Implementing graceful shutdown (#105)

* Initial commit for graceful shutdown

* fmt

* Add .vscode to gitignore

* Updates shutdown logic to use channels

* fmt

* fmt

* Adds shutdown timeout

* Fmt and updates tomls

* Updates readme

* fmt and updates log levels

* Update python tests to test shutdown

* merge changes

* Rename listener rx and update bash to be in line with master

* Update python test bash script ordering

* Adds error response message before shutdown

* Add details on shutdown event loop

* Fixes response length for error

* Adds handler for sigterm

* Uses ready for query function and fixes number of bytes

* fmt

Author: zainkabani

.circleci/pgcat.toml

@@ -17,6 +17,9 @@ connect_timeout = 100
 # How much time to give the health check query to return with a result (ms).
 healthcheck_timeout = 100
 
+# How much time to give clients during shutdown before forcibly killing client connections (ms).
+shutdown_timeout = 5000
+
 # For how long to ban a server if it fails a health check (seconds).
 ban_time = 60 # Seconds
 

.circleci/run_tests.sh

@@ -74,12 +74,12 @@ cd ../..
 
 #
 # Python tests
+# These tests will start and stop the pgcat server so it will need to be restarted after the tests
 #
-cd tests/python
-pip3 install -r requirements.txt
-python3 tests.py
-cd ../..
+pip3 install -r tests/python/requirements.txt
+python3 tests/python/tests.py
 
+start_pgcat "info"
 
 # Admin tests
 export PGPASSWORD=admin_pass

.gitignore

@@ -1,3 +1,4 @@
 .idea
 /target
 *.deb
+.vscode

README.md

@@ -47,6 +47,7 @@ psql -h 127.0.0.1 -p 6432 -c 'SELECT 1'
 | `pool_mode`             | The pool mode to use, i.e. `session` or `transaction`.                                                                                     | `transaction`                    |
 | `connect_timeout`       | Maximum time to establish a connection to a server (milliseconds). If reached, the server is banned and the next target is attempted.      | `5000`                           |
 | `healthcheck_timeout`   | Maximum time to pass a health check (`SELECT 1`, milliseconds). If reached, the server is banned and the next target is attempted.         | `1000`                           |
+| `shutdown_timeout`   | Maximum time to give clients during shutdown before forcibly killing client connections (ms).      | `60000`                           |
 | `ban_time`              | Ban time for a server (seconds). It won't be allowed to serve transactions until the ban expires; failover targets will be used instead.   | `60`                             |
 |                         |                                                                                                                                            |                                  |
 | **`user`**              |                                                                                                                                            |                                  |
@@ -250,6 +251,7 @@ The config can be reloaded by sending a `kill -s SIGHUP` to the process or by qu
 | `pool_mode`             | no                   |
 | `connect_timeout`       | yes                  |
 | `healthcheck_timeout`   | no                   |
+| `shutdown_timeout`      | no                   |
 | `ban_time`              | no                   |
 | `user`                  | yes                  |
 | `shards`                | yes                  |

examples/docker/pgcat.toml

@@ -17,6 +17,9 @@ connect_timeout = 5000
 # How much time to give `SELECT 1` health check query to return with a result (ms).
 healthcheck_timeout = 1000
 
+# How much time to give clients during shutdown before forcibly killing client connections (ms).
+shutdown_timeout = 60000
+
 # For how long to ban a server if it fails a health check (seconds).
 ban_time = 60 # seconds
 

pgcat.toml

@@ -17,6 +17,9 @@ connect_timeout = 5000
 # How much time to give the health check query to return with a result (ms).
 healthcheck_timeout = 1000
 
+# How much time to give clients during shutdown before forcibly killing client connections (ms).
+shutdown_timeout = 60000
+
 # For how long to ban a server if it fails a health check (seconds).
 ban_time = 60 # seconds
 

src/client.rs

@@ -4,6 +4,7 @@ use log::{debug, error, info, trace};
 use std::collections::HashMap;
 use tokio::io::{split, AsyncReadExt, BufReader, ReadHalf, WriteHalf};
 use tokio::net::TcpStream;
+use tokio::sync::broadcast::Receiver;
 
 use crate::admin::{generate_server_info_for_admin, handle_admin};
 use crate::config::get_config;
@@ -73,12 +74,15 @@ pub struct Client<S, T> {
     last_server_id: Option<i32>,
 
     target_pool: ConnectionPool,
+
+    shutdown_event_receiver: Receiver<()>,
 }
 
 /// Client entrypoint.
 pub async fn client_entrypoint(
     mut stream: TcpStream,
     client_server_map: ClientServerMap,
+    shutdown_event_receiver: Receiver<()>,
 ) -> Result<(), Error> {
     // Figure out if the client wants TLS or not.
     let addr = stream.peer_addr().unwrap();
@@ -97,7 +101,7 @@ pub async fn client_entrypoint(
                 write_all(&mut stream, yes).await?;
 
