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