checkpoint

* Add Discord link

* move it up

* :)

* hmm

* hmm

* :O

.circleci/config.yml

@@ -15,14 +15,34 @@ jobs:
           RUSTFLAGS: "-C instrument-coverage"
           LLVM_PROFILE_FILE: "pgcat-%m.profraw"
       - image: postgres:14
-        # auth:
-        #   username: mydockerhub-user
-        #   password: $DOCKERHUB_PASSWORD
+        command: ["postgres", "-p", "5432", "-c", "shared_preload_libraries=pg_stat_statements"]
         environment:
           POSTGRES_USER: postgres
           POSTGRES_DB: postgres
           POSTGRES_PASSWORD: postgres
           POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+      - image: postgres:14
+        command: ["postgres", "-p", "7432", "-c", "shared_preload_libraries=pg_stat_statements"]
+        environment:
+          POSTGRES_USER: postgres
+          POSTGRES_DB: postgres
+          POSTGRES_PASSWORD: postgres
+          POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+      - image: postgres:14
+        command: ["postgres", "-p", "8432", "-c", "shared_preload_libraries=pg_stat_statements"]
+        environment:
+          POSTGRES_USER: postgres
+          POSTGRES_DB: postgres
+          POSTGRES_PASSWORD: postgres
+          POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+      - image: postgres:14
+        command: ["postgres", "-p", "9432", "-c", "shared_preload_libraries=pg_stat_statements"]
+        environment:
+          POSTGRES_USER: postgres
+          POSTGRES_DB: postgres
+          POSTGRES_PASSWORD: postgres
+          POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+
     # Add steps to the job
     # See: https://circleci.com/docs/2.0/configuration-reference/#steps
     steps:

.circleci/run_tests.sh

@@ -3,6 +3,9 @@
 set -e
 set -o xtrace
 
+# non-zero exit code if we provide bad configs
+(! ./target/debug/pgcat "fake_configs"  2>/dev/null)
+
 # Start PgCat with a particular log level
 # for inspection.
 function start_pgcat() {
@@ -13,17 +16,20 @@ function start_pgcat() {
 
 # Setup the database with shards and user
 PGPASSWORD=postgres psql -e -h 127.0.0.1 -p 5432 -U postgres -f tests/sharding/query_routing_setup.sql
+PGPASSWORD=postgres psql -e -h 127.0.0.1 -p 7432 -U postgres -f tests/sharding/query_routing_setup.sql
+PGPASSWORD=postgres psql -e -h 127.0.0.1 -p 8432 -U postgres -f tests/sharding/query_routing_setup.sql
+PGPASSWORD=postgres psql -e -h 127.0.0.1 -p 9432 -U postgres -f tests/sharding/query_routing_setup.sql
 
 PGPASSWORD=sharding_user pgbench -h 127.0.0.1 -U sharding_user shard0 -i
 PGPASSWORD=sharding_user pgbench -h 127.0.0.1 -U sharding_user shard1 -i
 PGPASSWORD=sharding_user pgbench -h 127.0.0.1 -U sharding_user shard2 -i
 
 # Install Toxiproxy to simulate a downed/slow database
-wget -O toxiproxy-2.1.4.deb https://github.com/Shopify/toxiproxy/releases/download/v2.1.4/toxiproxy_2.1.4_amd64.deb
-sudo dpkg -i toxiproxy-2.1.4.deb
+wget -O toxiproxy-2.4.0.deb https://github.com/Shopify/toxiproxy/releases/download/v2.4.0/toxiproxy_2.4.0_linux_$(dpkg --print-architecture).deb
+sudo dpkg -i toxiproxy-2.4.0.deb
 
 # Start Toxiproxy
-toxiproxy-server &
+LOG_LEVEL=error toxiproxy-server &
 sleep 1
 
 # Create a database at port 5433, forward it to Postgres
@@ -84,7 +90,8 @@ kill -SIGHUP $(pgrep pgcat) # Reload config again
 cd tests/ruby
 sudo gem install bundler
 bundle install
-ruby tests.rb
+bundle exec ruby tests.rb || exit 1
+bundle exec rspec *_spec.rb || exit 1
 cd ../..
 
