Line data Source code
1 : use std::io;
2 : use std::net::SocketAddr;
3 : use std::time::Duration;
4 :
5 : use futures::{FutureExt, TryFutureExt};
6 : use itertools::Itertools;
7 : use postgres_client::tls::MakeTlsConnect;
8 : use postgres_client::{CancelToken, RawConnection};
9 : use postgres_protocol::message::backend::NoticeResponseBody;
10 : use pq_proto::StartupMessageParams;
11 : use rustls::pki_types::InvalidDnsNameError;
12 : use thiserror::Error;
13 : use tokio::net::TcpStream;
14 : use tracing::{debug, error, info, warn};
15 :
16 : use crate::auth::parse_endpoint_param;
17 : use crate::cancellation::CancelClosure;
18 : use crate::config::ComputeConfig;
19 : use crate::context::RequestContext;
20 : use crate::control_plane::client::ApiLockError;
21 : use crate::control_plane::errors::WakeComputeError;
22 : use crate::control_plane::messages::MetricsAuxInfo;
23 : use crate::error::{ReportableError, UserFacingError};
24 : use crate::metrics::{Metrics, NumDbConnectionsGuard};
25 : use crate::proxy::neon_option;
26 : use crate::tls::postgres_rustls::MakeRustlsConnect;
27 : use crate::types::Host;
28 :
29 : pub const COULD_NOT_CONNECT: &str = "Couldn't connect to compute node";
30 :
31 : #[derive(Debug, Error)]
32 : pub(crate) enum ConnectionError {
33 : /// This error doesn't seem to reveal any secrets; for instance,
34 : /// `postgres_client::error::Kind` doesn't contain ip addresses and such.
35 : #[error("{COULD_NOT_CONNECT}: {0}")]
36 : Postgres(#[from] postgres_client::Error),
37 :
38 : #[error("{COULD_NOT_CONNECT}: {0}")]
39 : CouldNotConnect(#[from] io::Error),
40 :
41 : #[error("{COULD_NOT_CONNECT}: {0}")]
42 : TlsError(#[from] InvalidDnsNameError),
43 :
44 : #[error("{COULD_NOT_CONNECT}: {0}")]
45 : WakeComputeError(#[from] WakeComputeError),
46 :
47 : #[error("error acquiring resource permit: {0}")]
48 : TooManyConnectionAttempts(#[from] ApiLockError),
49 : }
50 :
51 : impl UserFacingError for ConnectionError {
52 0 : fn to_string_client(&self) -> String {
53 0 : match self {
54 : // This helps us drop irrelevant library-specific prefixes.
55 : // TODO: propagate severity level and other parameters.
56 0 : ConnectionError::Postgres(err) => match err.as_db_error() {
57 0 : Some(err) => {
58 0 : let msg = err.message();
59 0 :
60 0 : if msg.starts_with("unsupported startup parameter: ")
61 0 : || msg.starts_with("unsupported startup parameter in options: ")
62 : {
63 0 : format!("{msg}. Please use unpooled connection or remove this parameter from the startup package. More details: https://neon.tech/docs/connect/connection-errors#unsupported-startup-parameter")
64 : } else {
65 0 : msg.to_owned()
66 : }
67 : }
68 0 : None => err.to_string(),
69 : },
70 0 : ConnectionError::WakeComputeError(err) => err.to_string_client(),
71 : ConnectionError::TooManyConnectionAttempts(_) => {
72 0 : "Failed to acquire permit to connect to the database. Too many database connection attempts are currently ongoing.".to_owned()
73 : }
74 0 : _ => COULD_NOT_CONNECT.to_owned(),
75 : }
76 0 : }
77 : }
78 :
79 : impl ReportableError for ConnectionError {
80 0 : fn get_error_kind(&self) -> crate::error::ErrorKind {
81 0 : match self {
82 0 : ConnectionError::Postgres(e) if e.as_db_error().is_some() => {
83 0 : crate::error::ErrorKind::Postgres
84 : }
85 0 : ConnectionError::Postgres(_) => crate::error::ErrorKind::Compute,
86 0 : ConnectionError::CouldNotConnect(_) => crate::error::ErrorKind::Compute,
87 0 : ConnectionError::TlsError(_) => crate::error::ErrorKind::Compute,
88 0 : ConnectionError::WakeComputeError(e) => e.get_error_kind(),
89 0 : ConnectionError::TooManyConnectionAttempts(e) => e.get_error_kind(),
90 : }
91 0 : }
92 : }
93 :
94 : /// A pair of `ClientKey` & `ServerKey` for `SCRAM-SHA-256`.
