Line data Source code
1 : use crate::{
2 : background_process,
3 : local_env::{LocalEnv, NeonStorageControllerConf},
4 : };
5 : use camino::{Utf8Path, Utf8PathBuf};
6 : use hyper0::Uri;
7 : use nix::unistd::Pid;
8 : use pageserver_api::{
9 : controller_api::{
10 : NodeConfigureRequest, NodeDescribeResponse, NodeRegisterRequest, TenantCreateRequest,
11 : TenantCreateResponse, TenantLocateResponse, TenantShardMigrateRequest,
12 : TenantShardMigrateResponse,
13 : },
14 : models::{
15 : TenantShardSplitRequest, TenantShardSplitResponse, TimelineCreateRequest, TimelineInfo,
16 : },
17 : shard::{ShardStripeSize, TenantShardId},
18 : };
19 : use pageserver_client::mgmt_api::ResponseErrorMessageExt;
20 : use postgres_backend::AuthType;
21 : use reqwest::Method;
22 : use serde::{de::DeserializeOwned, Deserialize, Serialize};
23 : use std::{fs, net::SocketAddr, path::PathBuf, str::FromStr, sync::OnceLock};
24 : use tokio::process::Command;
25 : use tracing::instrument;
26 : use url::Url;
27 : use utils::{
28 : auth::{encode_from_key_file, Claims, Scope},
29 : id::{NodeId, TenantId},
30 : };
31 : use whoami::username;
32 :
33 : pub struct StorageController {
34 : env: LocalEnv,
35 : private_key: Option<Vec<u8>>,
36 : public_key: Option<String>,
37 : client: reqwest::Client,
38 : config: NeonStorageControllerConf,
39 :
40 : // The listen addresses is learned when starting the storage controller,
41 : // hence the use of OnceLock to init it at the right time.
42 : listen: OnceLock<SocketAddr>,
43 : }
44 :
45 : const COMMAND: &str = "storage_controller";
46 :
47 : const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
48 :
49 : const DB_NAME: &str = "storage_controller";
50 :
51 : pub struct NeonStorageControllerStartArgs {
52 : pub instance_id: u8,
53 : pub base_port: Option<u16>,
54 : pub start_timeout: humantime::Duration,
55 : }
56 :
57 : impl NeonStorageControllerStartArgs {
58 0 : pub fn with_default_instance_id(start_timeout: humantime::Duration) -> Self {
59 0 : Self {
60 0 : instance_id: 1,
61 0 : base_port: None,
62 0 : start_timeout,
63 0 : }
64 0 : }
65 : }
66 :
67 : pub struct NeonStorageControllerStopArgs {
68 : pub instance_id: u8,
69 : pub immediate: bool,
70 : }
71 :
72 : impl NeonStorageControllerStopArgs {
73 0 : pub fn with_default_instance_id(immediate: bool) -> Self {
74 0 : Self {
75 0 : instance_id: 1,
76 0 : immediate,
77 0 : }
78 0 : }
79 : }
80 :
81 0 : #[derive(Serialize, Deserialize)]
82 : pub struct AttachHookRequest {
83 : pub tenant_shard_id: TenantShardId,
84 : pub node_id: Option<NodeId>,
85 : pub generation_override: Option<i32>,
86 : }
87 :
88 0 : #[derive(Serialize, Deserialize)]
89 : pub struct AttachHookResponse {
90 : pub gen: Option<u32>,
91 : }
92 :
93 0 : #[derive(Serialize, Deserialize)]
94 : pub struct InspectRequest {
95 : pub tenant_shard_id: TenantShardId,
96 : }
97 :
98 0 : #[derive(Serialize, Deserialize)]
99 : pub struct InspectResponse {
100 : pub attachment: Option<(u32, NodeId)>,
101 : }
102 :
103 : impl StorageController {
104 0 : pub fn from_env(env: &LocalEnv) -> Self {
105 0 : // Assume all pageservers have symmetric auth configuration: this service
106 0 : // expects to use one JWT token to talk to all of them.
107 0 : let ps_conf = env
108 0 : .pageservers
109 0 : .first()
110 0 : .expect("Config is validated to contain at least one pageserver");
111 0 : let (private_key, public_key) = match ps_conf.http_auth_type {
112 0 : AuthType::Trust => (None, None),
113 : AuthType::NeonJWT => {
114 0 : let private_key_path = env.get_private_key_path();
115 0 : let private_key = fs::read(private_key_path).expect("failed to read private key");
116 0 :
117 0 : // If pageserver auth is enabled, this implicitly enables auth for this service,
118 0 : // using the same credentials.
