Line data Source code
1 : use aws_config::environment::EnvironmentVariableCredentialsProvider;
2 : use aws_config::imds::credentials::ImdsCredentialsProvider;
3 : use aws_config::meta::credentials::CredentialsProviderChain;
4 : use aws_config::meta::region::RegionProviderChain;
5 : use aws_config::profile::ProfileFileCredentialsProvider;
6 : use aws_config::provider_config::ProviderConfig;
7 : use aws_config::web_identity_token::WebIdentityTokenCredentialsProvider;
8 : use futures::future::Either;
9 : use proxy::auth;
10 : use proxy::auth::backend::AuthRateLimiter;
11 : use proxy::auth::backend::MaybeOwned;
12 : use proxy::cancellation::CancelMap;
13 : use proxy::cancellation::CancellationHandler;
14 : use proxy::config::remote_storage_from_toml;
15 : use proxy::config::AuthenticationConfig;
16 : use proxy::config::CacheOptions;
17 : use proxy::config::HttpConfig;
18 : use proxy::config::ProjectInfoCacheOptions;
19 : use proxy::console;
20 : use proxy::context::parquet::ParquetUploadArgs;
21 : use proxy::http;
22 : use proxy::http::health_server::AppMetrics;
23 : use proxy::metrics::Metrics;
24 : use proxy::rate_limiter::EndpointRateLimiter;
25 : use proxy::rate_limiter::RateBucketInfo;
26 : use proxy::redis::cancellation_publisher::RedisPublisherClient;
27 : use proxy::redis::connection_with_credentials_provider::ConnectionWithCredentialsProvider;
28 : use proxy::redis::elasticache;
29 : use proxy::redis::notifications;
30 : use proxy::scram::threadpool::ThreadPool;
31 : use proxy::serverless::cancel_set::CancelSet;
32 : use proxy::serverless::GlobalConnPoolOptions;
33 : use proxy::usage_metrics;
34 :
35 : use anyhow::bail;
36 : use proxy::config::{self, ProxyConfig};
37 : use proxy::serverless;
38 : use remote_storage::RemoteStorageConfig;
39 : use std::net::SocketAddr;
40 : use std::pin::pin;
41 : use std::sync::Arc;
42 : use tokio::net::TcpListener;
43 : use tokio::sync::Mutex;
44 : use tokio::task::JoinSet;
45 : use tokio_util::sync::CancellationToken;
46 : use tracing::info;
47 : use tracing::warn;
48 : use tracing::Instrument;
49 : use utils::{project_build_tag, project_git_version, sentry_init::init_sentry};
50 :
51 : project_git_version!(GIT_VERSION);
52 : project_build_tag!(BUILD_TAG);
53 :
54 : use clap::{Parser, ValueEnum};
55 :
56 : #[global_allocator]
57 : static GLOBAL: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
58 :
59 20 : #[derive(Clone, Debug, ValueEnum)]
60 : enum AuthBackend {
61 : Console,
62 : #[cfg(feature = "testing")]
63 : Postgres,
64 : Link,
65 : }
66 :
67 : /// Neon proxy/router
68 4 : #[derive(Parser)]
69 : #[command(version = GIT_VERSION, about)]
70 : struct ProxyCliArgs {
71 : /// Name of the region this proxy is deployed in
72 2 : #[clap(long, default_value_t = String::new())]
73 0 : region: String,
74 : /// listen for incoming client connections on ip:port
75 : #[clap(short, long, default_value = "127.0.0.1:4432")]
76 0 : proxy: String,
77 2 : #[clap(value_enum, long, default_value_t = AuthBackend::Link)]
78 0 : auth_backend: AuthBackend,
79 : /// listen for management callback connection on ip:port
80 : #[clap(short, long, default_value = "127.0.0.1:7000")]
81 0 : mgmt: String,
82 : /// listen for incoming http connections (metrics, etc) on ip:port
83 : #[clap(long, default_value = "127.0.0.1:7001")]
84 0 : http: String,
85 : /// listen for incoming wss connections on ip:port
86 : #[clap(long)]
87 : wss: Option<String>,
88 : /// redirect unauthenticated users to the given uri in case of link auth
89 : #[clap(short, long, default_value = "http://localhost:3000/psql_session/")]
90 0 : uri: String,
91 : /// cloud API endpoint for authenticating users
92 : #[clap(
93 : short,
94 : long,
95 : default_value = "http://localhost:3000/authenticate_proxy_request/"
96 : )]
97 0 : auth_endpoint: String,