 #
@@ -92,7 +99,7 @@ cd ../..
 # These tests will start and stop the pgcat server so it will need to be restarted after the tests
 #
 pip3 install -r tests/python/requirements.txt
-python3 tests/python/tests.py
+python3 tests/python/tests.py || exit 1
 
 start_pgcat "info"
 
@@ -102,9 +109,9 @@ psql -U admin_user -e -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW STATS' > /dev/n
 psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'RELOAD' > /dev/null
 psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW CONFIG' > /dev/null
 psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW DATABASES' > /dev/null
-psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW LISTS' > /dev/null
-psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW POOLS' > /dev/null
-psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c 'SHOW VERSION' > /dev/null
+psql -U admin_user -h 127.0.0.1 -p 6432 -d pgcat -c 'SHOW LISTS' > /dev/null
+psql -U admin_user -h 127.0.0.1 -p 6432 -d pgcat -c 'SHOW POOLS' > /dev/null
+psql -U admin_user -h 127.0.0.1 -p 6432 -d pgcat -c 'SHOW VERSION' > /dev/null
 psql -U admin_user -h 127.0.0.1 -p 6432 -d pgbouncer -c "SET client_encoding TO 'utf8'" > /dev/null # will ignore
 (! psql -U admin_user -e -h 127.0.0.1 -p 6432 -d random_db -c 'SHOW STATS' > /dev/null)
 export PGPASSWORD=sharding_user
@@ -129,11 +136,14 @@ toxiproxy-cli toxic remove --toxicName latency_downstream postgres_replica
 start_pgcat "info"
 
 # Test session mode (and config reload)
-sed -i 's/pool_mode = "transaction"/pool_mode = "session"/' .circleci/pgcat.toml
+sed -i '0,/simple_db/s/pool_mode = "transaction"/pool_mode = "session"/' .circleci/pgcat.toml
 
 # Reload config test
 kill -SIGHUP $(pgrep pgcat)
 
+# Revert settings after reload. Makes test runs idempotent
+sed -i '0,/simple_db/s/pool_mode = "session"/pool_mode = "transaction"/' .circleci/pgcat.toml
+
 sleep 1
 
 # Prepared statements that will only work in session mode

Cargo.lock

@@ -159,6 +159,12 @@ dependencies = [
  "termcolor",
 ]
 
+[[package]]
+name = "exitcode"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "de853764b47027c2e862a995c34978ffa63c1501f2e15f987ba11bd4f9bba193"
+
 [[package]]
 name = "fnv"
 version = "1.0.7"
@@ -515,6 +521,7 @@ dependencies = [
  "bytes",
  "chrono",
  "env_logger",
+ "exitcode",
  "hmac",
  "hyper",
  "log",

Cargo.toml

@@ -33,3 +33,4 @@ tokio-rustls = "0.23"
 rustls-pemfile = "1"
 hyper = { version = "0.14", features = ["full"] }
 phf = { version = "0.10", features = ["macros"] }
+exitcode = "1.1.2"

README.md

@@ -1,8 +1,11 @@
-# PgCat
+![PgCat](./pgcat3.png)
+
+##### PgCat: PostgreSQL at petabyte scale
 
 [![CircleCI](https://circleci.com/gh/levkk/pgcat/tree/main.svg?style=svg)](https://circleci.com/gh/levkk/pgcat/tree/main)
-
-![PgCat](./pgcat3.png)
+<a href="https://discord.gg/DmyJP3qJ7U" target="_blank">
+    <img src="https://img.shields.io/discord/1013868243036930099" alt="Join our Discord!" />
+</a>
 
 PostgreSQL pooler (like PgBouncer) with sharding, load balancing and failover support.
 