119 0 : let public_key_path =
120 0 : camino::Utf8PathBuf::try_from(env.base_data_dir.join("auth_public_key.pem"))
121 0 : .unwrap();
122 :
123 : // This service takes keys as a string rather than as a path to a file/dir: read the key into memory.
124 0 : let public_key = if std::fs::metadata(&public_key_path)
125 0 : .expect("Can't stat public key")
126 0 : .is_dir()
127 : {
128 : // Our config may specify a directory: this is for the pageserver's ability to handle multiple
129 : // keys. We only use one key at a time, so, arbitrarily load the first one in the directory.
130 0 : let mut dir =
131 0 : std::fs::read_dir(&public_key_path).expect("Can't readdir public key path");
132 0 : let dent = dir
133 0 : .next()
134 0 : .expect("Empty key dir")
135 0 : .expect("Error reading key dir");
136 0 :
137 0 : std::fs::read_to_string(dent.path()).expect("Can't read public key")
138 : } else {
139 0 : std::fs::read_to_string(&public_key_path).expect("Can't read public key")
140 : };
141 0 : (Some(private_key), Some(public_key))
142 : }
143 : };
144 :
145 0 : Self {
146 0 : env: env.clone(),
147 0 : private_key,
148 0 : public_key,
149 0 : client: reqwest::ClientBuilder::new()
150 0 : .build()
151 0 : .expect("Failed to construct http client"),
152 0 : config: env.storage_controller.clone(),
153 0 : listen: OnceLock::default(),
154 0 : }
155 0 : }
156 :
157 0 : fn storage_controller_instance_dir(&self, instance_id: u8) -> PathBuf {
158 0 : self.env
159 0 : .base_data_dir
160 0 : .join(format!("storage_controller_{}", instance_id))
161 0 : }
162 :
163 0 : fn pid_file(&self, instance_id: u8) -> Utf8PathBuf {
164 0 : Utf8PathBuf::from_path_buf(
165 0 : self.storage_controller_instance_dir(instance_id)
166 0 : .join("storage_controller.pid"),
167 0 : )
168 0 : .expect("non-Unicode path")
169 0 : }
170 :
171 : /// PIDFile for the postgres instance used to store storage controller state
172 0 : fn postgres_pid_file(&self) -> Utf8PathBuf {
173 0 : Utf8PathBuf::from_path_buf(
174 0 : self.env
175 0 : .base_data_dir
176 0 : .join("storage_controller_postgres.pid"),
177 0 : )
178 0 : .expect("non-Unicode path")
179 0 : }
180 :
181 : /// Find the directory containing postgres subdirectories, such `bin` and `lib`
182 : ///
183 : /// This usually uses STORAGE_CONTROLLER_POSTGRES_VERSION of postgres, but will fall back
184 : /// to other versions if that one isn't found. Some automated tests create circumstances
185 : /// where only one version is available in pg_distrib_dir, such as `test_remote_extensions`.
186 0 : async fn get_pg_dir(&self, dir_name: &str) -> anyhow::Result<Utf8PathBuf> {
187 0 : let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 16, 15, 14];
188 :
189 0 : for v in prefer_versions {
190 0 : let path = Utf8PathBuf::from_path_buf(self.env.pg_dir(v, dir_name)?).unwrap();
191 0 : if tokio::fs::try_exists(&path).await? {
192 0 : return Ok(path);
193 0 : }
194 : }
195 :
196 : // Fall through
197 0 : anyhow::bail!(
198 0 : "Postgres directory '{}' not found in {}",
199 0 : dir_name,
200 0 : self.env.pg_distrib_dir.display(),
201 0 : );
202 0 : }
203 :
204 0 : pub async fn get_pg_bin_dir(&self) -> anyhow::Result<Utf8PathBuf> {
205 0 : self.get_pg_dir("bin").await
206 0 : }
207 :
208 0 : pub async fn get_pg_lib_dir(&self) -> anyhow::Result<Utf8PathBuf> {
209 0 : self.get_pg_dir("lib").await
210 0 : }
211 :
212 : /// Readiness check for our postgres process
213 0 : async fn pg_isready(&self, pg_bin_dir: &Utf8Path, postgres_port: u16) -> anyhow::Result<bool> {
214 0 : let bin_path = pg_bin_dir.join("pg_isready");
215 0 : let args = [
216 0 : "-h",
217 0 : "localhost",
218 0 : "-U",
219 0 : &username(),
220 0 : "-d",
221 0 : DB_NAME,
222 0 : "-p",
223 0 : &format!("{}", postgres_port),
224 0 : ];
225 0 : let exitcode = Command::new(bin_path).args(args).spawn()?.wait().await?;
226 :
227 0 : Ok(exitcode.success())
228 0 : }
229 :
230 : /// Create our database if it doesn't exist
231 : ///
232 : /// This function is equivalent to the `diesel setup` command in the diesel CLI. We implement
233 : /// the same steps by hand to avoid imposing a dependency on installing diesel-cli for developers
234 : /// who just want to run `cargo neon_local` without knowing about diesel.