98 : /// path to TLS key for client postgres connections
99 : ///
100 : /// tls-key and tls-cert are for backwards compatibility, we can put all certs in one dir
101 : #[clap(short = 'k', long, alias = "ssl-key")]
102 : tls_key: Option<String>,
103 : /// path to TLS cert for client postgres connections
104 : ///
105 : /// tls-key and tls-cert are for backwards compatibility, we can put all certs in one dir
106 : #[clap(short = 'c', long, alias = "ssl-cert")]
107 : tls_cert: Option<String>,
108 : /// path to directory with TLS certificates for client postgres connections
109 : #[clap(long)]
110 : certs_dir: Option<String>,
111 : /// timeout for the TLS handshake
112 : #[clap(long, default_value = "15s", value_parser = humantime::parse_duration)]
113 0 : handshake_timeout: tokio::time::Duration,
114 : /// http endpoint to receive periodic metric updates
115 : #[clap(long)]
116 : metric_collection_endpoint: Option<String>,
117 : /// how often metrics should be sent to a collection endpoint
118 : #[clap(long)]
119 : metric_collection_interval: Option<String>,
120 : /// cache for `wake_compute` api method (use `size=0` to disable)
121 : #[clap(long, default_value = config::CacheOptions::CACHE_DEFAULT_OPTIONS)]
122 0 : wake_compute_cache: String,
123 : /// lock for `wake_compute` api method. example: "shards=32,permits=4,epoch=10m,timeout=1s". (use `permits=0` to disable).
124 : #[clap(long, default_value = config::ConcurrencyLockOptions::DEFAULT_OPTIONS_WAKE_COMPUTE_LOCK)]
125 0 : wake_compute_lock: String,
126 : /// lock for `connect_compute` api method. example: "shards=32,permits=4,epoch=10m,timeout=1s". (use `permits=0` to disable).
127 : #[clap(long, default_value = config::ConcurrencyLockOptions::DEFAULT_OPTIONS_CONNECT_COMPUTE_LOCK)]
128 0 : connect_compute_lock: String,
129 : /// Allow self-signed certificates for compute nodes (for testing)
130 2 : #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
131 0 : allow_self_signed_compute: bool,
132 : #[clap(flatten)]
133 : sql_over_http: SqlOverHttpArgs,
134 : /// timeout for scram authentication protocol
135 : #[clap(long, default_value = "15s", value_parser = humantime::parse_duration)]
136 0 : scram_protocol_timeout: tokio::time::Duration,
137 : /// size of the threadpool for password hashing
138 2 : #[clap(long, default_value_t = 4)]
139 0 : scram_thread_pool_size: u8,
140 : /// Require that all incoming requests have a Proxy Protocol V2 packet **and** have an IP address associated.
141 2 : #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
142 0 : require_client_ip: bool,
143 : /// Disable dynamic rate limiter and store the metrics to ensure its production behaviour.
144 2 : #[clap(long, default_value_t = true, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
145 0 : disable_dynamic_rate_limiter: bool,
146 : /// Endpoint rate limiter max number of requests per second.
147 : ///
148 : /// Provided in the form '<Requests Per Second>@<Bucket Duration Size>'.
149 : /// Can be given multiple times for different bucket sizes.
150 12 : #[clap(long, default_values_t = RateBucketInfo::DEFAULT_ENDPOINT_SET)]
151 2 : endpoint_rps_limit: Vec<RateBucketInfo>,
152 : /// Wake compute rate limiter max number of requests per second.
153 12 : #[clap(long, default_values_t = RateBucketInfo::DEFAULT_SET)]
154 2 : wake_compute_limit: Vec<RateBucketInfo>,
155 : /// Whether the auth rate limiter actually takes effect (for testing)
156 2 : #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
157 0 : auth_rate_limit_enabled: bool,
158 : /// Authentication rate limiter max number of hashes per second.
159 12 : #[clap(long, default_values_t = RateBucketInfo::DEFAULT_AUTH_SET)]
160 2 : auth_rate_limit: Vec<RateBucketInfo>,
161 : /// The IP subnet to use when considering whether two IP addresses are considered the same.
