Line data Source code
1 : //! Code to manage compute endpoints
2 : //!
3 : //! In the local test environment, the data for each endpoint is stored in
4 : //!
5 : //! ```text
6 : //! .neon/endpoints/<endpoint id>
7 : //! ```
8 : //!
9 : //! Some basic information about the endpoint, like the tenant and timeline IDs,
10 : //! are stored in the `endpoint.json` file. The `endpoint.json` file is created
11 : //! when the endpoint is created, and doesn't change afterwards.
12 : //!
13 : //! The endpoint is managed by the `compute_ctl` binary. When an endpoint is
14 : //! started, we launch `compute_ctl` It synchronizes the safekeepers, downloads
15 : //! the basebackup from the pageserver to initialize the data directory, and
16 : //! finally launches the PostgreSQL process. It watches the PostgreSQL process
17 : //! until it exits.
18 : //!
19 : //! When an endpoint is created, a `postgresql.conf` file is also created in
20 : //! the endpoint's directory. The file can be modified before starting PostgreSQL.
21 : //! However, the `postgresql.conf` file in the endpoint directory is not used directly
22 : //! by PostgreSQL. It is passed to `compute_ctl`, and `compute_ctl` writes another
23 : //! copy of it in the data directory.
24 : //!
25 : //! Directory contents:
26 : //!
27 : //! ```text
28 : //! .neon/endpoints/main/
29 : //! compute.log - log output of `compute_ctl` and `postgres`
30 : //! endpoint.json - serialized `EndpointConf` struct
31 : //! postgresql.conf - postgresql settings
32 : //! spec.json - passed to `compute_ctl`
33 : //! pgdata/
34 : //! postgresql.conf - copy of postgresql.conf created by `compute_ctl`
35 : //! zenith.signal
36 : //! <other PostgreSQL files>
37 : //! ```
38 : //!
39 : use std::collections::BTreeMap;
40 : use std::net::SocketAddr;
41 : use std::net::TcpStream;
42 : use std::path::PathBuf;
43 : use std::process::Command;
44 : use std::str::FromStr;
45 : use std::sync::Arc;
46 : use std::time::Duration;
47 :
48 : use anyhow::{anyhow, bail, Context, Result};
49 : use compute_api::spec::Database;
50 : use compute_api::spec::PgIdent;
51 : use compute_api::spec::RemoteExtSpec;
52 : use compute_api::spec::Role;
53 : use nix::sys::signal::kill;
54 : use nix::sys::signal::Signal;
55 : use pageserver_api::shard::ShardStripeSize;
56 : use serde::{Deserialize, Serialize};
57 : use url::Host;
58 : use utils::id::{NodeId, TenantId, TimelineId};
59 :
60 : use crate::local_env::LocalEnv;
61 : use crate::postgresql_conf::PostgresConf;
62 : use crate::storage_controller::StorageController;
63 :
64 : use compute_api::responses::{ComputeState, ComputeStatus};
65 : use compute_api::spec::{Cluster, ComputeFeature, ComputeMode, ComputeSpec};
66 :
67 : // contents of a endpoint.json file
68 0 : #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug)]
69 : pub struct EndpointConf {
70 : endpoint_id: String,
71 : tenant_id: TenantId,
72 : timeline_id: TimelineId,
73 : mode: ComputeMode,
74 : pg_port: u16,
75 : http_port: u16,
76 : pg_version: u32,
77 : skip_pg_catalog_updates: bool,
78 : features: Vec<ComputeFeature>,
79 : }
80 :
81 : //
82 : // ComputeControlPlane
83 : //
84 : pub struct ComputeControlPlane {
85 : base_port: u16,
86 :
87 : // endpoint ID is the key
88 : pub endpoints: BTreeMap<String, Arc<Endpoint>>,
89 :
90 : env: LocalEnv,
91 : }
92 :
93 : impl ComputeControlPlane {
94 : // Load current endpoints from the endpoints/ subdirectories
95 0 : pub fn load(env: LocalEnv) -> Result<ComputeControlPlane> {
96 0 : let mut endpoints = BTreeMap::default();
97 0 : for endpoint_dir in std::fs::read_dir(env.endpoints_path())
98 0 : .with_context(|| format!("failed to list {}", env.endpoints_path().display()))?
99 : {
100 0 : let ep_res = Endpoint::from_dir_entry(endpoint_dir?, &env);
101 0 : let ep = match ep_res {
102 0 : Ok(ep) => ep,
103 0 : Err(e) => match e.downcast::<std::io::Error>() {
104 0 : Ok(e) => {
105 0 : // A parallel task could delete an endpoint while we have just scanned the directory
106 0 : if e.kind() == std::io::ErrorKind::NotFound {
107 0 : continue;
108 : } else {
109 0 : Err(e)?
