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::{IpAddr, Ipv4Addr, 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 postgres_backend::AuthType;
16 : use reqwest::Url;
17 : use serde::{Deserialize, Serialize};
18 : use utils::auth::{Claims, encode_from_key_file};
19 : use utils::id::{NodeId, TenantId, TenantTimelineId, TimelineId};
20 :
21 : use crate::pageserver::{PAGESERVER_REMOTE_STORAGE_DIR, PageServerNode};
22 : use crate::safekeeper::SafekeeperNode;
23 :
24 : pub const DEFAULT_PG_VERSION: u32 = 16;
25 :
26 : //
27 : // This data structures represents neon_local CLI config
28 : //
29 : // It is deserialized from the .neon/config file, or the config file passed
30 : // to 'neon_local init --config=<path>' option. See control_plane/simple.conf for
31 : // an example.
32 : //
33 : #[derive(PartialEq, Eq, Clone, Debug)]
34 : pub struct LocalEnv {
35 : // Base directory for all the nodes (the pageserver, safekeepers and
36 : // compute endpoints).
37 : //
38 : // This is not stored in the config file. Rather, this is the path where the
39 : // config file itself is. It is read from the NEON_REPO_DIR env variable which
40 : // must be an absolute path. If the env var is not set, $PWD/.neon is used.
41 : pub base_data_dir: PathBuf,
42 :
43 : // Path to postgres distribution. It's expected that "bin", "include",
44 : // "lib", "share" from postgres distribution are there. If at some point
45 : // in time we will be able to run against vanilla postgres we may split that
46 : // to four separate paths and match OS-specific installation layout.
47 : pub pg_distrib_dir: PathBuf,
48 :
49 : // Path to pageserver binary.
50 : pub neon_distrib_dir: PathBuf,
51 :
52 : // Default tenant ID to use with the 'neon_local' command line utility, when
53 : // --tenant_id is not explicitly specified.
54 : pub default_tenant_id: Option<TenantId>,
55 :
56 : // used to issue tokens during e.g pg start
57 : pub private_key_path: PathBuf,
58 :
59 : pub broker: NeonBroker,
60 :
61 : // Configuration for the storage controller (1 per neon_local environment)
62 : pub storage_controller: NeonStorageControllerConf,
63 :
64 : /// This Vec must always contain at least one pageserver
65 : /// Populdated by [`Self::load_config`] from the individual `pageserver.toml`s.
66 : /// NB: not used anymore except for informing users that they need to change their `.neon/config`.
67 : pub pageservers: Vec<PageServerConf>,
68 :
69 : pub safekeepers: Vec<SafekeeperConf>,
70 :
71 : // Control plane upcall API for pageserver: if None, we will not run storage_controller If set, this will
72 : // be propagated into each pageserver's configuration.
73 : pub control_plane_api: Url,
74 :
75 : // Control plane upcall APIs for storage controller. If set, this will be propagated into the
76 : // storage controller's configuration.
77 : pub control_plane_hooks_api: Option<Url>,
78 :
79 : /// Keep human-readable aliases in memory (and persist them to config), to hide ZId hex strings from the user.
80 : // A `HashMap<String, HashMap<TenantId, TimelineId>>` would be more appropriate here,
81 : // but deserialization into a generic toml object as `toml::Value::try_from` fails with an error.
82 : // https://toml.io/en/v1.0.0 does not contain a concept of "a table inside another table".
83 : pub branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
84 :
85 : /// Flag to generate SSL certificates for components that need it.
86 : /// Also generates root CA certificate that is used to sign all other certificates.
87 : pub generate_local_ssl_certs: bool,
88 : }
89 :
90 : /// On-disk state stored in `.neon/config`.
91 0 : #[derive(PartialEq, Eq, Clone, Debug, Default, Serialize, Deserialize)]
92 : #[serde(default, deny_unknown_fields)]
93 : pub struct OnDiskConfig {
94 : pub pg_distrib_dir: PathBuf,
95 : pub neon_distrib_dir: PathBuf,
96 : pub default_tenant_id: Option<TenantId>,
97 : pub private_key_path: PathBuf,
98 : pub broker: NeonBroker,
99 : pub storage_controller: NeonStorageControllerConf,
100 : #[serde(
101 : skip_serializing,
102 : deserialize_with = "fail_if_pageservers_field_specified"
103 : )]
104 : pub pageservers: Vec<PageServerConf>,
105 : pub safekeepers: Vec<SafekeeperConf>,
106 : pub control_plane_api: Option<Url>,
107 : pub control_plane_hooks_api: Option<Url>,
108 : pub control_plane_compute_hook_api: Option<Url>,
109 : branch_name_mappings: HashMap<String, Vec<(TenantId, TimelineId)>>,
110 : // Note: skip serializing because in compat tests old storage controller fails
111 : // to load new config file. May be removed after this field is in release branch.
112 : #[serde(skip_serializing_if = "std::ops::Not::not")]
113 : pub generate_local_ssl_certs: bool,
114 : }
115 :
116 0 : fn fail_if_pageservers_field_specified<'de, D>(_: D) -> Result<Vec<PageServerConf>, D::Error>
117 0 : where
118 0 : D: serde::Deserializer<'de>,
119 0 : {
120 0 : Err(serde::de::Error::custom(
121 0 : "The 'pageservers' field is no longer used; pageserver.toml is now authoritative; \
122 0 : Please remove the `pageservers` from your .neon/config.",
123 0 : ))
124 0 : }
125 :
126 : /// The description of the neon_local env to be initialized by `neon_local init --config`.