95 : pub(crate) type ScramKeys = postgres_client::config::ScramKeys<32>;
96 :
97 : /// A config for establishing a connection to compute node.
98 : /// Eventually, `postgres_client` will be replaced with something better.
99 : /// Newtype allows us to implement methods on top of it.
100 : #[derive(Clone)]
101 : pub(crate) struct ConnCfg(Box<postgres_client::Config>);
102 :
103 : /// Creation and initialization routines.
104 : impl ConnCfg {
105 10 : pub(crate) fn new(host: String, port: u16) -> Self {
106 10 : Self(Box::new(postgres_client::Config::new(host, port)))
107 10 : }
108 :
109 : /// Reuse password or auth keys from the other config.
110 4 : pub(crate) fn reuse_password(&mut self, other: Self) {
111 4 : if let Some(password) = other.get_password() {
112 4 : self.password(password);
113 4 : }
114 :
115 4 : if let Some(keys) = other.get_auth_keys() {
116 0 : self.auth_keys(keys);
117 4 : }
118 4 : }
119 :
120 0 : pub(crate) fn get_host(&self) -> Host {
121 0 : match self.0.get_host() {
122 0 : postgres_client::config::Host::Tcp(s) => s.into(),
123 0 : }
124 0 : }
125 :
126 : /// Apply startup message params to the connection config.
127 0 : pub(crate) fn set_startup_params(
128 0 : &mut self,
129 0 : params: &StartupMessageParams,
130 0 : arbitrary_params: bool,
131 0 : ) {
132 0 : if !arbitrary_params {
133 0 : self.set_param("client_encoding", "UTF8");
134 0 : }
135 0 : for (k, v) in params.iter() {
136 0 : match k {
137 : // Only set `user` if it's not present in the config.
138 : // Console redirect auth flow takes username from the console's response.
139 0 : "user" if self.user_is_set() => continue,
140 0 : "database" if self.db_is_set() => continue,
141 0 : "options" => {
142 0 : if let Some(options) = filtered_options(v) {
143 0 : self.set_param(k, &options);
144 0 : }
145 : }
146 0 : "user" | "database" | "application_name" | "replication" => {
147 0 : self.set_param(k, v);
148 0 : }
149 :
150 : // if we allow arbitrary params, then we forward them through.
151 : // this is a flag for a period of backwards compatibility
152 0 : k if arbitrary_params => {
153 0 : self.set_param(k, v);
154 0 : }
155 0 : _ => {}
156 : }
157 : }
158 0 : }
159 : }
160 :
161 : impl std::ops::Deref for ConnCfg {
162 : type Target = postgres_client::Config;
163 :
164 8 : fn deref(&self) -> &Self::Target {
165 8 : &self.0
166 8 : }
167 : }
168 :
169 : /// For now, let's make it easier to setup the config.
170 : impl std::ops::DerefMut for ConnCfg {
171 10 : fn deref_mut(&mut self) -> &mut Self::Target {
172 10 : &mut self.0
173 10 : }
174 : }
175 :
176 : impl ConnCfg {
177 : /// Establish a raw TCP connection to the compute node.