110 : }
111 : }
112 0 : Err(e) => Err(e)?,
113 : },
114 : };
115 0 : endpoints.insert(ep.endpoint_id.clone(), Arc::new(ep));
116 : }
117 :
118 0 : Ok(ComputeControlPlane {
119 0 : base_port: 55431,
120 0 : endpoints,
121 0 : env,
122 0 : })
123 0 : }
124 :
125 0 : fn get_port(&mut self) -> u16 {
126 0 : 1 + self
127 0 : .endpoints
128 0 : .values()
129 0 : .map(|ep| std::cmp::max(ep.pg_address.port(), ep.http_address.port()))
130 0 : .max()
131 0 : .unwrap_or(self.base_port)
132 0 : }
133 :
134 : #[allow(clippy::too_many_arguments)]
135 0 : pub fn new_endpoint(
136 0 : &mut self,
137 0 : endpoint_id: &str,
138 0 : tenant_id: TenantId,
139 0 : timeline_id: TimelineId,
140 0 : pg_port: Option<u16>,
141 0 : http_port: Option<u16>,
142 0 : pg_version: u32,
143 0 : mode: ComputeMode,
144 0 : skip_pg_catalog_updates: bool,
145 0 : ) -> Result<Arc<Endpoint>> {
146 0 : let pg_port = pg_port.unwrap_or_else(|| self.get_port());
147 0 : let http_port = http_port.unwrap_or_else(|| self.get_port() + 1);
148 0 : let ep = Arc::new(Endpoint {
149 0 : endpoint_id: endpoint_id.to_owned(),
150 0 : pg_address: SocketAddr::new("127.0.0.1".parse().unwrap(), pg_port),
151 0 : http_address: SocketAddr::new("127.0.0.1".parse().unwrap(), http_port),
152 0 : env: self.env.clone(),
153 0 : timeline_id,
154 0 : mode,
155 0 : tenant_id,
156 0 : pg_version,
157 0 : // We don't setup roles and databases in the spec locally, so we don't need to
158 0 : // do catalog updates. Catalog updates also include check availability
159 0 : // data creation. Yet, we have tests that check that size and db dump
160 0 : // before and after start are the same. So, skip catalog updates,
161 0 : // with this we basically test a case of waking up an idle compute, where
162 0 : // we also skip catalog updates in the cloud.
163 0 : skip_pg_catalog_updates,
164 0 : features: vec![],
165 0 : });
166 0 :
167 0 : ep.create_endpoint_dir()?;
168 : std::fs::write(
169 0 : ep.endpoint_path().join("endpoint.json"),
170 0 : serde_json::to_string_pretty(&EndpointConf {
171 0 : endpoint_id: endpoint_id.to_string(),
172 0 : tenant_id,
173 0 : timeline_id,
174 0 : mode,
175 0 : http_port,
176 0 : pg_port,
177 0 : pg_version,
178 0 : skip_pg_catalog_updates,
179 0 : features: vec![],
180 0 : })?,
181 0 : )?;
182 : std::fs::write(
183 0 : ep.endpoint_path().join("postgresql.conf"),
184 0 : ep.setup_pg_conf()?.to_string(),
185 0 : )?;
186 :
187 0 : self.endpoints
188 0 : .insert(ep.endpoint_id.clone(), Arc::clone(&ep));
189 0 :
190 0 : Ok(ep)
191 0 : }
192 :
193 0 : pub fn check_conflicting_endpoints(
194 0 : &self,
195 0 : mode: ComputeMode,
196 0 : tenant_id: TenantId,
197 0 : timeline_id: TimelineId,
198 0 : ) -> Result<()> {
199 0 : if matches!(mode, ComputeMode::Primary) {
200 : // this check is not complete, as you could have a concurrent attempt at
201 : // creating another primary, both reading the state before checking it here,
202 : // but it's better than nothing.
203 0 : let mut duplicates = self.endpoints.iter().filter(|(_k, v)| {
204 0 : v.tenant_id == tenant_id
205 0 : && v.timeline_id == timeline_id
206 0 : && v.mode == mode
207 0 : && v.status() != EndpointStatus::Stopped
208 0 : });
209 :
210 0 : if let Some((key, _)) = duplicates.next() {
211 0 : bail!("attempting to create a duplicate primary endpoint on tenant {tenant_id}, timeline {timeline_id}: endpoint {key:?} exists already. please don't do this, it is not supported.");
212 0 : }
213 0 : }
214 0 : Ok(())
215 0 : }
216 : }
217 :
218 : ///////////////////////////////////////////////////////////////////////////////
219 :
220 : #[derive(Debug)]
221 : pub struct Endpoint {
222 : /// used as the directory name
223 : endpoint_id: String,
224 : pub tenant_id: TenantId,
225 : pub timeline_id: TimelineId,
226 : pub mode: ComputeMode,
227 :
228 : // port and address of the Postgres server and `compute_ctl`'s HTTP API
229 : pub pg_address: SocketAddr,
230 : pub http_address: SocketAddr,
231 :
232 : // postgres major version in the format: 14, 15, etc.
