Line data Source code
1 : //
2 : // Main entry point for the safekeeper executable
3 : //
4 : use anyhow::{bail, Context, Result};
5 : use camino::{Utf8Path, Utf8PathBuf};
6 : use clap::{ArgAction, Parser};
7 : use futures::future::BoxFuture;
8 : use futures::stream::FuturesUnordered;
9 : use futures::{FutureExt, StreamExt};
10 : use remote_storage::RemoteStorageConfig;
11 : use sd_notify::NotifyState;
12 : use tokio::runtime::Handle;
13 : use tokio::signal::unix::{signal, SignalKind};
14 : use tokio::task::JoinError;
15 : use utils::logging::SecretString;
16 :
17 : use std::env::{var, VarError};
18 : use std::fs::{self, File};
19 : use std::io::{ErrorKind, Write};
20 : use std::str::FromStr;
21 : use std::sync::Arc;
22 : use std::time::Duration;
23 : use storage_broker::Uri;
24 :
25 : use tracing::*;
26 : use utils::pid_file;
27 :
28 : use metrics::set_build_info_metric;
29 : use safekeeper::defaults::{
30 : DEFAULT_CONTROL_FILE_SAVE_INTERVAL, DEFAULT_EVICTION_MIN_RESIDENT, DEFAULT_HEARTBEAT_TIMEOUT,
31 : DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES, DEFAULT_PARTIAL_BACKUP_CONCURRENCY,
32 : DEFAULT_PARTIAL_BACKUP_TIMEOUT, DEFAULT_PG_LISTEN_ADDR,
33 : };
34 : use safekeeper::http;
35 : use safekeeper::wal_service;
36 : use safekeeper::GlobalTimelines;
37 : use safekeeper::SafeKeeperConf;
38 : use safekeeper::{broker, WAL_SERVICE_RUNTIME};
39 : use safekeeper::{control_file, BROKER_RUNTIME};
40 : use safekeeper::{wal_backup, HTTP_RUNTIME};
41 : use storage_broker::DEFAULT_ENDPOINT;
42 : use utils::auth::{JwtAuth, Scope, SwappableJwtAuth};
43 : use utils::{
44 : id::NodeId,
45 : logging::{self, LogFormat},
46 : project_build_tag, project_git_version,
47 : sentry_init::init_sentry,
48 : tcp_listener,
49 : };
50 :
51 : const PID_FILE_NAME: &str = "safekeeper.pid";
52 : const ID_FILE_NAME: &str = "safekeeper.id";
53 :
54 : project_git_version!(GIT_VERSION);
55 : project_build_tag!(BUILD_TAG);
56 :
57 : const FEATURES: &[&str] = &[
58 : #[cfg(feature = "testing")]
59 : "testing",
60 : ];
61 :
62 0 : fn version() -> String {
63 0 : format!(
64 0 : "{GIT_VERSION} failpoints: {}, features: {:?}",
65 0 : fail::has_failpoints(),
66 0 : FEATURES,
67 0 : )
68 0 : }
69 :
70 : const ABOUT: &str = r#"
71 : A fleet of safekeepers is responsible for reliably storing WAL received from
72 : compute, passing it through consensus (mitigating potential computes brain
73 : split), and serving the hardened part further downstream to pageserver(s).
74 : "#;
75 :
76 2 : #[derive(Parser)]
77 : #[command(name = "Neon safekeeper", version = GIT_VERSION, about = ABOUT, long_about = None)]
78 : struct Args {
79 : /// Path to the safekeeper data directory.
80 : #[arg(short = 'D', long, default_value = "./")]
81 0 : datadir: Utf8PathBuf,
82 : /// Safekeeper node id.
83 : #[arg(long)]
84 : id: Option<u64>,
85 : /// Initialize safekeeper with given id and exit.
86 : #[arg(long)]
87 0 : init: bool,
88 : /// Listen endpoint for receiving/sending WAL in the form host:port.
89 : #[arg(short, long, default_value = DEFAULT_PG_LISTEN_ADDR)]
90 0 : listen_pg: String,
91 : /// Listen endpoint for receiving/sending WAL in the form host:port allowing
92 : /// only tenant scoped auth tokens. Pointless if auth is disabled.
