Line data Source code
1 : //!
2 : //! Postgres wrapper (`compute_ctl`) is intended to be run as a Docker entrypoint or as a `systemd`
3 : //! `ExecStart` option. It will handle all the `Neon` specifics during compute node
4 : //! initialization:
5 : //! - `compute_ctl` accepts cluster (compute node) specification as a JSON file.
6 : //! - Every start is a fresh start, so the data directory is removed and
7 : //! initialized again on each run.
8 : //! - If remote_extension_config is provided, it will be used to fetch extensions list
9 : //! and download `shared_preload_libraries` from the remote storage.
10 : //! - Next it will put configuration files into the `PGDATA` directory.
11 : //! - Sync safekeepers and get commit LSN.
12 : //! - Get `basebackup` from pageserver using the returned on the previous step LSN.
13 : //! - Try to start `postgres` and wait until it is ready to accept connections.
14 : //! - Check and alter/drop/create roles and databases.
15 : //! - Hang waiting on the `postmaster` process to exit.
16 : //!
17 : //! Also `compute_ctl` spawns two separate service threads:
18 : //! - `compute-monitor` checks the last Postgres activity timestamp and saves it
19 : //! into the shared `ComputeNode`;
20 : //! - `http-endpoint` runs a Hyper HTTP API server, which serves readiness and the
21 : //! last activity requests.
22 : //!
23 : //! If `AUTOSCALING` environment variable is set, `compute_ctl` will start the
24 : //! `vm-monitor` located in [`neon/libs/vm_monitor`]. For VM compute nodes,
25 : //! `vm-monitor` communicates with the VM autoscaling system. It coordinates
26 : //! downscaling and requests immediate upscaling under resource pressure.
27 : //!
28 : //! Usage example:
29 : //! ```sh
30 : //! compute_ctl -D /var/db/postgres/compute \
31 : //! -C 'postgresql://cloud_admin@localhost/postgres' \
32 : //! -S /var/db/postgres/specs/current.json \
33 : //! -b /usr/local/bin/postgres \
34 : //! -r http://pg-ext-s3-gateway \
35 : //! ```
36 : //!
37 : use std::collections::HashMap;
38 : use std::fs::File;
39 : use std::path::Path;
40 : use std::process::exit;
41 : use std::sync::atomic::Ordering;
42 : use std::sync::{mpsc, Arc, Condvar, Mutex, RwLock};
43 : use std::{thread, time::Duration};
44 :
45 : use anyhow::{Context, Result};
46 : use chrono::Utc;
47 : use clap::Arg;
48 : use signal_hook::consts::{SIGQUIT, SIGTERM};
49 : use signal_hook::{consts::SIGINT, iterator::Signals};
50 : use tracing::{error, info, warn};
51 : use url::Url;
52 :
53 : use compute_api::responses::ComputeStatus;
54 : use compute_api::spec::ComputeSpec;
55 :
56 : use compute_tools::compute::{
57 : forward_termination_signal, ComputeNode, ComputeState, ParsedSpec, PG_PID,
58 : };
59 : use compute_tools::configurator::launch_configurator;
60 : use compute_tools::extension_server::get_pg_version;
61 : use compute_tools::http::api::launch_http_server;
62 : use compute_tools::logger::*;
63 : use compute_tools::monitor::launch_monitor;
64 : use compute_tools::params::*;
65 : use compute_tools::spec::*;
66 : use compute_tools::swap::resize_swap;
67 :
68 : // this is an arbitrary build tag. Fine as a default / for testing purposes
69 : // in-case of not-set environment var
70 : const BUILD_TAG_DEFAULT: &str = "latest";
71 :
72 0 : fn main() -> Result<()> {
73 0 : let (build_tag, clap_args) = init()?;
74 :
75 0 : let (pg_handle, start_pg_result) = {
76 : // Enter startup tracing context
77 0 : let _startup_context_guard = startup_context_from_env();
78 :
79 0 : let cli_args = process_cli(&clap_args)?;
80 :
81 0 : let cli_spec = try_spec_from_cli(&clap_args, &cli_args)?;
82 :
83 0 : let wait_spec_result = wait_spec(build_tag, cli_args, cli_spec)?;
84 :
85 0 : start_postgres(&clap_args, wait_spec_result)?