233 : pg_version: u32,
234 :
235 : // These are not part of the endpoint as such, but the environment
236 : // the endpoint runs in.
237 : pub env: LocalEnv,
238 :
239 : // Optimizations
240 : skip_pg_catalog_updates: bool,
241 :
242 : // Feature flags
243 : features: Vec<ComputeFeature>,
244 : }
245 :
246 : #[derive(PartialEq, Eq)]
247 : pub enum EndpointStatus {
248 : Running,
249 : Stopped,
250 : Crashed,
251 : RunningNoPidfile,
252 : }
253 :
254 : impl std::fmt::Display for EndpointStatus {
255 0 : fn fmt(&self, writer: &mut std::fmt::Formatter) -> std::fmt::Result {
256 0 : let s = match self {
257 0 : Self::Running => "running",
258 0 : Self::Stopped => "stopped",
259 0 : Self::Crashed => "crashed",
260 0 : Self::RunningNoPidfile => "running, no pidfile",
261 : };
262 0 : write!(writer, "{}", s)
263 0 : }
264 : }
265 :
266 : impl Endpoint {
267 0 : fn from_dir_entry(entry: std::fs::DirEntry, env: &LocalEnv) -> Result<Endpoint> {
268 0 : if !entry.file_type()?.is_dir() {
269 0 : anyhow::bail!(
270 0 : "Endpoint::from_dir_entry failed: '{}' is not a directory",
271 0 : entry.path().display()
272 0 : );
273 0 : }
274 0 :
275 0 : // parse data directory name
276 0 : let fname = entry.file_name();
277 0 : let endpoint_id = fname.to_str().unwrap().to_string();
278 :
279 : // Read the endpoint.json file
280 0 : let conf: EndpointConf =
281 0 : serde_json::from_slice(&std::fs::read(entry.path().join("endpoint.json"))?)?;
282 :
283 0 : Ok(Endpoint {
284 0 : pg_address: SocketAddr::new("127.0.0.1".parse().unwrap(), conf.pg_port),
285 0 : http_address: SocketAddr::new("127.0.0.1".parse().unwrap(), conf.http_port),
286 0 : endpoint_id,
287 0 : env: env.clone(),
288 0 : timeline_id: conf.timeline_id,
289 0 : mode: conf.mode,
290 0 : tenant_id: conf.tenant_id,
291 0 : pg_version: conf.pg_version,
292 0 : skip_pg_catalog_updates: conf.skip_pg_catalog_updates,
293 0 : features: conf.features,
294 0 : })
295 0 : }
296 :
297 0 : fn create_endpoint_dir(&self) -> Result<()> {
298 0 : std::fs::create_dir_all(self.endpoint_path()).with_context(|| {
299 0 : format!(
300 0 : "could not create endpoint directory {}",
301 0 : self.endpoint_path().display()
302 0 : )
303 0 : })
304 0 : }
305 :
306 : // Generate postgresql.conf with default configuration
307 0 : fn setup_pg_conf(&self) -> Result<PostgresConf> {
308 0 : let mut conf = PostgresConf::new();
309 0 : conf.append("max_wal_senders", "10");
310 0 : conf.append("wal_log_hints", "off");
311 0 : conf.append("max_replication_slots", "10");
312 0 : conf.append("hot_standby", "on");
313 0 : conf.append("shared_buffers", "1MB");
314 0 : conf.append("fsync", "off");
315 0 : conf.append("max_connections", "100");
316 0 : conf.append("wal_level", "logical");
317 0 : // wal_sender_timeout is the maximum time to wait for WAL replication.
318 0 : // It also defines how often the walreciever will send a feedback message to the wal sender.
319 0 : conf.append("wal_sender_timeout", "5s");
320 0 : conf.append("listen_addresses", &self.pg_address.ip().to_string());
321 0 : conf.append("port", &self.pg_address.port().to_string());
322 0 : conf.append("wal_keep_size", "0");
323 0 : // walproposer panics when basebackup is invalid, it is pointless to restart in this case.
324 0 : conf.append("restart_after_crash", "off");
325 0 :
326 0 : // Load the 'neon' extension
327 0 : conf.append("shared_preload_libraries", "neon");
328 0 :
329 0 : conf.append_line("");
330 0 : // Replication-related configurations, such as WAL sending
331 0 : match &self.mode {
332 : ComputeMode::Primary => {
333 : // Configure backpressure
334 : // - Replication write lag depends on how fast the walreceiver can process incoming WAL.
