Line data Source code
1 : //! This module is responsible for locating and loading paths in a local setup.
2 : //!
3 : //! Now it also provides init method which acts like a stub for proper installation
4 : //! script which will use local paths.
5 :
6 : use std::collections::HashMap;
7 : use std::net::SocketAddr;
8 : use std::path::{Path, PathBuf};
9 : use std::process::{Command, Stdio};
10 : use std::time::Duration;
11 : use std::{env, fs};
12 :
13 : use anyhow::{Context, bail};
14 : use clap::ValueEnum;
15 : use pem::Pem;
16 : use postgres_backend::AuthType;
17 : use reqwest::{Certificate, Url};
18 : use serde::{Deserialize, Serialize};
19 : use utils::auth::encode_from_key_file;
20 : use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
21 :
22 : use crate::broker::StorageBroker;
23 : use crate::endpoint_storage::{
24 : ENDPOINT_STORAGE_DEFAULT_ADDR, ENDPOINT_STORAGE_REMOTE_STORAGE_DIR, EndpointStorage,
25 : };
26 : use crate::pageserver::{PAGESERVER_REMOTE_STORAGE_DIR, PageServerNode};
27 : use crate::safekeeper::SafekeeperNode;
28 :
29 : pub const DEFAULT_PG_VERSION: u32 = 17;
30 :
31 : //
32 : // This data structures represents neon_local CLI config
33 : //
34 : // It is deserialized from the .neon/config file, or the config file passed
35 : // to 'neon_local init --config=<path>' option. See control_plane/simple.conf for
36 : // an example.
37 : //
38 : #[derive(PartialEq, Eq, Clone, Debug)]
39 : pub struct LocalEnv {
40 : // Base directory for all the nodes (the pageserver, safekeepers and
41 : // compute endpoints).
42 : //
43 : // This is not stored in the config file. Rather, this is the path where the
44 : // config file itself is. It is read from the NEON_REPO_DIR env variable which
45 : // must be an absolute path. If the env var is not set, $PWD/.neon is used.
46 : pub base_data_dir: PathBuf,
47 :
48 : // Path to postgres distribution. It's expected that "bin", "include",
49 : // "lib", "share" from postgres distribution are there. If at some point
50 : // in time we will be able to run against vanilla postgres we may split that
51 : // to four separate paths and match OS-specific installation layout.
52 : pub pg_distrib_dir: PathBuf,
53 :
54 : // Path to pageserver binary.
55 : pub neon_distrib_dir: PathBuf,
56 :
57 : // Default tenant ID to use with the 'neon_local' command line utility, when
58 : // --tenant_id is not explicitly specified.
59 : pub default_tenant_id: Option<TenantId>,
60 :
61 : // used to issue tokens during e.g pg start
62 : pub private_key_path: PathBuf,
63 : /// Path to environment's public key
64 : pub public_key_path: PathBuf,
65 :
66 : pub broker: NeonBroker,
67 :
68 : // Configuration for the storage controller (1 per neon_local environment)
69 : pub storage_controller: NeonStorageControllerConf,
70 :
71 : /// This Vec must always contain at least one pageserver
72 : /// Populdated by [`Self::load_config`] from the individual `pageserver.toml`s.
73 : /// NB: not used anymore except for informing users that they need to change their `.neon/config`.
74 : pub pageservers: Vec<PageServerConf>,
75 :
76 : pub safekeepers: Vec<SafekeeperConf>,
77 :
78 : pub endpoint_storage: EndpointStorageConf,
79 :
80 : // Control plane upcall API for pageserver: if None, we will not run storage_controller If set, this will
81 : // be propagated into each pageserver's configuration.
82 : pub control_plane_api: Url,
83 :
84 : // Control plane upcall APIs for storage controller. If set, this will be propagated into the
85 : // storage controller's configuration.
86 : pub control_plane_hooks_api: Option<Url>,
87 :
88 : /// Keep human-readable aliases in memory (and persist them to config), to hide ZId hex strings from the user.
89 : // A `HashMap<String, HashMap<TenantId, TimelineId>>` would be more appropriate here,
90 : // but deserialization into a generic toml object as `toml::Value::try_from` fails with an error.
91 : // https://toml.io/en/v1.0.0 does not contain a concept of "a table inside another table".
92 : pub branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
93 :
94 : /// Flag to generate SSL certificates for components that need it.
95 : /// Also generates root CA certificate that is used to sign all other certificates.
96 : pub generate_local_ssl_certs: bool,
97 : }
98 :
99 : /// On-disk state stored in `.neon/config`.
100 0 : #[derive(PartialEq, Eq, Clone, Debug, Default, Serialize, Deserialize)]
101 : #[serde(default, deny_unknown_fields)]
102 : pub struct OnDiskConfig {
103 : pub pg_distrib_dir: PathBuf,
104 : pub neon_distrib_dir: PathBuf,
105 : pub default_tenant_id: Option<TenantId>,
106 : pub private_key_path: PathBuf,
107 : pub public_key_path: PathBuf,
108 : pub broker: NeonBroker,
109 : pub storage_controller: NeonStorageControllerConf,
110 : #[serde(
111 : skip_serializing,
112 : deserialize_with = "fail_if_pageservers_field_specified"
113 : )]
114 : pub pageservers: Vec<PageServerConf>,
115 : pub safekeepers: Vec<SafekeeperConf>,
116 : pub endpoint_storage: EndpointStorageConf,
117 : pub control_plane_api: Option<Url>,
118 : pub control_plane_hooks_api: Option<Url>,
119 : pub control_plane_compute_hook_api: Option<Url>,
120 : branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
121 : // Note: skip serializing because in compat tests old storage controller fails
122 : // to load new config file. May be removed after this field is in release branch.
123 : #[serde(skip_serializing_if = "std::ops::Not::not")]
124 : pub generate_local_ssl_certs: bool,
125 : }
126 :
127 0 : fn fail_if_pageservers_field_specified<'de, D>(_: D) -> Result<Vec<PageServerConf>, D::Error>
128 0 : where
129 0 : D: serde::Deserializer<'de>,
130 0 : {
131 0 : Err(serde::de::Error::custom(
132 0 : "The 'pageservers' field is no longer used; pageserver.toml is now authoritative; \
133 0 : Please remove the `pageservers` from your .neon/config.",
134 0 : ))
135 0 : }
136 :
137 : /// The description of the neon_local env to be initialized by `neon_local init --config`.