178 0 : async fn connect_raw(&self, timeout: Duration) -> io::Result<(SocketAddr, TcpStream, &str)> {
179 : use postgres_client::config::Host;
180 :
181 : // wrap TcpStream::connect with timeout
182 0 : let connect_with_timeout = |host, port| {
183 0 : tokio::time::timeout(timeout, TcpStream::connect((host, port))).map(
184 0 : move |res| match res {
185 0 : Ok(tcpstream_connect_res) => tcpstream_connect_res,
186 0 : Err(_) => Err(io::Error::new(
187 0 : io::ErrorKind::TimedOut,
188 0 : format!("exceeded connection timeout {timeout:?}"),
189 0 : )),
190 0 : },
191 0 : )
192 0 : };
193 :
194 0 : let connect_once = |host, port| {
195 0 : debug!("trying to connect to compute node at {host}:{port}");
196 0 : connect_with_timeout(host, port).and_then(|stream| async {
197 0 : let socket_addr = stream.peer_addr()?;
198 0 : let socket = socket2::SockRef::from(&stream);
199 0 : // Disable Nagle's algorithm to not introduce latency between
200 0 : // client and compute.
201 0 : socket.set_nodelay(true)?;
202 : // This prevents load balancer from severing the connection.
203 0 : socket.set_keepalive(true)?;
204 0 : Ok((socket_addr, stream))
205 0 : })
206 0 : };
207 :
208 : // We can't reuse connection establishing logic from `postgres_client` here,
209 : // because it has no means for extracting the underlying socket which we
210 : // require for our business.
211 0 : let port = self.0.get_port();
212 0 : let host = self.0.get_host();
213 0 :
214 0 : let host = match host {
215 0 : Host::Tcp(host) => host.as_str(),
216 0 : };
217 0 :
218 0 : match connect_once(host, port).await {
219 0 : Ok((sockaddr, stream)) => Ok((sockaddr, stream, host)),
220 0 : Err(err) => {
221 0 : warn!("couldn't connect to compute node at {host}:{port}: {err}");
222 0 : Err(err)
223 : }
224 : }
225 0 : }
226 : }
227 :
228 : type RustlsStream = <MakeRustlsConnect as MakeTlsConnect<tokio::net::TcpStream>>::Stream;
229 :
230 : pub(crate) struct PostgresConnection {
231 : /// Socket connected to a compute node.
232 : pub(crate) stream:
233 : postgres_client::maybe_tls_stream::MaybeTlsStream<tokio::net::TcpStream, RustlsStream>,
234 : /// PostgreSQL connection parameters.
235 : pub(crate) params: std::collections::HashMap<String, String>,
236 : /// Query cancellation token.
237 : pub(crate) cancel_closure: CancelClosure,
238 : /// Labels for proxy's metrics.
239 : pub(crate) aux: MetricsAuxInfo,
240 : /// Notices received from compute after authenticating
241 : pub(crate) delayed_notice: Vec<NoticeResponseBody>,
242 :
243 : _guage: NumDbConnectionsGuard<'static>,
244 : }
245 :
246 : impl ConnCfg {
247 : /// Connect to a corresponding compute node.
248 0 : pub(crate) async fn connect(
249 0 : &self,
250 0 : ctx: &RequestContext,
251 0 : aux: MetricsAuxInfo,
252 0 : config: &ComputeConfig,
253 0 : ) -> Result<PostgresConnection, ConnectionError> {
254 0 : let pause = ctx.latency_timer_pause(crate::metrics::Waiting::Compute);
255 0 : let (socket_addr, stream, host) = self.connect_raw(config.timeout).await?;
256 0 : drop(pause);
257 0 :
258 0 : let mut mk_tls = crate::tls::postgres_rustls::MakeRustlsConnect::new(config.tls.clone());
259 0 : let tls = <MakeRustlsConnect as MakeTlsConnect<tokio::net::TcpStream>>::make_tls_connect(
260 0 : &mut mk_tls,
261 0 : host,
262 0 : )?;
263 :
264 : // connect_raw() will not use TLS if sslmode is "disable"
265 0 : let pause = ctx.latency_timer_pause(crate::metrics::Waiting::Compute);
266 0 : let connection = self.0.connect_raw(stream, tls).await?;
267 0 : drop(pause);
268 0 :
269 0 : let RawConnection {
270 0 : stream,
271 0 : parameters,
272 0 : delayed_notice,
273 0 : process_id,
274 0 : secret_key,
275 0 : } = connection;
276 0 :
277 0 : tracing::Span::current().record("pid", tracing::field::display(process_id));
278 0 : let stream = stream.into_inner();
279 0 :
280 0 : // TODO: lots of useful info but maybe we can move it elsewhere (eg traces?)