127 0 : #[derive(Clone, Debug, Deserialize)]
128 : #[serde(deny_unknown_fields)]
129 : pub struct NeonLocalInitConf {
130 : // TODO: do we need this? Seems unused
131 : pub pg_distrib_dir: Option<PathBuf>,
132 : // TODO: do we need this? Seems unused
133 : pub neon_distrib_dir: Option<PathBuf>,
134 : pub default_tenant_id: TenantId,
135 : pub broker: NeonBroker,
136 : pub storage_controller: Option<NeonStorageControllerConf>,
137 : pub pageservers: Vec<NeonLocalInitPageserverConf>,
138 : pub safekeepers: Vec<SafekeeperConf>,
139 : pub control_plane_api: Option<Url>,
140 : pub control_plane_hooks_api: Option<Url>,
141 : pub generate_local_ssl_certs: bool,
142 : }
143 :
144 : /// Broker config for cluster internal communication.
145 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
146 : #[serde(default)]
147 : pub struct NeonBroker {
148 : /// Broker listen address for storage nodes coordination, e.g. '127.0.0.1:50051'.
149 : pub listen_addr: SocketAddr,
150 : }
151 :
152 : /// A part of storage controller's config the neon_local knows about.
153 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
154 : #[serde(default)]
155 : pub struct NeonStorageControllerConf {
156 : /// Heartbeat timeout before marking a node offline
157 : #[serde(with = "humantime_serde")]
158 : pub max_offline: Duration,
159 :
160 : #[serde(with = "humantime_serde")]
161 : pub max_warming_up: Duration,
162 :
163 : pub start_as_candidate: bool,
164 :
165 : /// Database url used when running multiple storage controller instances
166 : pub database_url: Option<SocketAddr>,
167 :
168 : /// Thresholds for auto-splitting a tenant into shards.
169 : pub split_threshold: Option<u64>,
170 : pub max_split_shards: Option<u8>,
171 : pub initial_split_threshold: Option<u64>,
172 : pub initial_split_shards: Option<u8>,
173 :
174 : pub max_secondary_lag_bytes: Option<u64>,
175 :
176 : #[serde(with = "humantime_serde")]
177 : pub heartbeat_interval: Duration,
178 :
179 : #[serde(with = "humantime_serde")]
180 : pub long_reconcile_threshold: Option<Duration>,
181 :
182 : pub use_https_pageserver_api: bool,
183 :
184 : pub timelines_onto_safekeepers: bool,
185 :
186 : pub use_https_safekeeper_api: bool,
187 :
188 : pub use_local_compute_notifications: bool,
189 : }
190 :
191 : impl NeonStorageControllerConf {
192 : // Use a shorter pageserver unavailability interval than the default to speed up tests.
193 : const DEFAULT_MAX_OFFLINE_INTERVAL: std::time::Duration = std::time::Duration::from_secs(10);
194 :
195 : const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
196 :
197 : // Very tight heartbeat interval to speed up tests
198 : const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(1000);
199 : }
200 :
201 : impl Default for NeonStorageControllerConf {
202 0 : fn default() -> Self {
203 0 : Self {
204 0 : max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
205 0 : max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
206 0 : start_as_candidate: false,
207 0 : database_url: None,
208 0 : split_threshold: None,
209 0 : max_split_shards: None,
210 0 : initial_split_threshold: None,
211 0 : initial_split_shards: None,
212 0 : max_secondary_lag_bytes: None,
213 0 : heartbeat_interval: Self::DEFAULT_HEARTBEAT_INTERVAL,
214 0 : long_reconcile_threshold: None,
215 0 : use_https_pageserver_api: false,
216 0 : timelines_onto_safekeepers: false,
217 0 : use_https_safekeeper_api: false,
218 0 : use_local_compute_notifications: true,
219 0 : }
220 0 : }
221 : }
222 :
223 : // Dummy Default impl to satisfy Deserialize derive.
224 : impl Default for NeonBroker {
225 0 : fn default() -> Self {
226 0 : NeonBroker {
227 0 : listen_addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::new(0, 0, 0, 0)), 0),
228 0 : }
229 0 : }
230 : }
231 :
232 : impl NeonBroker {
233 0 : pub fn client_url(&self) -> Url {
234 0 : Url::parse(&format!("http://{}", self.listen_addr)).expect("failed to construct url")
235 0 : }
236 : }
237 :
238 : // neon_local needs to know this subset of pageserver configuration.
239 : // For legacy reasons, this information is duplicated from `pageserver.toml` into `.neon/config`.
240 : // It can get stale if `pageserver.toml` is changed.
241 : // TODO(christian): don't store this at all in `.neon/config`, always load it from `pageserver.toml`
242 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
243 : #[serde(default, deny_unknown_fields)]
244 : pub struct PageServerConf {
245 : pub id: NodeId,
246 : pub listen_pg_addr: String,
247 : pub listen_http_addr: String,
248 : pub listen_https_addr: Option<String>,
249 : pub pg_auth_type: AuthType,
250 : pub http_auth_type: AuthType,
251 : pub no_sync: bool,
252 : }
253 :
254 : impl Default for PageServerConf {
255 0 : fn default() -> Self {
256 0 : Self {
257 0 : id: NodeId(0),
258 0 : listen_pg_addr: String::new(),
259 0 : listen_http_addr: String::new(),
260 0 : listen_https_addr: None,
261 0 : pg_auth_type: AuthType::Trust,
262 0 : http_auth_type: AuthType::Trust,
263 0 : no_sync: false,
264 0 : }
265 0 : }
266 : }
267 :
268 : /// The toml that can be passed to `neon_local init --config`.