@@ -15,7 +18,7 @@ PostgreSQL pooler (like PgBouncer) with sharding, load balancing and failover su
 | Session pooling                | :white_check_mark:          | Identical to PgBouncer.                                                                                                                               |
 | `COPY` support                 | :white_check_mark:          | Both `COPY TO` and `COPY FROM` are supported.                                                                                                         |
 | Query cancellation             | :white_check_mark:          | Supported both in transaction and session pooling modes.                                                                                              |
-| Load balancing of read queries | :white_check_mark:          | Using round-robin between replicas. Primary is included when `primary_reads_enabled` is enabled (default).                                            |
+| Load balancing of read queries | :white_check_mark:          | Using random between replicas. Primary is included when `primary_reads_enabled` is enabled (default).                                            |
 | Sharding                       | :white_check_mark:          | Transactions are sharded using `SET SHARD TO` and `SET SHARDING KEY TO` syntax extensions; see examples below.                                        |
 | Failover                       | :white_check_mark:          | Replicas are tested with a health check. If a health check fails, remaining replicas are attempted; see below for algorithm description and examples. |
 | Statistics                     | :white_check_mark:          | Statistics available in the admin database (`pgcat` and `pgbouncer`) with `SHOW STATS`, `SHOW POOLS` and others.                                      |
@@ -63,7 +66,7 @@ psql -h 127.0.0.1 -p 6432 -c 'SELECT 1'
 | `database`                   | The name of the database to connect to. This is the same on all servers that are part of one shard.                                        |                                  |
 |                              |                                                                                                                                            |                                  |
 | **`query_router`**           |                                                                                                                                            |                                  |
-| `default_role`               | Traffic is routed to this role by default (round-robin), unless the client specifies otherwise. Default is `any`, for any role available.  | `any`, `primary`, `replica`      |
+| `default_role`               | Traffic is routed to this role by default (random), unless the client specifies otherwise. Default is `any`, for any role available.  | `any`, `primary`, `replica`      |
 | `query_parser_enabled`       | Enable the query parser which will inspect incoming queries and route them to a primary or replicas.                                       | `false`                          |
 | `primary_reads_enabled`      | Enable this to allow read queries on the primary; otherwise read queries are routed to the replicas.                                       | `true`                           |
 
@@ -87,6 +90,14 @@ pgbench -t 1000 -p 6432 -h 127.0.0.1 --protocol extended
 
 See [sharding README](./tests/sharding/README.md) for sharding logic testing.
 
+Run `cargo test` to run Rust tests.
+
+Run the following commands to run Integration tests locally.
+```
+cd tests/docker/
+docker compose up --exit-code-from main # This will also produce coverage report under ./cov/
+```
+
 | **Feature**           | **Tested in CI**   | **Tested manually** | **Comments**                                                                                                             |
 |-----------------------|--------------------|---------------------|--------------------------------------------------------------------------------------------------------------------------|
 | Transaction pooling   | :white_check_mark: | :white_check_mark:  | Used by default for all tests.                                                                                           |
@@ -112,7 +123,7 @@ In transaction mode, a client talks to one server for the duration of a single t
 This mode is enabled by default.
 
 ### Load balancing of read queries
-All queries are load balanced against the configured servers using the round-robin algorithm. The most straight forward configuration example would be to put this pooler in front of several replicas and let it load balance all queries.
+All queries are load balanced against the configured servers using the random algorithm. The most straight forward configuration example would be to put this pooler in front of several replicas and let it load balance all queries.
 
 If the configuration includes a primary and replicas, the queries can be separated with the built-in query parser. The query parser will interpret the query and route all `SELECT` queries to a replica, while all other queries including explicit transactions will be routed to the primary.
 
@@ -151,18 +162,18 @@ Failover behavior can get pretty interesting (read complex) when multiple config
 
