Line data Source code
1 : use anyhow::{anyhow, Context};
2 : use camino::Utf8PathBuf;
3 : use clap::Parser;
4 : use diesel::Connection;
5 : use metrics::launch_timestamp::LaunchTimestamp;
6 : use metrics::BuildInfo;
7 : use std::path::PathBuf;
8 : use std::sync::Arc;
9 : use storage_controller::http::make_router;
10 : use storage_controller::metrics::preinitialize_metrics;
11 : use storage_controller::persistence::Persistence;
12 : use storage_controller::service::{
13 : Config, Service, MAX_UNAVAILABLE_INTERVAL_DEFAULT, RECONCILER_CONCURRENCY_DEFAULT,
14 : };
15 : use tokio::signal::unix::SignalKind;
16 : use tokio_util::sync::CancellationToken;
17 : use utils::auth::{JwtAuth, SwappableJwtAuth};
18 : use utils::logging::{self, LogFormat};
19 :
20 : use utils::sentry_init::init_sentry;
21 : use utils::{project_build_tag, project_git_version, tcp_listener};
22 :
23 : project_git_version!(GIT_VERSION);
24 : project_build_tag!(BUILD_TAG);
25 :
26 : use diesel_migrations::{embed_migrations, EmbeddedMigrations};
27 : pub const MIGRATIONS: EmbeddedMigrations = embed_migrations!("./migrations");
28 :
29 0 : #[derive(Parser)]
30 : #[command(author, version, about, long_about = None)]
31 : #[command(arg_required_else_help(true))]
32 : struct Cli {
33 : /// Host and port to listen on, like `127.0.0.1:1234`
34 : #[arg(short, long)]
35 0 : listen: std::net::SocketAddr,
36 :
37 : /// Public key for JWT authentication of clients
38 : #[arg(long)]
39 : public_key: Option<String>,
40 :
41 : /// Token for authenticating this service with the pageservers it controls
42 : #[arg(long)]
43 : jwt_token: Option<String>,
44 :
45 : /// Token for authenticating this service with the control plane, when calling
46 : /// the compute notification endpoint
47 : #[arg(long)]
48 : control_plane_jwt_token: Option<String>,
49 :
50 : /// URL to control plane compute notification endpoint
51 : #[arg(long)]
52 : compute_hook_url: Option<String>,
53 :
54 : /// Path to the .json file to store state (will be created if it doesn't exist)
55 : #[arg(short, long)]
56 : path: Option<Utf8PathBuf>,
57 :
58 : /// URL to connect to postgres, like postgresql://localhost:1234/storage_controller
59 : #[arg(long)]
60 : database_url: Option<String>,
61 :
62 : /// Flag to enable dev mode, which permits running without auth
63 : #[arg(long, default_value = "false")]
64 0 : dev: bool,
65 :
66 : /// Grace period before marking unresponsive pageserver offline
67 : #[arg(long)]
68 : max_unavailable_interval: Option<humantime::Duration>,
69 :
70 : /// Size threshold for automatically splitting shards (disabled by default)
71 : #[arg(long)]
72 : split_threshold: Option<u64>,
73 :
74 : /// Maximum number of reconcilers that may run in parallel
75 : #[arg(long)]
76 : reconciler_concurrency: Option<usize>,
77 :
78 : /// How long to wait for the initial database connection to be available.
79 : #[arg(long, default_value = "5s")]
80 0 : db_connect_timeout: humantime::Duration,
81 :
82 : /// `neon_local` sets this to the path of the neon_local repo dir.
83 : /// Only relevant for testing.
84 : // TODO: make `cfg(feature = "testing")`
85 : #[arg(long)]
86 : neon_local_repo_dir: Option<PathBuf>,
87 : }
88 :
89 : enum StrictMode {
90 : /// In strict mode, we will require that all secrets are loaded, i.e. security features
91 : /// may not be implicitly turned off by omitting secrets in the environment.
92 : Strict,
93 : /// In dev mode, secrets are optional, and omitting a particular secret will implicitly
94 : /// disable the auth related to it (e.g. no pageserver jwt key -> send unauthenticated
95 : /// requests, no public key -> don't authenticate incoming requests).
96 : Dev,
97 : }
98 :
99 : impl Default for StrictMode {
100 0 : fn default() -> Self {
101 0 : Self::Strict
102 0 : }
103 : }
104 :
105 : /// Secrets may either be provided on the command line (for testing), or loaded from AWS SecretManager: this
106 : /// type encapsulates the logic to decide which and do the loading.