235 : ///
236 : /// Returns the database url
237 0 : pub async fn setup_database(&self, postgres_port: u16) -> anyhow::Result<String> {
238 0 : let database_url = format!(
239 0 : "postgresql://{}@localhost:{}/{DB_NAME}",
240 0 : &username(),
241 0 : postgres_port
242 0 : );
243 :
244 0 : let pg_bin_dir = self.get_pg_bin_dir().await?;
245 0 : let createdb_path = pg_bin_dir.join("createdb");
246 0 : let output = Command::new(&createdb_path)
247 0 : .args([
248 0 : "-h",
249 0 : "localhost",
250 0 : "-p",
251 0 : &format!("{}", postgres_port),
252 0 : "-U",
253 0 : &username(),
254 0 : "-O",
255 0 : &username(),
256 0 : DB_NAME,
257 0 : ])
258 0 : .output()
259 0 : .await
260 0 : .expect("Failed to spawn createdb");
261 0 :
262 0 : if !output.status.success() {
263 0 : let stderr = String::from_utf8(output.stderr).expect("Non-UTF8 output from createdb");
264 0 : if stderr.contains("already exists") {
265 0 : tracing::info!("Database {DB_NAME} already exists");
266 : } else {
267 0 : anyhow::bail!("createdb failed with status {}: {stderr}", output.status);
268 : }
269 0 : }
270 :
271 0 : Ok(database_url)
272 0 : }
273 :
274 0 : pub async fn connect_to_database(
275 0 : &self,
276 0 : postgres_port: u16,
277 0 : ) -> anyhow::Result<(
278 0 : tokio_postgres::Client,
279 0 : tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
280 0 : )> {
281 0 : tokio_postgres::Config::new()
282 0 : .host("localhost")
283 0 : .port(postgres_port)
284 0 : // The user is the ambient operating system user name.
285 0 : // That is an impurity which we want to fix in => TODO https://github.com/neondatabase/neon/issues/8400
286 0 : //
287 0 : // Until we get there, use the ambient operating system user name.
288 0 : // Recent tokio-postgres versions default to this if the user isn't specified.
289 0 : // But tokio-postgres fork doesn't have this upstream commit:
290 0 : // https://github.com/sfackler/rust-postgres/commit/cb609be758f3fb5af537f04b584a2ee0cebd5e79
291 0 : // => we should rebase our fork => TODO https://github.com/neondatabase/neon/issues/8399
292 0 : .user(&username())
293 0 : .dbname(DB_NAME)
294 0 : .connect(tokio_postgres::NoTls)
295 0 : .await
296 0 : .map_err(anyhow::Error::new)
297 0 : }
298 :
299 0 : pub async fn start(&self, start_args: NeonStorageControllerStartArgs) -> anyhow::Result<()> {
300 0 : let instance_dir = self.storage_controller_instance_dir(start_args.instance_id);
301 0 : if let Err(err) = tokio::fs::create_dir(&instance_dir).await {
302 0 : if err.kind() != std::io::ErrorKind::AlreadyExists {
303 0 : panic!("Failed to create instance dir {instance_dir:?}");
304 0 : }
305 0 : }
306 :
307 0 : let (listen, postgres_port) = {
308 0 : if let Some(base_port) = start_args.base_port {
309 0 : (
310 0 : format!("127.0.0.1:{base_port}"),
311 0 : self.config
312 0 : .database_url
313 0 : .expect("--base-port requires NeonStorageControllerConf::database_url")
314 0 : .port(),
315 0 : )
316 : } else {
317 0 : let listen_url = self.env.control_plane_api.clone().unwrap();
318 0 :
319 0 : let listen = format!(
320 0 : "{}:{}",
321 0 : listen_url.host_str().unwrap(),
322 0 : listen_url.port().unwrap()
323 0 : );
324 0 :
325 0 : (listen, listen_url.port().unwrap() + 1)
326 : }
327 : };
328 :
329 0 : let socket_addr = listen
330 0 : .parse()
331 0 : .expect("listen address is a valid socket address");
332 0 : self.listen
333 0 : .set(socket_addr)
334 0 : .expect("StorageController::listen is only set here");
335 :
336 : // Do we remove the pid file on stop?