86 :
87 : // Startup is finished, exit the startup tracing span
88 : };
89 :
90 : // PostgreSQL is now running, if startup was successful. Wait until it exits.
91 0 : let wait_pg_result = wait_postgres(pg_handle)?;
92 :
93 0 : let delay_exit = cleanup_after_postgres_exit(start_pg_result)?;
94 :
95 0 : maybe_delay_exit(delay_exit);
96 0 :
97 0 : deinit_and_exit(wait_pg_result);
98 0 : }
99 :
100 0 : fn init() -> Result<(String, clap::ArgMatches)> {
101 0 : init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
102 :
103 0 : let mut signals = Signals::new([SIGINT, SIGTERM, SIGQUIT])?;
104 0 : thread::spawn(move || {
105 0 : for sig in signals.forever() {
106 0 : handle_exit_signal(sig);
107 0 : }
108 0 : });
109 0 :
110 0 : let build_tag = option_env!("BUILD_TAG")
111 0 : .unwrap_or(BUILD_TAG_DEFAULT)
112 0 : .to_string();
113 0 : info!("build_tag: {build_tag}");
114 :
115 0 : Ok((build_tag, cli().get_matches()))
116 0 : }
117 :
118 0 : fn process_cli(matches: &clap::ArgMatches) -> Result<ProcessCliResult> {
119 0 : let pgbin_default = "postgres";
120 0 : let pgbin = matches
121 0 : .get_one::<String>("pgbin")
122 0 : .map(|s| s.as_str())
123 0 : .unwrap_or(pgbin_default);
124 0 :
125 0 : let ext_remote_storage = matches
126 0 : .get_one::<String>("remote-ext-config")
127 0 : // Compatibility hack: if the control plane specified any remote-ext-config
128 0 : // use the default value for extension storage proxy gateway.
129 0 : // Remove this once the control plane is updated to pass the gateway URL
130 0 : .map(|conf| {
131 0 : if conf.starts_with("http") {
132 0 : conf.trim_end_matches('/')
133 : } else {
134 0 : "http://pg-ext-s3-gateway"
135 : }
136 0 : });
137 0 :
138 0 : let http_port = *matches
139 0 : .get_one::<u16>("http-port")
140 0 : .expect("http-port is required");
141 0 : let pgdata = matches
142 0 : .get_one::<String>("pgdata")
143 0 : .expect("PGDATA path is required");
144 0 : let connstr = matches
145 0 : .get_one::<String>("connstr")
146 0 : .expect("Postgres connection string is required");
147 0 : let spec_json = matches.get_one::<String>("spec");
148 0 : let spec_path = matches.get_one::<String>("spec-path");
149 0 : let resize_swap_on_bind = matches.get_flag("resize-swap-on-bind");
150 0 :
151 0 : Ok(ProcessCliResult {
152 0 : connstr,
153 0 : pgdata,
154 0 : pgbin,
155 0 : ext_remote_storage,
156 0 : http_port,
157 0 : spec_json,
158 0 : spec_path,
159 0 : resize_swap_on_bind,
160 0 : })
161 0 : }
162 :
163 : struct ProcessCliResult<'clap> {
164 : connstr: &'clap str,
165 : pgdata: &'clap str,
166 : pgbin: &'clap str,
167 : ext_remote_storage: Option<&'clap str>,
168 : http_port: u16,
169 : spec_json: Option<&'clap String>,
170 : spec_path: Option<&'clap String>,
171 : resize_swap_on_bind: bool,
172 : }
173 :
174 0 : fn startup_context_from_env() -> Option<opentelemetry::ContextGuard> {
175 0 : // Extract OpenTelemetry context for the startup actions from the
176 0 : // TRACEPARENT and TRACESTATE env variables, and attach it to the current
177 0 : // tracing context.
178 0 : //
179 0 : // This is used to propagate the context for the 'start_compute' operation
180 0 : // from the neon control plane. This allows linking together the wider
181 0 : // 'start_compute' operation that creates the compute container, with the
182 0 : // startup actions here within the container.