335 : // This lag determines latency of get_page_at_lsn. Speed of applying WAL is about 10MB/sec,
336 : // so to avoid expiration of 1 minute timeout, this lag should not be larger than 600MB.
337 : // Actually latency should be much smaller (better if < 1sec). But we assume that recently
338 : // updates pages are not requested from pageserver.
339 : // - Replication flush lag depends on speed of persisting data by checkpointer (creation of
340 : // delta/image layers) and advancing disk_consistent_lsn. Safekeepers are able to
341 : // remove/archive WAL only beyond disk_consistent_lsn. Too large a lag can cause long
342 : // recovery time (in case of pageserver crash) and disk space overflow at safekeepers.
343 : // - Replication apply lag depends on speed of uploading changes to S3 by uploader thread.
344 : // To be able to restore database in case of pageserver node crash, safekeeper should not
345 : // remove WAL beyond this point. Too large lag can cause space exhaustion in safekeepers
346 : // (if they are not able to upload WAL to S3).
347 0 : conf.append("max_replication_write_lag", "15MB");
348 0 : conf.append("max_replication_flush_lag", "10GB");
349 0 :
350 0 : if !self.env.safekeepers.is_empty() {
351 0 : // Configure Postgres to connect to the safekeepers
352 0 : conf.append("synchronous_standby_names", "walproposer");
353 0 :
354 0 : let safekeepers = self
355 0 : .env
356 0 : .safekeepers
357 0 : .iter()
358 0 : .map(|sk| format!("localhost:{}", sk.get_compute_port()))
359 0 : .collect::<Vec<String>>()
360 0 : .join(",");
361 0 : conf.append("neon.safekeepers", &safekeepers);
362 0 : } else {
363 0 : // We only use setup without safekeepers for tests,
364 0 : // and don't care about data durability on pageserver,
365 0 : // so set more relaxed synchronous_commit.
366 0 : conf.append("synchronous_commit", "remote_write");
367 0 :
368 0 : // Configure the node to stream WAL directly to the pageserver
369 0 : // This isn't really a supported configuration, but can be useful for
370 0 : // testing.
371 0 : conf.append("synchronous_standby_names", "pageserver");
372 0 : }
373 : }
374 0 : ComputeMode::Static(lsn) => {
375 0 : conf.append("recovery_target_lsn", &lsn.to_string());
376 0 : }
377 : ComputeMode::Replica => {
378 0 : assert!(!self.env.safekeepers.is_empty());
379 :
380 : // TODO: use future host field from safekeeper spec
381 : // Pass the list of safekeepers to the replica so that it can connect to any of them,
382 : // whichever is available.
383 0 : let sk_ports = self
384 0 : .env
385 0 : .safekeepers
386 0 : .iter()
387 0 : .map(|x| x.get_compute_port().to_string())
388 0 : .collect::<Vec<_>>()
389 0 : .join(",");
390 0 : let sk_hosts = vec!["localhost"; self.env.safekeepers.len()].join(",");
391 0 :
392 0 : let connstr = format!(
393 0 : "host={} port={} options='-c timeline_id={} tenant_id={}' application_name=replica replication=true",
394 0 : sk_hosts,
395 0 : sk_ports,
396 0 : &self.timeline_id.to_string(),
397 0 : &self.tenant_id.to_string(),
398 0 : );
399 0 :
400 0 : let slot_name = format!("repl_{}_", self.timeline_id);
401 0 : conf.append("primary_conninfo", connstr.as_str());
402 0 : conf.append("primary_slot_name", slot_name.as_str());
403 0 : conf.append("hot_standby", "on");
404 0 : // prefetching of blocks referenced in WAL doesn't make sense for us
405 0 : // Neon hot standby ignores pages that are not in the shared_buffers
406 0 : if self.pg_version >= 15 {
407 0 : conf.append("recovery_prefetch", "off");
408 0 : }
409 : }
410 : }
411 :
412 0 : Ok(conf)
413 0 : }
414 :
415 0 : pub fn endpoint_path(&self) -> PathBuf {