269 : /// This is a subset of the `pageserver.toml` configuration.
270 : // TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
271 0 : #[derive(Clone, Debug, serde::Deserialize, serde::Serialize)]
272 : pub struct NeonLocalInitPageserverConf {
273 : pub id: NodeId,
274 : pub listen_pg_addr: String,
275 : pub listen_http_addr: String,
276 : pub listen_https_addr: Option<String>,
277 : pub pg_auth_type: AuthType,
278 : pub http_auth_type: AuthType,
279 : #[serde(default, skip_serializing_if = "std::ops::Not::not")]
280 : pub no_sync: bool,
281 : #[serde(flatten)]
282 : pub other: HashMap<String, toml::Value>,
283 : }
284 :
285 : impl From<&NeonLocalInitPageserverConf> for PageServerConf {
286 0 : fn from(conf: &NeonLocalInitPageserverConf) -> Self {
287 0 : let NeonLocalInitPageserverConf {
288 0 : id,
289 0 : listen_pg_addr,
290 0 : listen_http_addr,
291 0 : listen_https_addr,
292 0 : pg_auth_type,
293 0 : http_auth_type,
294 0 : no_sync,
295 0 : other: _,
296 0 : } = conf;
297 0 : Self {
298 0 : id: *id,
299 0 : listen_pg_addr: listen_pg_addr.clone(),
300 0 : listen_http_addr: listen_http_addr.clone(),
301 0 : listen_https_addr: listen_https_addr.clone(),
302 0 : pg_auth_type: *pg_auth_type,
303 0 : http_auth_type: *http_auth_type,
304 0 : no_sync: *no_sync,
305 0 : }
306 0 : }
307 : }
308 :
309 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
310 : #[serde(default)]
311 : pub struct SafekeeperConf {
312 : pub id: NodeId,
313 : pub pg_port: u16,
314 : pub pg_tenant_only_port: Option<u16>,
315 : pub http_port: u16,
316 : pub https_port: Option<u16>,
317 : pub sync: bool,
318 : pub remote_storage: Option<String>,
319 : pub backup_threads: Option<u32>,
320 : pub auth_enabled: bool,
321 : pub listen_addr: Option<String>,
322 : }
323 :
324 : impl Default for SafekeeperConf {
325 0 : fn default() -> Self {
326 0 : Self {
327 0 : id: NodeId(0),
328 0 : pg_port: 0,
329 0 : pg_tenant_only_port: None,
330 0 : http_port: 0,
331 0 : https_port: None,
332 0 : sync: true,
333 0 : remote_storage: None,
334 0 : backup_threads: None,
335 0 : auth_enabled: false,
336 0 : listen_addr: None,
337 0 : }
338 0 : }
339 : }
340 :
341 : #[derive(Clone, Copy)]
342 : pub enum InitForceMode {
343 : MustNotExist,
344 : EmptyDirOk,
345 : RemoveAllContents,
346 : }
347 :
348 : impl ValueEnum for InitForceMode {
349 0 : fn value_variants<'a>() -> &'a [Self] {
350 0 : &[
351 0 : Self::MustNotExist,
352 0 : Self::EmptyDirOk,
353 0 : Self::RemoveAllContents,
354 0 : ]
355 0 : }
356 :
357 0 : fn to_possible_value(&self) -> Option<clap::builder::PossibleValue> {
358 0 : Some(clap::builder::PossibleValue::new(match self {
359 0 : InitForceMode::MustNotExist => "must-not-exist",
360 0 : InitForceMode::EmptyDirOk => "empty-dir-ok",
361 0 : InitForceMode::RemoveAllContents => "remove-all-contents",
362 : }))
363 0 : }
364 : }
365 :
366 : impl SafekeeperConf {
367 : /// Compute is served by port on which only tenant scoped tokens allowed, if
368 : /// it is configured.
369 0 : pub fn get_compute_port(&self) -> u16 {
370 0 : self.pg_tenant_only_port.unwrap_or(self.pg_port)
371 0 : }
372 : }
373 :
374 : impl LocalEnv {
375 0 : pub fn pg_distrib_dir_raw(&self) -> PathBuf {
376 0 : self.pg_distrib_dir.clone()
377 0 : }
378 :
379 0 : pub fn pg_distrib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
380 0 : let path = self.pg_distrib_dir.clone();
381 0 :
382 0 : #[allow(clippy::manual_range_patterns)]
383 0 : match pg_version {
384 0 : 14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
385 0 : _ => bail!("Unsupported postgres version: {}", pg_version),
386 : }
387 0 : }
388 :
389 0 : pub fn pg_dir(&self, pg_version: u32, dir_name: &str) -> anyhow::Result<PathBuf> {
390 0 : Ok(self.pg_distrib_dir(pg_version)?.join(dir_name))
391 0 : }
392 :
393 0 : pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
394 0 : self.pg_dir(pg_version, "bin")
395 0 : }
396 :
397 0 : pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
398 0 : self.pg_dir(pg_version, "lib")
399 0 : }
400 :
401 0 : pub fn pageserver_bin(&self) -> PathBuf {
402 0 : self.neon_distrib_dir.join("pageserver")
403 0 : }
404 :
405 0 : pub fn storage_controller_bin(&self) -> PathBuf {
406 0 : // Irrespective of configuration, storage controller binary is always
407 0 : // run from the same location as neon_local. This means that for compatibility
408 0 : // tests that run old pageserver/safekeeper, they still run latest storage controller.