162 2 : #[clap(long, default_value_t = 64)]
163 0 : auth_rate_limit_ip_subnet: u8,
164 : /// Redis rate limiter max number of requests per second.
165 12 : #[clap(long, default_values_t = RateBucketInfo::DEFAULT_SET)]
166 2 : redis_rps_limit: Vec<RateBucketInfo>,
167 : /// cache for `allowed_ips` (use `size=0` to disable)
168 : #[clap(long, default_value = config::CacheOptions::CACHE_DEFAULT_OPTIONS)]
169 0 : allowed_ips_cache: String,
170 : /// cache for `role_secret` (use `size=0` to disable)
171 : #[clap(long, default_value = config::CacheOptions::CACHE_DEFAULT_OPTIONS)]
172 0 : role_secret_cache: String,
173 : /// disable ip check for http requests. If it is too time consuming, it could be turned off.
174 2 : #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
175 0 : disable_ip_check_for_http: bool,
176 : /// redis url for notifications (if empty, redis_host:port will be used for both notifications and streaming connections)
177 : #[clap(long)]
178 : redis_notifications: Option<String>,
179 : /// redis host for streaming connections (might be different from the notifications host)
180 : #[clap(long)]
181 : redis_host: Option<String>,
182 : /// redis port for streaming connections (might be different from the notifications host)
183 : #[clap(long)]
184 : redis_port: Option<u16>,
185 : /// redis cluster name, used in aws elasticache
186 : #[clap(long)]
187 : redis_cluster_name: Option<String>,
188 : /// redis user_id, used in aws elasticache
189 : #[clap(long)]
190 : redis_user_id: Option<String>,
191 : /// aws region to retrieve credentials
192 2 : #[clap(long, default_value_t = String::new())]
193 0 : aws_region: String,
194 : /// cache for `project_info` (use `size=0` to disable)
195 : #[clap(long, default_value = config::ProjectInfoCacheOptions::CACHE_DEFAULT_OPTIONS)]
196 0 : project_info_cache: String,
197 : /// cache for all valid endpoints
198 : #[clap(long, default_value = config::EndpointCacheConfig::CACHE_DEFAULT_OPTIONS)]
199 0 : endpoint_cache_config: String,
200 : #[clap(flatten)]
201 : parquet_upload: ParquetUploadArgs,
202 :
203 : /// interval for backup metric collection
204 : #[clap(long, default_value = "10m", value_parser = humantime::parse_duration)]
205 0 : metric_backup_collection_interval: std::time::Duration,
206 : /// remote storage configuration for backup metric collection
207 : /// Encoded as toml (same format as pageservers), eg
208 : /// `{bucket_name='the-bucket',bucket_region='us-east-1',prefix_in_bucket='proxy',endpoint='http://minio:9000'}`
209 : #[clap(long, value_parser = remote_storage_from_toml)]
210 : metric_backup_collection_remote_storage: Option<RemoteStorageConfig>,
211 : /// chunk size for backup metric collection
212 : /// Size of each event is no more than 400 bytes, so 2**22 is about 200MB before the compression.
213 : #[clap(long, default_value = "4194304")]
214 0 : metric_backup_collection_chunk_size: usize,
215 : /// Whether to retry the connection to the compute node
216 : #[clap(long, default_value = config::RetryConfig::CONNECT_TO_COMPUTE_DEFAULT_VALUES)]
217 0 : connect_to_compute_retry: String,
218 : /// Whether to retry the wake_compute request
219 : #[clap(long, default_value = config::RetryConfig::WAKE_COMPUTE_DEFAULT_VALUES)]
220 0 : wake_compute_retry: String,
221 : }
222 :
223 4 : #[derive(clap::Args, Clone, Copy, Debug)]
224 : struct SqlOverHttpArgs {
225 : /// timeout for http connection requests
226 : #[clap(long, default_value = "15s", value_parser = humantime::parse_duration)]
227 0 : sql_over_http_timeout: tokio::time::Duration,
228 :
229 : /// Whether the SQL over http pool is opt-in
230 2 : #[clap(long, default_value_t = true, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
231 0 : sql_over_http_pool_opt_in: bool,
232 :
233 : /// How many connections to pool for each endpoint. Excess connections are discarded
234 2 : #[clap(long, default_value_t = 20)]
235 0 : sql_over_http_pool_max_conns_per_endpoint: usize,
236 :
237 : /// How many connections to pool for each endpoint. Excess connections are discarded
238 2 : #[clap(long, default_value_t = 20000)]
239 0 : sql_over_http_pool_max_total_conns: usize,
240 :
241 : /// How long pooled connections should remain idle for before closing
242 : #[clap(long, default_value = "5m", value_parser = humantime::parse_duration)]
243 0 : sql_over_http_idle_timeout: tokio::time::Duration,
244 :
245 : /// Duration each shard will wait on average before a GC sweep.