 | **Query**                 | **`SET SERVER ROLE TO`** | **`query_parser_enabled`** | **`primary_reads_enabled`** | **Target state** | **Outcome**                                                                                                                                                          |
 |---------------------------|--------------------------|----------------------------|-----------------------------|------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| Read query, i.e. `SELECT` | unset (any)              | false                      | false                       | up               | Query is routed to the first instance in the round-robin loop.                                                                                                       |
-| Read query                | unset (any)              | true                       | false                       | up               | Query is routed to the first replica instance in the round-robin loop.                                                                                               |
-| Read query                | unset (any)              | true                       | true                        | up               | Query is routed to the first instance in the round-robin loop.                                                                                                       |
-| Read query                | replica                  | false                      | false                       | up               | Query is routed to the first replica instance in the round-robin loop.                                                                                               |
+| Read query, i.e. `SELECT` | unset (any)              | false                      | false                       | up               | Query is routed to the first instance in the random loop.                                                                                                       |
+| Read query                | unset (any)              | true                       | false                       | up               | Query is routed to the first replica instance in the random loop.                                                                                               |
+| Read query                | unset (any)              | true                       | true                        | up               | Query is routed to the first instance in the random loop.                                                                                                       |
+| Read query                | replica                  | false                      | false                       | up               | Query is routed to the first replica instance in the random loop.                                                                                               |
 | Read query                | primary                  | false                      | false                       | up               | Query is routed to the primary.                                                                                                                                      |
-| Read query                | unset (any)              | false                      | false                       | down             | First instance is banned for reads. Next target in the round-robin loop is attempted.                                                                                |
-| Read query                | unset (any)              | true                       | false                       | down             | First replica instance is banned. Next replica instance is attempted in the round-robin loop.                                                                        |
-| Read query                | unset (any)              | true                       | true                        | down             | First instance (even if primary) is banned for reads. Next instance is attempted in the round-robin loop.                                                            |
-| Read query                | replica                  | false                      | false                       | down             | First replica instance is banned. Next replica instance is attempted in the round-robin loop.                                                                        |
+| Read query                | unset (any)              | false                      | false                       | down             | First instance is banned for reads. Next target in the random loop is attempted.                                                                                |
+| Read query                | unset (any)              | true                       | false                       | down             | First replica instance is banned. Next replica instance is attempted in the random loop.                                                                        |
+| Read query                | unset (any)              | true                       | true                        | down             | First instance (even if primary) is banned for reads. Next instance is attempted in the random loop.                                                            |
+| Read query                | replica                  | false                      | false                       | down             | First replica instance is banned. Next replica instance is attempted in the random loop.                                                                        |
 | Read query                | primary                  | false                      | false                       | down             | The query is attempted against the primary and fails. The client receives an error.                                                                                  |
 |                           |                          |                            |                             |                  |                                                                                                                                                                      |