183 0 : //
184 0 : // There is no standard for passing context in env variables, but a lot of
185 0 : // tools use TRACEPARENT/TRACESTATE, so we use that convention too. See
186 0 : // https://github.com/open-telemetry/opentelemetry-specification/issues/740
187 0 : //
188 0 : // Switch to the startup context here, and exit it once the startup has
189 0 : // completed and Postgres is up and running.
190 0 : //
191 0 : // If this pod is pre-created without binding it to any particular endpoint
192 0 : // yet, this isn't the right place to enter the startup context. In that
193 0 : // case, the control plane should pass the tracing context as part of the
194 0 : // /configure API call.
195 0 : //
196 0 : // NOTE: This is supposed to only cover the *startup* actions. Once
197 0 : // postgres is configured and up-and-running, we exit this span. Any other
198 0 : // actions that are performed on incoming HTTP requests, for example, are
199 0 : // performed in separate spans.
200 0 : //
201 0 : // XXX: If the pod is restarted, we perform the startup actions in the same
202 0 : // context as the original startup actions, which probably doesn't make
203 0 : // sense.
204 0 : let mut startup_tracing_carrier: HashMap<String, String> = HashMap::new();
205 0 : if let Ok(val) = std::env::var("TRACEPARENT") {
206 0 : startup_tracing_carrier.insert("traceparent".to_string(), val);
207 0 : }
208 0 : if let Ok(val) = std::env::var("TRACESTATE") {
209 0 : startup_tracing_carrier.insert("tracestate".to_string(), val);
210 0 : }
211 0 : if !startup_tracing_carrier.is_empty() {
212 : use opentelemetry::propagation::TextMapPropagator;
213 : use opentelemetry::sdk::propagation::TraceContextPropagator;
214 0 : let guard = TraceContextPropagator::new()
215 0 : .extract(&startup_tracing_carrier)
216 0 : .attach();
217 0 : info!("startup tracing context attached");
218 0 : Some(guard)
219 : } else {
220 0 : None
221 : }
222 0 : }
223 :
224 0 : fn try_spec_from_cli(
225 0 : matches: &clap::ArgMatches,
226 0 : ProcessCliResult {
227 0 : spec_json,
228 0 : spec_path,
229 0 : ..
230 0 : }: &ProcessCliResult,
231 0 : ) -> Result<CliSpecParams> {
232 0 : let compute_id = matches.get_one::<String>("compute-id");
233 0 : let control_plane_uri = matches.get_one::<String>("control-plane-uri");
234 0 :
235 0 : let spec;
236 0 : let mut live_config_allowed = false;
237 0 : match spec_json {
238 : // First, try to get cluster spec from the cli argument
239 0 : Some(json) => {
240 0 : info!("got spec from cli argument {}", json);
241 0 : spec = Some(serde_json::from_str(json)?);
242 : }
243 : None => {
244 : // Second, try to read it from the file if path is provided
245 0 : if let Some(sp) = spec_path {
246 0 : let path = Path::new(sp);
247 0 : let file = File::open(path)?;
248 0 : spec = Some(serde_json::from_reader(file)?);
249 0 : live_config_allowed = true;
250 0 : } else if let Some(id) = compute_id {
251 0 : if let Some(cp_base) = control_plane_uri {
252 0 : live_config_allowed = true;
253 0 : spec = match get_spec_from_control_plane(cp_base, id) {
254 0 : Ok(s) => s,
255 0 : Err(e) => {
256 0 : error!("cannot get response from control plane: {}", e);
257 0 : panic!("neither spec nor confirmation that compute is in the Empty state was received");
258 : }
259 : };
260 : } else {
261 0 : panic!("must specify both --control-plane-uri and --compute-id or none");
262 : }
263 : } else {
264 0 : panic!(
265 0 : "compute spec should be provided by one of the following ways: \
266 0 : --spec OR --spec-path OR --control-plane-uri and --compute-id"
267 0 : );
268 : }
269 : }
270 : };
271 :
272 0 : Ok(CliSpecParams {
273 0 : spec,
274 0 : live_config_allowed,
275 0 : })
276 0 : }
277 :
278 : struct CliSpecParams {
279 : /// If a spec was provided via CLI or file, the [`ComputeSpec`]
280 : spec: Option<ComputeSpec>,
281 : live_config_allowed: bool,
282 : }
283 :
284 0 : fn wait_spec(
285 0 : build_tag: String,
286 0 : ProcessCliResult {
287 0 : connstr,
288 0 : pgdata,
289 0 : pgbin,
290 0 : ext_remote_storage,
291 0 : resize_swap_on_bind,
292 0 : http_port,
293 0 : ..
