LCOV - code coverage report
Current view: top level - safekeeper/src/bin - safekeeper.rs (source / functions) Coverage Total Hit
Test: 792183ae0ef4f1f8b22e9ac7e8748740ab73f873.info Lines: 1.8 % 331 6
Test Date: 2024-06-26 01:04:33 Functions: 5.6 % 71 4

            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 toml_edit::Document;
      16              : use utils::logging::SecretString;
      17              : 
      18              : use std::env::{var, VarError};
      19              : use std::fs::{self, File};
      20              : use std::io::{ErrorKind, Write};
      21              : use std::str::FromStr;
      22              : use std::sync::Arc;
      23              : use std::time::Duration;
      24              : use storage_broker::Uri;
      25              : 
      26              : use tracing::*;
      27              : use utils::pid_file;
      28              : 
      29              : use metrics::set_build_info_metric;
      30              : use safekeeper::defaults::{
      31              :     DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES,
      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              :     #[arg(long)]
     176            0 :     partial_backup_enabled: bool,
     177              :     /// Controls how long backup will wait until uploading the partial segment.
     178              :     #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_PARTIAL_BACKUP_TIMEOUT, verbatim_doc_comment)]
     179            0 :     partial_backup_timeout: Duration,
     180              :     /// Disable task to push messages to broker every second. Supposed to
     181              :     /// be used in tests.
     182              :     #[arg(long)]
     183            0 :     disable_periodic_broker_push: bool,
     184              : }
     185              : 
     186              : // Like PathBufValueParser, but allows empty string.
     187            0 : fn opt_pathbuf_parser(s: &str) -> Result<Utf8PathBuf, String> {
     188            0 :     Ok(Utf8PathBuf::from_str(s).unwrap())
     189            0 : }
     190              : 
     191              : #[tokio::main(flavor = "current_thread")]
     192            0 : async fn main() -> anyhow::Result<()> {
     193            0 :     // We want to allow multiple occurences of the same arg (taking the last) so
     194            0 :     // that neon_local could generate command with defaults + overrides without
     195            0 :     // getting 'argument cannot be used multiple times' error. This seems to be
     196            0 :     // impossible with pure Derive API, so convert struct to Command, modify it,
     197            0 :     // parse arguments, and then fill the struct back.
     198            0 :     let cmd = <Args as clap::CommandFactory>::command()
     199            0 :         .args_override_self(true)
     200            0 :         .version(version());
     201            0 :     let mut matches = cmd.get_matches();
     202            0 :     let mut args = <Args as clap::FromArgMatches>::from_arg_matches_mut(&mut matches)?;
     203            0 : 
     204            0 :     // I failed to modify opt_pathbuf_parser to return Option<PathBuf> in
     205            0 :     // reasonable time, so turn empty string into option post factum.
     206            0 :     if let Some(pb) = &args.pg_auth_public_key_path {
     207            0 :         if pb.as_os_str().is_empty() {
     208            0 :             args.pg_auth_public_key_path = None;
     209            0 :         }
     210            0 :     }
     211            0 :     if let Some(pb) = &args.pg_tenant_only_auth_public_key_path {
     212            0 :         if pb.as_os_str().is_empty() {
     213            0 :             args.pg_tenant_only_auth_public_key_path = None;
     214            0 :         }
     215            0 :     }
     216            0 :     if let Some(pb) = &args.http_auth_public_key_path {
     217            0 :         if pb.as_os_str().is_empty() {
     218            0 :             args.http_auth_public_key_path = None;
     219            0 :         }
     220            0 :     }
     221            0 : 
     222            0 :     if let Some(addr) = args.dump_control_file {
     223            0 :         let state = control_file::FileStorage::load_control_file(addr)?;
     224            0 :         let json = serde_json::to_string(&state)?;
     225            0 :         print!("{json}");
     226            0 :         return Ok(());
     227            0 :     }
     228            0 : 
     229            0 :     // important to keep the order of:
     230            0 :     // 1. init logging
     231            0 :     // 2. tracing panic hook
     232            0 :     // 3. sentry
     233            0 :     logging::init(
     234            0 :         LogFormat::from_config(&args.log_format)?,
     235            0 :         logging::TracingErrorLayerEnablement::Disabled,
     236            0 :         logging::Output::Stdout,
     237            0 :     )?;
     238            0 :     logging::replace_panic_hook_with_tracing_panic_hook().forget();
     239            0 :     info!("version: {GIT_VERSION}");
     240            0 :     info!("buld_tag: {BUILD_TAG}");
     241            0 : 
     242            0 :     let args_workdir = &args.datadir;
     243            0 :     let workdir = args_workdir.canonicalize_utf8().with_context(|| {
     244            0 :         format!("Failed to get the absolute path for input workdir {args_workdir:?}")
     245            0 :     })?;
     246            0 : 
     247            0 :     // Change into the data directory.