138 0 : #[derive(Clone, Debug, Deserialize)]
139 : #[serde(deny_unknown_fields)]
140 : pub struct NeonLocalInitConf {
141 : // TODO: do we need this? Seems unused
142 : pub pg_distrib_dir: Option<PathBuf>,
143 : // TODO: do we need this? Seems unused
144 : pub neon_distrib_dir: Option<PathBuf>,
145 : pub default_tenant_id: TenantId,
146 : pub broker: NeonBroker,
147 : pub storage_controller: Option<NeonStorageControllerConf>,
148 : pub pageservers: Vec<NeonLocalInitPageserverConf>,
149 : pub safekeepers: Vec<SafekeeperConf>,
150 : pub endpoint_storage: EndpointStorageConf,
151 : pub control_plane_api: Option<Url>,
152 : pub control_plane_hooks_api: Option<Url>,
153 : pub generate_local_ssl_certs: bool,
154 : }
155 :
156 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
157 : #[serde(default)]
158 : pub struct EndpointStorageConf {
159 : pub listen_addr: SocketAddr,
160 : }
161 :
162 : /// Broker config for cluster internal communication.
163 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug, Default)]
164 : #[serde(default)]
165 : pub struct NeonBroker {
166 : /// Broker listen HTTP address for storage nodes coordination, e.g. '127.0.0.1:50051'.
167 : /// At least one of listen_addr or listen_https_addr must be set.
168 : pub listen_addr: Option<SocketAddr>,
169 : /// Broker listen HTTPS address for storage nodes coordination, e.g. '127.0.0.1:50051'.
170 : /// At least one of listen_addr or listen_https_addr must be set.
171 : /// listen_https_addr is preferred over listen_addr in neon_local.
172 : pub listen_https_addr: Option<SocketAddr>,
173 : }
174 :
175 : /// A part of storage controller's config the neon_local knows about.
176 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
177 : #[serde(default)]
178 : pub struct NeonStorageControllerConf {
179 : /// Heartbeat timeout before marking a node offline
180 : #[serde(with = "humantime_serde")]
181 : pub max_offline: Duration,
182 :
183 : #[serde(with = "humantime_serde")]
184 : pub max_warming_up: Duration,
185 :
186 : pub start_as_candidate: bool,
187 :
188 : /// Database url used when running multiple storage controller instances
189 : pub database_url: Option<SocketAddr>,
190 :
191 : /// Thresholds for auto-splitting a tenant into shards.
192 : pub split_threshold: Option<u64>,
193 : pub max_split_shards: Option<u8>,
194 : pub initial_split_threshold: Option<u64>,
195 : pub initial_split_shards: Option<u8>,
196 :
197 : pub max_secondary_lag_bytes: Option<u64>,
198 :
199 : #[serde(with = "humantime_serde")]
200 : pub heartbeat_interval: Duration,
201 :
202 : #[serde(with = "humantime_serde")]
203 : pub long_reconcile_threshold: Option<Duration>,
204 :
205 : pub use_https_pageserver_api: bool,
206 :
207 : pub timelines_onto_safekeepers: bool,
208 :
209 : pub use_https_safekeeper_api: bool,
210 :
211 : pub use_local_compute_notifications: bool,
212 : }
213 :
214 : impl NeonStorageControllerConf {
215 : // Use a shorter pageserver unavailability interval than the default to speed up tests.
216 : const DEFAULT_MAX_OFFLINE_INTERVAL: std::time::Duration = std::time::Duration::from_secs(10);
217 :
218 : const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
219 :
220 : // Very tight heartbeat interval to speed up tests
221 : const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(1000);
222 : }
223 :
224 : impl Default for NeonStorageControllerConf {
225 0 : fn default() -> Self {
226 0 : Self {
227 0 : max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
228 0 : max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
229 0 : start_as_candidate: false,
230 0 : database_url: None,
231 0 : split_threshold: None,
232 0 : max_split_shards: None,
233 0 : initial_split_threshold: None,
234 0 : initial_split_shards: None,
235 0 : max_secondary_lag_bytes: None,
236 0 : heartbeat_interval: Self::DEFAULT_HEARTBEAT_INTERVAL,
237 0 : long_reconcile_threshold: None,
238 0 : use_https_pageserver_api: false,
239 0 : timelines_onto_safekeepers: false,
240 0 : use_https_safekeeper_api: false,
241 0 : use_local_compute_notifications: true,
242 0 : }
243 0 : }
244 : }
245 :
246 : impl Default for EndpointStorageConf {
247 0 : fn default() -> Self {
248 0 : Self {
249 0 : listen_addr: ENDPOINT_STORAGE_DEFAULT_ADDR,
250 0 : }
251 0 : }
252 : }
253 :
254 : impl NeonBroker {
255 0 : pub fn client_url(&self) -> Url {
256 0 : let url = if let Some(addr) = self.listen_https_addr {
257 0 : format!("https://{}", addr)
258 : } else {
259 0 : format!(
260 0 : "http://{}",
261 0 : self.listen_addr
262 0 : .expect("at least one address should be set")
263 0 : )
264 : };
265 :
266 0 : Url::parse(&url).expect("failed to construct url")
267 0 : }
268 : }
269 :
270 : // neon_local needs to know this subset of pageserver configuration.
271 : // For legacy reasons, this information is duplicated from `pageserver.toml` into `.neon/config`.
272 : // It can get stale if `pageserver.toml` is changed.