93 : #[arg(long, default_value = None, verbatim_doc_comment)]
94 : listen_pg_tenant_only: Option<String>,
95 : /// Listen http endpoint for management and metrics in the form host:port.
96 : #[arg(long, default_value = DEFAULT_HTTP_LISTEN_ADDR)]
97 0 : listen_http: String,
98 : /// Advertised endpoint for receiving/sending WAL in the form host:port. If not
99 : /// specified, listen_pg is used to advertise instead.
100 : #[arg(long, default_value = None)]
101 : advertise_pg: Option<String>,
102 : /// Availability zone of the safekeeper.
103 : #[arg(long)]
104 : availability_zone: Option<String>,
105 : /// Do not wait for changes to be written safely to disk. Unsafe.
106 : #[arg(short, long)]
107 0 : no_sync: bool,
108 : /// Dump control file at path specified by this argument and exit.
109 : #[arg(long)]
110 : dump_control_file: Option<Utf8PathBuf>,
111 : /// Broker endpoint for storage nodes coordination in the form
112 : /// http[s]://host:port. In case of https schema TLS is connection is
113 : /// established; plaintext otherwise.
114 : #[arg(long, default_value = DEFAULT_ENDPOINT, verbatim_doc_comment)]
115 0 : broker_endpoint: Uri,
116 : /// Broker keepalive interval.
117 : #[arg(long, value_parser= humantime::parse_duration, default_value = storage_broker::DEFAULT_KEEPALIVE_INTERVAL)]
118 0 : broker_keepalive_interval: Duration,
119 : /// Peer safekeeper is considered dead after not receiving heartbeats from
120 : /// it during this period passed as a human readable duration.
121 : #[arg(long, value_parser= humantime::parse_duration, default_value = DEFAULT_HEARTBEAT_TIMEOUT, verbatim_doc_comment)]
122 0 : heartbeat_timeout: Duration,
123 : /// Enable/disable peer recovery.
124 : #[arg(long, default_value = "false", action=ArgAction::Set)]
125 0 : peer_recovery: bool,
126 : /// Remote storage configuration for WAL backup (offloading to s3) as TOML
127 : /// inline table, e.g.
128 : /// {max_concurrent_syncs = 17, max_sync_errors = 13, bucket_name = "<BUCKETNAME>", bucket_region = "<REGION>", concurrency_limit = 119}
129 : /// Safekeeper offloads WAL to
130 : /// [prefix_in_bucket/]<tenant_id>/<timeline_id>/<segment_file>, mirroring
131 : /// structure on the file system.
132 : #[arg(long, value_parser = parse_remote_storage, verbatim_doc_comment)]
133 : remote_storage: Option<RemoteStorageConfig>,
134 : /// Safekeeper won't be elected for WAL offloading if it is lagging for more than this value in bytes
135 2 : #[arg(long, default_value_t = DEFAULT_MAX_OFFLOADER_LAG_BYTES)]
136 0 : max_offloader_lag: u64,
137 : /// Number of max parallel WAL segments to be offloaded to remote storage.
138 : #[arg(long, default_value = "5")]
139 0 : wal_backup_parallel_jobs: usize,
140 : /// Disable WAL backup to s3. When disabled, safekeeper removes WAL ignoring
141 : /// WAL backup horizon.
142 : #[arg(long)]
143 0 : disable_wal_backup: bool,
144 : /// If given, enables auth on incoming connections to WAL service endpoint
145 : /// (--listen-pg). Value specifies path to a .pem public key used for
146 : /// validations of JWT tokens. Empty string is allowed and means disabling
147 : /// auth.
148 : #[arg(long, verbatim_doc_comment, value_parser = opt_pathbuf_parser)]
149 : pg_auth_public_key_path: Option<Utf8PathBuf>,
150 : /// If given, enables auth on incoming connections to tenant only WAL
151 : /// service endpoint (--listen-pg-tenant-only). Value specifies path to a
152 : /// .pem public key used for validations of JWT tokens. Empty string is
153 : /// allowed and means disabling auth.