     248            0 :     std::env::set_current_dir(&workdir)?;
     249            0 : 
     250            0 :     // Set or read our ID.
     251            0 :     let id = set_id(&workdir, args.id.map(NodeId))?;
     252            0 :     if args.init {
     253            0 :         return Ok(());
     254            0 :     }
     255            0 : 
     256            0 :     let pg_auth = match args.pg_auth_public_key_path.as_ref() {
     257            0 :         None => {
     258            0 :             info!("pg auth is disabled");
     259            0 :             None
     260            0 :         }
     261            0 :         Some(path) => {
     262            0 :             info!("loading pg auth JWT key from {path}");
     263            0 :             Some(Arc::new(
     264            0 :                 JwtAuth::from_key_path(path).context("failed to load the auth key")?,
     265            0 :             ))
     266            0 :         }
     267            0 :     };
     268            0 :     let pg_tenant_only_auth = match args.pg_tenant_only_auth_public_key_path.as_ref() {
     269            0 :         None => {
     270            0 :             info!("pg tenant only auth is disabled");
     271            0 :             None
     272            0 :         }
     273            0 :         Some(path) => {
     274            0 :             info!("loading pg tenant only auth JWT key from {path}");
     275            0 :             Some(Arc::new(
     276            0 :                 JwtAuth::from_key_path(path).context("failed to load the auth key")?,
     277            0 :             ))
     278            0 :         }
     279            0 :     };
     280            0 :     let http_auth = match args.http_auth_public_key_path.as_ref() {
     281            0 :         None => {
     282            0 :             info!("http auth is disabled");
     283            0 :             None
     284            0 :         }
     285            0 :         Some(path) => {
     286            0 :             info!("loading http auth JWT key(s) from {path}");
     287            0 :             let jwt_auth = JwtAuth::from_key_path(path).context("failed to load the auth key")?;
     288            0 :             Some(Arc::new(SwappableJwtAuth::new(jwt_auth)))
     289            0 :         }
     290            0 :     };
     291            0 : 
     292            0 :     // Load JWT auth token to connect to other safekeepers for pull_timeline.
     293            0 :     let sk_auth_token = match var("SAFEKEEPER_AUTH_TOKEN") {
     294            0 :         Ok(v) => {
     295            0 :             info!("loaded JWT token for authentication with safekeepers");
     296            0 :             Some(SecretString::from(v))
     297            0 :         }
     298            0 :         Err(VarError::NotPresent) => {
     299            0 :             info!("no JWT token for authentication with safekeepers detected");
     300            0 :             None
     301            0 :         }
     302            0 :         Err(_) => {
     303            0 :             warn!("JWT token for authentication with safekeepers is not unicode");
     304            0 :             None
     305            0 :         }
     306            0 :     };
     307            0 : 
     308            0 :     let conf = SafeKeeperConf {
     309            0 :         workdir,
     310            0 :         my_id: id,
     311            0 :         listen_pg_addr: args.listen_pg,
     312            0 :         listen_pg_addr_tenant_only: args.listen_pg_tenant_only,
     313            0 :         listen_http_addr: args.listen_http,
     314            0 :         advertise_pg_addr: args.advertise_pg,
     315            0 :         availability_zone: args.availability_zone,
     316            0 :         no_sync: args.no_sync,
     317            0 :         broker_endpoint: args.broker_endpoint,
     318            0 :         broker_keepalive_interval: args.broker_keepalive_interval,
     319            0 :         heartbeat_timeout: args.heartbeat_timeout,
     320            0 :         peer_recovery_enabled: args.peer_recovery,
     321            0 :         remote_storage: args.remote_storage,
     322            0 :         max_offloader_lag_bytes: args.max_offloader_lag,
     323            0 :         wal_backup_enabled: !args.disable_wal_backup,
     324            0 :         backup_parallel_jobs: args.wal_backup_parallel_jobs,
     325            0 :         pg_auth,
     326            0 :         pg_tenant_only_auth,
