : secondary_conf: Some(conf),
193 0 : shard_number: self.tenant_shard_id.shard_number.0,
194 0 : shard_count: self.tenant_shard_id.shard_count.literal(),
195 0 : shard_stripe_size: self.shard_identity.stripe_size.0,
196 0 : tenant_conf: tenant_conf.into(),
197 0 : }
198 0 : }
199 :
200 0 : pub(crate) fn get_tenant_shard_id(&self) -> &TenantShardId {
201 0 : &self.tenant_shard_id
202 0 : }
203 :
204 0 : pub(crate) fn get_layers_for_eviction(self: &Arc<Self>) -> (DiskUsageEvictionInfo, usize) {
205 0 : self.detail.lock().unwrap().get_layers_for_eviction(self)
206 0 : }
207 :
208 : /// Cancellation safe, but on cancellation the eviction will go through
209 0 : #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(), timeline_id=%timeline_id, name=%name))]
210 : pub(crate) async fn evict_layer(self: &Arc<Self>, timeline_id: TimelineId, name: LayerName) {
211 : debug_assert_current_span_has_tenant_id();
212 :
213 : let guard = match self.gate.enter() {
214 : Ok(g) => g,
215 : Err(_) => {
216 : tracing::debug!("Dropping layer evictions, secondary tenant shutting down",);
217 : return;
218 : }
219 : };
220 :
221 : let now = SystemTime::now();
222 : tracing::info!("Evicting secondary layer");
223 :
224 : let this = self.clone();
225 :
226 : // spawn it to be cancellation safe
227 0 : tokio::task::spawn_blocking(move || {
228 0 : let _guard = guard;
229 0 :
230 0 : // Update the timeline's state. This does not have to be synchronized with
231 0 : // the download process, because:
232 0 : // - If downloader is racing with us to remove a file (e.g. because it is
233 0 : // removed from heatmap), then our mutual .remove() operations will both
234 0 : // succeed.
235 0 : // - If downloader is racing with us to download the object (this would require
236 0 : // multiple eviction iterations to race with multiple download iterations), then
237 0 : // if we remove it from the state, the worst that happens is the downloader
238 0 : // downloads it again before re-inserting, or we delete the file but it remains
239 0 : // in the state map (in which case it will be downloaded if this secondary
240 0 : // tenant transitions to attached and tries to access it)
241 0 : //
242 0 : // The important assumption here is that the secondary timeline state does not
243 0 : // have to 100% match what is on disk, because it's a best-effort warming
244 0 : // of the cache.
245 0 : let mut detail = this.detail.lock().unwrap();
246 0 : if let Some(removed) =
247 0 : detail.evict_layer(name, &timeline_id, now, &this.resident_size_metric)
248 0 : {
249 0 : // We might race with removal of the same layer during downloads, so finding the layer we
250 0 : // were trying to remove is optional. Only issue the disk I/O to remove it if we found it.
251 0 : removed.remove_blocking();
252 0 : }
253 0 : })
254 : .await
255 : .expect("secondary eviction should not have panicked");
256 : }
257 : }
258 :
259 : /// The SecondaryController is a pseudo-rpc client for administrative control of secondary mode downloads,
260 : /// and heatmap uploads. This is not a hot data path: it's used for:
261 : /// - Live migrations, where we want to ensure a migration destination has the freshest possible
262 : /// content before trying to cut over.
263 : /// - Tests, where we want to immediately upload/download for a particular tenant.
264 : ///
265 : /// In normal operations, outside of migrations, uploads & downloads are autonomous and not driven by this interface.