246 : /// A longer time will causes sweeps to take longer but will interfere less frequently.
247 : #[clap(long, default_value = "10m", value_parser = humantime::parse_duration)]
248 0 : sql_over_http_pool_gc_epoch: tokio::time::Duration,
249 :
250 : /// How many shards should the global pool have. Must be a power of two.
251 : /// More shards will introduce less contention for pool operations, but can
252 : /// increase memory used by the pool
253 2 : #[clap(long, default_value_t = 128)]
254 0 : sql_over_http_pool_shards: usize,
255 :
256 2 : #[clap(long, default_value_t = 10000)]
257 0 : sql_over_http_client_conn_threshold: u64,
258 :
259 2 : #[clap(long, default_value_t = 64)]
260 0 : sql_over_http_cancel_set_shards: usize,
261 : }
262 :
263 : #[tokio::main]
264 0 : async fn main() -> anyhow::Result<()> {
265 0 : let _logging_guard = proxy::logging::init().await?;
266 0 : let _panic_hook_guard = utils::logging::replace_panic_hook_with_tracing_panic_hook();
267 0 : let _sentry_guard = init_sentry(Some(GIT_VERSION.into()), &[]);
268 0 :
269 0 : info!("Version: {GIT_VERSION}");
270 0 : info!("Build_tag: {BUILD_TAG}");
271 0 : let neon_metrics = ::metrics::NeonMetrics::new(::metrics::BuildInfo {
272 0 : revision: GIT_VERSION,
273 0 : build_tag: BUILD_TAG,
274 0 : });
275 0 :
276 0 : let jemalloc = match proxy::jemalloc::MetricRecorder::new() {
277 0 : Ok(t) => Some(t),
278 0 : Err(e) => {
279 0 : tracing::error!(error = ?e, "could not start jemalloc metrics loop");
280 0 : None
281 0 : }
282 0 : };
283 0 :
284 0 : let args = ProxyCliArgs::parse();
285 0 : let config = build_config(&args)?;
286 0 :
287 0 : info!("Authentication backend: {}", config.auth_backend);
288 0 : info!("Using region: {}", config.aws_region);
289 0 :
290 0 : let region_provider = RegionProviderChain::default_provider().or_else(&*config.aws_region); // Replace with your Redis region if needed
291 0 : let provider_conf =
292 0 : ProviderConfig::without_region().with_region(region_provider.region().await);
293 0 : let aws_credentials_provider = {
294 0 : // uses "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY"
295 0 : CredentialsProviderChain::first_try("env", EnvironmentVariableCredentialsProvider::new())
296 0 : // uses "AWS_PROFILE" / `aws sso login --profile <profile>`
297 0 : .or_else(
298 0 : "profile-sso",
299 0 : ProfileFileCredentialsProvider::builder()
300 0 : .configure(&provider_conf)
301 0 : .build(),
302 0 : )
303 0 : // uses "AWS_WEB_IDENTITY_TOKEN_FILE", "AWS_ROLE_ARN", "AWS_ROLE_SESSION_NAME"
304 0 : // needed to access remote extensions bucket
305 0 : .or_else(
306 0 : "token",
307 0 : WebIdentityTokenCredentialsProvider::builder()
308 0 : .configure(&provider_conf)
309 0 : .build(),
310 0 : )
311 0 : // uses imds v2
312 0 : .or_else("imds", ImdsCredentialsProvider::builder().build())
313 0 : };
314 0 : let elasticache_credentials_provider = Arc::new(elasticache::CredentialsProvider::new(
315 0 : elasticache::AWSIRSAConfig::new(
316 0 : config.aws_region.clone(),
317 0 : args.redis_cluster_name,
318 0 : args.redis_user_id,
319 0 : ),
320 0 : aws_credentials_provider,
321 0 : ));
322 0 : let regional_redis_client = match (args.redis_host, args.redis_port) {
323 0 : (Some(host), Some(port)) => Some(
324 0 : ConnectionWithCredentialsProvider::new_with_credentials_provider(
325 0 : host,
326 0 : port,
327 0 : elasticache_credentials_provider.clone(),
328 0 : ),
329 0 : ),
330 0 : (None, None) => {
331 0 : warn!("Redis events from console are disabled");
332 0 : None
333 0 : }
334 0 : _ => {
335 0 : bail!("redis-host and redis-port must be specified together");
336 0 : }
337 0 : };
