Line data Source code
1 : #![recursion_limit = "300"]
2 : #![deny(clippy::undocumented_unsafe_blocks)]
3 :
4 : mod auth;
5 : pub mod basebackup;
6 : pub mod config;
7 : pub mod consumption_metrics;
8 : pub mod context;
9 : pub mod control_plane_client;
10 : pub mod deletion_queue;
11 : pub mod disk_usage_eviction_task;
12 : pub mod http;
13 : pub mod import_datadir;
14 : pub use pageserver_api::keyspace;
15 : pub mod aux_file;
16 : pub mod metrics;
17 : pub mod page_cache;
18 : pub mod page_service;
19 : pub mod pgdatadir_mapping;
20 : pub mod repository;
21 : pub mod span;
22 : pub(crate) mod statvfs;
23 : pub mod task_mgr;
24 : pub mod tenant;
25 : pub mod trace;
26 : pub mod utilization;
27 : pub mod virtual_file;
28 : pub mod walingest;
29 : pub mod walrecord;
30 : pub mod walredo;
31 :
32 : use crate::task_mgr::TaskKind;
33 : use camino::Utf8Path;
34 : use deletion_queue::DeletionQueue;
35 : use tenant::mgr::TenantManager;
36 : use tracing::info;
37 :
38 : /// Current storage format version
39 : ///
40 : /// This is embedded in the header of all the layer files.
41 : /// If you make any backwards-incompatible changes to the storage
42 : /// format, bump this!
43 : /// Note that TimelineMetadata uses its own version number to track
44 : /// backwards-compatible changes to the metadata format.
45 : pub const STORAGE_FORMAT_VERSION: u16 = 3;
46 :
47 : pub const DEFAULT_PG_VERSION: u32 = 15;
48 :
49 : // Magic constants used to identify different kinds of files
50 : pub const IMAGE_FILE_MAGIC: u16 = 0x5A60;
51 : pub const DELTA_FILE_MAGIC: u16 = 0x5A61;
52 :
53 : static ZERO_PAGE: bytes::Bytes = bytes::Bytes::from_static(&[0u8; 8192]);
54 :
55 : pub use crate::metrics::preinitialize_metrics;
56 :
57 0 : #[tracing::instrument(skip_all, fields(%exit_code))]
58 : pub async fn shutdown_pageserver(
59 : tenant_manager: &TenantManager,
60 : deletion_queue: Option<DeletionQueue>,
61 : exit_code: i32,
62 : ) {
63 : use std::time::Duration;
64 : // Shut down the libpq endpoint task. This prevents new connections from
65 : // being accepted.
66 : timed(
67 : task_mgr::shutdown_tasks(Some(TaskKind::LibpqEndpointListener), None, None),
68 : "shutdown LibpqEndpointListener",
69 : Duration::from_secs(1),
70 : )
71 : .await;
72 :
73 : // Shut down all the tenants. This flushes everything to disk and kills
74 : // the checkpoint and GC tasks.
75 : timed(
76 : tenant_manager.shutdown(),
77 : "shutdown all tenants",
78 : Duration::from_secs(5),
79 : )
80 : .await;
81 :
82 : // Shut down any page service tasks: any in-progress work for particular timelines or tenants
83 : // should already have been canclled via mgr::shutdown_all_tenants
84 : timed(
85 : task_mgr::shutdown_tasks(Some(TaskKind::PageRequestHandler), None, None),
86 : "shutdown PageRequestHandlers",
87 : Duration::from_secs(1),
88 : )
89 : .await;
90 :
91 : // Best effort to persist any outstanding deletions, to avoid leaking objects
92 : if let Some(mut deletion_queue) = deletion_queue {
93 : deletion_queue.shutdown(Duration::from_secs(5)).await;
94 : }
95 :
96 : // Shut down the HTTP endpoint last, so that you can still check the server's
97 : // status while it's shutting down.
98 : // FIXME: We should probably stop accepting commands like attach/detach earlier.
99 : timed(
100 : task_mgr::shutdown_tasks(Some(TaskKind::HttpEndpointListener), None, None),
101 : "shutdown http",
102 : Duration::from_secs(1),
103 : )
104 : .await;
105 :
106 : // There should be nothing left, but let's be sure
107 : timed(
108 : task_mgr::shutdown_tasks(None, None, None),
109 : "shutdown leftovers",
110 : Duration::from_secs(1),
111 : )
112 : .await;
113 : info!("Shut down successfully completed");
114 : std::process::exit(exit_code);
115 : }
116 :
117 : /// The name of the metadata file pageserver creates per timeline.
118 : /// Full path: `tenants/<tenant_id>/timelines/<timeline_id>/metadata`.
119 : pub const METADATA_FILE_NAME: &str = "metadata";
120 :
121 : /// Per-tenant configuration file.
122 : /// Full path: `tenants/<tenant_id>/config`.
123 : pub(crate) const TENANT_CONFIG_NAME: &str = "config";
124 :
125 : /// Per-tenant configuration file.
126 : /// Full path: `tenants/<tenant_id>/config`.
127 : pub(crate) const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
128 :
129 : /// Per-tenant copy of their remote heatmap, downloaded into the local
130 : /// tenant path while in secondary mode.
131 : pub(crate) const TENANT_HEATMAP_BASENAME: &str = "heatmap-v1.json";
132 :
133 : /// A suffix used for various temporary files. Any temporary files found in the
134 : /// data directory at pageserver startup can be automatically removed.