294 0 : }: ProcessCliResult,
295 0 : CliSpecParams {
296 0 : spec,
297 0 : live_config_allowed,
298 0 : }: CliSpecParams,
299 0 : ) -> Result<WaitSpecResult> {
300 0 : let mut new_state = ComputeState::new();
301 : let spec_set;
302 :
303 0 : if let Some(spec) = spec {
304 0 : let pspec = ParsedSpec::try_from(spec).map_err(|msg| anyhow::anyhow!(msg))?;
305 0 : info!("new pspec.spec: {:?}", pspec.spec);
306 0 : new_state.pspec = Some(pspec);
307 0 : spec_set = true;
308 0 : } else {
309 0 : spec_set = false;
310 0 : }
311 0 : let compute_node = ComputeNode {
312 0 : connstr: Url::parse(connstr).context("cannot parse connstr as a URL")?,
313 0 : pgdata: pgdata.to_string(),
314 0 : pgbin: pgbin.to_string(),
315 0 : pgversion: get_pg_version(pgbin),
316 0 : live_config_allowed,
317 0 : state: Mutex::new(new_state),
318 0 : state_changed: Condvar::new(),
319 0 : ext_remote_storage: ext_remote_storage.map(|s| s.to_string()),
320 0 : ext_download_progress: RwLock::new(HashMap::new()),
321 0 : build_tag,
322 0 : };
323 0 : let compute = Arc::new(compute_node);
324 0 :
325 0 : // If this is a pooled VM, prewarm before starting HTTP server and becoming
326 0 : // available for binding. Prewarming helps Postgres start quicker later,
327 0 : // because QEMU will already have its memory allocated from the host, and
328 0 : // the necessary binaries will already be cached.
329 0 : if !spec_set {
330 0 : compute.prewarm_postgres()?;
331 0 : }
332 :
333 : // Launch http service first, so that we can serve control-plane requests
334 : // while configuration is still in progress.
335 0 : let _http_handle =
336 0 : launch_http_server(http_port, &compute).expect("cannot launch http endpoint thread");
337 0 :
338 0 : if !spec_set {
339 : // No spec provided, hang waiting for it.
340 0 : info!("no compute spec provided, waiting");
341 :
342 0 : let mut state = compute.state.lock().unwrap();
343 0 : while state.status != ComputeStatus::ConfigurationPending {
344 0 : state = compute.state_changed.wait(state).unwrap();
345 0 :
346 0 : if state.status == ComputeStatus::ConfigurationPending {
347 0 : info!("got spec, continue configuration");
348 : // Spec is already set by the http server handler.
349 0 : break;
350 0 : }
351 : }
352 :
353 : // Record for how long we slept waiting for the spec.
354 0 : let now = Utc::now();
355 0 : state.metrics.wait_for_spec_ms = now
356 0 : .signed_duration_since(state.start_time)
357 0 : .to_std()
358 0 : .unwrap()
359 0 : .as_millis() as u64;
360 0 :
361 0 : // Reset start time, so that the total startup time that is calculated later will
362 0 : // not include the time that we waited for the spec.