281 0 : info!(
282 0 : cold_start_info = ctx.cold_start_info().as_str(),
283 0 : "connected to compute node at {host} ({socket_addr}) sslmode={:?}",
284 0 : self.0.get_ssl_mode()
285 : );
286 :
287 : // NB: CancelToken is supposed to hold socket_addr, but we use connect_raw.
288 : // Yet another reason to rework the connection establishing code.
289 0 : let cancel_closure = CancelClosure::new(
290 0 : socket_addr,
291 0 : CancelToken {
292 0 : socket_config: None,
293 0 : ssl_mode: self.0.get_ssl_mode(),
294 0 : process_id,
295 0 : secret_key,
296 0 : },
297 0 : vec![],
298 0 : host.to_string(),
299 0 : );
300 0 :
301 0 : let connection = PostgresConnection {
302 0 : stream,
303 0 : params: parameters,
304 0 : delayed_notice,
305 0 : cancel_closure,
306 0 : aux,
307 0 : _guage: Metrics::get().proxy.db_connections.guard(ctx.protocol()),
308 0 : };
309 0 :
310 0 : Ok(connection)
311 0 : }
312 : }
313 :
314 : /// Retrieve `options` from a startup message, dropping all proxy-secific flags.
315 6 : fn filtered_options(options: &str) -> Option<String> {
316 6 : #[allow(unstable_name_collisions)]
317 6 : let options: String = StartupMessageParams::parse_options_raw(options)
318 14 : .filter(|opt| parse_endpoint_param(opt).is_none() && neon_option(opt).is_none())
319 6 : .intersperse(" ") // TODO: use impl from std once it's stabilized
320 6 : .collect();
321 6 :
322 6 : // Don't even bother with empty options.
323 6 : if options.is_empty() {
324 3 : return None;
325 3 : }
326 3 :
327 3 : Some(options)
328 6 : }
329 :
330 : #[cfg(test)]
331 : mod tests {
332 : use super::*;
333 :
334 : #[test]
335 1 : fn test_filtered_options() {
336 1 : // Empty options is unlikely to be useful anyway.
337 1 : let params = "";
338 1 : assert_eq!(filtered_options(params), None);
339 :
340 : // It's likely that clients will only use options to specify endpoint/project.
341 1 : let params = "project=foo";
342 1 : assert_eq!(filtered_options(params), None);
343 :
344 : // Same, because unescaped whitespaces are no-op.
345 1 : let params = " project=foo ";
346 1 : assert_eq!(filtered_options(params).as_deref(), None);
347 :
348 1 : let params = r"\ project=foo \ ";
349 1 : assert_eq!(filtered_options(params).as_deref(), Some(r"\ \ "));
350 :
351 1 : let params = "project = foo";
352 1 : assert_eq!(filtered_options(params).as_deref(), Some("project = foo"));
353 :
354 1 : let params = "project = foo neon_endpoint_type:read_write neon_lsn:0/2 neon_proxy_params_compat:true";
355 1 : assert_eq!(filtered_options(params).as_deref(), Some("project = foo"));
356 1 : }
357 : }
|