154 : #[arg(long, verbatim_doc_comment, value_parser = opt_pathbuf_parser)]
155 : pg_tenant_only_auth_public_key_path: Option<Utf8PathBuf>,
156 : /// If given, enables auth on incoming connections to http management
157 : /// service endpoint (--listen-http). Value specifies path to a .pem public
158 : /// key used for validations of JWT tokens. Empty string is allowed and
159 : /// means disabling auth.
160 : #[arg(long, verbatim_doc_comment, value_parser = opt_pathbuf_parser)]
161 : http_auth_public_key_path: Option<Utf8PathBuf>,
162 : /// Format for logging, either 'plain' or 'json'.
163 : #[arg(long, default_value = "plain")]
164 0 : log_format: String,
165 : /// Run everything in single threaded current thread runtime, might be
166 : /// useful for debugging.
167 : #[arg(long)]
168 0 : current_thread_runtime: bool,
169 : /// Keep horizon for walsenders, i.e. don't remove WAL segments that are
170 : /// still needed for existing replication connection.
171 : #[arg(long)]
172 0 : walsenders_keep_horizon: bool,
173 : /// Enable partial backup. If disabled, safekeeper will not upload partial
174 : /// segments to remote storage.
175 : /// TODO: now partial backup is always enabled, remove this flag.
176 : #[arg(long)]
177 0 : partial_backup_enabled: bool,
178 : /// Controls how long backup will wait until uploading the partial segment.
179 : #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_PARTIAL_BACKUP_TIMEOUT, verbatim_doc_comment)]
180 0 : partial_backup_timeout: Duration,
181 : /// Disable task to push messages to broker every second. Supposed to
182 : /// be used in tests.
183 : #[arg(long)]
184 0 : disable_periodic_broker_push: bool,
185 : /// Enable automatic switching to offloaded state.
186 : #[arg(long)]
187 0 : enable_offload: bool,
188 : /// Delete local WAL files after offloading. When disabled, they will be left on disk.
189 : #[arg(long)]
190 0 : delete_offloaded_wal: bool,
191 : /// Pending updates to control file will be automatically saved after this interval.
192 : #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_CONTROL_FILE_SAVE_INTERVAL)]
193 0 : control_file_save_interval: Duration,
194 : /// Number of allowed concurrent uploads of partial segments to remote storage.
195 : #[arg(long, default_value = DEFAULT_PARTIAL_BACKUP_CONCURRENCY)]
196 0 : partial_backup_concurrency: usize,
197 : /// How long a timeline must be resident before it is eligible for eviction.
198 : /// Usually, timeline eviction has to wait for `partial_backup_timeout` before being eligible for eviction,
199 : /// but if a timeline is un-evicted and then _not_ written to, it would immediately flap to evicting again,
200 : /// if it weren't for `eviction_min_resident` preventing that.
201 : #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_EVICTION_MIN_RESIDENT)]
202 0 : eviction_min_resident: Duration,
203 : }
204 :
205 : // Like PathBufValueParser, but allows empty string.
206 0 : fn opt_pathbuf_parser(s: &str) -> Result<Utf8PathBuf, String> {
207 0 : Ok(Utf8PathBuf::from_str(s).unwrap())
208 0 : }
209 :
210 : #[tokio::main(flavor = "current_thread")]
211 0 : async fn main() -> anyhow::Result<()> {
212 0 : // We want to allow multiple occurences of the same arg (taking the last) so
213 0 : // that neon_local could generate command with defaults + overrides without
214 0 : // getting 'argument cannot be used multiple times' error. This seems to be
215 0 : // impossible with pure Derive API, so convert struct to Command, modify it,
216 0 : // parse arguments, and then fill the struct back.
217 0 : let cmd = <Args as clap::CommandFactory>::command()
218 0 : .args_override_self(true)
219 0 : .version(version());
220 0 : let mut matches = cmd.get_matches();
221 0 : let mut args = <Args as clap::FromArgMatches>::from_arg_matches_mut(&mut matches)?;
222 0 :
223 0 : // I failed to modify opt_pathbuf_parser to return Option<PathBuf> in
224 0 : // reasonable time, so turn empty string into option post factum.