409 0 : let neon_local_bin_dir = env::current_exe().unwrap().parent().unwrap().to_owned();
410 0 : neon_local_bin_dir.join("storage_controller")
411 0 : }
412 :
413 0 : pub fn safekeeper_bin(&self) -> PathBuf {
414 0 : self.neon_distrib_dir.join("safekeeper")
415 0 : }
416 :
417 0 : pub fn storage_broker_bin(&self) -> PathBuf {
418 0 : self.neon_distrib_dir.join("storage_broker")
419 0 : }
420 :
421 0 : pub fn endpoints_path(&self) -> PathBuf {
422 0 : self.base_data_dir.join("endpoints")
423 0 : }
424 :
425 0 : pub fn pageserver_data_dir(&self, pageserver_id: NodeId) -> PathBuf {
426 0 : self.base_data_dir
427 0 : .join(format!("pageserver_{pageserver_id}"))
428 0 : }
429 :
430 0 : pub fn safekeeper_data_dir(&self, data_dir_name: &str) -> PathBuf {
431 0 : self.base_data_dir.join("safekeepers").join(data_dir_name)
432 0 : }
433 :
434 0 : pub fn get_pageserver_conf(&self, id: NodeId) -> anyhow::Result<&PageServerConf> {
435 0 : if let Some(conf) = self.pageservers.iter().find(|node| node.id == id) {
436 0 : Ok(conf)
437 : } else {
438 0 : let have_ids = self
439 0 : .pageservers
440 0 : .iter()
441 0 : .map(|node| format!("{}:{}", node.id, node.listen_http_addr))
442 0 : .collect::<Vec<_>>();
443 0 : let joined = have_ids.join(",");
444 0 : bail!("could not find pageserver {id}, have ids {joined}")
445 : }
446 0 : }
447 :
448 0 : pub fn ssl_ca_cert_path(&self) -> Option<PathBuf> {
449 0 : if self.generate_local_ssl_certs {
450 0 : Some(self.base_data_dir.join("rootCA.crt"))
451 : } else {
452 0 : None
453 : }
454 0 : }
455 :
456 0 : pub fn ssl_ca_key_path(&self) -> Option<PathBuf> {
457 0 : if self.generate_local_ssl_certs {
458 0 : Some(self.base_data_dir.join("rootCA.key"))
459 : } else {
460 0 : None
461 : }
462 0 : }
463 :
464 0 : pub fn generate_ssl_ca_cert(&self) -> anyhow::Result<()> {
465 0 : let cert_path = self.ssl_ca_cert_path().unwrap();
466 0 : let key_path = self.ssl_ca_key_path().unwrap();
467 0 : if !fs::exists(cert_path.as_path())? {
468 0 : generate_ssl_ca_cert(cert_path.as_path(), key_path.as_path())?;
469 0 : }
470 0 : Ok(())
471 0 : }
472 :
473 0 : pub fn generate_ssl_cert(&self, cert_path: &Path, key_path: &Path) -> anyhow::Result<()> {
474 0 : self.generate_ssl_ca_cert()?;
475 0 : generate_ssl_cert(
476 0 : cert_path,
477 0 : key_path,
478 0 : self.ssl_ca_cert_path().unwrap().as_path(),
479 0 : self.ssl_ca_key_path().unwrap().as_path(),
480 0 : )
481 0 : }
482 :
483 : /// Inspect the base data directory and extract the instance id and instance directory path
484 : /// for all storage controller instances
485 0 : pub async fn storage_controller_instances(&self) -> std::io::Result<Vec<(u8, PathBuf)>> {
486 0 : let mut instances = Vec::default();
487 :
488 0 : let dir = std::fs::read_dir(self.base_data_dir.clone())?;
489 0 : for dentry in dir {
490 0 : let dentry = dentry?;
491 0 : let is_dir = dentry.metadata()?.is_dir();
492 0 : let filename = dentry.file_name().into_string().unwrap();
493 0 : let parsed_instance_id = match filename.strip_prefix("storage_controller_") {
494 0 : Some(suffix) => suffix.parse::<u8>().ok(),
495 0 : None => None,
496 : };
497 :
498 0 : let is_instance_dir = is_dir && parsed_instance_id.is_some();
499 :
500 0 : if !is_instance_dir {
501 0 : continue;
502 0 : }
503 0 :
504 0 : instances.push((
505 0 : parsed_instance_id.expect("Checked previously"),
506 0 : dentry.path(),
507 0 : ));
508 : }
509 :
510 0 : Ok(instances)
511 0 : }
512 :
513 0 : pub fn register_branch_mapping(
514 0 : &mut self,
515 0 : branch_name: String,
516 0 : tenant_id: TenantId,
517 0 : timeline_id: TimelineId,
518 0 : ) -> anyhow::Result<()> {
519 0 : let existing_values = self
520 0 : .branch_name_mappings
521 0 : .entry(branch_name.clone())
522 0 : .or_default();
523 0 :
524 0 : let existing_ids = existing_values
525 0 : .iter()
526 0 : .find(|(existing_tenant_id, _)| existing_tenant_id == &tenant_id);
527 :
528 0 : if let Some((_, old_timeline_id)) = existing_ids {
529 0 : if old_timeline_id == &timeline_id {
530 0 : Ok(())
531 : } else {
532 0 : bail!(
533 0 : "branch '{branch_name}' is already mapped to timeline {old_timeline_id}, cannot map to another timeline {timeline_id}"
534 0 : );
535 : }
536 : } else {
537 0 : existing_values.push((tenant_id, timeline_id));
538 0 : Ok(())
539 : }
540 0 : }
541 :
542 0 : pub fn get_branch_timeline_id(
543 0 : &self,
544 0 : branch_name: &str,
545 0 : tenant_id: TenantId,
546 0 : ) -> Option<TimelineId> {
547 0 : self.branch_name_mappings
548 0 : .get(branch_name)?