-| Write query e.g. `INSERT` | unset (any)              | false                      | false                       | up               | The query is attempted against the first available instance in the round-robin loop. If the instance is a replica, the query fails and the client receives an error. |
+| Write query e.g. `INSERT` | unset (any)              | false                      | false                       | up               | The query is attempted against the first available instance in the random loop. If the instance is a replica, the query fails and the client receives an error. |
 | Write query               | unset (any)              | true                       | false                       | up               | The query is routed to the primary.                                                                                                                                  |
 | Write query               | unset (any)              | true                       | true                        | up               | The query is routed to the primary.                                                                                                                                  |
 | Write query               | primary                  | false                      | false                       | up               | The query is routed to the primary.                                                                                                                                  |
@@ -447,7 +458,7 @@ Always good to have a base line.
 
 ```
 $ pgbench -t 1000 -c 16 -j 2 -p 5432 -h 127.0.0.1 -S --protocol extended shard0
-Password: 
+Password:
 starting vacuum...end.
 transaction type: <builtin: select only>
 scaling factor: 1
@@ -461,7 +472,7 @@ tps = 139443.955722 (including connections establishing)
 tps = 142314.859075 (excluding connections establishing)
 
 $ pgbench -t 1000 -c 32 -j 2 -p 5432 -h 127.0.0.1 -S --protocol extended shard0
-Password: 
+Password:
 starting vacuum...end.
 transaction type: <builtin: select only>
 scaling factor: 1
@@ -475,7 +486,7 @@ tps = 150644.840891 (including connections establishing)
 tps = 152218.499430 (excluding connections establishing)
 
 $ pgbench -t 1000 -c 64 -j 2 -p 5432 -h 127.0.0.1 -S --protocol extended shard0
-Password: 
+Password:
 starting vacuum...end.
 transaction type: <builtin: select only>
 scaling factor: 1
@@ -489,7 +500,7 @@ tps = 152517.663404 (including connections establishing)
 tps = 153319.188482 (excluding connections establishing)
 
 $ pgbench -t 1000 -c 128 -j 2 -p 5432 -h 127.0.0.1 -S --protocol extended shard0
-Password: 
+Password:
 starting vacuum...end.
 transaction type: <builtin: select only>
 scaling factor: 1

src/admin.rs

@@ -265,11 +265,11 @@ where
     for (_, pool) in get_all_pools() {
         let pool_config = pool.settings.clone();
         for shard in 0..pool.shards() {
-            let database_name = &pool_config.shards[&shard.to_string()].database;
+            let database_name = &pool.address(shard, 0).database;
             for server in 0..pool.servers(shard) {
                 let address = pool.address(shard, server);
                 let pool_state = pool.pool_state(shard, server);
-                let banned = pool.is_banned(address, shard, Some(address.role));
+                let banned = pool.is_banned(address, Some(address.role));
 
                 res.put(data_row(&vec![
                     address.name(),                         // name

src/client.rs

@@ -5,13 +5,14 @@ use std::collections::HashMap;
 use tokio::io::{split, AsyncReadExt, BufReader, ReadHalf, WriteHalf};
 use tokio::net::TcpStream;
 use tokio::sync::broadcast::Receiver;
+use tokio::sync::mpsc::Sender;
 
 use crate::admin::{generate_server_info_for_admin, handle_admin};
 use crate::config::{get_config, Address};
 use crate::constants::*;
 use crate::errors::Error;
 use crate::messages::*;
-use crate::pool::{get_pool, ClientServerMap, ConnectionPool};
+use crate::pool::{get_pool, ClientServerMap, ConnectionPool, PoolMode};
 use crate::query_router::{Command, QueryRouter};
 use crate::server::Server;
 use crate::stats::{get_reporter, Reporter};
@@ -73,21 +74,32 @@ pub struct Client<S, T> {
     /// Last server process id we talked to.
     last_server_id: Option<i32>,
 
+    /// Connected to server
+    connected_to_server: bool,
+
     /// Name of the server pool for this client (This comes from the database name in the connection string)
-    target_pool_name: String,
+    pool_name: String,
 
     /// Postgres user for this client (This comes from the user in the connection string)
-    target_user_name: String,
+    username: String,
+
+    /// Application name for this client (defaults to pgcat)
+    application_name: String,
 
     /// Used to notify clients about an impending shutdown
-    shutdown_event_receiver: Receiver<()>,
+    shutdown: Receiver<()>,
+
+    // Sharding key column position
+    sharding_key_column: Option<usize>,
 }
 
 /// Client entrypoint.
 pub async fn client_entrypoint(
     mut stream: TcpStream,
     client_server_map: ClientServerMap,
-    shutdown_event_receiver: Receiver<()>,
+    shutdown: Receiver<()>,
+    drain: Sender<i32>,
+    admin_only: bool,
 ) -> Result<(), Error> {
     // Figure out if the client wants TLS or not.
     let addr = stream.peer_addr().unwrap();
@@ -106,11 +118,21 @@ pub async fn client_entrypoint(
                 write_all(&mut stream, yes).await?;
 
                 // Negotiate TLS.
-                match startup_tls(stream, client_server_map, shutdown_event_receiver).await {
+                match startup_tls(stream, client_server_map, shutdown, admin_only).await {
                     Ok(mut client) => {
                         info!("Client {:?} connected (TLS)", addr);
 
-                        client.handle().await
+                        if !client.is_admin() {
+                            let _ = drain.send(1).await;
+                        }
+
+                        let result = client.handle().await;
+
+                        if !client.is_admin() {
+                            let _ = drain.send(-1).await;
+                        }
+
+                        result
                     }
                     Err(err) => Err(err),
                 }
@@ -136,14 +158,25 @@ pub async fn client_entrypoint(
                             addr,
                             bytes,