363 0 : state.start_time = now;
364 0 : }
365 :
366 0 : Ok(WaitSpecResult {
367 0 : compute,
368 0 : http_port,
369 0 : resize_swap_on_bind,
370 0 : })
371 0 : }
372 :
373 : struct WaitSpecResult {
374 : compute: Arc<ComputeNode>,
375 : // passed through from ProcessCliResult
376 : http_port: u16,
377 : resize_swap_on_bind: bool,
378 : }
379 :
380 0 : fn start_postgres(
381 0 : // need to allow unused because `matches` is only used if target_os = "linux"
382 0 : #[allow(unused_variables)] matches: &clap::ArgMatches,
383 0 : WaitSpecResult {
384 0 : compute,
385 0 : http_port,
386 0 : resize_swap_on_bind,
387 0 : }: WaitSpecResult,
388 0 : ) -> Result<(Option<PostgresHandle>, StartPostgresResult)> {
389 0 : // We got all we need, update the state.
390 0 : let mut state = compute.state.lock().unwrap();
391 0 : state.status = ComputeStatus::Init;
392 0 : compute.state_changed.notify_all();
393 0 :
394 0 : info!(
395 0 : "running compute with features: {:?}",
396 0 : state.pspec.as_ref().unwrap().spec.features
397 : );
398 : // before we release the mutex, fetch the swap size (if any) for later.
399 0 : let swap_size_bytes = state.pspec.as_ref().unwrap().spec.swap_size_bytes;
400 0 : drop(state);
401 0 :
402 0 : // Launch remaining service threads
403 0 : let _monitor_handle = launch_monitor(&compute);
404 0 : let _configurator_handle = launch_configurator(&compute);
405 0 :
406 0 : let mut prestartup_failed = false;
407 0 : let mut delay_exit = false;
408 :
409 : // Resize swap to the desired size if the compute spec says so
410 0 : if let (Some(size_bytes), true) = (swap_size_bytes, resize_swap_on_bind) {
411 : // To avoid 'swapoff' hitting postgres startup, we need to run resize-swap to completion
412 : // *before* starting postgres.
413 : //
414 : // In theory, we could do this asynchronously if SkipSwapon was enabled for VMs, but this
415 : // carries a risk of introducing hard-to-debug issues - e.g. if postgres sometimes gets
416 : // OOM-killed during startup because swap wasn't available yet.
417 0 : match resize_swap(size_bytes) {
418 : Ok(()) => {
419 0 : let size_gib = size_bytes as f32 / (1 << 20) as f32; // just for more coherent display.
420 0 : info!(%size_bytes, %size_gib, "resized swap");
421 : }
422 0 : Err(err) => {
423 0 : let err = err.context("failed to resize swap");
424 0 : error!("{err:#}");
425 :
426 : // Mark compute startup as failed; don't try to start postgres, and report this
427 : // error to the control plane when it next asks.
428 0 : prestartup_failed = true;
429 0 : let mut state = compute.state.lock().unwrap();
430 0 : state.error = Some(format!("{err:?}"));
431 0 : state.status = ComputeStatus::Failed;
432 0 : compute.state_changed.notify_all();
433 0 : delay_exit = true;
434 : }
435 : }
436 0 : }
437 :
438 0 : let extension_server_port: u16 = http_port;
439 0 :
440 0 : // Start Postgres
441 0 : let mut pg = None;
442 0 : if !prestartup_failed {
443 0 : pg = match compute.start_compute(extension_server_port) {
444 0 : Ok(pg) => Some(pg),
445 0 : Err(err) => {
446 0 : error!("could not start the compute node: {:#}", err);
447 0 : let mut state = compute.state.lock().unwrap();
448 0 : state.error = Some(format!("{:?}", err));
449 0 : state.status = ComputeStatus::Failed;
450 0 : // Notify others that Postgres failed to start. In case of configuring the
451 0 : // empty compute, it's likely that API handler is still waiting for compute
452 0 : // state change. With this we will notify it that compute is in Failed state,
453 0 : // so control plane will know about it earlier and record proper error instead
454 0 : // of timeout.
455 0 : compute.state_changed.notify_all();
456 0 : drop(state); // unlock
457 0 : delay_exit = true;
458 0 : None
459 : }
460 : };
461 : } else {
462 0 : warn!("skipping postgres startup because pre-startup step failed");
463 : }
464 :
465 : // Start the vm-monitor if directed to. The vm-monitor only runs on linux
466 : // because it requires cgroups.