225 0 : if let Some(pb) = &args.pg_auth_public_key_path {
226 0 : if pb.as_os_str().is_empty() {
227 0 : args.pg_auth_public_key_path = None;
228 0 : }
229 0 : }
230 0 : if let Some(pb) = &args.pg_tenant_only_auth_public_key_path {
231 0 : if pb.as_os_str().is_empty() {
232 0 : args.pg_tenant_only_auth_public_key_path = None;
233 0 : }
234 0 : }
235 0 : if let Some(pb) = &args.http_auth_public_key_path {
236 0 : if pb.as_os_str().is_empty() {
237 0 : args.http_auth_public_key_path = None;
238 0 : }
239 0 : }
240 0 :
241 0 : if let Some(addr) = args.dump_control_file {
242 0 : let state = control_file::FileStorage::load_control_file(addr)?;
243 0 : let json = serde_json::to_string(&state)?;
244 0 : print!("{json}");
245 0 : return Ok(());
246 0 : }
247 0 :
248 0 : // important to keep the order of:
249 0 : // 1. init logging
250 0 : // 2. tracing panic hook
251 0 : // 3. sentry
252 0 : logging::init(
253 0 : LogFormat::from_config(&args.log_format)?,
254 0 : logging::TracingErrorLayerEnablement::Disabled,
255 0 : logging::Output::Stdout,
256 0 : )?;
257 0 : logging::replace_panic_hook_with_tracing_panic_hook().forget();
258 0 : info!("version: {GIT_VERSION}");
259 0 : info!("buld_tag: {BUILD_TAG}");
260 0 :
261 0 : let args_workdir = &args.datadir;
262 0 : let workdir = args_workdir.canonicalize_utf8().with_context(|| {
263 0 : format!("Failed to get the absolute path for input workdir {args_workdir:?}")
264 0 : })?;
265 0 :
266 0 : // Change into the data directory.
267 0 : std::env::set_current_dir(&workdir)?;
268 0 :
269 0 : // Set or read our ID.
270 0 : let id = set_id(&workdir, args.id.map(NodeId))?;
271 0 : if args.init {
272 0 : return Ok(());
273 0 : }
274 0 :
275 0 : let pg_auth = match args.pg_auth_public_key_path.as_ref() {
276 0 : None => {
277 0 : info!("pg auth is disabled");
278 0 : None
279 0 : }
280 0 : Some(path) => {
281 0 : info!("loading pg auth JWT key from {path}");
282 0 : Some(Arc::new(
283 0 : JwtAuth::from_key_path(path).context("failed to load the auth key")?,
284 0 : ))
285 0 : }
286 0 : };
287 0 : let pg_tenant_only_auth = match args.pg_tenant_only_auth_public_key_path.as_ref() {
288 0 : None => {
289 0 : info!("pg tenant only auth is disabled");
290 0 : None
291 0 : }
292 0 : Some(path) => {
293 0 : info!("loading pg tenant only auth JWT key from {path}");
294 0 : Some(Arc::new(
295 0 : JwtAuth::from_key_path(path).context("failed to load the auth key")?,
296 0 : ))
297 0 : }
298 0 : };
299 0 : let http_auth = match args.http_auth_public_key_path.as_ref() {
300 0 : None => {
301 0 : info!("http auth is disabled");
302 0 : None
303 0 : }
304 0 : Some(path) => {
305 0 : info!("loading http auth JWT key(s) from {path}");
306 0 : let jwt_auth = JwtAuth::from_key_path(path).context("failed to load the auth key")?;
307 0 : Some(Arc::new(SwappableJwtAuth::new(jwt_auth)))
308 0 : }
309 0 : };
310 0 :
311 0 : // Load JWT auth token to connect to other safekeepers for pull_timeline.