338 0 : let redis_notifications_client = if let Some(url) = args.redis_notifications {
339 0 : Some(ConnectionWithCredentialsProvider::new_with_static_credentials(url))
340 0 : } else {
341 0 : regional_redis_client.clone()
342 0 : };
343 0 :
344 0 : // Check that we can bind to address before further initialization
345 0 : let http_address: SocketAddr = args.http.parse()?;
346 0 : info!("Starting http on {http_address}");
347 0 : let http_listener = TcpListener::bind(http_address).await?.into_std()?;
348 0 :
349 0 : let mgmt_address: SocketAddr = args.mgmt.parse()?;
350 0 : info!("Starting mgmt on {mgmt_address}");
351 0 : let mgmt_listener = TcpListener::bind(mgmt_address).await?;
352 0 :
353 0 : let proxy_address: SocketAddr = args.proxy.parse()?;
354 0 : info!("Starting proxy on {proxy_address}");
355 0 : let proxy_listener = TcpListener::bind(proxy_address).await?;
356 0 : let cancellation_token = CancellationToken::new();
357 0 :
358 0 : let cancel_map = CancelMap::default();
359 0 :
360 0 : let redis_publisher = match ®ional_redis_client {
361 0 : Some(redis_publisher) => Some(Arc::new(Mutex::new(RedisPublisherClient::new(
362 0 : redis_publisher.clone(),
363 0 : args.region.clone(),
364 0 : &config.redis_rps_limit,
365 0 : )?))),
366 0 : None => None,
367 0 : };
368 0 : let cancellation_handler = Arc::new(CancellationHandler::<
369 0 : Option<Arc<tokio::sync::Mutex<RedisPublisherClient>>>,
370 0 : >::new(
371 0 : cancel_map.clone(),
372 0 : redis_publisher,
373 0 : proxy::metrics::CancellationSource::FromClient,
374 0 : ));
375 0 :
376 0 : let mut endpoint_rps_limit = args.endpoint_rps_limit.clone();
377 0 : RateBucketInfo::validate(&mut endpoint_rps_limit)?;
378 0 : let endpoint_rate_limiter = Arc::new(EndpointRateLimiter::new(endpoint_rps_limit));
379 0 :
380 0 : // client facing tasks. these will exit on error or on cancellation
381 0 : // cancellation returns Ok(())
382 0 : let mut client_tasks = JoinSet::new();
383 0 : client_tasks.spawn(proxy::proxy::task_main(
384 0 : config,
385 0 : proxy_listener,
386 0 : cancellation_token.clone(),
387 0 : cancellation_handler.clone(),
388 0 : endpoint_rate_limiter.clone(),
389 0 : ));
390 0 :
391 0 : // TODO: rename the argument to something like serverless.
392 0 : // It now covers more than just websockets, it also covers SQL over HTTP.
393 0 : if let Some(serverless_address) = args.wss {
394 0 : let serverless_address: SocketAddr = serverless_address.parse()?;
395 0 : info!("Starting wss on {serverless_address}");
396 0 : let serverless_listener = TcpListener::bind(serverless_address).await?;
397 0 :
398 0 : client_tasks.spawn(serverless::task_main(
399 0 : config,
400 0 : serverless_listener,
401 0 : cancellation_token.clone(),
402 0 : cancellation_handler.clone(),
403 0 : endpoint_rate_limiter.clone(),
404 0 : ));
405 0 : }
406 0 :
407 0 : client_tasks.spawn(proxy::context::parquet::worker(
408 0 : cancellation_token.clone(),
409 0 : args.parquet_upload,
410 0 : ));
411 0 :
412 0 : // maintenance tasks. these never return unless there's an error
413 0 : let mut maintenance_tasks = JoinSet::new();
414 0 : maintenance_tasks.spawn(proxy::handle_signals(cancellation_token.clone()));
415 0 : maintenance_tasks.spawn(http::health_server::task_main(
416 0 : http_listener,
417 0 : AppMetrics {
418 0 : jemalloc,
419 0 : neon_metrics,
420 0 : proxy: proxy::metrics::Metrics::get(),
421 0 : },
422 0 : ));
423 0 : maintenance_tasks.spawn(console::mgmt::task_main(mgmt_listener));
424 0 :
425 0 : if let Some(metrics_config) = &config.metric_collection {
426 0 : // TODO: Add gc regardles of the metric collection being enabled.