467 : cfg_if::cfg_if! {
468 : if #[cfg(target_os = "linux")] {
469 : use std::env;
470 : use tokio_util::sync::CancellationToken;
471 0 : let vm_monitor_addr = matches
472 0 : .get_one::<String>("vm-monitor-addr")
473 0 : .expect("--vm-monitor-addr should always be set because it has a default arg");
474 0 : let file_cache_connstr = matches.get_one::<String>("filecache-connstr");
475 0 : let cgroup = matches.get_one::<String>("cgroup");
476 :
477 : // Only make a runtime if we need to.
478 : // Note: it seems like you can make a runtime in an inner scope and
479 : // if you start a task in it it won't be dropped. However, make it
480 : // in the outermost scope just to be safe.
481 0 : let rt = if env::var_os("AUTOSCALING").is_some() {
482 0 : Some(
483 0 : tokio::runtime::Builder::new_multi_thread()
484 0 : .worker_threads(4)
485 0 : .enable_all()
486 0 : .build()
487 0 : .expect("failed to create tokio runtime for monitor")
488 0 : )
489 : } else {
490 0 : None
491 : };
492 :
493 : // This token is used internally by the monitor to clean up all threads
494 0 : let token = CancellationToken::new();
495 0 :
496 0 : let vm_monitor = rt.as_ref().map(|rt| {
497 0 : rt.spawn(vm_monitor::start(
498 0 : Box::leak(Box::new(vm_monitor::Args {
499 0 : cgroup: cgroup.cloned(),
500 0 : pgconnstr: file_cache_connstr.cloned(),
501 0 : addr: vm_monitor_addr.clone(),
502 0 : })),
503 0 : token.clone(),
504 0 : ))
505 0 : });
506 0 : }
507 0 : }
508 0 :
509 0 : Ok((
510 0 : pg,
511 0 : StartPostgresResult {
512 0 : delay_exit,
513 0 : compute,
514 0 : #[cfg(target_os = "linux")]
515 0 : rt,
516 0 : #[cfg(target_os = "linux")]
517 0 : token,
518 0 : #[cfg(target_os = "linux")]
519 0 : vm_monitor,
520 0 : },
521 0 : ))
522 0 : }
523 :
524 : type PostgresHandle = (std::process::Child, std::thread::JoinHandle<()>);
525 :
526 : struct StartPostgresResult {
527 : delay_exit: bool,
528 : // passed through from WaitSpecResult
529 : compute: Arc<ComputeNode>,
530 :
531 : #[cfg(target_os = "linux")]
532 : rt: Option<tokio::runtime::Runtime>,
533 : #[cfg(target_os = "linux")]
534 : token: tokio_util::sync::CancellationToken,
535 : #[cfg(target_os = "linux")]
536 : vm_monitor: Option<tokio::task::JoinHandle<Result<()>>>,
537 : }
538 :
539 0 : fn wait_postgres(pg: Option<PostgresHandle>) -> Result<WaitPostgresResult> {
540 0 : // Wait for the child Postgres process forever. In this state Ctrl+C will
541 0 : // propagate to Postgres and it will be shut down as well.
542 0 : let mut exit_code = None;
543 0 : if let Some((mut pg, logs_handle)) = pg {
544 0 : let ecode = pg
545 0 : .wait()
546 0 : .expect("failed to start waiting on Postgres process");
547 0 : PG_PID.store(0, Ordering::SeqCst);
548 0 :
549 0 : // Process has exited, so we can join the logs thread.
550 0 : let _ = logs_handle
551 0 : .join()
552 0 : .map_err(|e| tracing::error!("log thread panicked: {:?}", e));
553 0 :
554 0 : info!("Postgres exited with code {}, shutting down", ecode);
555 0 : exit_code = ecode.code()
556 0 : }
557 :
558 0 : Ok(WaitPostgresResult { exit_code })
559 0 : }
560 :
561 : struct WaitPostgresResult {
562 : exit_code: Option<i32>,
563 : }
564 :
565 0 : fn cleanup_after_postgres_exit(
566 0 : StartPostgresResult {
567 0 : mut delay_exit,
568 0 : compute,
569 0 : #[cfg(target_os = "linux")]
570 0 : vm_monitor,
571 0 : #[cfg(target_os = "linux")]
572 0 : token,
573 0 : #[cfg(target_os = "linux")]
574 0 : rt,
575 0 : }: StartPostgresResult,
576 0 : ) -> Result<bool> {
577 : // Terminate the vm_monitor so it releases the file watcher on
578 : // /sys/fs/cgroup/neon-postgres.
