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 : mut deletion_queue: 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 : deletion_queue.shutdown(Duration::from_secs(5)).await;
93 :
94 : // Shut down the HTTP endpoint last, so that you can still check the server's
95 : // status while it's shutting down.
96 : // FIXME: We should probably stop accepting commands like attach/detach earlier.
97 : timed(
98 : task_mgr::shutdown_tasks(Some(TaskKind::HttpEndpointListener), None, None),
99 : "shutdown http",
100 : Duration::from_secs(1),
101 : )
102 : .await;
103 :
104 : // There should be nothing left, but let's be sure
105 : timed(
106 : task_mgr::shutdown_tasks(None, None, None),
107 : "shutdown leftovers",
108 : Duration::from_secs(1),
109 : )
110 : .await;
111 : info!("Shut down successfully completed");
112 : std::process::exit(exit_code);
113 : }
114 :
115 : /// Per-tenant configuration file.
116 : /// Full path: `tenants/<tenant_id>/config`.
117 : pub(crate) const TENANT_CONFIG_NAME: &str = "config";
118 :
119 : /// Per-tenant configuration file.
120 : /// Full path: `tenants/<tenant_id>/config`.
121 : pub(crate) const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";
122 :
123 : /// Per-tenant copy of their remote heatmap, downloaded into the local
124 : /// tenant path while in secondary mode.
125 : pub(crate) const TENANT_HEATMAP_BASENAME: &str = "heatmap-v1.json";
126 :
127 : /// A suffix used for various temporary files. Any temporary files found in the
128 : /// data directory at pageserver startup can be automatically removed.
129 : pub(crate) const TEMP_FILE_SUFFIX: &str = "___temp";
130 :
131 : /// A marker file to mark that a timeline directory was not fully initialized.
132 : /// If a timeline directory with this marker is encountered at pageserver startup,
133 : /// the timeline directory and the marker file are both removed.
134 : /// Full path: `tenants/<tenant_id>/timelines/<timeline_id>___uninit`.
135 : pub(crate) const TIMELINE_UNINIT_MARK_SUFFIX: &str = "___uninit";
136 :
137 : pub(crate) const TIMELINE_DELETE_MARK_SUFFIX: &str = "___delete";
138 :
139 8 : pub fn is_temporary(path: &Utf8Path) -> bool {
140 8 : match path.file_name() {
141 8 : Some(name) => name.ends_with(TEMP_FILE_SUFFIX),
142 0 : None => false,
143 : }
144 8 : }
145 :
146 16 : fn ends_with_suffix(path: &Utf8Path, suffix: &str) -> bool {
147 16 : match path.file_name() {
148 16 : Some(name) => name.ends_with(suffix),
149 0 : None => false,
150 : }
151 16 : }
152 :
153 : // FIXME: DO NOT ADD new query methods like this, which will have a next step of parsing timelineid
154 : // from the directory name. Instead create type "UninitMark(TimelineId)" and only parse it once
155 : // from the name.
156 :
157 8 : pub(crate) fn is_uninit_mark(path: &Utf8Path) -> bool {
158 8 : ends_with_suffix(path, TIMELINE_UNINIT_MARK_SUFFIX)
159 8 : }
160 :
161 8 : pub(crate) fn is_delete_mark(path: &Utf8Path) -> bool {
162 8 : ends_with_suffix(path, TIMELINE_DELETE_MARK_SUFFIX)
163 8 : }
164 :
165 : /// During pageserver startup, we need to order operations not to exhaust tokio worker threads by
166 : /// blocking.
167 : ///
168 : /// The instances of this value exist only during startup, otherwise `None` is provided, meaning no
169 : /// delaying is needed.
170 : #[derive(Clone)]
171 : pub struct InitializationOrder {
172 : /// Each initial tenant load task carries this until it is done loading timelines from remote storage
173 : pub initial_tenant_load_remote: Option<utils::completion::Completion>,
174 :
175 : /// Each initial tenant load task carries this until completion.
176 : pub initial_tenant_load: Option<utils::completion::Completion>,
177 :
178 : /// Barrier for when we can start any background jobs.
179 : ///
180 : /// This can be broken up later on, but right now there is just one class of a background job.
181 : pub background_jobs_can_start: utils::completion::Barrier,
182 : }
183 :
184 : /// Time the future with a warning when it exceeds a threshold.
185 4 : async fn timed<Fut: std::future::Future>(
186 4 : fut: Fut,
187 4 : name: &str,
188 4 : warn_at: std::time::Duration,
189 4 : ) -> <Fut as std::future::Future>::Output {
190 4 : let started = std::time::Instant::now();
191 4 :
192 4 : let mut fut = std::pin::pin!(fut);
193 4 :
194 4 : match tokio::time::timeout(warn_at, &mut fut).await {
195 2 : Ok(ret) => {
196 2 : tracing::info!(
197 : stage = name,
198 0 : elapsed_ms = started.elapsed().as_millis(),
199 0 : "completed"
200 : );
201 2 : ret
202 : }
203 : Err(_) => {
204 2 : tracing::info!(
205 : stage = name,
206 0 : elapsed_ms = started.elapsed().as_millis(),
207 0 : "still waiting, taking longer than expected..."
208 : );
209 :
210 2 : let ret = fut.await;
211 :
212 : // this has a global allowed_errors
213 2 : tracing::warn!(
214 : stage = name,
215 0 : elapsed_ms = started.elapsed().as_millis(),
216 0 : "completed, took longer than expected"
217 : );
218 :
219 2 : ret
220 : }
221 : }
222 4 : }
223 :
224 : #[cfg(test)]
225 : mod timed_tests {
226 : use super::timed;
227 : use std::time::Duration;
228 :
229 : #[tokio::test]
230 2 : async fn timed_completes_when_inner_future_completes() {
231 2 : // A future that completes on time should have its result returned
232 2 : let r1 = timed(
233 2 : async move {
234 2 : tokio::time::sleep(Duration::from_millis(10)).await;
235 2 : 123
236 2 : },
237 2 : "test 1",
238 2 : Duration::from_millis(50),
239 2 : )
240 2 : .await;
241 2 : assert_eq!(r1, 123);
242 2 :
243 2 : // A future that completes too slowly should also have its result returned
244 2 : let r1 = timed(
245 2 : async move {
246 6 : tokio::time::sleep(Duration::from_millis(50)).await;
247 2 : 456
248 2 : },
249 2 : "test 1",
250 2 : Duration::from_millis(10),
251 2 : )
252 4 : .await;
253 2 : assert_eq!(r1, 456);
254 2 : }
255 : }
|