549 0 : .iter()
550 0 : .find(|(mapped_tenant_id, _)| mapped_tenant_id == &tenant_id)
551 0 : .map(|&(_, timeline_id)| timeline_id)
552 0 : }
553 :
554 0 : pub fn timeline_name_mappings(&self) -> HashMap<TenantTimelineId, String> {
555 0 : self.branch_name_mappings
556 0 : .iter()
557 0 : .flat_map(|(name, tenant_timelines)| {
558 0 : tenant_timelines.iter().map(|&(tenant_id, timeline_id)| {
559 0 : (TenantTimelineId::new(tenant_id, timeline_id), name.clone())
560 0 : })
561 0 : })
562 0 : .collect()
563 0 : }
564 :
565 : /// Construct `Self` from on-disk state.
566 0 : pub fn load_config(repopath: &Path) -> anyhow::Result<Self> {
567 0 : if !repopath.exists() {
568 0 : bail!(
569 0 : "Neon config is not found in {}. You need to run 'neon_local init' first",
570 0 : repopath.to_str().unwrap()
571 0 : );
572 0 : }
573 :
574 : // TODO: check that it looks like a neon repository
575 :
576 : // load and parse file
577 0 : let config_file_contents = fs::read_to_string(repopath.join("config"))?;
578 0 : let on_disk_config: OnDiskConfig = toml::from_str(config_file_contents.as_str())?;
579 0 : let mut env = {
580 0 : let OnDiskConfig {
581 0 : pg_distrib_dir,
582 0 : neon_distrib_dir,
583 0 : default_tenant_id,
584 0 : private_key_path,
585 0 : broker,
586 0 : storage_controller,
587 0 : pageservers,
588 0 : safekeepers,
589 0 : control_plane_api,
590 0 : control_plane_hooks_api,
591 0 : control_plane_compute_hook_api: _,
592 0 : branch_name_mappings,
593 0 : generate_local_ssl_certs,
594 0 : } = on_disk_config;
595 0 : LocalEnv {
596 0 : base_data_dir: repopath.to_owned(),
597 0 : pg_distrib_dir,
598 0 : neon_distrib_dir,
599 0 : default_tenant_id,
600 0 : private_key_path,
601 0 : broker,
602 0 : storage_controller,
603 0 : pageservers,
604 0 : safekeepers,
605 0 : control_plane_api: control_plane_api.unwrap(),
606 0 : control_plane_hooks_api,
607 0 : branch_name_mappings,
608 0 : generate_local_ssl_certs,
609 0 : }
610 0 : };
611 0 :
612 0 : // The source of truth for pageserver configuration is the pageserver.toml.
613 0 : assert!(
614 0 : env.pageservers.is_empty(),
615 0 : "we ensure this during deserialization"
616 : );
617 0 : env.pageservers = {
618 0 : let iter = std::fs::read_dir(repopath).context("open dir")?;
619 0 : let mut pageservers = Vec::new();
620 0 : for res in iter {
621 0 : let dentry = res?;
622 : const PREFIX: &str = "pageserver_";
623 0 : let dentry_name = dentry
624 0 : .file_name()
625 0 : .into_string()
626 0 : .ok()
627 0 : .with_context(|| format!("non-utf8 dentry: {:?}", dentry.path()))
628 0 : .unwrap();
629 0 : if !dentry_name.starts_with(PREFIX) {
630 0 : continue;
631 0 : }
632 0 : if !dentry.file_type().context("determine file type")?.is_dir() {
633 0 : anyhow::bail!("expected a directory, got {:?}", dentry.path());
634 0 : }
635 0 : let id = dentry_name[PREFIX.len()..]