107 : struct Secrets {
108 : database_url: String,
109 : public_key: Option<JwtAuth>,
110 : jwt_token: Option<String>,
111 : control_plane_jwt_token: Option<String>,
112 : }
113 :
114 : impl Secrets {
115 : const DATABASE_URL_ENV: &'static str = "DATABASE_URL";
116 : const PAGESERVER_JWT_TOKEN_ENV: &'static str = "PAGESERVER_JWT_TOKEN";
117 : const CONTROL_PLANE_JWT_TOKEN_ENV: &'static str = "CONTROL_PLANE_JWT_TOKEN";
118 : const PUBLIC_KEY_ENV: &'static str = "PUBLIC_KEY";
119 :
120 : /// Load secrets from, in order of preference:
121 : /// - CLI args if database URL is provided on the CLI
122 : /// - Environment variables if DATABASE_URL is set.
123 : /// - AWS Secrets Manager secrets
124 0 : async fn load(args: &Cli) -> anyhow::Result<Self> {
125 0 : let Some(database_url) =
126 0 : Self::load_secret(&args.database_url, Self::DATABASE_URL_ENV).await
127 : else {
128 0 : anyhow::bail!(
129 0 : "Database URL is not set (set `--database-url`, or `DATABASE_URL` environment)"
130 0 : )
131 : };
132 :
133 0 : let public_key = match Self::load_secret(&args.public_key, Self::PUBLIC_KEY_ENV).await {
134 0 : Some(v) => Some(JwtAuth::from_key(v).context("Loading public key")?),
135 0 : None => None,
136 : };
137 :
138 0 : let this = Self {
139 0 : database_url,
140 0 : public_key,
141 0 : jwt_token: Self::load_secret(&args.jwt_token, Self::PAGESERVER_JWT_TOKEN_ENV).await,
142 0 : control_plane_jwt_token: Self::load_secret(
143 0 : &args.control_plane_jwt_token,
144 0 : Self::CONTROL_PLANE_JWT_TOKEN_ENV,
145 0 : )
146 0 : .await,
147 : };
148 :
149 0 : Ok(this)
150 0 : }
151 :
152 0 : async fn load_secret(cli: &Option<String>, env_name: &str) -> Option<String> {
153 0 : if let Some(v) = cli {
154 0 : Some(v.clone())
155 0 : } else if let Ok(v) = std::env::var(env_name) {
156 0 : Some(v)
157 : } else {
158 0 : None
159 : }
160 0 : }
161 : }
162 :
163 : /// Execute the diesel migrations that are built into this binary
164 0 : async fn migration_run(database_url: &str) -> anyhow::Result<()> {
165 : use diesel::PgConnection;
166 : use diesel_migrations::{HarnessWithOutput, MigrationHarness};
167 0 : let mut conn = PgConnection::establish(database_url)?;
168 :
169 0 : HarnessWithOutput::write_to_stdout(&mut conn)
170 0 : .run_pending_migrations(MIGRATIONS)
171 0 : .map(|_| ())
172 0 : .map_err(|e| anyhow::anyhow!(e))?;
173 :
174 0 : Ok(())
175 0 : }
176 :
177 0 : fn main() -> anyhow::Result<()> {
178 0 : let default_panic = std::panic::take_hook();
179 0 : std::panic::set_hook(Box::new(move |info| {
180 0 : default_panic(info);
181 0 : std::process::exit(1);
182 0 : }));
183 0 :
184 0 : let _sentry_guard = init_sentry(Some(GIT_VERSION.into()), &[]);
185 0 :
186 0 : tokio::runtime::Builder::new_current_thread()
187 0 : // We use spawn_blocking for database operations, so require approximately
188 0 : // as many blocking threads as we will open database connections.
189 0 : .max_blocking_threads(Persistence::MAX_CONNECTIONS as usize)
190 0 : .enable_all()
191 0 : .build()
192 0 : .unwrap()
193 0 : .block_on(async_main())
194 0 : }
195 :
196 0 : async fn async_main() -> anyhow::Result<()> {
197 0 : let launch_ts = Box::leak(Box::new(LaunchTimestamp::generate()));
198 0 :
199 0 : logging::init(
200 0 : LogFormat::Plain,
201 0 : logging::TracingErrorLayerEnablement::Disabled,
202 0 : logging::Output::Stdout,
203 0 : )?;
204 :
205 0 : preinitialize_metrics();
206 0 :
207 0 : let args = Cli::parse();
208 0 : tracing::info!(
209 0 : "version: {}, launch_timestamp: {}, build_tag {}, state at {}, listening on {}",
210 0 : GIT_VERSION,
211 0 : launch_ts.to_string(),
212 0 : BUILD_TAG,
213 0 : args.path.as_ref().unwrap_or(&Utf8PathBuf::from("<none>")),
214 : args.listen
215 : );
216 :
217 0 : let build_info = BuildInfo {
218 0 : revision: GIT_VERSION,
219 0 : build_tag: BUILD_TAG,
220 0 : };
221 :
222 0 : let strict_mode = if args.dev {
223 0 : StrictMode::Dev
224 : } else {
225 0 : StrictMode::Strict
226 : };
227 :
228 0 : let secrets = Secrets::load(&args).await?;
229 :
230 : // Validate required secrets and arguments are provided in strict mode
231 0 : match strict_mode {
232 : StrictMode::Strict
233 0 : if (secrets.public_key.is_none()
234 0 : || secrets.jwt_token.is_none()
235 0 : || secrets.control_plane_jwt_token.is_none()) =>
236 0 : {
237 0 : // Production systems should always have secrets configured: if public_key was not set
238 0 : // then we would implicitly disable auth.