312 0 : let sk_auth_token = match var("SAFEKEEPER_AUTH_TOKEN") {
313 0 : Ok(v) => {
314 0 : info!("loaded JWT token for authentication with safekeepers");
315 0 : Some(SecretString::from(v))
316 0 : }
317 0 : Err(VarError::NotPresent) => {
318 0 : info!("no JWT token for authentication with safekeepers detected");
319 0 : None
320 0 : }
321 0 : Err(_) => {
322 0 : warn!("JWT token for authentication with safekeepers is not unicode");
323 0 : None
324 0 : }
325 0 : };
326 0 :
327 0 : let conf = SafeKeeperConf {
328 0 : workdir,
329 0 : my_id: id,
330 0 : listen_pg_addr: args.listen_pg,
331 0 : listen_pg_addr_tenant_only: args.listen_pg_tenant_only,
332 0 : listen_http_addr: args.listen_http,
333 0 : advertise_pg_addr: args.advertise_pg,
334 0 : availability_zone: args.availability_zone,
335 0 : no_sync: args.no_sync,
336 0 : broker_endpoint: args.broker_endpoint,
337 0 : broker_keepalive_interval: args.broker_keepalive_interval,
338 0 : heartbeat_timeout: args.heartbeat_timeout,
339 0 : peer_recovery_enabled: args.peer_recovery,
340 0 : remote_storage: args.remote_storage,
341 0 : max_offloader_lag_bytes: args.max_offloader_lag,
342 0 : wal_backup_enabled: !args.disable_wal_backup,
343 0 : backup_parallel_jobs: args.wal_backup_parallel_jobs,
344 0 : pg_auth,
345 0 : pg_tenant_only_auth,
346 0 : http_auth,
347 0 : sk_auth_token,
348 0 : current_thread_runtime: args.current_thread_runtime,
349 0 : walsenders_keep_horizon: args.walsenders_keep_horizon,
350 0 : partial_backup_enabled: true,
351 0 : partial_backup_timeout: args.partial_backup_timeout,
352 0 : disable_periodic_broker_push: args.disable_periodic_broker_push,
353 0 : enable_offload: args.enable_offload,
354 0 : delete_offloaded_wal: args.delete_offloaded_wal,
355 0 : control_file_save_interval: args.control_file_save_interval,
356 0 : partial_backup_concurrency: args.partial_backup_concurrency,
357 0 : eviction_min_resident: args.eviction_min_resident,
358 0 : };
359 0 :
360 0 : // initialize sentry if SENTRY_DSN is provided
361 0 : let _sentry_guard = init_sentry(
362 0 : Some(GIT_VERSION.into()),
363 0 : &[("node_id", &conf.my_id.to_string())],
364 0 : );
365 0 : start_safekeeper(conf).await
366 0 : }
367 :
368 : /// Result of joining any of main tasks: upper error means task failed to
369 : /// complete, e.g. panicked, inner is error produced by task itself.
370 : type JoinTaskRes = Result<anyhow::Result<()>, JoinError>;
371 :
372 0 : async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
373 0 : // Prevent running multiple safekeepers on the same directory
374 0 : let lock_file_path = conf.workdir.join(PID_FILE_NAME);
375 0 : let lock_file =
376 0 : pid_file::claim_for_current_process(&lock_file_path).context("claim pid file")?;
377 0 : info!("claimed pid file at {lock_file_path:?}");
378 :
379 : // ensure that the lock file is held even if the main thread of the process is panics
380 : // we need to release the lock file only when the current process is gone
381 0 : std::mem::forget(lock_file);
382 0 :
383 0 : info!("starting safekeeper WAL service on {}", conf.listen_pg_addr);
384 0 : let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
385 0 : error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
386 0 : e
387 0 : })?;
388 :
389 0 : let pg_listener_tenant_only =
390 0 : if let Some(listen_pg_addr_tenant_only) = &conf.listen_pg_addr_tenant_only {
391 0 : info!(
392 0 : "starting safekeeper tenant scoped WAL service on {}",
393 : listen_pg_addr_tenant_only
394 : );
395 0 : let listener = tcp_listener::bind(listen_pg_addr_tenant_only.clone()).map_err(|e| {
396 0 : error!(
397 0 : "failed to bind to address {}: {}",
398 : listen_pg_addr_tenant_only, e
399 : );
400 0 : e
401 0 : })?;
402 0 : Some(listener)
403 : } else {
404 0 : None
405 : };
406 :
407 0 : info!(
408 0 : "starting safekeeper HTTP service on {}",
409 : conf.listen_http_addr
410 : );
411 0 : let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
412 0 : error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
413 0 : e
414 0 : })?;
415 :
416 : // Register metrics collector for active timelines. It's important to do this
417 : // after daemonizing, otherwise process collector will be upset.