636 0 : .parse::<NodeId>()
637 0 : .with_context(|| format!("parse id from {:?}", dentry.path()))?;
638 : // TODO(christian): use pageserver_api::config::ConfigToml (PR #7656)
639 0 : #[derive(serde::Serialize, serde::Deserialize)]
640 : // (allow unknown fields, unlike PageServerConf)
641 : struct PageserverConfigTomlSubset {
642 : listen_pg_addr: String,
643 : listen_http_addr: String,
644 : listen_https_addr: Option<String>,
645 : pg_auth_type: AuthType,
646 : http_auth_type: AuthType,
647 : #[serde(default)]
648 : no_sync: bool,
649 : }
650 0 : let config_toml_path = dentry.path().join("pageserver.toml");
651 0 : let config_toml: PageserverConfigTomlSubset = toml_edit::de::from_str(
652 0 : &std::fs::read_to_string(&config_toml_path)
653 0 : .with_context(|| format!("read {:?}", config_toml_path))?,
654 : )
655 0 : .context("parse pageserver.toml")?;
656 0 : let identity_toml_path = dentry.path().join("identity.toml");
657 0 : #[derive(serde::Serialize, serde::Deserialize)]
658 : struct IdentityTomlSubset {
659 : id: NodeId,
660 : }
661 0 : let identity_toml: IdentityTomlSubset = toml_edit::de::from_str(
662 0 : &std::fs::read_to_string(&identity_toml_path)
663 0 : .with_context(|| format!("read {:?}", identity_toml_path))?,
664 : )
665 0 : .context("parse identity.toml")?;
666 : let PageserverConfigTomlSubset {
667 0 : listen_pg_addr,
668 0 : listen_http_addr,
669 0 : listen_https_addr,
670 0 : pg_auth_type,
671 0 : http_auth_type,
672 0 : no_sync,
673 0 : } = config_toml;
674 0 : let IdentityTomlSubset {
675 0 : id: identity_toml_id,
676 0 : } = identity_toml;
677 0 : let conf = PageServerConf {
678 : id: {
679 0 : anyhow::ensure!(
680 0 : identity_toml_id == id,
681 0 : "id mismatch: identity.toml:id={identity_toml_id} pageserver_(.*) id={id}",
682 : );
683 0 : id
684 0 : },
685 0 : listen_pg_addr,
686 0 : listen_http_addr,
687 0 : listen_https_addr,
688 0 : pg_auth_type,
689 0 : http_auth_type,
690 0 : no_sync,
691 0 : };
692 0 : pageservers.push(conf);
693 : }
694 0 : pageservers
695 0 : };
696 0 :
697 0 : Ok(env)
698 0 : }
699 :
700 0 : pub fn persist_config(&self) -> anyhow::Result<()> {
701 0 : Self::persist_config_impl(
702 0 : &self.base_data_dir,
703 0 : &OnDiskConfig {
704 0 : pg_distrib_dir: self.pg_distrib_dir.clone(),
705 0 : neon_distrib_dir: self.neon_distrib_dir.clone(),
706 0 : default_tenant_id: self.default_tenant_id,
707 0 : private_key_path: self.private_key_path.clone(),
708 0 : broker: self.broker.clone(),
709 0 : storage_controller: self.storage_controller.clone(),
710 0 : pageservers: vec![], // it's skip_serializing anyway
711 0 : safekeepers: self.safekeepers.clone(),
712 0 : control_plane_api: Some(self.control_plane_api.clone()),
713 0 : control_plane_hooks_api: self.control_plane_hooks_api.clone(),
714 0 : control_plane_compute_hook_api: None,
715 0 : branch_name_mappings: self.branch_name_mappings.clone(),
716 0 : generate_local_ssl_certs: self.generate_local_ssl_certs,
717 0 : },
718 0 : )
719 0 : }
720 :
721 0 : pub fn persist_config_impl(base_path: &Path, config: &OnDiskConfig) -> anyhow::Result<()> {
722 0 : let conf_content = &toml::to_string_pretty(config)?;
723 0 : let target_config_path = base_path.join("config");
724 0 : fs::write(&target_config_path, conf_content).with_context(|| {
725 0 : format!(
726 0 : "Failed to write config file into path '{}'",
727 0 : target_config_path.display()
728 0 : )
729 0 : })
730 0 : }
731 :
732 : // this function is used only for testing purposes in CLI e g generate tokens during init
733 0 : pub fn generate_auth_token(&self, claims: &Claims) -> anyhow::Result<String> {
734 0 : let private_key_path = self.get_private_key_path();
735 0 : let key_data = fs::read(private_key_path)?;
736 0 : encode_from_key_file(claims, &key_data)
737 0 : }
738 :
739 0 : pub fn get_private_key_path(&self) -> PathBuf {
740 0 : if self.private_key_path.is_absolute() {
741 0 : self.private_key_path.to_path_buf()
742 : } else {
743 0 : self.base_data_dir.join(&self.private_key_path)
744 : }
745 0 : }
746 :
747 : /// Materialize the [`NeonLocalInitConf`] to disk. Called during [`neon_local init`].
748 0 : pub fn init(conf: NeonLocalInitConf, force: &InitForceMode) -> anyhow::Result<()> {
749 0 : let base_path = base_path();
750 0 : assert_ne!(base_path, Path::new(""));
751 0 : let base_path = &base_path;
752 0 :
753 0 : // create base_path dir
754 0 : if base_path.exists() {
755 0 : match force {
756 : InitForceMode::MustNotExist => {
757 0 : bail!(
758 0 : "directory '{}' already exists. Perhaps already initialized?",
759 0 : base_path.display()
760 0 : );
761 : }
762 : InitForceMode::EmptyDirOk => {
763 0 : if let Some(res) = std::fs::read_dir(base_path)?.next() {
764 0 : res.context("check if directory is empty")?;
765 0 : anyhow::bail!("directory not empty: {base_path:?}");
766 0 : }
767 : }
768 : InitForceMode::RemoveAllContents => {
769 0 : println!("removing all contents of '{}'", base_path.display());
770 : // instead of directly calling `remove_dir_all`, we keep the original dir but removing
771 : // all contents inside. This helps if the developer symbol links another directory (i.e.,
772 : // S3 local SSD) to the `.neon` base directory.