337 0 : let pg_started = self.is_postgres_running().await?;
338 0 : let pg_lib_dir = self.get_pg_lib_dir().await?;
339 :
340 0 : if !pg_started {
341 : // Start a vanilla Postgres process used by the storage controller for persistence.
342 0 : let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
343 0 : .unwrap()
344 0 : .join("storage_controller_db");
345 0 : let pg_bin_dir = self.get_pg_bin_dir().await?;
346 0 : let pg_log_path = pg_data_path.join("postgres.log");
347 0 :
348 0 : if !tokio::fs::try_exists(&pg_data_path).await? {
349 0 : let initdb_args = [
350 0 : "--pgdata",
351 0 : pg_data_path.as_ref(),
352 0 : "--username",
353 0 : &username(),
354 0 : "--no-sync",
355 0 : "--no-instructions",
356 0 : ];
357 0 : tracing::info!(
358 0 : "Initializing storage controller database with args: {:?}",
359 : initdb_args
360 : );
361 :
362 : // Initialize empty database
363 0 : let initdb_path = pg_bin_dir.join("initdb");
364 0 : let mut child = Command::new(&initdb_path)
365 0 : .envs(vec![
366 0 : ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
367 0 : ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
368 0 : ])
369 0 : .args(initdb_args)
370 0 : .spawn()
371 0 : .expect("Failed to spawn initdb");
372 0 : let status = child.wait().await?;
373 0 : if !status.success() {
374 0 : anyhow::bail!("initdb failed with status {status}");
375 0 : }
376 0 : };
377 :
378 : // Write a minimal config file:
379 : // - Specify the port, since this is chosen dynamically
380 : // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
381 : // the storage controller we don't want a slow local disk to interfere with that.
382 : //
383 : // NB: it's important that we rewrite this file on each start command so we propagate changes
384 : // from `LocalEnv`'s config file (`.neon/config`).
385 0 : tokio::fs::write(
386 0 : &pg_data_path.join("postgresql.conf"),
387 0 : format!("port = {}\nfsync=off\n", postgres_port),
388 0 : )
389 0 : .await?;
390 :
391 0 : println!("Starting storage controller database...");
392 0 : let db_start_args = [
393 0 : "-w",
394 0 : "-D",
395 0 : pg_data_path.as_ref(),
396 0 : "-l",
397 0 : pg_log_path.as_ref(),
398 0 : "-U",
399 0 : &username(),
400 0 : "start",
401 0 : ];
402 0 : tracing::info!(
403 0 : "Starting storage controller database with args: {:?}",
404 : db_start_args
405 : );
406 :
407 0 : background_process::start_process(
408 0 : "storage_controller_db",
409 0 : &self.env.base_data_dir,
410 0 : pg_bin_dir.join("pg_ctl").as_std_path(),
411 0 : db_start_args,
412 0 : vec![
413 0 : ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
414 0 : ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
415 0 : ],
416 0 : background_process::InitialPidFile::Create(self.postgres_pid_file()),
417 0 : &start_args.start_timeout,
418 0 : || self.pg_isready(&pg_bin_dir, postgres_port),
419 0 : )
420 0 : .await?;
421 :
422 0 : self.setup_database(postgres_port).await?;
423 0 : }
424 :
425 0 : let database_url = format!("postgresql://localhost:{}/{DB_NAME}", postgres_port);
426 0 :
427 0 : // We support running a startup SQL script to fiddle with the database before we launch storcon.
428 0 : // This is used by the test suite.