418 0 : let timeline_collector = safekeeper::metrics::TimelineCollector::new();
419 0 : metrics::register_internal(Box::new(timeline_collector))?;
420 :
421 0 : wal_backup::init_remote_storage(&conf).await;
422 :
423 : // Keep handles to main tasks to die if any of them disappears.
424 0 : let mut tasks_handles: FuturesUnordered<BoxFuture<(String, JoinTaskRes)>> =
425 0 : FuturesUnordered::new();
426 0 :
427 0 : // Start wal backup launcher before loading timelines as we'll notify it
428 0 : // through the channel about timelines which need offloading, not draining
429 0 : // the channel would cause deadlock.
430 0 : let current_thread_rt = conf
431 0 : .current_thread_runtime
432 0 : .then(|| Handle::try_current().expect("no runtime in main"));
433 0 :
434 0 : // Load all timelines from disk to memory.
435 0 : GlobalTimelines::init(conf.clone()).await?;
436 :
437 0 : let conf_ = conf.clone();
438 0 : // Run everything in current thread rt, if asked.
439 0 : if conf.current_thread_runtime {
440 0 : info!("running in current thread runtime");
441 0 : }
442 :
443 0 : let wal_service_handle = current_thread_rt
444 0 : .as_ref()
445 0 : .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
446 0 : .spawn(wal_service::task_main(
447 0 : conf_,
448 0 : pg_listener,
449 0 : Scope::SafekeeperData,
450 0 : ))
451 0 : // wrap with task name for error reporting
452 0 : .map(|res| ("WAL service main".to_owned(), res));
453 0 : tasks_handles.push(Box::pin(wal_service_handle));
454 0 :
455 0 : let timeline_housekeeping_handle = current_thread_rt
456 0 : .as_ref()
457 0 : .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
458 0 : .spawn(async move {
459 : const TOMBSTONE_TTL: Duration = Duration::from_secs(3600 * 24);
460 : loop {
461 0 : tokio::time::sleep(TOMBSTONE_TTL).await;
462 0 : GlobalTimelines::housekeeping(&TOMBSTONE_TTL);
463 : }
464 0 : })
465 0 : .map(|res| ("Timeline map housekeeping".to_owned(), res));
466 0 : tasks_handles.push(Box::pin(timeline_housekeeping_handle));
467 :
468 0 : if let Some(pg_listener_tenant_only) = pg_listener_tenant_only {
469 0 : let conf_ = conf.clone();
470 0 : let wal_service_handle = current_thread_rt
471 0 : .as_ref()
472 0 : .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
473 0 : .spawn(wal_service::task_main(
474 0 : conf_,
475 0 : pg_listener_tenant_only,
476 0 : Scope::Tenant,
477 0 : ))
478 0 : // wrap with task name for error reporting
479 0 : .map(|res| ("WAL service tenant only main".to_owned(), res));
480 0 : tasks_handles.push(Box::pin(wal_service_handle));
481 0 : }
482 :
483 0 : let conf_ = conf.clone();
484 0 : let http_handle = current_thread_rt
485 0 : .as_ref()
486 0 : .unwrap_or_else(|| HTTP_RUNTIME.handle())
487 0 : .spawn(http::task_main(conf_, http_listener))
488 0 : .map(|res| ("HTTP service main".to_owned(), res));
489 0 : tasks_handles.push(Box::pin(http_handle));
490 0 :
491 0 : let conf_ = conf.clone();
492 0 : let broker_task_handle = current_thread_rt
493 0 : .as_ref()
494 0 : .unwrap_or_else(|| BROKER_RUNTIME.handle())
495 0 : .spawn(broker::task_main(conf_).instrument(info_span!("broker")))
496 0 : .map(|res| ("broker main".to_owned(), res));
497 0 : tasks_handles.push(Box::pin(broker_task_handle));
498 0 :
499 0 : set_build_info_metric(GIT_VERSION, BUILD_TAG);
500 :
501 : // TODO: update tokio-stream, convert to real async Stream with
502 : // SignalStream, map it to obtain missing signal name, combine streams into
503 : // single stream we can easily sit on.