273 : // TODO(christian): don't store this at all in `.neon/config`, always load it from `pageserver.toml`
274 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
275 : #[serde(default, deny_unknown_fields)]
276 : pub struct PageServerConf {
277 : pub id: NodeId,
278 : pub listen_pg_addr: String,
279 : pub listen_http_addr: String,
280 : pub listen_https_addr: Option<String>,
281 : pub listen_grpc_addr: Option<String>,
282 : pub pg_auth_type: AuthType,
283 : pub http_auth_type: AuthType,
284 : pub grpc_auth_type: AuthType,
285 : pub no_sync: bool,
286 : }
287 :
288 : impl Default for PageServerConf {
289 0 : fn default() -> Self {
290 0 : Self {
291 0 : id: NodeId(0),
292 0 : listen_pg_addr: String::new(),
293 0 : listen_http_addr: String::new(),
294 0 : listen_https_addr: None,
295 0 : listen_grpc_addr: None,
296 0 : pg_auth_type: AuthType::Trust,
297 0 : http_auth_type: AuthType::Trust,
298 0 : grpc_auth_type: AuthType::Trust,
299 0 : no_sync: false,
300 0 : }
301 0 : }
302 : }
303 :
304 : /// The toml that can be passed to `neon_local init --config`.
305 : /// This is a subset of the `pageserver.toml` configuration.
306 : // TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
307 0 : #[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
308 : pub struct NeonLocalInitPageserverConf {
309 : pub id: NodeId,
310 : pub listen_pg_addr: String,
311 : pub listen_http_addr: String,
312 : pub listen_https_addr: Option<String>,
313 : pub listen_grpc_addr: Option<String>,
314 : pub pg_auth_type: AuthType,
315 : pub http_auth_type: AuthType,
316 : pub grpc_auth_type: AuthType,
317 : #[serde(default, skip_serializing_if = "std::ops::Not::not")]
318 : pub no_sync: bool,
319 : #[serde(flatten)]
320 : pub other: HashMap<String, toml::Value>,
321 : }
322 :
323 : impl From<&NeonLocalInitPageserverConf> for PageServerConf {
324 0 : fn from(conf: &NeonLocalInitPageserverConf) -> Self {
325 0 : let NeonLocalInitPageserverConf {
326 0 : id,
327 0 : listen_pg_addr,
328 0 : listen_http_addr,
329 0 : listen_https_addr,
330 0 : listen_grpc_addr,
331 0 : pg_auth_type,
332 0 : http_auth_type,
333 0 : grpc_auth_type,
334 0 : no_sync,
335 0 : other: _,
336 0 : } = conf;
337 0 : Self {
338 0 : id: *id,
339 0 : listen_pg_addr: listen_pg_addr.clone(),
340 0 : listen_http_addr: listen_http_addr.clone(),
341 0 : listen_https_addr: listen_https_addr.clone(),
342 0 : listen_grpc_addr: listen_grpc_addr.clone(),
343 0 : pg_auth_type: *pg_auth_type,
344 0 : grpc_auth_type: *grpc_auth_type,
345 0 : http_auth_type: *http_auth_type,
346 0 : no_sync: *no_sync,
347 0 : }
348 0 : }
349 : }
350 :
351 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
352 : #[serde(default)]
353 : pub struct SafekeeperConf {
354 : pub id: NodeId,
355 : pub pg_port: u16,
356 : pub pg_tenant_only_port: Option<u16>,
357 : pub http_port: u16,
358 : pub https_port: Option<u16>,
359 : pub sync: bool,
360 : pub remote_storage: Option<String>,
361 : pub backup_threads: Option<u32>,
362 : pub auth_enabled: bool,
363 : pub listen_addr: Option<String>,
364 : }
365 :
366 : impl Default for SafekeeperConf {
367 0 : fn default() -> Self {
368 0 : Self {
369 0 : id: NodeId(0),
370 0 : pg_port: 0,
371 0 : pg_tenant_only_port: None,
372 0 : http_port: 0,
373 0 : https_port: None,
374 0 : sync: true,
375 0 : remote_storage: None,
376 0 : backup_threads: None,
377 0 : auth_enabled: false,
378 0 : listen_addr: None,
379 0 : }
380 0 : }
381 : }
382 :
383 : #[derive(Clone, Copy)]
384 : pub enum InitForceMode {
385 : MustNotExist,
386 : EmptyDirOk,
387 : RemoveAllContents,
388 : }
389 :
390 : impl ValueEnum for InitForceMode {
391 0 : fn value_variants<'a>() -> &'a [Self] {
392 0 : &[
393 0 : Self::MustNotExist,
394 0 : Self::EmptyDirOk,
395 0 : Self::RemoveAllContents,
396 0 : ]
397 0 : }
398 :
399 0 : fn to_possible_value(&self) -> Option<clap::builder::PossibleValue> {
400 0 : Some(clap::builder::PossibleValue::new(match self {
401 0 : InitForceMode::MustNotExist => "must-not-exist",
402 0 : InitForceMode::EmptyDirOk => "empty-dir-ok",
403 0 : InitForceMode::RemoveAllContents => "remove-all-contents",
404 : }))
405 0 : }
406 : }
407 :
408 : impl SafekeeperConf {
409 : /// Compute is served by port on which only tenant scoped tokens allowed, if
410 : /// it is configured.
411 0 : pub fn get_compute_port(&self) -> u16 {
412 0 : self.pg_tenant_only_port.unwrap_or(self.pg_port)
413 0 : }
414 : }
415 :
416 : impl LocalEnv {
417 0 : pub fn pg_distrib_dir_raw(&self) -> PathBuf {
418 0 : self.pg_distrib_dir.clone()
419 0 : }
420 :
421 0 : pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
422 0 : let path = self.pg_distrib_dir.clone();
423 0 :
424 0 : #[allow(clippy::manual_range_patterns)]
425 0 : match pg_version {
426 0 : 14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
427 0 : _ => bail!("Unsupported postgres version: {}", pg_version),
428 : }
429 0 : }
430 :
431 0 : pub fn pg_dir(&self, pg_version: u32, dir_name: &str) -> anyhow::Result<PathBuf> {
432 0 : Ok(self.pg_distrib_dir(pg_version)?.join(dir_name))
433 0 : }
434 :
435 0 : pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
436 0 : self.pg_dir(pg_version, "bin")
437 0 : }
438 :
439 0 : pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
440 0 : self.pg_dir(pg_version, "lib")
441 0 : }
442 :
443 0 : pub fn endpoint_storage_bin(&self) -> PathBuf {
444 0 : self.neon_distrib_dir.join("endpoint_storage")
445 0 : }
446 :
447 0 : pub fn pageserver_bin(&self) -> PathBuf {
448 0 : self.neon_distrib_dir.join("pageserver")
449 0 : }
450 :
451 0 : pub fn storage_controller_bin(&self) -> PathBuf {
452 0 : // Irrespective of configuration, storage controller binary is always
453 0 : // run from the same location as neon_local. This means that for compatibility
454 0 : // tests that run old pageserver/safekeeper, they still run latest storage controller.