239 0 : anyhow::bail!(
240 0 : "Insecure config! One or more secrets is not set. This is only permitted in `--dev` mode"
241 0 : );
242 : }
243 0 : StrictMode::Strict if args.compute_hook_url.is_none() => {
244 0 : // Production systems should always have a compute hook set, to prevent falling
245 0 : // back to trying to use neon_local.
246 0 : anyhow::bail!(
247 0 : "`--compute-hook-url` is not set: this is only permitted in `--dev` mode"
248 0 : );
249 : }
250 : StrictMode::Strict => {
251 0 : tracing::info!("Starting in strict mode: configuration is OK.")
252 : }
253 : StrictMode::Dev => {
254 0 : tracing::warn!("Starting in dev mode: this may be an insecure configuration.")
255 : }
256 : }
257 :
258 0 : let config = Config {
259 0 : jwt_token: secrets.jwt_token,
260 0 : control_plane_jwt_token: secrets.control_plane_jwt_token,
261 0 : compute_hook_url: args.compute_hook_url,
262 0 : max_unavailable_interval: args
263 0 : .max_unavailable_interval
264 0 : .map(humantime::Duration::into)
265 0 : .unwrap_or(MAX_UNAVAILABLE_INTERVAL_DEFAULT),
266 0 : reconciler_concurrency: args
267 0 : .reconciler_concurrency
268 0 : .unwrap_or(RECONCILER_CONCURRENCY_DEFAULT),
269 0 : split_threshold: args.split_threshold,
270 0 : neon_local_repo_dir: args.neon_local_repo_dir,
271 0 : };
272 0 :
273 0 : // After loading secrets & config, but before starting anything else, apply database migrations
274 0 : Persistence::await_connection(&secrets.database_url, args.db_connect_timeout.into()).await?;
275 :
276 0 : migration_run(&secrets.database_url)
277 0 : .await
278 0 : .context("Running database migrations")?;
279 :
280 0 : let json_path = args.path;
281 0 : let persistence = Arc::new(Persistence::new(secrets.database_url, json_path.clone()));
282 :
283 0 : let service = Service::spawn(config, persistence.clone()).await?;
284 :
285 0 : let http_listener = tcp_listener::bind(args.listen)?;
286 :
287 0 : let auth = secrets
288 0 : .public_key
289 0 : .map(|jwt_auth| Arc::new(SwappableJwtAuth::new(jwt_auth)));
290 0 : let router = make_router(service.clone(), auth, build_info)
291 0 : .build()
292 0 : .map_err(|err| anyhow!(err))?;
293 0 : let router_service = utils::http::RouterService::new(router).unwrap();
294 0 :
295 0 : // Start HTTP server
296 0 : let server_shutdown = CancellationToken::new();
297 0 : let server = hyper::Server::from_tcp(http_listener)?
298 0 : .serve(router_service)
299 0 : .with_graceful_shutdown({
300 0 : let server_shutdown = server_shutdown.clone();
301 0 : async move {
302 0 : server_shutdown.cancelled().await;
303 0 : }
304 0 : });
305 0 : tracing::info!("Serving on {0}", args.listen);
306 0 : let server_task = tokio::task::spawn(server);
307 :
308 : // Wait until we receive a signal
309 0 : let mut sigint = tokio::signal::unix::signal(SignalKind::interrupt())?;
310 0 : let mut sigquit = tokio::signal::unix::signal(SignalKind::quit())?;
311 0 : let mut sigterm = tokio::signal::unix::signal(SignalKind::terminate())?;
312 : tokio::select! {
313 : _ = sigint.recv() => {},
314 : _ = sigterm.recv() => {},
315 : _ = sigquit.recv() => {},
316 : }
317 0 : tracing::info!("Terminating on signal");
318 :
319 0 : if json_path.is_some() {
320 : // Write out a JSON dump on shutdown: this is used in compat tests to avoid passing
321 : // full postgres dumps around.
322 0 : if let Err(e) = persistence.write_tenants_json().await {
323 0 : tracing::error!("Failed to write JSON on shutdown: {e}")
324 0 : }
325 0 : }
326 :
327 : // Stop HTTP server first, so that we don't have to service requests
328 : // while shutting down Service
329 0 : server_shutdown.cancel();
330 0 : if let Err(e) = server_task.await {
331 0 : tracing::error!("Error joining HTTP server task: {e}")
332 0 : }
333 0 : tracing::info!("Joined HTTP server task");
334 :
335 0 : service.shutdown().await;
336 0 : tracing::info!("Service shutdown complete");
337 :
338 0 : std::process::exit(0);
339 0 : }
|