504 0 : let mut sigquit_stream = signal(SignalKind::quit())?;
505 0 : let mut sigint_stream = signal(SignalKind::interrupt())?;
506 0 : let mut sigterm_stream = signal(SignalKind::terminate())?;
507 :
508 : // Notify systemd that we are ready. This is important as currently loading
509 : // timelines takes significant time (~30s in busy regions).
510 0 : if let Err(e) = sd_notify::notify(true, &[NotifyState::Ready]) {
511 0 : warn!("systemd notify failed: {:?}", e);
512 0 : }
513 :
514 : tokio::select! {
515 : Some((task_name, res)) = tasks_handles.next()=> {
516 : error!("{} task failed: {:?}, exiting", task_name, res);
517 : std::process::exit(1);
518 : }
519 : // On any shutdown signal, log receival and exit. Additionally, handling
520 : // SIGQUIT prevents coredump.
521 : _ = sigquit_stream.recv() => info!("received SIGQUIT, terminating"),
522 : _ = sigint_stream.recv() => info!("received SIGINT, terminating"),
523 : _ = sigterm_stream.recv() => info!("received SIGTERM, terminating")
524 :
525 : };
526 0 : std::process::exit(0);
527 0 : }
528 :
529 : /// Determine safekeeper id.
530 0 : fn set_id(workdir: &Utf8Path, given_id: Option<NodeId>) -> Result<NodeId> {
531 0 : let id_file_path = workdir.join(ID_FILE_NAME);
532 0 :
533 0 : let my_id: NodeId;
534 0 : // If file with ID exists, read it in; otherwise set one passed.
535 0 : match fs::read(&id_file_path) {
536 0 : Ok(id_serialized) => {
537 0 : my_id = NodeId(
538 0 : std::str::from_utf8(&id_serialized)
539 0 : .context("failed to parse safekeeper id")?
540 0 : .parse()
541 0 : .context("failed to parse safekeeper id")?,
542 : );
543 0 : if let Some(given_id) = given_id {
544 0 : if given_id != my_id {
545 0 : bail!(
546 0 : "safekeeper already initialized with id {}, can't set {}",
547 0 : my_id,
548 0 : given_id
549 0 : );
550 0 : }
551 0 : }
552 0 : info!("safekeeper ID {}", my_id);
553 : }
554 0 : Err(error) => match error.kind() {
555 : ErrorKind::NotFound => {
556 0 : my_id = if let Some(given_id) = given_id {
557 0 : given_id
558 : } else {
559 0 : bail!("safekeeper id is not specified");
560 : };
561 0 : let mut f = File::create(&id_file_path)
562 0 : .with_context(|| format!("Failed to create id file at {id_file_path:?}"))?;
563 0 : f.write_all(my_id.to_string().as_bytes())?;
564 0 : f.sync_all()?;
565 0 : info!("initialized safekeeper id {}", my_id);
566 : }
567 : _ => {
568 0 : return Err(error.into());
569 : }
570 : },
571 : }
572 0 : Ok(my_id)
573 0 : }
574 :
575 0 : fn parse_remote_storage(storage_conf: &str) -> anyhow::Result<RemoteStorageConfig> {
576 0 : RemoteStorageConfig::from_toml(&storage_conf.parse()?)
577 0 : }
578 :
579 : #[test]
580 2 : fn verify_cli() {
581 2 : use clap::CommandFactory;
582 2 : Args::command().debug_assert()
583 2 : }
|