455 0 : let neon_local_bin_dir = env::current_exe().unwrap().parent().unwrap().to_owned();
456 0 : neon_local_bin_dir.join("storage_controller")
457 0 : }
458 :
459 0 : pub fn safekeeper_bin(&self) -> PathBuf {
460 0 : self.neon_distrib_dir.join("safekeeper")
461 0 : }
462 :
463 0 : pub fn storage_broker_bin(&self) -> PathBuf {
464 0 : self.neon_distrib_dir.join("storage_broker")
465 0 : }
466 :
467 0 : pub fn endpoints_path(&self) -> PathBuf {
468 0 : self.base_data_dir.join("endpoints")
469 0 : }
470 :
471 0 : pub fn storage_broker_data_dir(&self) -> PathBuf {
472 0 : self.base_data_dir.join("storage_broker")
473 0 : }
474 :
475 0 : pub fn pageserver_data_dir(&self, pageserver_id: NodeId) -> PathBuf {
476 0 : self.base_data_dir
477 0 : .join(format!("pageserver_{pageserver_id}"))
478 0 : }
479 :
480 0 : pub fn safekeeper_data_dir(&self, data_dir_name: &str) -> PathBuf {
481 0 : self.base_data_dir.join("safekeepers").join(data_dir_name)
482 0 : }
483 :
484 0 : pub fn endpoint_storage_data_dir(&self) -> PathBuf {
485 0 : self.base_data_dir.join("endpoint_storage")
486 0 : }
487 :
488 0 : pub fn get_pageserver_conf(&self, id: NodeId) -> anyhow::Result<&PageServerConf> {
489 0 : if let Some(conf) = self.pageservers.iter().find(|node| node.id == id) {
490 0 : Ok(conf)
491 : } else {
492 0 : let have_ids = self
493 0 : .pageservers
494 0 : .iter()
495 0 : .map(|node| format!("{}:{}", node.id, node.listen_http_addr))
496 0 : .collect::<Vec<_>>();
497 0 : let joined = have_ids.join(",");
498 0 : bail!("could not find pageserver {id}, have ids {joined}")
499 : }
500 0 : }
501 :
502 0 : pub fn ssl_ca_cert_path(&self) -> Option<PathBuf> {
503 0 : if self.generate_local_ssl_certs {
504 0 : Some(self.base_data_dir.join("rootCA.crt"))
505 : } else {
506 0 : None
507 : }
508 0 : }
509 :
510 0 : pub fn ssl_ca_key_path(&self) -> Option<PathBuf> {
511 0 : if self.generate_local_ssl_certs {
512 0 : Some(self.base_data_dir.join("rootCA.key"))
513 : } else {
514 0 : None
515 : }
516 0 : }
517 :
518 0 : pub fn generate_ssl_ca_cert(&self) -> anyhow::Result<()> {
519 0 : let cert_path = self.ssl_ca_cert_path().unwrap();
520 0 : let key_path = self.ssl_ca_key_path().unwrap();
521 0 : if !fs::exists(cert_path.as_path())? {
522 0 : generate_ssl_ca_cert(cert_path.as_path(), key_path.as_path())?;
523 0 : }
524 0 : Ok(())
525 0 : }
526 :
527 0 : pub fn generate_ssl_cert(&self, cert_path: &Path, key_path: &Path) -> anyhow::Result<()> {
528 0 : self.generate_ssl_ca_cert()?;
529 0 : generate_ssl_cert(
530 0 : cert_path,
531 0 : key_path,
532 0 : self.ssl_ca_cert_path().unwrap().as_path(),
533 0 : self.ssl_ca_key_path().unwrap().as_path(),
534 0 : )
535 0 : }
536 :
537 : /// Creates HTTP client with local SSL CA certificates.
538 0 : pub fn create_http_client(&self) -> reqwest::Client {
539 0 : let ssl_ca_certs = self.ssl_ca_cert_path().map(|ssl_ca_file| {
540 0 : let buf = std::fs::read(ssl_ca_file).expect("SSL CA file should exist");
541 0 : Certificate::from_pem_bundle(&buf).expect("SSL CA file should be valid")
542 0 : });
543 0 :
544 0 : let mut http_client = reqwest::Client::builder();
545 0 : for ssl_ca_cert in ssl_ca_certs.unwrap_or_default() {
546 0 : http_client = http_client.add_root_certificate(ssl_ca_cert);
547 0 : }
548 :
549 0 : http_client
550 0 : .build()
551 0 : .expect("HTTP client should construct with no error")
552 0 : }
553 :
554 : /// Inspect the base data directory and extract the instance id and instance directory path
555 : /// for all storage controller instances
556 0 : pub async fn storage_controller_instances(&self) -> std::io::Result<Vec<(u8, PathBuf)>> {
557 0 : let mut instances = Vec::default();
558 :
559 0 : let dir = std::fs::read_dir(self.base_data_dir.clone())?;
560 0 : for dentry in dir {
561 0 : let dentry = dentry?;
562 0 : let is_dir = dentry.metadata()?.is_dir();
563 0 : let filename = dentry.file_name().into_string().unwrap();
564 0 : let parsed_instance_id = match filename.strip_prefix("storage_controller_") {
565 0 : Some(suffix) => suffix.parse::<u8>().ok(),
566 0 : None => None,
567 : };
568 :
569 0 : let is_instance_dir = is_dir && parsed_instance_id.is_some();
570 :
571 0 : if !is_instance_dir {
572 0 : continue;
573 0 : }
574 0 :
575 0 : instances.push((
576 0 : parsed_instance_id.expect("Checked previously"),
577 0 : dentry.path(),
578 0 : ));
579 : }
580 :
581 0 : Ok(instances)
582 0 : }
583 :
584 0 : pub fn register_branch_mapping(
585 0 : &mut self,
586 0 : branch_name: String,
587 0 : tenant_id: TenantId,
588 0 : timeline_id: TimelineId,
589 0 : ) -> anyhow::Result<()> {
590 0 : let existing_values = self
591 0 : .branch_name_mappings
592 0 : .entry(branch_name.clone())
593 0 : .or_default();
594 0 :
595 0 : let existing_ids = existing_values
596 0 : .iter()
597 0 : .find(|(existing_tenant_id, _)| existing_tenant_id == &tenant_id);
598 :
599 0 : if let Some((_, old_timeline_id)) = existing_ids {
600 0 : if old_timeline_id == &timeline_id {
601 0 : Ok(())
602 : } else {
603 0 : bail!(
604 0 : "branch '{branch_name}' is already mapped to timeline {old_timeline_id}, cannot map to another timeline {timeline_id}"
605 0 : );
606 : }
607 : } else {
608 0 : existing_values.push((tenant_id, timeline_id));
609 0 : Ok(())
610 : }
611 0 : }
612 :
613 0 : pub fn get_branch_timeline_id(
614 0 : &self,
615 0 : branch_name: &str,
616 0 : tenant_id: TenantId,
617 0 : ) -> Option<TimelineId> {
618 0 : self.branch_name_mappings
619 0 : .get(branch_name)?