429 0 : let startup_script_path = self
430 0 : .env
431 0 : .base_data_dir
432 0 : .join("storage_controller_db.startup.sql");
433 0 : let startup_script = match tokio::fs::read_to_string(&startup_script_path).await {
434 0 : Ok(script) => {
435 0 : tokio::fs::remove_file(startup_script_path).await?;
436 0 : script
437 : }
438 0 : Err(e) => {
439 0 : if e.kind() == std::io::ErrorKind::NotFound {
440 : // always run some startup script so that this code path doesn't bit rot
441 0 : "BEGIN; COMMIT;".to_string()
442 : } else {
443 0 : anyhow::bail!("Failed to read startup script: {e}")
444 : }
445 : }
446 : };
447 0 : let (mut client, conn) = self.connect_to_database(postgres_port).await?;
448 0 : let conn = tokio::spawn(conn);
449 0 : let tx = client.build_transaction();
450 0 : let tx = tx.start().await?;
451 0 : tx.batch_execute(&startup_script).await?;
452 0 : tx.commit().await?;
453 0 : drop(client);
454 0 : conn.await??;
455 :
456 0 : let listen = self
457 0 : .listen
458 0 : .get()
459 0 : .expect("cell is set earlier in this function");
460 0 : let address_for_peers = Uri::builder()
461 0 : .scheme("http")
462 0 : .authority(format!("{}:{}", listen.ip(), listen.port()))
463 0 : .path_and_query("")
464 0 : .build()
465 0 : .unwrap();
466 0 :
467 0 : let mut args = vec![
468 0 : "-l",
469 0 : &listen.to_string(),
470 0 : "--dev",
471 0 : "--database-url",
472 0 : &database_url,
473 0 : "--max-offline-interval",
474 0 : &humantime::Duration::from(self.config.max_offline).to_string(),
475 0 : "--max-warming-up-interval",
476 0 : &humantime::Duration::from(self.config.max_warming_up).to_string(),
477 0 : "--heartbeat-interval",
478 0 : &humantime::Duration::from(self.config.heartbeat_interval).to_string(),
479 0 : "--address-for-peers",
480 0 : &address_for_peers.to_string(),
481 0 : ]
482 0 : .into_iter()
483 0 : .map(|s| s.to_string())
484 0 : .collect::<Vec<_>>();
485 0 :
486 0 : if self.config.start_as_candidate {
487 0 : args.push("--start-as-candidate".to_string());
488 0 : }
489 :
490 0 : if let Some(private_key) = &self.private_key {
491 0 : let claims = Claims::new(None, Scope::PageServerApi);
492 0 : let jwt_token =
493 0 : encode_from_key_file(&claims, private_key).expect("failed to generate jwt token");
494 0 : args.push(format!("--jwt-token={jwt_token}"));
495 0 :
496 0 : let peer_claims = Claims::new(None, Scope::Admin);
497 0 : let peer_jwt_token = encode_from_key_file(&peer_claims, private_key)
498 0 : .expect("failed to generate jwt token");
499 0 : args.push(format!("--peer-jwt-token={peer_jwt_token}"));
500 0 : }
501 :
502 0 : if let Some(public_key) = &self.public_key {
503 0 : args.push(format!("--public-key=\"{public_key}\""));
504 0 : }
505 :
506 0 : if let Some(control_plane_compute_hook_api) = &self.env.control_plane_compute_hook_api {
507 0 : args.push(format!(
508 0 : "--compute-hook-url={control_plane_compute_hook_api}"
509 0 : ));
510 0 : }
511 :
512 0 : if let Some(split_threshold) = self.config.split_threshold.as_ref() {
513 0 : args.push(format!("--split-threshold={split_threshold}"))
514 0 : }
515 :
516 0 : if let Some(lag) = self.config.max_secondary_lag_bytes.as_ref() {
517 0 : args.push(format!("--max-secondary-lag-bytes={lag}"))
518 0 : }
519 :
520 0 : if let Some(threshold) = self.config.long_reconcile_threshold {
521 0 : args.push(format!(
522 0 : "--long-reconcile-threshold={}",