416 0 : self.env.endpoints_path().join(&self.endpoint_id)
417 0 : }
418 :
419 0 : pub fn pgdata(&self) -> PathBuf {
420 0 : self.endpoint_path().join("pgdata")
421 0 : }
422 :
423 0 : pub fn status(&self) -> EndpointStatus {
424 0 : let timeout = Duration::from_millis(300);
425 0 : let has_pidfile = self.pgdata().join("postmaster.pid").exists();
426 0 : let can_connect = TcpStream::connect_timeout(&self.pg_address, timeout).is_ok();
427 0 :
428 0 : match (has_pidfile, can_connect) {
429 0 : (true, true) => EndpointStatus::Running,
430 0 : (false, false) => EndpointStatus::Stopped,
431 0 : (true, false) => EndpointStatus::Crashed,
432 0 : (false, true) => EndpointStatus::RunningNoPidfile,
433 : }
434 0 : }
435 :
436 0 : fn pg_ctl(&self, args: &[&str], auth_token: &Option<String>) -> Result<()> {
437 0 : let pg_ctl_path = self.env.pg_bin_dir(self.pg_version)?.join("pg_ctl");
438 0 : let mut cmd = Command::new(&pg_ctl_path);
439 0 : cmd.args(
440 0 : [
441 0 : &[
442 0 : "-D",
443 0 : self.pgdata().to_str().unwrap(),
444 0 : "-w", //wait till pg_ctl actually does what was asked
445 0 : ],
446 0 : args,
447 0 : ]
448 0 : .concat(),
449 0 : )
450 0 : .env_clear()
451 0 : .env(
452 0 : "LD_LIBRARY_PATH",
453 0 : self.env.pg_lib_dir(self.pg_version)?.to_str().unwrap(),
454 0 : )
455 0 : .env(
456 0 : "DYLD_LIBRARY_PATH",
457 0 : self.env.pg_lib_dir(self.pg_version)?.to_str().unwrap(),
458 : );
459 :
460 : // Pass authentication token used for the connections to pageserver and safekeepers
461 0 : if let Some(token) = auth_token {
462 0 : cmd.env("NEON_AUTH_TOKEN", token);
463 0 : }
464 :
465 0 : let pg_ctl = cmd
466 0 : .output()
467 0 : .context(format!("{} failed", pg_ctl_path.display()))?;
468 0 : if !pg_ctl.status.success() {
469 0 : anyhow::bail!(
470 0 : "pg_ctl failed, exit code: {}, stdout: {}, stderr: {}",
471 0 : pg_ctl.status,
472 0 : String::from_utf8_lossy(&pg_ctl.stdout),
473 0 : String::from_utf8_lossy(&pg_ctl.stderr),
474 0 : );
475 0 : }
476 0 :
477 0 : Ok(())
478 0 : }
479 :
480 0 : fn wait_for_compute_ctl_to_exit(&self, send_sigterm: bool) -> Result<()> {
481 0 : // TODO use background_process::stop_process instead: https://github.com/neondatabase/neon/pull/6482
482 0 : let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
483 0 : let pid: u32 = std::fs::read_to_string(pidfile_path)?.parse()?;
484 0 : let pid = nix::unistd::Pid::from_raw(pid as i32);
485 0 : if send_sigterm {
486 0 : kill(pid, Signal::SIGTERM).ok();
487 0 : }
488 0 : crate::background_process::wait_until_stopped("compute_ctl", pid)?;
489 0 : Ok(())
490 0 : }
491 :
492 0 : fn read_postgresql_conf(&self) -> Result<String> {
493 0 : // Slurp the endpoints/<endpoint id>/postgresql.conf file into
494 0 : // memory. We will include it in the spec file that we pass to
495 0 : // `compute_ctl`, and `compute_ctl` will write it to the postgresql.conf
496 0 : // in the data directory.
497 0 : let postgresql_conf_path = self.endpoint_path().join("postgresql.conf");
498 0 : match std::fs::read(&postgresql_conf_path) {
499 0 : Ok(content) => Ok(String::from_utf8(content)?),
500 0 : Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok("".to_string()),
501 0 : Err(e) => Err(anyhow::Error::new(e).context(format!(
502 0 : "failed to read config file in {}",
503 0 : postgresql_conf_path.to_str().unwrap()
504 0 : ))),
505 : }
506 0 : }
507 :
508 0 : fn build_pageserver_connstr(pageservers: &[(Host, u16)]) -> String {
509 0 : pageservers
510 0 : .iter()
511 0 : .map(|(host, port)| format!("postgresql://no_user@{host}:{port}"))
512 0 : .collect::<Vec<_>>()
513 0 : .join(",")
514 0 : }
515 :
516 : /// Map safekeepers ids to the actual connection strings.