     327            0 :         http_auth,
     328            0 :         sk_auth_token,
     329            0 :         current_thread_runtime: args.current_thread_runtime,
     330            0 :         walsenders_keep_horizon: args.walsenders_keep_horizon,
     331            0 :         partial_backup_enabled: args.partial_backup_enabled,
     332            0 :         partial_backup_timeout: args.partial_backup_timeout,
     333            0 :         disable_periodic_broker_push: args.disable_periodic_broker_push,
     334            0 :     };
     335            0 : 
     336            0 :     // initialize sentry if SENTRY_DSN is provided
     337            0 :     let _sentry_guard = init_sentry(
     338            0 :         Some(GIT_VERSION.into()),
     339            0 :         &[("node_id", &conf.my_id.to_string())],
     340            0 :     );
     341            0 :     start_safekeeper(conf).await
     342            0 : }
     343              : 
     344              : /// Result of joining any of main tasks: upper error means task failed to
     345              : /// complete, e.g. panicked, inner is error produced by task itself.
     346              : type JoinTaskRes = Result<anyhow::Result<()>, JoinError>;
     347              : 
     348            0 : async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     349            0 :     // Prevent running multiple safekeepers on the same directory
     350            0 :     let lock_file_path = conf.workdir.join(PID_FILE_NAME);
     351            0 :     let lock_file =
     352            0 :         pid_file::claim_for_current_process(&lock_file_path).context("claim pid file")?;
     353            0 :     info!("claimed pid file at {lock_file_path:?}");
     354              : 
     355              :     // ensure that the lock file is held even if the main thread of the process is panics
     356              :     // we need to release the lock file only when the current process is gone
     357            0 :     std::mem::forget(lock_file);
     358            0 : 
     359            0 :     info!("starting safekeeper WAL service on {}", conf.listen_pg_addr);
     360            0 :     let pg_listener = tcp_listener::bind(conf.listen_pg_addr.clone()).map_err(|e| {
     361            0 :         error!("failed to bind to address {}: {}", conf.listen_pg_addr, e);
     362            0 :         e
     363            0 :     })?;
     364              : 
     365            0 :     let pg_listener_tenant_only =
     366            0 :         if let Some(listen_pg_addr_tenant_only) = &conf.listen_pg_addr_tenant_only {
     367            0 :             info!(
     368            0 :                 "starting safekeeper tenant scoped WAL service on {}",
     369              :                 listen_pg_addr_tenant_only
     370              :             );
     371            0 :             let listener = tcp_listener::bind(listen_pg_addr_tenant_only.clone()).map_err(|e| {
     372            0 :                 error!(
     373            0 :                     "failed to bind to address {}: {}",
     374              :                     listen_pg_addr_tenant_only, e
     375              :                 );
     376            0 :                 e
     377            0 :             })?;
     378            0 :             Some(listener)
     379              :         } else {
     380            0 :             None
     381              :         };
     382              : 
     383            0 :     info!(
     384            0 :         "starting safekeeper HTTP service on {}",
     385              :         conf.listen_http_addr
     386              :     );
     387            0 :     let http_listener = tcp_listener::bind(conf.listen_http_addr.clone()).map_err(|e| {
     388            0 :         error!("failed to bind to address {}: {}", conf.listen_http_addr, e);
     389            0 :         e
     390            0 :     })?;
     391              : 
     392              :     // Register metrics collector for active timelines. It's important to do this
     393              :     // after daemonizing, otherwise process collector will be upset.
     394            0 :     let timeline_collector = safekeeper::metrics::TimelineCollector::new();
     395            0 :     metrics::register_internal(Box::new(timeline_collector))?;
     396              : 
     397            0 :     wal_backup::init_remote_storage(&conf);
     398            0 : 
     399            0 :     // Keep handles to main tasks to die if any of them disappears.