523 0 : humantime::Duration::from(threshold)
524 0 : ))
525 0 : }
526 :
527 0 : args.push(format!(
528 0 : "--neon-local-repo-dir={}",
529 0 : self.env.base_data_dir.display()
530 0 : ));
531 0 :
532 0 : background_process::start_process(
533 0 : COMMAND,
534 0 : &instance_dir,
535 0 : &self.env.storage_controller_bin(),
536 0 : args,
537 0 : vec![
538 0 : ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
539 0 : ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
540 0 : ],
541 0 : background_process::InitialPidFile::Create(self.pid_file(start_args.instance_id)),
542 0 : &start_args.start_timeout,
543 0 : || async {
544 0 : match self.ready().await {
545 0 : Ok(_) => Ok(true),
546 0 : Err(_) => Ok(false),
547 : }
548 0 : },
549 0 : )
550 0 : .await?;
551 :
552 0 : Ok(())
553 0 : }
554 :
555 0 : pub async fn stop(&self, stop_args: NeonStorageControllerStopArgs) -> anyhow::Result<()> {
556 0 : background_process::stop_process(
557 0 : stop_args.immediate,
558 0 : COMMAND,
559 0 : &self.pid_file(stop_args.instance_id),
560 0 : )?;
561 :
562 0 : let storcon_instances = self.env.storage_controller_instances().await?;
563 0 : for (instance_id, instanced_dir_path) in storcon_instances {
564 0 : if instance_id == stop_args.instance_id {
565 0 : continue;
566 0 : }
567 0 :
568 0 : let pid_file = instanced_dir_path.join("storage_controller.pid");
569 0 : let pid = tokio::fs::read_to_string(&pid_file)
570 0 : .await
571 0 : .map_err(|err| {
572 0 : anyhow::anyhow!("Failed to read storcon pid file at {pid_file:?}: {err}")
573 0 : })?
574 0 : .parse::<i32>()
575 0 : .expect("pid is valid i32");
576 :
577 0 : let other_proc_alive = !background_process::process_has_stopped(Pid::from_raw(pid))?;
578 0 : if other_proc_alive {
579 : // There is another storage controller instance running, so we return
580 : // and leave the database running.
581 0 : return Ok(());
582 0 : }
583 : }
584 :
585 0 : let pg_data_path = self.env.base_data_dir.join("storage_controller_db");
586 0 : let pg_bin_dir = self.get_pg_bin_dir().await?;
587 :
588 0 : println!("Stopping storage controller database...");
589 0 : let pg_stop_args = ["-D", &pg_data_path.to_string_lossy(), "stop"];
590 0 : let stop_status = Command::new(pg_bin_dir.join("pg_ctl"))
591 0 : .args(pg_stop_args)
592 0 : .spawn()?
593 0 : .wait()
594 0 : .await?;
595 0 : if !stop_status.success() {
596 0 : match self.is_postgres_running().await {
597 : Ok(false) => {
598 0 : println!("Storage controller database is already stopped");
599 0 : return Ok(());
600 : }
601 : Ok(true) => {
602 0 : anyhow::bail!("Failed to stop storage controller database");
603 : }
604 0 : Err(err) => {
605 0 : anyhow::bail!("Failed to stop storage controller database: {err}");
606 : }
607 : }
608 0 : }
609 0 :
610 0 : Ok(())
611 0 : }
612 :
613 0 : async fn is_postgres_running(&self) -> anyhow::Result<bool> {
614 0 : let pg_data_path = self.env.base_data_dir.join("storage_controller_db");
615 0 : let pg_bin_dir = self.get_pg_bin_dir().await?;
616 :
617 0 : let pg_status_args = ["-D", &pg_data_path.to_string_lossy(), "status"];
618 0 : let status_exitcode = Command::new(pg_bin_dir.join("pg_ctl"))
619 0 : .args(pg_status_args)
620 0 : .spawn()?
621 0 : .wait()
622 0 : .await?;
623 :
624 : // pg_ctl status returns this exit code if postgres is not running: in this case it is
625 : // fine that stop failed. Otherwise it is an error that stop failed.