579 : // Note: the vm-monitor only runs on linux because it requires cgroups.
580 : cfg_if::cfg_if! {
581 : if #[cfg(target_os = "linux")] {
582 0 : if let Some(handle) = vm_monitor {
583 0 : // Kills all threads spawned by the monitor
584 0 : token.cancel();
585 0 : // Kills the actual task running the monitor
586 0 : handle.abort();
587 0 :
588 0 : // If handle is some, rt must have been used to produce it, and
589 0 : // hence is also some
590 0 : rt.unwrap().shutdown_timeout(Duration::from_secs(2));
591 0 : }
592 : }
593 : }
594 :
595 : // Maybe sync safekeepers again, to speed up next startup
596 0 : let compute_state = compute.state.lock().unwrap().clone();
597 0 : let pspec = compute_state.pspec.as_ref().expect("spec must be set");
598 0 : if matches!(pspec.spec.mode, compute_api::spec::ComputeMode::Primary) {
599 0 : info!("syncing safekeepers on shutdown");
600 0 : let storage_auth_token = pspec.storage_auth_token.clone();
601 0 : let lsn = compute.sync_safekeepers(storage_auth_token)?;
602 0 : info!("synced safekeepers at lsn {lsn}");
603 0 : }
604 :
605 0 : let mut state = compute.state.lock().unwrap();
606 0 : if state.status == ComputeStatus::TerminationPending {
607 0 : state.status = ComputeStatus::Terminated;
608 0 : compute.state_changed.notify_all();
609 0 : // we were asked to terminate gracefully, don't exit to avoid restart
610 0 : delay_exit = true
611 0 : }
612 0 : drop(state);
613 :
614 0 : if let Err(err) = compute.check_for_core_dumps() {
615 0 : error!("error while checking for core dumps: {err:?}");
616 0 : }
617 :
618 0 : Ok(delay_exit)
619 0 : }
620 :
621 0 : fn maybe_delay_exit(delay_exit: bool) {
622 0 : // If launch failed, keep serving HTTP requests for a while, so the cloud
623 0 : // control plane can get the actual error.
624 0 : if delay_exit {
625 0 : info!("giving control plane 30s to collect the error before shutdown");
626 0 : thread::sleep(Duration::from_secs(30));
627 0 : }
628 0 : }
629 :
630 0 : fn deinit_and_exit(WaitPostgresResult { exit_code }: WaitPostgresResult) -> ! {
631 0 : // Shutdown trace pipeline gracefully, so that it has a chance to send any
632 0 : // pending traces before we exit. Shutting down OTEL tracing provider may
633 0 : // hang for quite some time, see, for example:
634 0 : // - https://github.com/open-telemetry/opentelemetry-rust/issues/868
635 0 : // - and our problems with staging https://github.com/neondatabase/cloud/issues/3707#issuecomment-1493983636
636 0 : //
637 0 : // Yet, we want computes to shut down fast enough, as we may need a new one
638 0 : // for the same timeline ASAP. So wait no longer than 2s for the shutdown to
639 0 : // complete, then just error out and exit the main thread.