427 0 : maintenance_tasks.spawn(usage_metrics::task_main(metrics_config));
428 0 : client_tasks.spawn(usage_metrics::task_backup(
429 0 : &metrics_config.backup_metric_collection_config,
430 0 : cancellation_token.clone(),
431 0 : ));
432 0 : }
433 0 :
434 0 : if let auth::BackendType::Console(api, _) = &config.auth_backend {
435 0 : if let proxy::console::provider::ConsoleBackend::Console(api) = &**api {
436 0 : match (redis_notifications_client, regional_redis_client.clone()) {
437 0 : (None, None) => {}
438 0 : (client1, client2) => {
439 0 : let cache = api.caches.project_info.clone();
440 0 : if let Some(client) = client1 {
441 0 : maintenance_tasks.spawn(notifications::task_main(
442 0 : client,
443 0 : cache.clone(),
444 0 : cancel_map.clone(),
445 0 : args.region.clone(),
446 0 : ));
447 0 : }
448 0 : if let Some(client) = client2 {
449 0 : maintenance_tasks.spawn(notifications::task_main(
450 0 : client,
451 0 : cache.clone(),
452 0 : cancel_map.clone(),
453 0 : args.region.clone(),
454 0 : ));
455 0 : }
456 0 : maintenance_tasks.spawn(async move { cache.clone().gc_worker().await });
457 0 : }
458 0 : }
459 0 : if let Some(regional_redis_client) = regional_redis_client {
460 0 : let cache = api.caches.endpoints_cache.clone();
461 0 : let con = regional_redis_client;
462 0 : let span = tracing::info_span!("endpoints_cache");
463 0 : maintenance_tasks.spawn(
464 0 : async move { cache.do_read(con, cancellation_token.clone()).await }
465 0 : .instrument(span),
466 0 : );
467 0 : }
468 0 : }
469 0 : }
470 0 :
471 0 : let maintenance = loop {
472 0 : // get one complete task
473 0 : match futures::future::select(
474 0 : pin!(maintenance_tasks.join_next()),
475 0 : pin!(client_tasks.join_next()),
476 0 : )
477 0 : .await
478 0 : {
479 0 : // exit immediately on maintenance task completion
480 0 : Either::Left((Some(res), _)) => break proxy::flatten_err(res)?,
481 0 : // exit with error immediately if all maintenance tasks have ceased (should be caught by branch above)
482 0 : Either::Left((None, _)) => bail!("no maintenance tasks running. invalid state"),
483 0 : // exit immediately on client task error
484 0 : Either::Right((Some(res), _)) => proxy::flatten_err(res)?,
485 0 : // exit if all our client tasks have shutdown gracefully
486 0 : Either::Right((None, _)) => return Ok(()),
487 0 : }
488 0 : };
489 0 :
490 0 : // maintenance tasks return Infallible success values, this is an impossible value
491 0 : // so this match statically ensures that there are no possibilities for that value
492 0 : match maintenance {}
493 0 : }
494 :
495 : /// ProxyConfig is created at proxy startup, and lives forever.