620 0 : .iter()
621 0 : .find(|(mapped_tenant_id, _)| mapped_tenant_id == &tenant_id)
622 0 : .map(|&(_, timeline_id)| timeline_id)
623 0 : }
624 :
625 0 : pub fn timeline_name_mappings(&self) -> HashMap<TenantTimelineId, String> {
626 0 : self.branch_name_mappings
627 0 : .iter()
628 0 : .flat_map(|(name, tenant_timelines)| {
629 0 : tenant_timelines.iter().map(|&(tenant_id, timeline_id)| {
630 0 : (TenantTimelineId::new(tenant_id, timeline_id), name.clone())
631 0 : })
632 0 : })
633 0 : .collect()
634 0 : }
635 :
636 : /// Construct `Self` from on-disk state.
637 0 : pub fn load_config(repopath: &Path) -> anyhow::Result<Self> {
638 0 : if !repopath.exists() {
639 0 : bail!(
640 0 : "Neon config is not found in {}. You need to run 'neon_local init' first",
641 0 : repopath.to_str().unwrap()
642 0 : );
643 0 : }
644 :
645 : // TODO: check that it looks like a neon repository
646 :
647 : // load and parse file
648 0 : let config_file_contents = fs::read_to_string(repopath.join("config"))?;
649 0 : let on_disk_config: OnDiskConfig = toml::from_str(config_file_contents.as_str())?;
650 0 : let mut env = {
651 0 : let OnDiskConfig {
652 0 : pg_distrib_dir,
653 0 : neon_distrib_dir,
654 0 : default_tenant_id,
655 0 : private_key_path,
656 0 : public_key_path,
657 0 : broker,
658 0 : storage_controller,
659 0 : pageservers,
660 0 : safekeepers,
661 0 : control_plane_api,
662 0 : control_plane_hooks_api,
663 0 : control_plane_compute_hook_api: _,
664 0 : branch_name_mappings,
665 0 : generate_local_ssl_certs,
666 0 : endpoint_storage,
667 0 : } = on_disk_config;
668 0 : LocalEnv {
669 0 : base_data_dir: repopath.to_owned(),
670 0 : pg_distrib_dir,
671 0 : neon_distrib_dir,
672 0 : default_tenant_id,
673 0 : private_key_path,
674 0 : public_key_path,
675 0 : broker,
676 0 : storage_controller,
677 0 : pageservers,
678 0 : safekeepers,
679 0 : control_plane_api: control_plane_api.unwrap(),
680 0 : control_plane_hooks_api,
681 0 : branch_name_mappings,
682 0 : generate_local_ssl_certs,
683 0 : endpoint_storage,
684 0 : }
685 0 : };
686 0 :
687 0 : // The source of truth for pageserver configuration is the pageserver.toml.
688 0 : assert!(
689 0 : env.pageservers.is_empty(),
690 0 : "we ensure this during deserialization"
691 : );
692 0 : env.pageservers = {
693 0 : let iter = std::fs::read_dir(repopath).context("open dir")?;
694 0 : let mut pageservers = Vec::new();
695 0 : for res in iter {
696 0 : let dentry = res?;
697 : const PREFIX: &str = "pageserver_";
698 0 : let dentry_name = dentry
699 0 : .file_name()
700 0 : .into_string()
701 0 : .ok()
702 0 : .with_context(|| format!("non-utf8 dentry: {:?}", dentry.path()))
703 0 : .unwrap();
704 0 : if !dentry_name.starts_with(PREFIX) {
705 0 : continue;
706 0 : }
707 0 : if !dentry.file_type().context("determine file type")?.is_dir() {
708 0 : anyhow::bail!("expected a directory, got {:?}", dentry.path());
709 0 : }
710 0 : let id = dentry_name[PREFIX.len()..]