135 : pub(crate) const TEMP_FILE_SUFFIX: &str = "___temp";
136 :
137 : /// A marker file to mark that a timeline directory was not fully initialized.
138 : /// If a timeline directory with this marker is encountered at pageserver startup,
139 : /// the timeline directory and the marker file are both removed.
140 : /// Full path: `tenants/<tenant_id>/timelines/<timeline_id>___uninit`.
141 : pub(crate) const TIMELINE_UNINIT_MARK_SUFFIX: &str = "___uninit";
142 :
143 : pub(crate) const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
144 :
145 : /// A marker file to prevent pageserver from loading a certain tenant on restart.
146 : /// Different from [`TIMELINE_UNINIT_MARK_SUFFIX`] due to semantics of the corresponding
147 : /// `ignore` management API command, that expects the ignored tenant to be properly loaded
148 : /// into pageserver's memory before being ignored.
149 : /// Full path: `tenants/<tenant_id>/___ignored_tenant`.
150 : pub const IGNORED_TENANT_FILE_NAME: &str = "___ignored_tenant";
151 :
152 8 : pub fn is_temporary(path: &Utf8Path) -> bool {
153 8 : match path.file_name() {
154 8 : Some(name) => name.ends_with(TEMP_FILE_SUFFIX),
155 0 : None => false,
156 : }
157 8 : }
158 :
159 16 : fn ends_with_suffix(path: &Utf8Path, suffix: &str) -> bool {
160 16 : match path.file_name() {
161 16 : Some(name) => name.ends_with(suffix),
162 0 : None => false,
163 : }
164 16 : }
165 :
166 : // FIXME: DO NOT ADD new query methods like this, which will have a next step of parsing timelineid
167 : // from the directory name. Instead create type "UninitMark(TimelineId)" and only parse it once
168 : // from the name.
169 :
170 8 : pub(crate) fn is_uninit_mark(path: &Utf8Path) -> bool {
171 8 : ends_with_suffix(path, TIMELINE_UNINIT_MARK_SUFFIX)
172 8 : }
173 :
174 8 : pub(crate) fn is_delete_mark(path: &Utf8Path) -> bool {
175 8 : ends_with_suffix(path, TIMELINE_DELETE_MARK_SUFFIX)
176 8 : }
177 :
178 : /// During pageserver startup, we need to order operations not to exhaust tokio worker threads by
179 : /// blocking.
180 : ///
181 : /// The instances of this value exist only during startup, otherwise `None` is provided, meaning no
182 : /// delaying is needed.
183 : #[derive(Clone)]
184 : pub struct InitializationOrder {
185 : /// Each initial tenant load task carries this until it is done loading timelines from remote storage
186 : pub initial_tenant_load_remote: Option<utils::completion::Completion>,
187 :
188 : /// Each initial tenant load task carries this until completion.
189 : pub initial_tenant_load: Option<utils::completion::Completion>,
190 :
191 : /// Barrier for when we can start any background jobs.
192 : ///
193 : /// This can be broken up later on, but right now there is just one class of a background job.
194 : pub background_jobs_can_start: utils::completion::Barrier,
195 : }
196 :
197 : /// Time the future with a warning when it exceeds a threshold.
198 4 : async fn timed<Fut: std::future::Future>(
199 4 : fut: Fut,
200 4 : name: &str,
201 4 : warn_at: std::time::Duration,
202 4 : ) -> <Fut as std::future::Future>::Output {
203 4 : let started = std::time::Instant::now();
204 4 :
205 4 : let mut fut = std::pin::pin!(fut);
206 4 :
207 4 : match tokio::time::timeout(warn_at, &mut fut).await {
208 2 : Ok(ret) => {
209 2 : tracing::info!(
210 : stage = name,
211 0 : elapsed_ms = started.elapsed().as_millis(),
212 0 : "completed"
213 : );
214 2 : ret
215 : }
216 : Err(_) => {
217 2 : tracing::info!(
218 : stage = name,
219 0 : elapsed_ms = started.elapsed().as_millis(),
220 0 : "still waiting, taking longer than expected..."
221 : );
222 :
223 2 : let ret = fut.await;
224 :
225 : // this has a global allowed_errors
226 2 : tracing::warn!(
227 : stage = name,
228 0 : elapsed_ms = started.elapsed().as_millis(),
229 0 : "completed, took longer than expected"
230 : );
231 :
232 2 : ret
233 : }
234 : }
235 4 : }
236 :
237 : #[cfg(test)]
238 : mod timed_tests {
239 : use super::timed;
240 : use std::time::Duration;
241 :
242 : #[tokio::test]
243 2 : async fn timed_completes_when_inner_future_completes() {
244 2 : // A future that completes on time should have its result returned
245 2 : let r1 = timed(
246 2 : async move {
247 2 : tokio::time::sleep(Duration::from_millis(10)).await;
248 2 : 123
249 2 : },
250 2 : "test 1",
251 2 : Duration::from_millis(50),
252 2 : )
253 2 : .await;
254 2 : assert_eq!(r1, 123);
255 2 :
256 2 : // A future that completes too slowly should also have its result returned
257 2 : let r1 = timed(
258 2 : async move {
259 6 : tokio::time::sleep(Duration::from_millis(50)).await;
260 2 : 456
261 2 : },
262 2 : "test 1",
263 2 : Duration::from_millis(10),
264 2 : )
265 4 : .await;
266 2 : assert_eq!(r1, 456);
267 2 : }
268 : }
|