                             client_server_map,
-                            shutdown_event_receiver,
+                            shutdown,
+                            admin_only,
                         )
                         .await
                         {
                             Ok(mut client) => {
                                 info!("Client {:?} connected (plain)", addr);
 
-                                client.handle().await
+                                if !client.is_admin() {
+                                    let _ = drain.send(1).await;
+                                }
+
+                                let result = client.handle().await;
+
+                                if !client.is_admin() {
+                                    let _ = drain.send(-1).await;
+                                }
+
+                                result
                             }
                             Err(err) => Err(err),
                         }
@@ -166,14 +199,25 @@ pub async fn client_entrypoint(
                 addr,
                 bytes,
                 client_server_map,
-                shutdown_event_receiver,
+                shutdown,
+                admin_only,
             )
             .await
             {
                 Ok(mut client) => {
                     info!("Client {:?} connected (plain)", addr);
 
-                    client.handle().await
+                    if !client.is_admin() {
+                        let _ = drain.send(1).await;
+                    }
+
+                    let result = client.handle().await;
+
+                    if !client.is_admin() {
+                        let _ = drain.send(-1).await;
+                    }
+
+                    result
                 }
                 Err(err) => Err(err),
             }
@@ -184,20 +228,21 @@ 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,
-                shutdown_event_receiver,
-            )
-            .await
-            {
+            match Client::cancel(read, write, addr, bytes, client_server_map, shutdown).await {
                 Ok(mut client) => {
                     info!("Client {:?} issued a cancel query request", addr);
 
-                    client.handle().await
+                    if !client.is_admin() {
+                        let _ = drain.send(1).await;
+                    }
+
+                    let result = client.handle().await;
+
+                    if !client.is_admin() {
+                        let _ = drain.send(-1).await;
+                    }
+
+                    result
                 }
 
                 Err(err) => Err(err),
@@ -250,7 +295,8 @@ where
 pub async fn startup_tls(
     stream: TcpStream,
     client_server_map: ClientServerMap,
-    shutdown_event_receiver: Receiver<()>,
+    shutdown: Receiver<()>,
+    admin_only: bool,
 ) -> Result<Client<ReadHalf<TlsStream<TcpStream>>, WriteHalf<TlsStream<TcpStream>>>, Error> {
     // Negotiate TLS.
     let tls = Tls::new()?;
@@ -280,7 +326,8 @@ pub async fn startup_tls(
                 addr,
                 bytes,
                 client_server_map,
-                shutdown_event_receiver,
+                shutdown,
+                admin_only,
             )
             .await
         }
@@ -295,6 +342,10 @@ where
     S: tokio::io::AsyncRead + std::marker::Unpin,
     T: tokio::io::AsyncWrite + std::marker::Unpin,
 {
+    pub fn is_admin(&self) -> bool {
+        self.admin
+    }
+
     /// Handle Postgres client startup after TLS negotiation is complete
     /// or over plain text.
     pub async fn startup(
@@ -303,29 +354,49 @@ where
         addr: std::net::SocketAddr,
         bytes: BytesMut, // The rest of the startup message.
         client_server_map: ClientServerMap,
-        shutdown_event_receiver: Receiver<()>,
+        shutdown: Receiver<()>,
+        admin_only: bool,
     ) -> Result<Client<S, T>, Error> {
         let config = get_config();
         let stats = get_reporter();
-
-        trace!("Got StartupMessage");
         let parameters = parse_startup(bytes.clone())?;
-        let target_pool_name = match parameters.get("database") {
+
+        // These two parameters are mandatory by the protocol.
+        let pool_name = match parameters.get("database") {
             Some(db) => db,
             None => return Err(Error::ClientError),
         };
 
-        let target_user_name = match parameters.get("user") {
+        let username = match parameters.get("user") {
             Some(user) => user,
             None => return Err(Error::ClientError),
         };
 
+        let application_name = match parameters.get("application_name") {
+            Some(application_name) => application_name,
+            None => "pgcat",
+        };
+
         let admin = ["pgcat", "pgbouncer"]
             .iter()
-            .filter(|db| *db == &target_pool_name)
+            .filter(|db| *db == &pool_name)
             .count()
             == 1;
 
+        // Kick any client that's not admin while we're in admin-only mode.
+        if !admin && admin_only {
+            debug!(
+                "Rejecting non-admin connection to {} when in admin only mode",
+                pool_name
+            );
+            error_response_terminal(
+                &mut write,
+                &format!("terminating connection due to administrator command"),
+            )
+            .await?;
+            return Err(Error::ShuttingDown);
+        }
+
         // Generate random backend ID and secret key
         let process_id: i32 = rand::random();
         let secret_key: i32 = rand::random();
@@ -357,46 +428,55 @@ where
             Err(_) => return Err(Error::SocketError),
         };
 
+        // Authenticate admin user.
         let (transaction_mode, server_info) = if admin {
-            let correct_user = config.general.admin_username.as_str();
-            let correct_password = config.general.admin_password.as_str();
-
             // Compare server and client hashes.
-            let password_hash = md5_hash_password(correct_user, correct_password, &salt);
+            let password_hash = md5_hash_password(
+                &config.general.admin_username,
+                &config.general.admin_password,
+                &salt,
+            );
+