711 0 : .parse::<NodeId>()
712 0 : .with_context(|| format!("parse id from {:?}", dentry.path()))?;
713 : // TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
714 0 : #[derive(serde::Serialize, serde::Deserialize)]
715 : // (allow unknown fields, unlike PageServerConf)
716 : struct PageserverConfigTomlSubset {
717 : listen_pg_addr: String,
718 : listen_http_addr: String,
719 : listen_https_addr: Option<String>,
720 : listen_grpc_addr: Option<String>,
721 : pg_auth_type: AuthType,
722 : http_auth_type: AuthType,
723 : grpc_auth_type: AuthType,
724 : #[serde(default)]
725 : no_sync: bool,
726 : }
727 0 : let config_toml_path = dentry.path().join("pageserver.toml");
728 0 : let config_toml: PageserverConfigTomlSubset = toml_edit::de::from_str(
729 0 : &std::fs::read_to_string(&config_toml_path)
730 0 : .with_context(|| format!("read {:?}", config_toml_path))?,
731 : )
732 0 : .context("parse pageserver.toml")?;
733 0 : let identity_toml_path = dentry.path().join("identity.toml");
734 0 : #[derive(serde::Serialize, serde::Deserialize)]
735 : struct IdentityTomlSubset {
736 : id: NodeId,
737 : }
738 0 : let identity_toml: IdentityTomlSubset = toml_edit::de::from_str(
739 0 : &std::fs::read_to_string(&identity_toml_path)
740 0 : .with_context(|| format!("read {:?}", identity_toml_path))?,
741 : )
742 0 : .context("parse identity.toml")?;
743 : let PageserverConfigTomlSubset {
744 0 : listen_pg_addr,
745 0 : listen_http_addr,
746 0 : listen_https_addr,
747 0 : listen_grpc_addr,
748 0 : pg_auth_type,
749 0 : http_auth_type,
750 0 : grpc_auth_type,
751 0 : no_sync,
752 0 : } = config_toml;
753 0 : let IdentityTomlSubset {
754 0 : id: identity_toml_id,
755 0 : } = identity_toml;
756 0 : let conf = PageServerConf {
757 : id: {
758 0 : anyhow::ensure!(
759 0 : identity_toml_id == id,
760 0 : "id mismatch: identity.toml:id={identity_toml_id} pageserver_(.*) id={id}",
761 : );
762 0 : id
763 0 : },
764 0 : listen_pg_addr,
765 0 : listen_http_addr,
766 0 : listen_https_addr,
767 0 : listen_grpc_addr,
768 0 : pg_auth_type,
769 0 : http_auth_type,
770 0 : grpc_auth_type,
771 0 : no_sync,
772 0 : };
773 0 : pageservers.push(conf);
774 : }
775 0 : pageservers
776 0 : };
777 0 :
778 0 : Ok(env)
779 0 : }
780 :
781 0 : pub fn persist_config(&self) -> anyhow::Result<()> {
782 0 : Self::persist_config_impl(
783 0 : &self.base_data_dir,
784 0 : &OnDiskConfig {
785 0 : pg_distrib_dir: self.pg_distrib_dir.clone(),
786 0 : neon_distrib_dir: self.neon_distrib_dir.clone(),
787 0 : default_tenant_id: self.default_tenant_id,
788 0 : private_key_path: self.private_key_path.clone(),
789 0 : public_key_path: self.public_key_path.clone(),
790 0 : broker: self.broker.clone(),
791 0 : storage_controller: self.storage_controller.clone(),
792 0 : pageservers: vec![], // it's skip_serializing anyway
793 0 : safekeepers: self.safekeepers.clone(),
794 0 : control_plane_api: Some(self.control_plane_api.clone()),
795 0 : control_plane_hooks_api: self.control_plane_hooks_api.clone(),
796 0 : control_plane_compute_hook_api: None,
797 0 : branch_name_mappings: self.branch_name_mappings.clone(),
798 0 : generate_local_ssl_certs: self.generate_local_ssl_certs,
799 0 : endpoint_storage: self.endpoint_storage.clone(),
800 0 : },
801 0 : )
802 0 : }
803 :
804 0 : pub fn persist_config_impl(base_path: &Path, config: &OnDiskConfig) -> anyhow::Result<()> {
805 0 : let conf_content = &toml::to_string_pretty(config)?;
806 0 : let target_config_path = base_path.join("config");
807 0 : fs::write(&target_config_path, conf_content).with_context(|| {
808 0 : format!(
809 0 : "Failed to write config file into path '{}'",
810 0 : target_config_path.display()
811 0 : )
812 0 : })
813 0 : }
814 :
815 : // this function is used only for testing purposes in CLI e g generate tokens during init
816 0 : pub fn generate_auth_token<S: Serialize>(&self, claims: &S) -> anyhow::Result<String> {
817 0 : let key = self.read_private_key()?;
818 0 : encode_from_key_file(claims, &key)
819 0 : }
820 :
821 : /// Get the path to the private key.
822 0 : pub fn get_private_key_path(&self) -> PathBuf {
823 0 : if self.private_key_path.is_absolute() {
824 0 : self.private_key_path.to_path_buf()
825 : } else {
826 0 : self.base_data_dir.join(&self.private_key_path)
827 : }
828 0 : }
829 :
830 : /// Get the path to the public key.
831 0 : pub fn get_public_key_path(&self) -> PathBuf {
832 0 : if self.public_key_path.is_absolute() {
833 0 : self.public_key_path.to_path_buf()
834 : } else {
835 0 : self.base_data_dir.join(&self.public_key_path)
836 : }
837 0 : }
838 :
839 : /// Read the contents of the private key file.
840 0 : pub fn read_private_key(&self) -> anyhow::Result<Pem> {
841 0 : let private_key_path = self.get_private_key_path();
842 0 : let pem = pem::parse(fs::read(private_key_path)?)?;
843 0 : Ok(pem)
844 0 : }
845 :
846 : /// Read the contents of the public key file.
847 0 : pub fn read_public_key(&self) -> anyhow::Result<Pem> {
848 0 : let public_key_path = self.get_public_key_path();
849 0 : let pem = pem::parse(fs::read(public_key_path)?)?;
850 0 : Ok(pem)
851 0 : }
852 :
853 : /// Materialize the [`NeonLocalInitConf`] to disk. Called during [`neon_local init`].
854 0 : pub fn init(conf: NeonLocalInitConf, force: &InitForceMode) -> anyhow::Result<()> {
855 0 : let base_path = base_path();
856 0 : assert_ne!(base_path, Path::new(""));
857 0 : let base_path = &base_path;
858 0 :
859 0 : // create base_path dir
860 0 : if base_path.exists() {
861 0 : match force {
862 : InitForceMode::MustNotExist => {
863 0 : bail!(
864 0 : "directory '{}' already exists. Perhaps already initialized?",
865 0 : base_path.display()
866 0 : );
867 : }
868 : InitForceMode::EmptyDirOk => {
869 0 : if let Some(res) = std::fs::read_dir(base_path)?.next() {
870 0 : res.context("check if directory is empty")?;
871 0 : anyhow::bail!("directory not empty: {base_path:?}");
872 0 : }
873 : }
874 : InitForceMode::RemoveAllContents => {
875 0 : println!("removing all contents of '{}'", base_path.display());
876 : // instead of directly calling `remove_dir_all`, we keep the original dir but removing
877 : // all contents inside. This helps if the developer symbol links another directory (i.e.,
878 : // S3 local SSD) to the `.neon` base directory.