773 0 : for entry in std::fs::read_dir(base_path)? {
774 0 : let entry = entry?;
775 0 : let path = entry.path();
776 0 : if path.is_dir() {
777 0 : fs::remove_dir_all(&path)?;
778 : } else {
779 0 : fs::remove_file(&path)?;
780 : }
781 : }
782 : }
783 : }
784 0 : }
785 0 : if !base_path.exists() {
786 0 : fs::create_dir(base_path)?;
787 0 : }
788 :
789 : let NeonLocalInitConf {
790 0 : pg_distrib_dir,
791 0 : neon_distrib_dir,
792 0 : default_tenant_id,
793 0 : broker,
794 0 : storage_controller,
795 0 : pageservers,
796 0 : safekeepers,
797 0 : control_plane_api,
798 0 : generate_local_ssl_certs,
799 0 : control_plane_hooks_api,
800 0 : } = conf;
801 0 :
802 0 : // Find postgres binaries.
803 0 : // Follow POSTGRES_DISTRIB_DIR if set, otherwise look in "pg_install".
804 0 : // Note that later in the code we assume, that distrib dirs follow the same pattern
805 0 : // for all postgres versions.
806 0 : let pg_distrib_dir = pg_distrib_dir.unwrap_or_else(|| {
807 0 : if let Some(postgres_bin) = env::var_os("POSTGRES_DISTRIB_DIR") {
808 0 : postgres_bin.into()
809 : } else {
810 0 : let cwd = env::current_dir().unwrap();
811 0 : cwd.join("pg_install")
812 : }
813 0 : });
814 0 :
815 0 : // Find neon binaries.
816 0 : let neon_distrib_dir = neon_distrib_dir
817 0 : .unwrap_or_else(|| env::current_exe().unwrap().parent().unwrap().to_owned());
818 0 :
819 0 : // Generate keypair for JWT.
820 0 : //
821 0 : // The keypair is only needed if authentication is enabled in any of the
822 0 : // components. For convenience, we generate the keypair even if authentication
823 0 : // is not enabled, so that you can easily enable it after the initialization
824 0 : // step.
825 0 : generate_auth_keys(
826 0 : base_path.join("auth_private_key.pem").as_path(),
827 0 : base_path.join("auth_public_key.pem").as_path(),
828 0 : )
829 0 : .context("generate auth keys")?;
830 0 : let private_key_path = PathBuf::from("auth_private_key.pem");
831 0 :
832 0 : // create the runtime type because the remaining initialization code below needs
833 0 : // a LocalEnv instance op operation
834 0 : // TODO: refactor to avoid this, LocalEnv should only be constructed from on-disk state
835 0 : let env = LocalEnv {
836 0 : base_data_dir: base_path.clone(),
837 0 : pg_distrib_dir,
838 0 : neon_distrib_dir,
839 0 : default_tenant_id: Some(default_tenant_id),
840 0 : private_key_path,
841 0 : broker,
842 0 : storage_controller: storage_controller.unwrap_or_default(),
843 0 : pageservers: pageservers.iter().map(Into::into).collect(),
844 0 : safekeepers,
845 0 : control_plane_api: control_plane_api.unwrap(),
846 0 : control_plane_hooks_api,
847 0 : branch_name_mappings: Default::default(),
848 0 : generate_local_ssl_certs,
849 0 : };
850 0 :
851 0 : if generate_local_ssl_certs {
852 0 : env.generate_ssl_ca_cert()?;
853 0 : }
854 :
855 : // create endpoints dir
856 0 : fs::create_dir_all(env.endpoints_path())?;
857 :
858 : // create safekeeper dirs
859 0 : for safekeeper in &env.safekeepers {
860 0 : fs::create_dir_all(SafekeeperNode::datadir_path_by_id(&env, safekeeper.id))?;
861 0 : SafekeeperNode::from_env(&env, safekeeper)
862 0 : .initialize()
863 0 : .context("safekeeper init failed")?;
864 : }
865 :
866 : // initialize pageserver state
867 0 : for (i, ps) in pageservers.into_iter().enumerate() {
868 0 : let runtime_ps = &env.pageservers[i];
869 0 : assert_eq!(&PageServerConf::from(&ps), runtime_ps);
870 0 : fs::create_dir(env.pageserver_data_dir(ps.id))?;
871 0 : PageServerNode::from_env(&env, runtime_ps)
872 0 : .initialize(ps)
873 0 : .context("pageserver init failed")?;
874 : }
875 :
876 : // setup remote remote location for default LocalFs remote storage
877 0 : std::fs::create_dir_all(env.base_data_dir.join(PAGESERVER_REMOTE_STORAGE_DIR))?;
878 :
879 0 : env.persist_config()
880 0 : }
881 : }
882 :
883 0 : pub fn base_path() -> PathBuf {
884 0 : let path = match std::env::var_os("NEON_REPO_DIR") {
885 0 : Some(val) => {
886 0 : let path = PathBuf::from(val);
887 0 : if !path.is_absolute() {
888 : // repeat the env var in the error because our default is always absolute
889 0 : panic!("NEON_REPO_DIR must be an absolute path, got {path:?}");
890 0 : }
891 0 : path
892 : }
893 : None => {
894 0 : let pwd = std::env::current_dir()
895 0 : // technically this can fail but it's quite unlikeley
896 0 : .expect("determine current directory");