626 : const PG_STATUS_NOT_RUNNING: i32 = 3;
627 : const PG_NO_DATA_DIR: i32 = 4;
628 : const PG_STATUS_RUNNING: i32 = 0;
629 0 : match status_exitcode.code() {
630 0 : Some(PG_STATUS_NOT_RUNNING) => Ok(false),
631 0 : Some(PG_NO_DATA_DIR) => Ok(false),
632 0 : Some(PG_STATUS_RUNNING) => Ok(true),
633 0 : Some(code) => Err(anyhow::anyhow!(
634 0 : "pg_ctl status returned unexpected status code: {:?}",
635 0 : code
636 0 : )),
637 0 : None => Err(anyhow::anyhow!("pg_ctl status returned no status code")),
638 : }
639 0 : }
640 :
641 0 : fn get_claims_for_path(path: &str) -> anyhow::Result<Option<Claims>> {
642 0 : let category = match path.find('/') {
643 0 : Some(idx) => &path[..idx],
644 0 : None => path,
645 : };
646 :
647 0 : match category {
648 0 : "status" | "ready" => Ok(None),
649 0 : "control" | "debug" => Ok(Some(Claims::new(None, Scope::Admin))),
650 0 : "v1" => Ok(Some(Claims::new(None, Scope::PageServerApi))),
651 0 : _ => Err(anyhow::anyhow!("Failed to determine claims for {}", path)),
652 : }
653 0 : }
654 :
655 : /// Simple HTTP request wrapper for calling into storage controller
656 0 : async fn dispatch<RQ, RS>(
657 0 : &self,
658 0 : method: reqwest::Method,
659 0 : path: String,
660 0 : body: Option<RQ>,
661 0 : ) -> anyhow::Result<RS>
662 0 : where
663 0 : RQ: Serialize + Sized,
664 0 : RS: DeserializeOwned + Sized,
665 0 : {
666 : // In the special case of the `storage_controller start` subcommand, we wish
667 : // to use the API endpoint of the newly started storage controller in order
668 : // to pass the readiness check. In this scenario [`Self::listen`] will be set
669 : // (see [`Self::start`]).
670 : //
671 : // Otherwise, we infer the storage controller api endpoint from the configured
672 : // control plane API.
673 0 : let url = if let Some(socket_addr) = self.listen.get() {
674 0 : Url::from_str(&format!(
675 0 : "http://{}:{}/{path}",
676 0 : socket_addr.ip().to_canonical(),
677 0 : socket_addr.port()
678 0 : ))
679 0 : .unwrap()
680 : } else {
681 : // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
682 : // for general purpose API access.
683 0 : let listen_url = self.env.control_plane_api.clone().unwrap();
684 0 : Url::from_str(&format!(
685 0 : "http://{}:{}/{path}",
686 0 : listen_url.host_str().unwrap(),
687 0 : listen_url.port().unwrap()
688 0 : ))
689 0 : .unwrap()
690 : };
691 :
692 0 : let mut builder = self.client.request(method, url);
693 0 : if let Some(body) = body {
694 0 : builder = builder.json(&body)
695 0 : }
696 0 : if let Some(private_key) = &self.private_key {
697 0 : println!("Getting claims for path {}", path);
698 0 : if let Some(required_claims) = Self::get_claims_for_path(&path)? {
699 0 : println!("Got claims {:?} for path {}", required_claims, path);
700 0 : let jwt_token = encode_from_key_file(&required_claims, private_key)?;
701 0 : builder = builder.header(
702 0 : reqwest::header::AUTHORIZATION,
703 0 : format!("Bearer {jwt_token}"),
704 0 : );
705 0 : }
706 0 : }
707 :
708 0 : let response = builder.send().await?;
709 0 : let response = response.error_from_body().await?;
710 :
711 0 : Ok(response
712 0 : .json()
713 0 : .await
714 0 : .map_err(pageserver_client::mgmt_api::Error::ReceiveBody)?)