     400            0 :     let mut tasks_handles: FuturesUnordered<BoxFuture<(String, JoinTaskRes)>> =
     401            0 :         FuturesUnordered::new();
     402            0 : 
     403            0 :     // Start wal backup launcher before loading timelines as we'll notify it
     404            0 :     // through the channel about timelines which need offloading, not draining
     405            0 :     // the channel would cause deadlock.
     406            0 :     let current_thread_rt = conf
     407            0 :         .current_thread_runtime
     408            0 :         .then(|| Handle::try_current().expect("no runtime in main"));
     409            0 : 
     410            0 :     // Load all timelines from disk to memory.
     411            0 :     GlobalTimelines::init(conf.clone()).await?;
     412              : 
     413            0 :     let conf_ = conf.clone();
     414            0 :     // Run everything in current thread rt, if asked.
     415            0 :     if conf.current_thread_runtime {
     416            0 :         info!("running in current thread runtime");
     417            0 :     }
     418              : 
     419            0 :     let wal_service_handle = current_thread_rt
     420            0 :         .as_ref()
     421            0 :         .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
     422            0 :         .spawn(wal_service::task_main(
     423            0 :             conf_,
     424            0 :             pg_listener,
     425            0 :             Scope::SafekeeperData,
     426            0 :         ))
     427            0 :         // wrap with task name for error reporting
     428            0 :         .map(|res| ("WAL service main".to_owned(), res));
     429            0 :     tasks_handles.push(Box::pin(wal_service_handle));
     430              : 
     431            0 :     if let Some(pg_listener_tenant_only) = pg_listener_tenant_only {
     432            0 :         let conf_ = conf.clone();
     433            0 :         let wal_service_handle = current_thread_rt
     434            0 :             .as_ref()
     435            0 :             .unwrap_or_else(|| WAL_SERVICE_RUNTIME.handle())
     436            0 :             .spawn(wal_service::task_main(
     437            0 :                 conf_,
     438            0 :                 pg_listener_tenant_only,
     439            0 :                 Scope::Tenant,
     440            0 :             ))
     441            0 :             // wrap with task name for error reporting
     442            0 :             .map(|res| ("WAL service tenant only main".to_owned(), res));
     443            0 :         tasks_handles.push(Box::pin(wal_service_handle));
     444            0 :     }
     445              : 
     446            0 :     let conf_ = conf.clone();
     447            0 :     let http_handle = current_thread_rt
     448            0 :         .as_ref()
     449            0 :         .unwrap_or_else(|| HTTP_RUNTIME.handle())
     450            0 :         .spawn(http::task_main(conf_, http_listener))
     451            0 :         .map(|res| ("HTTP service main".to_owned(), res));
     452            0 :     tasks_handles.push(Box::pin(http_handle));
     453            0 : 
     454            0 :     let conf_ = conf.clone();
     455            0 :     let broker_task_handle = current_thread_rt
     456            0 :         .as_ref()
     457            0 :         .unwrap_or_else(|| BROKER_RUNTIME.handle())
     458            0 :         .spawn(broker::task_main(conf_).instrument(info_span!("broker")))
     459            0 :         .map(|res| ("broker main".to_owned(), res));
     460            0 :     tasks_handles.push(Box::pin(broker_task_handle));
     461            0 : 
     462            0 :     set_build_info_metric(GIT_VERSION, BUILD_TAG);
     463              : 
     464              :     // TODO: update tokio-stream, convert to real async Stream with
     465              :     // SignalStream, map it to obtain missing signal name, combine streams into
     466              :     // single stream we can easily sit on.
     467            0 :     let mut sigquit_stream = signal(SignalKind::quit())?;
     468            0 :     let mut sigint_stream = signal(SignalKind::interrupt())?;
     469            0 :     let mut sigterm_stream = signal(SignalKind::terminate())?;
     470              : 
     471              :     // Notify systemd that we are ready. This is important as currently loading
     472              :     // timelines takes significant time (~30s in busy regions).