897 0 : let pwd_abs = pwd.canonicalize().expect("canonicalize current directory");
898 0 : pwd_abs.join(".neon")
899 : }
900 : };
901 0 : assert!(path.is_absolute());
902 0 : path
903 0 : }
904 :
905 : /// Generate a public/private key pair for JWT authentication
906 0 : fn generate_auth_keys(private_key_path: &Path, public_key_path: &Path) -> anyhow::Result<()> {
907 : // Generate the key pair
908 : //
909 : // openssl genpkey -algorithm ed25519 -out auth_private_key.pem
910 0 : let keygen_output = Command::new("openssl")
911 0 : .arg("genpkey")
912 0 : .args(["-algorithm", "ed25519"])
913 0 : .args(["-out", private_key_path.to_str().unwrap()])
914 0 : .stdout(Stdio::null())
915 0 : .output()
916 0 : .context("failed to generate auth private key")?;
917 0 : if !keygen_output.status.success() {
918 0 : bail!(
919 0 : "openssl failed: '{}'",
920 0 : String::from_utf8_lossy(&keygen_output.stderr)
921 0 : );
922 0 : }
923 : // Extract the public key from the private key file
924 : //
925 : // openssl pkey -in auth_private_key.pem -pubout -out auth_public_key.pem
926 0 : let keygen_output = Command::new("openssl")
927 0 : .arg("pkey")
928 0 : .args(["-in", private_key_path.to_str().unwrap()])
929 0 : .arg("-pubout")
930 0 : .args(["-out", public_key_path.to_str().unwrap()])
931 0 : .output()
932 0 : .context("failed to extract public key from private key")?;
933 0 : if !keygen_output.status.success() {
934 0 : bail!(
935 0 : "openssl failed: '{}'",
936 0 : String::from_utf8_lossy(&keygen_output.stderr)
937 0 : );
938 0 : }
939 0 : Ok(())
940 0 : }
941 :
942 0 : fn generate_ssl_ca_cert(cert_path: &Path, key_path: &Path) -> anyhow::Result<()> {
943 : // openssl req -x509 -newkey rsa:2048 -nodes -subj "/CN=Neon Local CA" -days 36500 \
944 : // -out rootCA.crt -keyout rootCA.key
945 0 : let keygen_output = Command::new("openssl")
946 0 : .args([
947 0 : "req", "-x509", "-newkey", "rsa:2048", "-nodes", "-days", "36500",
948 0 : ])
949 0 : .args(["-subj", "/CN=Neon Local CA"])
950 0 : .args(["-out", cert_path.to_str().unwrap()])
951 0 : .args(["-keyout", key_path.to_str().unwrap()])
952 0 : .output()
953 0 : .context("failed to generate CA certificate")?;
954 0 : if !keygen_output.status.success() {
955 0 : bail!(
956 0 : "openssl failed: '{}'",
957 0 : String::from_utf8_lossy(&keygen_output.stderr)
958 0 : );
959 0 : }
960 0 : Ok(())
961 0 : }
962 :
963 0 : fn generate_ssl_cert(
964 0 : cert_path: &Path,
965 0 : key_path: &Path,
966 0 : ca_cert_path: &Path,
967 0 : ca_key_path: &Path,
968 0 : ) -> anyhow::Result<()> {
969 0 : // Generate Certificate Signing Request (CSR).
970 0 : let mut csr_path = cert_path.to_path_buf();
971 0 : csr_path.set_extension(".csr");
972 :
973 : // openssl req -new -nodes -newkey rsa:2048 -keyout server.key -out server.csr \
974 : // -subj "/CN=localhost" -addext "subjectAltName=DNS:localhost,IP:127.0.0.1"
975 0 : let keygen_output = Command::new("openssl")
976 0 : .args(["req", "-new", "-nodes"])
977 0 : .args(["-newkey", "rsa:2048"])
978 0 : .args(["-subj", "/CN=localhost"])
979 0 : .args(["-addext", "subjectAltName=DNS:localhost,IP:127.0.0.1"])
980 0 : .args(["-keyout", key_path.to_str().unwrap()])
981 0 : .args(["-out", csr_path.to_str().unwrap()])
982 0 : .output()
983 0 : .context("failed to generate CSR")?;
984 0 : if !keygen_output.status.success() {
985 0 : bail!(
986 0 : "openssl failed: '{}'",
987 0 : String::from_utf8_lossy(&keygen_output.stderr)
988 0 : );
989 0 : }
990 :
991 : // Sign CSR with CA key.
992 : //
993 : // openssl x509 -req -in server.csr -CA rootCA.crt -CAkey rootCA.key -CAcreateserial \
994 : // -out server.crt -days 36500 -copy_extensions copyall
995 0 : let keygen_output = Command::new("openssl")
996 0 : .args(["x509", "-req"])
997 0 : .args(["-in", csr_path.to_str().unwrap()])
998 0 : .args(["-CA", ca_cert_path.to_str().unwrap()])
999 0 : .args(["-CAkey", ca_key_path.to_str().unwrap()])
1000 0 : .arg("-CAcreateserial")
1001 0 : .args(["-out", cert_path.to_str().unwrap()])
1002 0 : .args(["-days", "36500"])
1003 0 : .args(["-copy_extensions", "copyall"])
1004 0 : .output()
1005 0 : .context("failed to sign CSR")?;
1006 0 : if !keygen_output.status.success() {
1007 0 : bail!(
1008 0 : "openssl failed: '{}'",
1009 0 : String::from_utf8_lossy(&keygen_output.stderr)
1010 0 : );
1011 0 : }
1012 0 :
1013 0 : // Remove CSR file as it's not needed anymore.
1014 0 : fs::remove_file(csr_path)?;
1015 :
1016 0 : Ok(())
1017 0 : }
|