715 0 : }
716 :
717 : /// Call into the attach_hook API, for use before handing out attachments to pageservers
718 0 : #[instrument(skip(self))]
719 : pub async fn attach_hook(
720 : &self,
721 : tenant_shard_id: TenantShardId,
722 : pageserver_id: NodeId,
723 : ) -> anyhow::Result<Option<u32>> {
724 : let request = AttachHookRequest {
725 : tenant_shard_id,
726 : node_id: Some(pageserver_id),
727 : generation_override: None,
728 : };
729 :
730 : let response = self
731 : .dispatch::<_, AttachHookResponse>(
732 : Method::POST,
733 : "debug/v1/attach-hook".to_string(),
734 : Some(request),
735 : )
736 : .await?;
737 :
738 : Ok(response.gen)
739 : }
740 :
741 0 : #[instrument(skip(self))]
742 : pub async fn inspect(
743 : &self,
744 : tenant_shard_id: TenantShardId,
745 : ) -> anyhow::Result<Option<(u32, NodeId)>> {
746 : let request = InspectRequest { tenant_shard_id };
747 :
748 : let response = self
749 : .dispatch::<_, InspectResponse>(
750 : Method::POST,
751 : "debug/v1/inspect".to_string(),
752 : Some(request),
753 : )
754 : .await?;
755 :
756 : Ok(response.attachment)
757 : }
758 :
759 0 : #[instrument(skip(self))]
760 : pub async fn tenant_create(
761 : &self,
762 : req: TenantCreateRequest,
763 : ) -> anyhow::Result<TenantCreateResponse> {
764 : self.dispatch(Method::POST, "v1/tenant".to_string(), Some(req))
765 : .await
766 : }
767 :
768 0 : #[instrument(skip(self))]
769 : pub async fn tenant_import(&self, tenant_id: TenantId) -> anyhow::Result<TenantCreateResponse> {
770 : self.dispatch::<(), TenantCreateResponse>(
771 : Method::POST,
772 : format!("debug/v1/tenant/{tenant_id}/import"),
773 : None,
774 : )
775 : .await
776 : }
777 :
778 0 : #[instrument(skip(self))]
779 : pub async fn tenant_locate(&self, tenant_id: TenantId) -> anyhow::Result<TenantLocateResponse> {
780 : self.dispatch::<(), _>(
781 : Method::GET,
782 : format!("debug/v1/tenant/{tenant_id}/locate"),
783 : None,
784 : )
785 : .await
786 : }
787 :
788 0 : #[instrument(skip(self))]
789 : pub async fn tenant_migrate(
790 : &self,
791 : tenant_shard_id: TenantShardId,
792 : node_id: NodeId,
793 : ) -> anyhow::Result<TenantShardMigrateResponse> {
794 : self.dispatch(
795 : Method::PUT,
796 : format!("control/v1/tenant/{tenant_shard_id}/migrate"),
797 : Some(TenantShardMigrateRequest {
798 : tenant_shard_id,
799 : node_id,
800 : }),
801 : )
802 : .await
803 : }
804 :
805 0 : #[instrument(skip(self), fields(%tenant_id, %new_shard_count))]
806 : pub async fn tenant_split(
807 : &self,
808 : tenant_id: TenantId,
809 : new_shard_count: u8,
810 : new_stripe_size: Option<ShardStripeSize>,
811 : ) -> anyhow::Result<TenantShardSplitResponse> {
812 : self.dispatch(
813 : Method::PUT,
814 : format!("control/v1/tenant/{tenant_id}/shard_split"),
815 : Some(TenantShardSplitRequest {
816 : new_shard_count,
817 : new_stripe_size,
818 : }),
819 : )
820 : .await
821 : }
822 :
823 0 : #[instrument(skip_all, fields(node_id=%req.node_id))]
824 : pub async fn node_register(&self, req: NodeRegisterRequest) -> anyhow::Result<()> {
825 : self.dispatch::<_, ()>(Method::POST, "control/v1/node".to_string(), Some(req))
826 : .await
827 : }
828 :
829 0 : #[instrument(skip_all, fields(node_id=%req.node_id))]
830 : pub async fn node_configure(&self, req: NodeConfigureRequest) -> anyhow::Result<()> {
831 : self.dispatch::<_, ()>(
832 : Method::PUT,
833 : format!("control/v1/node/{}/config", req.node_id),
834 : Some(req),
835 : )
836 : .await
837 : }
838 :
839 0 : pub async fn node_list(&self) -> anyhow::Result<Vec<NodeDescribeResponse>> {
840 0 : self.dispatch::<(), Vec<NodeDescribeResponse>>(
841 0 : Method::GET,
842 0 : "control/v1/node".to_string(),
843 0 : None,
844 0 : )
845 0 : .await
846 0 : }
847 :
848 0 : #[instrument(skip(self))]
849 : pub async fn ready(&self) -> anyhow::Result<()> {
850 : self.dispatch::<(), ()>(Method::GET, "ready".to_string(), None)
851 : .await
852 : }
853 :
854 0 : #[instrument(skip_all, fields(%tenant_id, timeline_id=%req.new_timeline_id))]
855 : pub async fn tenant_timeline_create(
856 : &self,
857 : tenant_id: TenantId,
858 : req: TimelineCreateRequest,
859 : ) -> anyhow::Result<TimelineInfo> {
860 : self.dispatch(
861 : Method::POST,
862 : format!("v1/tenant/{tenant_id}/timeline"),
863 : Some(req),
864 : )
865 : .await
866 : }
867 : }
|