496 0 : fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
497 0 : let thread_pool = ThreadPool::new(args.scram_thread_pool_size);
498 0 : Metrics::install(thread_pool.metrics.clone());
499 :
500 0 : let tls_config = match (&args.tls_key, &args.tls_cert) {
501 0 : (Some(key_path), Some(cert_path)) => Some(config::configure_tls(
502 0 : key_path,
503 0 : cert_path,
504 0 : args.certs_dir.as_ref(),
505 0 : )?),
506 0 : (None, None) => None,
507 0 : _ => bail!("either both or neither tls-key and tls-cert must be specified"),
508 : };
509 :
510 0 : if args.allow_self_signed_compute {
511 0 : warn!("allowing self-signed compute certificates");
512 0 : }
513 0 : let backup_metric_collection_config = config::MetricBackupCollectionConfig {
514 0 : interval: args.metric_backup_collection_interval,
515 0 : remote_storage_config: args.metric_backup_collection_remote_storage.clone(),
516 0 : chunk_size: args.metric_backup_collection_chunk_size,
517 0 : };
518 :
519 0 : let metric_collection = match (
520 0 : &args.metric_collection_endpoint,
521 0 : &args.metric_collection_interval,
522 : ) {
523 0 : (Some(endpoint), Some(interval)) => Some(config::MetricCollectionConfig {
524 0 : endpoint: endpoint.parse()?,
525 0 : interval: humantime::parse_duration(interval)?,
526 0 : backup_metric_collection_config,
527 : }),
528 0 : (None, None) => None,
529 0 : _ => bail!(
530 0 : "either both or neither metric-collection-endpoint \
531 0 : and metric-collection-interval must be specified"
532 0 : ),
533 : };
534 0 : if !args.disable_dynamic_rate_limiter {
535 0 : bail!("dynamic rate limiter should be disabled");
536 0 : }
537 :
538 0 : let auth_backend = match &args.auth_backend {
539 : AuthBackend::Console => {
540 0 : let wake_compute_cache_config: CacheOptions = args.wake_compute_cache.parse()?;
541 0 : let project_info_cache_config: ProjectInfoCacheOptions =
542 0 : args.project_info_cache.parse()?;
543 0 : let endpoint_cache_config: config::EndpointCacheConfig =
544 0 : args.endpoint_cache_config.parse()?;
545 :
546 0 : info!("Using NodeInfoCache (wake_compute) with options={wake_compute_cache_config:?}");
547 0 : info!(
548 0 : "Using AllowedIpsCache (wake_compute) with options={project_info_cache_config:?}"
549 : );
550 0 : info!("Using EndpointCacheConfig with options={endpoint_cache_config:?}");
551 0 : let caches = Box::leak(Box::new(console::caches::ApiCaches::new(
552 0 : wake_compute_cache_config,
553 0 : project_info_cache_config,
554 0 : endpoint_cache_config,
555 0 : )));
556 :
557 : let config::ConcurrencyLockOptions {
558 0 : shards,
559 0 : limiter,
560 0 : epoch,
561 0 : timeout,
562 0 : } = args.wake_compute_lock.parse()?;
563 0 : info!(?limiter, shards, ?epoch, "Using NodeLocks (wake_compute)");
564 0 : let locks = Box::leak(Box::new(console::locks::ApiLocks::new(
565 0 : "wake_compute_lock",
566 0 : limiter,
567 0 : shards,
568 0 : timeout,
569 0 : epoch,
570 0 : &Metrics::get().wake_compute_lock,
571 0 : )?));
572 0 : tokio::spawn(locks.garbage_collect_worker());
573 :
574 0 : let url = args.auth_endpoint.parse()?;
575 0 : let endpoint = http::Endpoint::new(url, http::new_client());
576 0 :
577 0 : let mut wake_compute_rps_limit = args.wake_compute_limit.clone();
578 0 : RateBucketInfo::validate(&mut wake_compute_rps_limit)?;
579 0 : let wake_compute_endpoint_rate_limiter =
580 0 : Arc::new(EndpointRateLimiter::new(wake_compute_rps_limit));
581 0 : let api = console::provider::neon::Api::new(
582 0 : endpoint,
583 0 : caches,
584 0 : locks,
585 0 : wake_compute_endpoint_rate_limiter,
586 0 : );
587 0 : let api = console::provider::ConsoleBackend::Console(api);
588 0 : auth::BackendType::Console(MaybeOwned::Owned(api), ())
589 : }
590 : #[cfg(feature = "testing")]