                 // Negotiate TLS.
-                match startup_tls(stream, client_server_map).await {
+                match startup_tls(stream, client_server_map, shutdown_event_receiver).await {
                     Ok(mut client) => {
                         info!("Client {:?} connected (TLS)", addr);
 
@@ -121,7 +125,16 @@ pub async fn client_entrypoint(
                         let (read, write) = split(stream);
 
                         // Continue with regular startup.
-                        match Client::startup(read, write, addr, bytes, client_server_map).await {
+                        match Client::startup(
+                            read,
+                            write,
+                            addr,
+                            bytes,
+                            client_server_map,
+                            shutdown_event_receiver,
+                        )
+                        .await
+                        {
                             Ok(mut client) => {
                                 info!("Client {:?} connected (plain)", addr);
 
@@ -142,7 +155,16 @@ pub async fn client_entrypoint(
             let (read, write) = split(stream);
 
             // Continue with regular startup.
-            match Client::startup(read, write, addr, bytes, client_server_map).await {
+            match Client::startup(
+                read,
+                write,
+                addr,
+                bytes,
+                client_server_map,
+                shutdown_event_receiver,
+            )
+            .await
+            {
                 Ok(mut client) => {
                     info!("Client {:?} connected (plain)", addr);
 
@@ -157,7 +179,16 @@ pub async fn client_entrypoint(
             let (read, write) = split(stream);
 
             // Continue with cancel query request.
-            match Client::cancel(read, write, addr, bytes, client_server_map).await {
+            match Client::cancel(
+                read,
+                write,
+                addr,
+                bytes,
+                client_server_map,
+                shutdown_event_receiver,
+            )
+            .await
+            {
                 Ok(mut client) => {
                     info!("Client {:?} issued a cancel query request", addr);
 
@@ -214,6 +245,7 @@ where
 pub async fn startup_tls(
     stream: TcpStream,
     client_server_map: ClientServerMap,
+    shutdown_event_receiver: Receiver<()>,
 ) -> Result<Client<ReadHalf<TlsStream<TcpStream>>, WriteHalf<TlsStream<TcpStream>>>, Error> {
     // Negotiate TLS.
     let tls = Tls::new()?;
@@ -237,7 +269,15 @@ pub async fn startup_tls(
         Ok((ClientConnectionType::Startup, bytes)) => {
             let (read, write) = split(stream);
 
-            Client::startup(read, write, addr, bytes, client_server_map).await
+            Client::startup(
+                read,
+                write,
+                addr,
+                bytes,
+                client_server_map,
+                shutdown_event_receiver,
+            )
+            .await
         }
 
         // Bad Postgres client.
@@ -258,6 +298,7 @@ where
         addr: std::net::SocketAddr,
         bytes: BytesMut, // The rest of the startup message.
         client_server_map: ClientServerMap,
+        shutdown_event_receiver: Receiver<()>,
     ) -> Result<Client<S, T>, Error> {
         let config = get_config();
         let stats = get_reporter();
@@ -384,6 +425,7 @@ where
             last_address_id: None,
             last_server_id: None,
             target_pool: target_pool,
+            shutdown_event_receiver: shutdown_event_receiver,
         });
     }
 
@@ -394,6 +436,7 @@ where
         addr: std::net::SocketAddr,
         mut bytes: BytesMut, // The rest of the startup message.
         client_server_map: ClientServerMap,
+        shutdown_event_receiver: Receiver<()>,
     ) -> Result<Client<S, T>, Error> {
         let process_id = bytes.get_i32();
         let secret_key = bytes.get_i32();
@@ -413,6 +456,7 @@ where
             last_address_id: None,
             last_server_id: None,
             target_pool: ConnectionPool::default(),
+            shutdown_event_receiver: shutdown_event_receiver,
         });
     }
 
@@ -467,7 +511,14 @@ where
             // We can parse it here before grabbing a server from the pool,
             // in case the client is sending some custom protocol messages, e.g.
             // SET SHARDING KEY TO 'bigint';
-            let mut message = read_message(&mut self.read).await?;
+
+            let mut message = tokio::select! {
+                _ = self.shutdown_event_receiver.recv() => {
+                    error_response_terminal(&mut self.write, &format!("terminating connection due to administrator command")).await?;
+                    return Ok(())
+                },
+                message_result = read_message(&mut self.read) => message_result?
+            };
 
             // Get a pool instance referenced by the most up-to-date
             // pointer. This ensures we always read the latest config

src/config.rs

@@ -119,6 +119,7 @@ pub struct General {
     pub port: i16,
     pub connect_timeout: u64,
     pub healthcheck_timeout: u64,
+    pub shutdown_timeout: u64,
     pub ban_time: i64,
     pub autoreload: bool,
     pub tls_certificate: Option<String>,
@@ -134,6 +135,7 @@ impl Default for General {
             port: 5432,
             connect_timeout: 5000,
             healthcheck_timeout: 1000,
+            shutdown_timeout: 60000,
             ban_time: 60,
             autoreload: false,
             tls_certificate: None,
@@ -273,6 +275,10 @@ impl From<&Config> for std::collections::HashMap<String, String> {
                 "healthcheck_timeout".to_string(),
                 config.general.healthcheck_timeout.to_string(),
             ),
+            (
+                "shutdown_timeout".to_string(),
+                config.general.shutdown_timeout.to_string(),
+            ),
             ("ban_time".to_string(), config.general.ban_time.to_string()),
         ];
 