266 : pub struct SecondaryController {
267 : upload_req_tx: tokio::sync::mpsc::Sender<CommandRequest<UploadCommand>>,
268 : download_req_tx: tokio::sync::mpsc::Sender<CommandRequest<DownloadCommand>>,
269 : }
270 :
271 : impl SecondaryController {
272 0 : async fn dispatch<T>(
273 0 : &self,
274 0 : queue: &tokio::sync::mpsc::Sender<CommandRequest<T>>,
275 0 : payload: T,
276 0 : ) -> anyhow::Result<()> {
277 0 : let (response_tx, response_rx) = tokio::sync::oneshot::channel();
278 0 :
279 0 : queue
280 0 : .send(CommandRequest {
281 0 : payload,
282 0 : response_tx,
283 0 : })
284 0 : .await
285 0 : .map_err(|_| anyhow::anyhow!("Receiver shut down"))?;
286 :
287 0 : let response = response_rx
288 0 : .await
289 0 : .map_err(|_| anyhow::anyhow!("Request dropped"))?;
290 :
291 0 : response.result
292 0 : }
293 :
294 0 : pub async fn upload_tenant(&self, tenant_shard_id: TenantShardId) -> anyhow::Result<()> {
295 0 : self.dispatch(&self.upload_req_tx, UploadCommand::Upload(tenant_shard_id))
296 0 : .await
297 0 : }
298 0 : pub async fn download_tenant(&self, tenant_shard_id: TenantShardId) -> anyhow::Result<()> {
299 0 : self.dispatch(
300 0 : &self.download_req_tx,
301 0 : DownloadCommand::Download(tenant_shard_id),
302 0 : )
303 0 : .await
304 0 : }
305 : }
306 :
307 : pub struct GlobalTasks {
308 : cancel: CancellationToken,
309 : uploader: JoinHandle<()>,
310 : downloader: JoinHandle<()>,
311 : }
312 :
313 : impl GlobalTasks {
314 : /// Caller is responsible for requesting shutdown via the cancellation token that was
315 : /// passed to [`spawn_tasks`].
316 : ///
317 : /// # Panics
318 : ///
319 : /// This method panics if that token is not cancelled.
320 : /// This is low-risk because we're calling this during process shutdown, so, a panic
321 : /// will be informative but not cause undue downtime.
322 0 : pub async fn wait(self) {
323 0 : let Self {
324 0 : cancel,
325 0 : uploader,
326 0 : downloader,
327 0 : } = self;
328 0 : assert!(
329 0 : cancel.is_cancelled(),
330 0 : "must cancel cancellation token, otherwise the tasks will not shut down"
331 : );
332 :
333 0 : let (uploader, downloader) = futures::future::join(uploader, downloader).await;
334 0 : uploader.expect(
335 0 : "unreachable: exit_on_panic_or_error would catch the panic and exit the process",
336 0 : );
337 0 : downloader.expect(
338 0 : "unreachable: exit_on_panic_or_error would catch the panic and exit the process",
339 0 : );
340 0 : }
341 : }
342 :
343 0 : pub fn spawn_tasks(
344 0 : tenant_manager: Arc<TenantManager>,
345 0 : remote_storage: GenericRemoteStorage,
346 0 : background_jobs_can_start: Barrier,
347 0 : cancel: CancellationToken,
348 0 : ) -> (SecondaryController, GlobalTasks) {
349 0 : let mgr_clone = tenant_manager.clone();
350 0 : let storage_clone = remote_storage.clone();
351 0 : let bg_jobs_clone = background_jobs_can_start.clone();
352 0 :
353 0 : let (download_req_tx, download_req_rx) =
354 0 : tokio::sync::mpsc::channel::<CommandRequest<DownloadCommand>>(16);
355 0 : let (upload_req_tx, upload_req_rx) =
356 0 : tokio::sync::mpsc::channel::<CommandRequest<UploadCommand>>(16);
357 0 :
358 0 : let cancel_clone = cancel.clone();
359 0 : let downloader = BACKGROUND_RUNTIME.spawn(task_mgr::exit_on_panic_or_error(
360 0 : "secondary tenant downloads",
361 0 : async move {
362 0 : downloader_task(
363 0 : mgr_clone,
364 0 : storage_clone,
365 0 : download_req_rx,
366 0 : bg_jobs_clone,
367 0 : cancel_clone,
368 0 : RequestContext::new(
369 0 : TaskKind::SecondaryDownloads,
370 0 : crate::context::DownloadBehavior::Download,
371 0 : ),
372 0 : )
373 0 : .await;
374 0 : anyhow::Ok(())
375 0 : },
376 0 : ));
377 0 :
378 0 : let cancel_clone = cancel.clone();
379 0 : let uploader = BACKGROUND_RUNTIME.spawn(task_mgr::exit_on_panic_or_error(
380 0 : "heatmap uploads",
381 0 : async move {
382 0 : heatmap_uploader_task(
383 0 : tenant_manager,
384 0 : remote_storage,
385 0 : upload_req_rx,
386 0 : background_jobs_can_start,
387 0 : cancel_clone,
388 0 : )
389 0 : .await;
390 0 : anyhow::Ok(())
391 0 : },
392 0 : ));
393 0 :
394 0 : (
395 0 : SecondaryController {
396 0 : upload_req_tx,
397 0 : download_req_tx,
398 0 : },
399 0 : GlobalTasks {
400 0 : cancel,
401 0 : uploader,
402 0 : downloader,
403 0 : },
404 0 : )
405 0 : }
|