     473            0 :     if let Err(e) = sd_notify::notify(true, &[NotifyState::Ready]) {
     474            0 :         warn!("systemd notify failed: {:?}", e);
     475            0 :     }
     476              : 
     477              :     tokio::select! {
     478              :         Some((task_name, res)) = tasks_handles.next()=> {
     479              :             error!("{} task failed: {:?}, exiting", task_name, res);
     480              :             std::process::exit(1);
     481              :         }
     482              :         // On any shutdown signal, log receival and exit. Additionally, handling
     483              :         // SIGQUIT prevents coredump.
     484              :         _ = sigquit_stream.recv() => info!("received SIGQUIT, terminating"),
     485              :         _ = sigint_stream.recv() => info!("received SIGINT, terminating"),
     486              :         _ = sigterm_stream.recv() => info!("received SIGTERM, terminating")
     487              : 
     488              :     };
     489            0 :     std::process::exit(0);
     490            0 : }
     491              : 
     492              : /// Determine safekeeper id.
     493            0 : fn set_id(workdir: &Utf8Path, given_id: Option<NodeId>) -> Result<NodeId> {
     494            0 :     let id_file_path = workdir.join(ID_FILE_NAME);
     495            0 : 
     496            0 :     let my_id: NodeId;
     497            0 :     // If file with ID exists, read it in; otherwise set one passed.
     498            0 :     match fs::read(&id_file_path) {
     499            0 :         Ok(id_serialized) => {
     500            0 :             my_id = NodeId(
     501            0 :                 std::str::from_utf8(&id_serialized)
     502            0 :                     .context("failed to parse safekeeper id")?
     503            0 :                     .parse()
     504            0 :                     .context("failed to parse safekeeper id")?,
     505              :             );
     506            0 :             if let Some(given_id) = given_id {
     507            0 :                 if given_id != my_id {
     508            0 :                     bail!(
     509            0 :                         "safekeeper already initialized with id {}, can't set {}",
     510            0 :                         my_id,
     511            0 :                         given_id
     512            0 :                     );
     513            0 :                 }
     514            0 :             }
     515            0 :             info!("safekeeper ID {}", my_id);
     516              :         }
     517            0 :         Err(error) => match error.kind() {
     518              :             ErrorKind::NotFound => {
     519            0 :                 my_id = if let Some(given_id) = given_id {
     520            0 :                     given_id
     521              :                 } else {
     522            0 :                     bail!("safekeeper id is not specified");
     523              :                 };
     524            0 :                 let mut f = File::create(&id_file_path)
     525            0 :                     .with_context(|| format!("Failed to create id file at {id_file_path:?}"))?;
     526            0 :                 f.write_all(my_id.to_string().as_bytes())?;
     527            0 :                 f.sync_all()?;
     528            0 :                 info!("initialized safekeeper id {}", my_id);
     529              :             }
     530              :             _ => {
     531            0 :                 return Err(error.into());
     532              :             }
     533              :         },
     534              :     }
     535            0 :     Ok(my_id)
     536            0 : }
     537              : 
     538              : // Parse RemoteStorage from TOML table.
     539            0 : fn parse_remote_storage(storage_conf: &str) -> anyhow::Result<RemoteStorageConfig> {
     540            0 :     // funny toml doesn't consider plain inline table as valid document, so wrap in a key to parse
     541            0 :     let storage_conf_toml = format!("remote_storage = {storage_conf}");
     542            0 :     let parsed_toml = storage_conf_toml.parse::<Document>()?; // parse
     543            0 :     let (_, storage_conf_parsed_toml) = parsed_toml.iter().next().unwrap(); // and strip key off again
     544            0 :     RemoteStorageConfig::from_toml(storage_conf_parsed_toml).and_then(|parsed_config| {
     545            0 :         // XXX: Don't print the original toml here, there might be some sensitive data
     546            0 :         parsed_config.context("Incorrectly parsed remote storage toml as no remote storage config")
     547            0 :     })
     548            0 : }
     549              : 
     550              : #[test]
     551            2 : fn verify_cli() {
     552            2 :     use clap::CommandFactory;
     553            2 :     Args::command().debug_assert()
     554            2 : }
        

Generated by: LCOV version 2.1-beta