879 0 : for entry in std::fs::read_dir(base_path)? {
880 0 : let entry = entry?;
881 0 : let path = entry.path();
882 0 : if path.is_dir() {
883 0 : fs::remove_dir_all(&path)?;
884 : } else {
885 0 : fs::remove_file(&path)?;
886 : }
887 : }
888 : }
889 : }
890 0 : }
891 0 : if !base_path.exists() {
892 0 : fs::create_dir(base_path)?;
893 0 : }
894 :
895 : let NeonLocalInitConf {
896 0 : pg_distrib_dir,
897 0 : neon_distrib_dir,
898 0 : default_tenant_id,
899 0 : broker,
900 0 : storage_controller,
901 0 : pageservers,
902 0 : safekeepers,
903 0 : control_plane_api,
904 0 : generate_local_ssl_certs,
905 0 : control_plane_hooks_api,
906 0 : endpoint_storage,
907 0 : } = conf;
908 0 :
909 0 : // Find postgres binaries.
910 0 : // Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "pg_install".
911 0 : // Note that later in the code we assume, that distrib dirs follow the same pattern
912 0 : // for all postgres versions.
913 0 : let pg_distrib_dir = pg_distrib_dir.unwrap_or_else(|| {
914 0 : if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
915 0 : postgres_bin.into()
916 : } else {
917 0 : let cwd = env::current_dir().unwrap();
918 0 : cwd.join("pg_install")
919 : }
920 0 : });
921 0 :
922 0 : // Find neon binaries.
923 0 : let neon_distrib_dir = neon_distrib_dir
924 0 : .unwrap_or_else(|| env::current_exe().unwrap().parent().unwrap().to_owned());
925 0 :
926 0 : // Generate keypair for JWT.
927 0 : //
928 0 : // The keypair is only needed if authentication is enabled in any of the
929 0 : // components. For convenience, we generate the keypair even if authentication
930 0 : // is not enabled, so that you can easily enable it after the initialization
931 0 : // step.
932 0 : generate_auth_keys(
933 0 : base_path.join("auth_private_key.pem").as_path(),
934 0 : base_path.join("auth_public_key.pem").as_path(),
935 0 : )
936 0 : .context("generate auth keys")?;
937 0 : let private_key_path = PathBuf::from("auth_private_key.pem");
938 0 : let public_key_path = PathBuf::from("auth_public_key.pem");
939 0 :
940 0 : // create the runtime type because the remaining initialization code below needs
941 0 : // a LocalEnv instance op operation
942 0 : // TODO: refactor to avoid this, LocalEnv should only be constructed from on-disk state
943 0 : let env = LocalEnv {
944 0 : base_data_dir: base_path.clone(),
945 0 : pg_distrib_dir,
946 0 : neon_distrib_dir,
947 0 : default_tenant_id: Some(default_tenant_id),
948 0 : private_key_path,
949 0 : public_key_path,
950 0 : broker,
951 0 : storage_controller: storage_controller.unwrap_or_default(),
952 0 : pageservers: pageservers.iter().map(Into::into).collect(),
953 0 : safekeepers,
954 0 : control_plane_api: control_plane_api.unwrap(),
955 0 : control_plane_hooks_api,
956 0 : branch_name_mappings: Default::default(),
957 0 : generate_local_ssl_certs,
958 0 : endpoint_storage,
959 0 : };
960 0 :
961 0 : if generate_local_ssl_certs {
962 0 : env.generate_ssl_ca_cert()?;
963 0 : }
964 :
965 : // create endpoints dir
966 0 : fs::create_dir_all(env.endpoints_path())?;
967 :
968 : // create storage broker dir
969 0 : fs::create_dir_all(env.storage_broker_data_dir())?;
970 0 : StorageBroker::from_env(&env)
971 0 : .initialize()
972 0 : .context("storage broker init failed")?;
973 :
974 : // create safekeeper dirs
975 0 : for safekeeper in &env.safekeepers {
976 0 : fs::create_dir_all(SafekeeperNode::datadir_path_by_id(&env, safekeeper.id))?;
977 0 : SafekeeperNode::from_env(&env, safekeeper)
978 0 : .initialize()
979 0 : .context("safekeeper init failed")?;
980 : }
981 :
982 : // initialize pageserver state
983 0 : for (i, ps) in pageservers.into_iter().enumerate() {
984 0 : let runtime_ps = &env.pageservers[i];
985 0 : assert_eq!(&PageServerConf::from(&ps), runtime_ps);
986 0 : fs::create_dir(env.pageserver_data_dir(ps.id))?;
987 0 : PageServerNode::from_env(&env, runtime_ps)
988 0 : .initialize(ps)
989 0 : .context("pageserver init failed")?;
990 : }
991 :
992 0 : EndpointStorage::from_env(&env)
993 0 : .init()
994 0 : .context("object storage init failed")?;
995 :
996 : // setup remote remote location for default LocalFs remote storage
997 0 : std::fs::create_dir_all(env.base_data_dir.join(PAGESERVER_REMOTE_STORAGE_DIR))?;
998 0 : std::fs::create_dir_all(env.base_data_dir.join(ENDPOINT_STORAGE_REMOTE_STORAGE_DIR))?;
999 :
1000 0 : env.persist_config()
1001 0 : }
1002 : }
1003 :
1004 0 : pub fn base_path() -> PathBuf {
1005 0 : let path = match std::env::var_os("NEON_REPO_DIR") {
1006 0 : Some(val) => {
1007 0 : let path = PathBuf::from(val);
1008 0 : if !path.is_absolute() {
1009 : // repeat the env var in the error because our default is always absolute
1010 0 : panic!("NEON_REPO_DIR must be an absolute path, got {path:?}");
1011 0 : }