591 : AuthBackend::Postgres => {
592 0 : let url = args.auth_endpoint.parse()?;
593 0 : let api = console::provider::mock::Api::new(url);
594 0 : let api = console::provider::ConsoleBackend::Postgres(api);
595 0 : auth::BackendType::Console(MaybeOwned::Owned(api), ())
596 : }
597 : AuthBackend::Link => {
598 0 : let url = args.uri.parse()?;
599 0 : auth::BackendType::Link(MaybeOwned::Owned(url), ())
600 : }
601 : };
602 :
603 : let config::ConcurrencyLockOptions {
604 0 : shards,
605 0 : limiter,
606 0 : epoch,
607 0 : timeout,
608 0 : } = args.connect_compute_lock.parse()?;
609 0 : info!(
610 : ?limiter,
611 : shards,
612 : ?epoch,
613 0 : "Using NodeLocks (connect_compute)"
614 : );
615 0 : let connect_compute_locks = console::locks::ApiLocks::new(
616 0 : "connect_compute_lock",
617 0 : limiter,
618 0 : shards,
619 0 : timeout,
620 0 : epoch,
621 0 : &Metrics::get().proxy.connect_compute_lock,
622 0 : )?;
623 :
624 0 : let http_config = HttpConfig {
625 0 : request_timeout: args.sql_over_http.sql_over_http_timeout,
626 0 : pool_options: GlobalConnPoolOptions {
627 0 : max_conns_per_endpoint: args.sql_over_http.sql_over_http_pool_max_conns_per_endpoint,
628 0 : gc_epoch: args.sql_over_http.sql_over_http_pool_gc_epoch,
629 0 : pool_shards: args.sql_over_http.sql_over_http_pool_shards,
630 0 : idle_timeout: args.sql_over_http.sql_over_http_idle_timeout,
631 0 : opt_in: args.sql_over_http.sql_over_http_pool_opt_in,
632 0 : max_total_conns: args.sql_over_http.sql_over_http_pool_max_total_conns,
633 0 : },
634 0 : cancel_set: CancelSet::new(args.sql_over_http.sql_over_http_cancel_set_shards),
635 0 : client_conn_threshold: args.sql_over_http.sql_over_http_client_conn_threshold,
636 0 : };
637 0 : let authentication_config = AuthenticationConfig {
638 0 : thread_pool,
639 0 : scram_protocol_timeout: args.scram_protocol_timeout,
640 0 : rate_limiter_enabled: args.auth_rate_limit_enabled,
641 0 : rate_limiter: AuthRateLimiter::new(args.auth_rate_limit.clone()),
642 0 : rate_limit_ip_subnet: args.auth_rate_limit_ip_subnet,
643 0 : };
644 0 :
645 0 : let mut redis_rps_limit = args.redis_rps_limit.clone();
646 0 : RateBucketInfo::validate(&mut redis_rps_limit)?;
647 :
648 0 : let config = Box::leak(Box::new(ProxyConfig {
649 0 : tls_config,
650 0 : auth_backend,
651 0 : metric_collection,
652 0 : allow_self_signed_compute: args.allow_self_signed_compute,
653 0 : http_config,
654 0 : authentication_config,
655 0 : require_client_ip: args.require_client_ip,
656 0 : disable_ip_check_for_http: args.disable_ip_check_for_http,
657 0 : redis_rps_limit,
658 0 : handshake_timeout: args.handshake_timeout,
659 0 : region: args.region.clone(),
660 0 : aws_region: args.aws_region.clone(),
661 0 : wake_compute_retry_config: config::RetryConfig::parse(&args.wake_compute_retry)?,
662 0 : connect_compute_locks,
663 0 : connect_to_compute_retry_config: config::RetryConfig::parse(
664 0 : &args.connect_to_compute_retry,
665 0 : )?,
666 : }));
667 :
668 0 : tokio::spawn(config.connect_compute_locks.garbage_collect_worker());
669 0 :
670 0 : Ok(config)
671 0 : }
672 :
673 : #[cfg(test)]
674 : mod tests {
675 : use std::time::Duration;
676 :
677 : use clap::Parser;
678 : use proxy::rate_limiter::RateBucketInfo;
679 :
680 : #[test]
681 2 : fn parse_endpoint_rps_limit() {
682 2 : let config = super::ProxyCliArgs::parse_from([
683 2 : "proxy",
684 2 : "--endpoint-rps-limit",
685 2 : "100@1s",
686 2 : "--endpoint-rps-limit",
687 2 : "20@30s",
688 2 : ]);
689 2 :
690 2 : assert_eq!(
691 2 : config.endpoint_rps_limit,
692 2 : vec![
693 2 : RateBucketInfo::new(100, Duration::from_secs(1)),
694 2 : RateBucketInfo::new(20, Duration::from_secs(30)),
695 2 : ]
696 2 : );
697 2 : }
698 : }
|