@@ -290,6 +296,7 @@ impl Config {
             self.general.healthcheck_timeout
         );
         info!("Connection timeout: {}ms", self.general.connect_timeout);
+        info!("Shutdown timeout: {}ms", self.general.shutdown_timeout);
         match self.general.tls_certificate.clone() {
             Some(tls_certificate) => {
                 info!("TLS certificate: {}", tls_certificate);

src/main.rs

@@ -40,13 +40,13 @@ use log::{debug, error, info};
 use parking_lot::Mutex;
 use tokio::net::TcpListener;
 use tokio::{
-    signal,
     signal::unix::{signal as unix_signal, SignalKind},
     sync::mpsc,
 };
 
 use std::collections::HashMap;
 use std::sync::Arc;
+use tokio::sync::broadcast;
 
 mod admin;
 mod client;
@@ -139,24 +139,52 @@ async fn main() {
 
     info!("Waiting for clients");
 
+    let (shutdown_event_tx, mut shutdown_event_rx) = broadcast::channel::<()>(1);
+
+    let shutdown_event_tx_clone = shutdown_event_tx.clone();
+
     // Client connection loop.
     tokio::task::spawn(async move {
+        // Creates event subscriber for shutdown event, this is dropped when shutdown event is broadcast
+        let mut listener_shutdown_event_rx = shutdown_event_tx_clone.subscribe();
         loop {
             let client_server_map = client_server_map.clone();
 
-            let (socket, addr) = match listener.accept().await {
-                Ok((socket, addr)) => (socket, addr),
-                Err(err) => {
-                    error!("{:?}", err);
-                    continue;
+            // Listen for shutdown event and client connection at the same time
+            let (socket, addr) = tokio::select! {
+                _ = listener_shutdown_event_rx.recv() => {
+                    // Exits client connection loop which drops listener, listener_shutdown_event_rx and shutdown_event_tx_clone
+                    break;
+                }
+
+                listener_response = listener.accept() => {
+                    match listener_response {
+                        Ok((socket, addr)) => (socket, addr),
+                        Err(err) => {
+                            error!("{:?}", err);
+                            continue;
+                        }
+                    }
                 }
             };
 
+            // Used to signal shutdown
+            let client_shutdown_handler_rx = shutdown_event_tx_clone.subscribe();
+
+            // Used to signal that the task has completed
+            let dummy_tx = shutdown_event_tx_clone.clone();
+
             // Handle client.
             tokio::task::spawn(async move {
                 let start = chrono::offset::Utc::now().naive_utc();
 
-                match client::client_entrypoint(socket, client_server_map).await {
+                match client::client_entrypoint(
+                    socket,
+                    client_server_map,
+                    client_shutdown_handler_rx,
+                )
+                .await
+                {
                     Ok(_) => {
                         let duration = chrono::offset::Utc::now().naive_utc() - start;
 
@@ -171,6 +199,8 @@ async fn main() {
                         debug!("Client disconnected with error {:?}", err);
                     }
                 };
+                // Drop this transmitter so receiver knows that the task is completed
+                drop(dummy_tx);
             });
         }
     });
@@ -214,13 +244,41 @@ async fn main() {
         });
     }
 
-    // Exit on Ctrl-C (SIGINT) and SIGTERM.
     let mut term_signal = unix_signal(SignalKind::terminate()).unwrap();
+    let mut interrupt_signal = unix_signal(SignalKind::interrupt()).unwrap();
 
     tokio::select! {
-        _ = signal::ctrl_c() => (),
+        // Initiate graceful shutdown sequence on sig int
+        _ = interrupt_signal.recv() => {
+            info!("Got SIGINT, waiting for client connection drain now");
+
+            // Broadcast that client tasks need to finish
+            shutdown_event_tx.send(()).unwrap();
+            // Closes transmitter
+            drop(shutdown_event_tx);
+
+            // This is in a loop because the first event that the receiver receives will be the shutdown event
+            // This is not what we are waiting for instead, we want the receiver to send an error once all senders are closed which is reached after the shutdown event is received
+            loop {
+                match tokio::time::timeout(
+                    tokio::time::Duration::from_millis(config.general.shutdown_timeout),
+                    shutdown_event_rx.recv(),
+                )
+                .await
+                {
+                    Ok(res) => match res {
+                        Ok(_) => {}
+                        Err(_) => break,
+                    },
+                    Err(_) => {
+                        info!("Timed out while waiting for clients to shutdown");
+                        break;
+                    }
+                }
+            }
+        },
         _ = term_signal.recv() => (),
-    };
+    }
 
     info!("Shutting down...");
 }