517 0 : fn build_safekeepers_connstrs(&self, sk_ids: Vec<NodeId>) -> Result<Vec<String>> {
518 0 : let mut safekeeper_connstrings = Vec::new();
519 0 : if self.mode == ComputeMode::Primary {
520 0 : for sk_id in sk_ids {
521 0 : let sk = self
522 0 : .env
523 0 : .safekeepers
524 0 : .iter()
525 0 : .find(|node| node.id == sk_id)
526 0 : .ok_or_else(|| anyhow!("safekeeper {sk_id} does not exist"))?;
527 0 : safekeeper_connstrings.push(format!("127.0.0.1:{}", sk.get_compute_port()));
528 : }
529 0 : }
530 0 : Ok(safekeeper_connstrings)
531 0 : }
532 :
533 0 : pub async fn start(
534 0 : &self,
535 0 : auth_token: &Option<String>,
536 0 : safekeepers: Vec<NodeId>,
537 0 : pageservers: Vec<(Host, u16)>,
538 0 : remote_ext_config: Option<&String>,
539 0 : shard_stripe_size: usize,
540 0 : create_test_user: bool,
541 0 : ) -> Result<()> {
542 0 : if self.status() == EndpointStatus::Running {
543 0 : anyhow::bail!("The endpoint is already running");
544 0 : }
545 :
546 0 : let postgresql_conf = self.read_postgresql_conf()?;
547 :
548 : // We always start the compute node from scratch, so if the Postgres
549 : // data dir exists from a previous launch, remove it first.
550 0 : if self.pgdata().exists() {
551 0 : std::fs::remove_dir_all(self.pgdata())?;
552 0 : }
553 :
554 0 : let pageserver_connstring = Self::build_pageserver_connstr(&pageservers);
555 0 : assert!(!pageserver_connstring.is_empty());
556 :
557 0 : let safekeeper_connstrings = self.build_safekeepers_connstrs(safekeepers)?;
558 :
559 : // check for file remote_extensions_spec.json
560 : // if it is present, read it and pass to compute_ctl
561 0 : let remote_extensions_spec_path = self.endpoint_path().join("remote_extensions_spec.json");
562 0 : let remote_extensions_spec = std::fs::File::open(remote_extensions_spec_path);
563 : let remote_extensions: Option<RemoteExtSpec>;
564 :
565 0 : if let Ok(spec_file) = remote_extensions_spec {
566 0 : remote_extensions = serde_json::from_reader(spec_file).ok();
567 0 : } else {
568 0 : remote_extensions = None;
569 0 : };
570 :
571 : // Create spec file
572 0 : let spec = ComputeSpec {
573 0 : skip_pg_catalog_updates: self.skip_pg_catalog_updates,
574 0 : format_version: 1.0,
575 0 : operation_uuid: None,
576 0 : features: self.features.clone(),
577 0 : swap_size_bytes: None,
578 0 : disk_quota_bytes: None,
579 0 : cluster: Cluster {
580 0 : cluster_id: None, // project ID: not used
581 0 : name: None, // project name: not used
582 0 : state: None,
583 0 : roles: if create_test_user {
584 0 : vec![Role {
585 0 : name: PgIdent::from_str("test").unwrap(),
586 0 : encrypted_password: None,
587 0 : options: None,
588 0 : }]
589 : } else {
590 0 : Vec::new()
591 : },
592 0 : databases: if create_test_user {
593 0 : vec![Database {
594 0 : name: PgIdent::from_str("neondb").unwrap(),
595 0 : owner: PgIdent::from_str("test").unwrap(),
596 0 : options: None,
597 0 : restrict_conn: false,
598 0 : invalid: false,
599 0 : }]
600 : } else {
601 0 : Vec::new()
602 : },
603 0 : settings: None,
604 0 : postgresql_conf: Some(postgresql_conf),
605 0 : },
606 0 : delta_operations: None,
607 0 : tenant_id: Some(self.tenant_id),
608 0 : timeline_id: Some(self.timeline_id),
609 0 : mode: self.mode,
610 0 : pageserver_connstring: Some(pageserver_connstring),
611 0 : safekeeper_connstrings,
612 0 : storage_auth_token: auth_token.clone(),
613 0 : remote_extensions,
614 0 : pgbouncer_settings: None,
615 0 : shard_stripe_size: Some(shard_stripe_size),
616 0 : local_proxy_config: None,
617 0 : };
618 0 : let spec_path = self.endpoint_path().join("spec.json");
619 0 : std::fs::write(spec_path, serde_json::to_string_pretty(&spec)?)?;
620 :
621 : // Open log file. We'll redirect the stdout and stderr of `compute_ctl` to it.
622 0 : let logfile = std::fs::OpenOptions::new()
623 0 : .create(true)
624 0 : .append(true)
625 0 : .open(self.endpoint_path().join("compute.log"))?;
626 :
627 : // Launch compute_ctl
628 0 : let conn_str = self.connstr("cloud_admin", "postgres");
629 0 : println!("Starting postgres node at '{}'", conn_str);
630 0 : if create_test_user {
631 0 : let conn_str = self.connstr("test", "neondb");
632 0 : println!("Also at '{}'", conn_str);
633 0 : }
634 0 : let mut cmd = Command::new(self.env.neon_distrib_dir.join("compute_ctl"));
635 0 : cmd.args(["--http-port", &self.http_address.port().to_string()])
636 0 : .args(["--pgdata", self.pgdata().to_str().unwrap()])
637 0 : .args(["--connstr", &conn_str])
638 0 : .args([
639 0 : "--spec-path",
640 0 : self.endpoint_path().join("spec.json").to_str().unwrap(),
641 0 : ])
642 0 : .args([
643 0 : "--pgbin",
644 0 : self.env
645 0 : .pg_bin_dir(self.pg_version)?