1012 0 : path
1013 : }
1014 : None => {
1015 0 : let pwd = std::env::current_dir()
1016 0 : // technically this can fail but it's quite unlikeley
1017 0 : .expect("determine current directory");
1018 0 : let pwd_abs = pwd.canonicalize().expect("canonicalize current directory");
1019 0 : pwd_abs.join(".neon")
1020 : }
1021 : };
1022 0 : assert!(path.is_absolute());
1023 0 : path
1024 0 : }
1025 :
1026 : /// Generate a public/private key pair for JWT authentication
1027 0 : fn generate_auth_keys(private_key_path: &Path, public_key_path: &Path) -> anyhow::Result<()> {
1028 : // Generate the key pair
1029 : //
1030 : // openssl genpkey -algorithm ed25519 -out auth_private_key.pem
1031 0 : let keygen_output = Command::new("openssl")
1032 0 : .arg("genpkey")
1033 0 : .args(["-algorithm", "ed25519"])
1034 0 : .args(["-out", private_key_path.to_str().unwrap()])
1035 0 : .stdout(Stdio::null())
1036 0 : .output()
1037 0 : .context("failed to generate auth private key")?;
1038 0 : if !keygen_output.status.success() {
1039 0 : bail!(
1040 0 : "openssl failed: '{}'",
1041 0 : String::from_utf8_lossy(&keygen_output.stderr)
1042 0 : );
1043 0 : }
1044 :
1045 : // Extract the public key from the private key file
1046 : //
1047 : // openssl pkey -in auth_private_key.pem -pubout -out auth_public_key.pem
1048 0 : let keygen_output = Command::new("openssl")
1049 0 : .arg("pkey")
1050 0 : .args(["-in", private_key_path.to_str().unwrap()])
1051 0 : .arg("-pubout")
1052 0 : .args(["-out", public_key_path.to_str().unwrap()])
1053 0 : .output()
1054 0 : .context("failed to extract public key from private key")?;
1055 0 : if !keygen_output.status.success() {
1056 0 : bail!(
1057 0 : "openssl failed: '{}'",
1058 0 : String::from_utf8_lossy(&keygen_output.stderr)
1059 0 : );
1060 0 : }
1061 0 :
1062 0 : Ok(())
1063 0 : }
1064 :
1065 0 : fn generate_ssl_ca_cert(cert_path: &Path, key_path: &Path) -> anyhow::Result<()> {
1066 : // openssl req -x509 -newkey rsa:2048 -nodes -subj "/CN=Neon Local CA" -days 36500 \
1067 : // -out rootCA.crt -keyout rootCA.key
1068 0 : let keygen_output = Command::new("openssl")
1069 0 : .args([
1070 0 : "req", "-x509", "-newkey", "ed25519", "-nodes", "-days", "36500",
1071 0 : ])
1072 0 : .args(["-subj", "/CN=Neon Local CA"])
1073 0 : .args(["-out", cert_path.to_str().unwrap()])
1074 0 : .args(["-keyout", key_path.to_str().unwrap()])
1075 0 : .output()
1076 0 : .context("failed to generate CA certificate")?;
1077 0 : if !keygen_output.status.success() {
1078 0 : bail!(
1079 0 : "openssl failed: '{}'",
1080 0 : String::from_utf8_lossy(&keygen_output.stderr)
1081 0 : );
1082 0 : }
1083 0 : Ok(())
1084 0 : }
1085 :
1086 0 : fn generate_ssl_cert(
1087 0 : cert_path: &Path,
1088 0 : key_path: &Path,
1089 0 : ca_cert_path: &Path,
1090 0 : ca_key_path: &Path,
1091 0 : ) -> anyhow::Result<()> {
1092 0 : // Generate Certificate Signing Request (CSR).
1093 0 : let mut csr_path = cert_path.to_path_buf();
1094 0 : csr_path.set_extension(".csr");
1095 :
1096 : // openssl req -new -nodes -newkey rsa:2048 -keyout server.key -out server.csr \
1097 : // -subj "/CN=localhost" -addext "subjectAltName=DNS:localhost,IP:127.0.0.1"
1098 0 : let keygen_output = Command::new("openssl")
1099 0 : .args(["req", "-new", "-nodes"])
1100 0 : .args(["-newkey", "ed25519"])
1101 0 : .args(["-subj", "/CN=localhost"])
1102 0 : .args(["-addext", "subjectAltName=DNS:localhost,IP:127.0.0.1"])
1103 0 : .args(["-keyout", key_path.to_str().unwrap()])
1104 0 : .args(["-out", csr_path.to_str().unwrap()])
1105 0 : .output()
1106 0 : .context("failed to generate CSR")?;
1107 0 : if !keygen_output.status.success() {
1108 0 : bail!(
1109 0 : "openssl failed: '{}'",
1110 0 : String::from_utf8_lossy(&keygen_output.stderr)
1111 0 : );
1112 0 : }
1113 :
1114 : // Sign CSR with CA key.
1115 : //
1116 : // openssl x509 -req -in server.csr -CA rootCA.crt -CAkey rootCA.key -CAcreateserial \
1117 : // -out server.crt -days 36500 -copy_extensions copyall
1118 0 : let keygen_output = Command::new("openssl")
1119 0 : .args(["x509", "-req"])
1120 0 : .args(["-in", csr_path.to_str().unwrap()])
1121 0 : .args(["-CA", ca_cert_path.to_str().unwrap()])
1122 0 : .args(["-CAkey", ca_key_path.to_str().unwrap()])
1123 0 : .arg("-CAcreateserial")
1124 0 : .args(["-out", cert_path.to_str().unwrap()])
1125 0 : .args(["-days", "36500"])
1126 0 : .args(["-copy_extensions", "copyall"])
1127 0 : .output()
1128 0 : .context("failed to sign CSR")?;
1129 0 : if !keygen_output.status.success() {
1130 0 : bail!(
1131 0 : "openssl failed: '{}'",
1132 0 : String::from_utf8_lossy(&keygen_output.stderr)
1133 0 : );
1134 0 : }
1135 0 :
1136 0 : // Remove CSR file as it's not needed anymore.
1137 0 : fs::remove_file(csr_path)?;
1138 :
1139 0 : Ok(())
1140 0 : }
|