640 0 : info!("shutting down tracing");
641 0 : let (sender, receiver) = mpsc::channel();
642 0 : let _ = thread::spawn(move || {
643 0 : tracing_utils::shutdown_tracing();
644 0 : sender.send(()).ok()
645 0 : });
646 0 : let shutdown_res = receiver.recv_timeout(Duration::from_millis(2000));
647 0 : if shutdown_res.is_err() {
648 0 : error!("timed out while shutting down tracing, exiting anyway");
649 0 : }
650 :
651 0 : info!("shutting down");
652 0 : exit(exit_code.unwrap_or(1))
653 : }
654 :
655 2 : fn cli() -> clap::Command {
656 2 : // Env variable is set by `cargo`
657 2 : let version = option_env!("CARGO_PKG_VERSION").unwrap_or("unknown");
658 2 : clap::Command::new("compute_ctl")
659 2 : .version(version)
660 2 : .arg(
661 2 : Arg::new("http-port")
662 2 : .long("http-port")
663 2 : .value_name("HTTP_PORT")
664 2 : .default_value("3080")
665 2 : .value_parser(clap::value_parser!(u16))
666 2 : .required(false),
667 2 : )
668 2 : .arg(
669 2 : Arg::new("connstr")
670 2 : .short('C')
671 2 : .long("connstr")
672 2 : .value_name("DATABASE_URL")
673 2 : .required(true),
674 2 : )
675 2 : .arg(
676 2 : Arg::new("pgdata")
677 2 : .short('D')
678 2 : .long("pgdata")
679 2 : .value_name("DATADIR")
680 2 : .required(true),
681 2 : )
682 2 : .arg(
683 2 : Arg::new("pgbin")
684 2 : .short('b')
685 2 : .long("pgbin")
686 2 : .default_value("postgres")
687 2 : .value_name("POSTGRES_PATH"),
688 2 : )
689 2 : .arg(
690 2 : Arg::new("spec")
691 2 : .short('s')
692 2 : .long("spec")
693 2 : .value_name("SPEC_JSON"),
694 2 : )
695 2 : .arg(
696 2 : Arg::new("spec-path")
697 2 : .short('S')
698 2 : .long("spec-path")
699 2 : .value_name("SPEC_PATH"),
700 2 : )
701 2 : .arg(
702 2 : Arg::new("compute-id")
703 2 : .short('i')
704 2 : .long("compute-id")
705 2 : .value_name("COMPUTE_ID"),
706 2 : )
707 2 : .arg(
708 2 : Arg::new("control-plane-uri")
709 2 : .short('p')
710 2 : .long("control-plane-uri")
711 2 : .value_name("CONTROL_PLANE_API_BASE_URI"),
712 2 : )
713 2 : .arg(
714 2 : Arg::new("remote-ext-config")
715 2 : .short('r')
716 2 : .long("remote-ext-config")
717 2 : .value_name("REMOTE_EXT_CONFIG"),
718 2 : )
719 2 : // TODO(fprasx): we currently have default arguments because the cloud PR
720 2 : // to pass them in hasn't been merged yet. We should get rid of them once
721 2 : // the PR is merged.
722 2 : .arg(
723 2 : Arg::new("vm-monitor-addr")
724 2 : .long("vm-monitor-addr")
725 2 : .default_value("0.0.0.0:10301")
726 2 : .value_name("VM_MONITOR_ADDR"),
727 2 : )
728 2 : .arg(
729 2 : Arg::new("cgroup")
730 2 : .long("cgroup")
731 2 : .default_value("neon-postgres")
732 2 : .value_name("CGROUP"),
733 2 : )
734 2 : .arg(
735 2 : Arg::new("filecache-connstr")
736 2 : .long("filecache-connstr")
737 2 : .default_value(
738 2 : "host=localhost port=5432 dbname=postgres user=cloud_admin sslmode=disable application_name=vm-monitor",
739 2 : )
740 2 : .value_name("FILECACHE_CONNSTR"),
741 2 : )
742 2 : .arg(
743 2 : Arg::new("resize-swap-on-bind")
744 2 : .long("resize-swap-on-bind")
745 2 : .action(clap::ArgAction::SetTrue),
746 2 : )
747 2 : }
748 :
749 : /// When compute_ctl is killed, send also termination signal to sync-safekeepers
750 : /// to prevent leakage. TODO: it is better to convert compute_ctl to async and
751 : /// wait for termination which would be easy then.
752 0 : fn handle_exit_signal(sig: i32) {
753 0 : info!("received {sig} termination signal");
754 0 : forward_termination_signal();
755 0 : exit(1);
756 : }
757 :
758 : #[test]
759 2 : fn verify_cli() {
760 2 : cli().debug_assert()
761 2 : }
|