646 0 : .join("postgres")
647 0 : .to_str()
648 0 : .unwrap(),
649 0 : ])
650 0 : .stdin(std::process::Stdio::null())
651 0 : .stderr(logfile.try_clone()?)
652 0 : .stdout(logfile);
653 :
654 0 : if let Some(remote_ext_config) = remote_ext_config {
655 0 : cmd.args(["--remote-ext-config", remote_ext_config]);
656 0 : }
657 :
658 0 : let child = cmd.spawn()?;
659 : // set up a scopeguard to kill & wait for the child in case we panic or bail below
660 0 : let child = scopeguard::guard(child, |mut child| {
661 0 : println!("SIGKILL & wait the started process");
662 0 : (|| {
663 0 : // TODO: use another signal that can be caught by the child so it can clean up any children it spawned
664 0 : child.kill().context("SIGKILL child")?;
665 0 : child.wait().context("wait() for child process")?;
666 0 : anyhow::Ok(())
667 0 : })()
668 0 : .with_context(|| format!("scopeguard kill&wait child {child:?}"))
669 0 : .unwrap();
670 0 : });
671 0 :
672 0 : // Write down the pid so we can wait for it when we want to stop
673 0 : // TODO use background_process::start_process instead: https://github.com/neondatabase/neon/pull/6482
674 0 : let pid = child.id();
675 0 : let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
676 0 : std::fs::write(pidfile_path, pid.to_string())?;
677 :
678 : // Wait for it to start
679 0 : let mut attempt = 0;
680 : const ATTEMPT_INTERVAL: Duration = Duration::from_millis(100);
681 : const MAX_ATTEMPTS: u32 = 10 * 90; // Wait up to 1.5 min
682 : loop {
683 0 : attempt += 1;
684 0 : match self.get_status().await {
685 0 : Ok(state) => {
686 0 : match state.status {
687 : ComputeStatus::Init => {
688 0 : if attempt == MAX_ATTEMPTS {
689 0 : bail!("compute startup timed out; still in Init state");
690 0 : }
691 : // keep retrying
692 : }
693 : ComputeStatus::Running => {
694 : // All good!
695 0 : break;
696 : }
697 : ComputeStatus::Failed => {
698 0 : bail!(
699 0 : "compute startup failed: {}",
700 0 : state
701 0 : .error
702 0 : .as_deref()
703 0 : .unwrap_or("<no error from compute_ctl>")
704 0 : );
705 : }
706 : ComputeStatus::Empty
707 : | ComputeStatus::ConfigurationPending
708 : | ComputeStatus::Configuration
709 : | ComputeStatus::TerminationPending
710 : | ComputeStatus::Terminated => {
711 0 : bail!("unexpected compute status: {:?}", state.status)
712 : }
713 : }
714 : }
715 0 : Err(e) => {
716 0 : if attempt == MAX_ATTEMPTS {
717 0 : return Err(e).context("timed out waiting to connect to compute_ctl HTTP");
718 0 : }
719 : }
720 : }
721 0 : tokio::time::sleep(ATTEMPT_INTERVAL).await;
722 : }
723 :
724 : // disarm the scopeguard, let the child outlive this function (and neon_local invoction)
725 0 : drop(scopeguard::ScopeGuard::into_inner(child));
726 0 :
727 0 : Ok(())
728 0 : }
729 :
730 : // Call the /status HTTP API
731 0 : pub async fn get_status(&self) -> Result<ComputeState> {
732 0 : let client = reqwest::Client::new();
733 :
734 0 : let response = client
735 0 : .request(
736 0 : reqwest::Method::GET,
737 0 : format!(
738 0 : "http://{}:{}/status",
739 0 : self.http_address.ip(),
740 0 : self.http_address.port()
741 0 : ),
742 0 : )
743 0 : .send()
744 0 : .await?;
745 :
746 : // Interpret the response
747 0 : let status = response.status();
748 0 : if !(status.is_client_error() || status.is_server_error()) {
749 0 : Ok(response.json().await?)
750 : } else {
751 : // reqwest does not export its error construction utility functions, so let's craft the message ourselves
752 0 : let url = response.url().to_owned();
753 0 : let msg = match response.text().await {
754 0 : Ok(err_body) => format!("Error: {}", err_body),
755 0 : Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
756 : };
757 0 : Err(anyhow::anyhow!(msg))
758 : }
759 0 : }
760 :
761 0 : pub async fn reconfigure(
762 0 : &self,
763 0 : mut pageservers: Vec<(Host, u16)>,
764 0 : stripe_size: Option<ShardStripeSize>,
765 0 : safekeepers: Option<Vec<NodeId>>,
766 0 : ) -> Result<()> {
767 0 : let mut spec: ComputeSpec = {
768 0 : let spec_path = self.endpoint_path().join("spec.json");
769 0 : let file = std::fs::File::open(spec_path)?;
770 0 : serde_json::from_reader(file)?
771 : };
772 :
773 0 : let postgresql_conf = self.read_postgresql_conf()?;
774 0 : spec.cluster.postgresql_conf = Some(postgresql_conf);
775 0 :
776 0 : // If we weren't given explicit pageservers, query the storage controller
777 0 : if pageservers.is_empty() {
778 0 : let storage_controller = StorageController::from_env(&self.env);
779 0 : let locate_result = storage_controller.tenant_locate(self.tenant_id).await?;
780 0 : pageservers = locate_result
781 0 : .shards
782 0 : .into_iter()
783 0 : .map(|shard| {
784 0 : (
785 0 : Host::parse(&shard.listen_pg_addr)
786 0 : .expect("Storage controller reported bad hostname"),
787 0 : shard.listen_pg_port,
788 0 : )
789 0 : })
790 0 : .collect::<Vec<_>>();
791 0 : }
792 :
793 0 : let pageserver_connstr = Self::build_pageserver_connstr(&pageservers);
794 0 : assert!(!pageserver_connstr.is_empty());
795 0 : spec.pageserver_connstring = Some(pageserver_connstr);
796 0 : if stripe_size.is_some() {
797 0 : spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
798 0 : }
799 :
800 : // If safekeepers are not specified, don't change them.
801 0 : if let Some(safekeepers) = safekeepers {
802 0 : let safekeeper_connstrings = self.build_safekeepers_connstrs(safekeepers)?;
803 0 : spec.safekeeper_connstrings = safekeeper_connstrings;
804 0 : }
805 :
806 0 : let client = reqwest::Client::builder()
807 0 : .timeout(Duration::from_secs(30))
808 0 : .build()
809 0 : .unwrap();
810 0 : let response = client
811 0 : .post(format!(
812 0 : "http://{}:{}/configure",
813 0 : self.http_address.ip(),
814 0 : self.http_address.port()
815 0 : ))
816 0 : .body(format!(
817 0 : "{{\"spec\":{}}}",
818 0 : serde_json::to_string_pretty(&spec)?
819 : ))
820 0 : .send()
821 0 : .await?;
822 :
823 0 : let status = response.status();
824 0 : if !(status.is_client_error() || status.is_server_error()) {
825 0 : Ok(())
826 : } else {
827 0 : let url = response.url().to_owned();
828 0 : let msg = match response.text().await {
829 0 : Ok(err_body) => format!("Error: {}", err_body),
830 0 : Err(_) => format!("Http error ({}) at {}.", status.as_u16(), url),
831 : };
832 0 : Err(anyhow::anyhow!(msg))
833 : }
834 0 : }
835 :
836 0 : pub fn stop(&self, mode: &str, destroy: bool) -> Result<()> {
837 0 : self.pg_ctl(&["-m", mode, "stop"], &None)?;
838 :
839 : // Also wait for the compute_ctl process to die. It might have some
840 : // cleanup work to do after postgres stops, like syncing safekeepers,
841 : // etc.
842 : //
843 : // If destroying or stop mode is immediate, send it SIGTERM before
844 : // waiting. Sometimes we do *not* want this cleanup: tests intentionally
845 : // do stop when majority of safekeepers is down, so sync-safekeepers
846 : // would hang otherwise. This could be a separate flag though.
847 0 : let send_sigterm = destroy || mode == "immediate";
848 0 : self.wait_for_compute_ctl_to_exit(send_sigterm)?;
849 0 : if destroy {
850 0 : println!(
851 0 : "Destroying postgres data directory '{}'",
852 0 : self.pgdata().to_str().unwrap()
853 0 : );
854 0 : std::fs::remove_dir_all(self.endpoint_path())?;
855 0 : }
856 0 : Ok(())
857 0 : }
858 :
859 0 : pub fn connstr(&self, user: &str, db_name: &str) -> String {
860 0 : format!(
861 0 : "postgresql://{}@{}:{}/{}",
862 0 : user,
863 0 : self.pg_address.ip(),
864 0 : self.pg_address.port(),
865 0 : db_name
866 0 : )
867 0 : }
868 : }
|