             if password_hash != password_response {
                 debug!("Password authentication failed");
-                wrong_password(&mut write, target_user_name).await?;
+                wrong_password(&mut write, username).await?;
+
                 return Err(Error::ClientError);
             }
 
             (false, generate_server_info_for_admin())
-        } else {
-            let target_pool = match get_pool(target_pool_name.clone(), target_user_name.clone()) {
+        }
+        // Authenticate normal user.
+        else {
+            let pool = match get_pool(pool_name.clone(), username.clone()) {
                 Some(pool) => pool,
                 None => {
                     error_response(
                         &mut write,
                         &format!(
                             "No pool configured for database: {:?}, user: {:?}",
-                            target_pool_name, target_user_name
+                            pool_name, username
                         ),
                     )
                     .await?;
+
                     return Err(Error::ClientError);
                 }
             };
-            let transaction_mode = target_pool.settings.pool_mode == "transaction";
-            let server_info = target_pool.server_info();
+
             // Compare server and client hashes.
-            let correct_password = target_pool.settings.user.password.as_str();
-            let password_hash = md5_hash_password(&target_user_name, correct_password, &salt);
+            let password_hash = md5_hash_password(&username, &pool.settings.user.password, &salt);
 
             if password_hash != password_response {
                 debug!("Password authentication failed");
-                wrong_password(&mut write, &target_user_name).await?;
+                wrong_password(&mut write, username).await?;
+
                 return Err(Error::ClientError);
             }
-            (transaction_mode, server_info)
+
+            let transaction_mode = pool.settings.pool_mode == PoolMode::Transaction;
+
+            (transaction_mode, pool.server_info())
         };
 
         debug!("Password authentication successful");
@@ -408,27 +488,27 @@ where
 
         trace!("Startup OK");
 
-        // Split the read and write streams
-        // so we can control buffering.
-
         return Ok(Client {
             read: BufReader::new(read),
             write: write,
             addr,
             buffer: BytesMut::with_capacity(8196),
             cancel_mode: false,
-            transaction_mode: transaction_mode,
-            process_id: process_id,
-            secret_key: secret_key,
-            client_server_map: client_server_map,
+            transaction_mode,
+            process_id,
+            secret_key,
+            client_server_map,
             parameters: parameters.clone(),
             stats: stats,
             admin: admin,
             last_address_id: None,
             last_server_id: None,
-            target_pool_name: target_pool_name.clone(),
-            target_user_name: target_user_name.clone(),
-            shutdown_event_receiver: shutdown_event_receiver,
+            pool_name: pool_name.clone(),
+            username: username.clone(),
+            application_name: application_name.to_string(),
+            shutdown,
+            connected_to_server: false,
+            sharding_key_column: None,
         });
     }
 
@@ -439,7 +519,7 @@ where
         addr: std::net::SocketAddr,
         mut bytes: BytesMut, // The rest of the startup message.
         client_server_map: ClientServerMap,
-        shutdown_event_receiver: Receiver<()>,
+        shutdown: Receiver<()>,
     ) -> Result<Client<S, T>, Error> {
         let process_id = bytes.get_i32();
         let secret_key = bytes.get_i32();
@@ -450,17 +530,20 @@ where
             buffer: BytesMut::with_capacity(8196),
             cancel_mode: true,
             transaction_mode: false,
-            process_id: process_id,
-            secret_key: secret_key,
-            client_server_map: client_server_map,
+            process_id,
+            secret_key,
+            client_server_map,
             parameters: HashMap::new(),
             stats: get_reporter(),
             admin: false,
             last_address_id: None,
             last_server_id: None,
-            target_pool_name: String::from("undefined"),
-            target_user_name: String::from("undefined"),
-            shutdown_event_receiver: shutdown_event_receiver,
+            pool_name: String::from("undefined"),
+            username: String::from("undefined"),
+            application_name: String::from("undefined"),
+            shutdown,
+            connected_to_server: false,
+            sharding_key_column: None,
         });
     }
 
@@ -481,7 +564,7 @@ where
                         process_id.clone(),
                         secret_key.clone(),
                         address.clone(),
-                        port.clone(),
+                        *port,
                     ),
 
                     // The client doesn't know / got the wrong server,
@@ -493,13 +576,12 @@ where
             // Opens a new separate connection to the server, sends the backend_id
             // and secret_key and then closes it for security reasons. No other interactions
             // take place.
-            return Ok(Server::cancel(&address, &port, process_id, secret_key).await?);
+            return Ok(Server::cancel(&address, port, process_id, secret_key).await?);
         }
 
         // The query router determines where the query is going to go,
         // e.g. primary, replica, which shard.
         let mut query_router = QueryRouter::new();
-        let mut round_robin = rand::random();
 
         // Our custom protocol loop.
         // We expect the client to either start a transaction with regular queries
@@ -517,17 +599,40 @@ where
             // SET SHARDING KEY TO 'bigint';
 
             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(())
+                _ = self.shutdown.recv() => {
+                    if !self.admin {
+                        error_response_terminal(
+                            &mut self.write,
+                            &format!("terminating connection due to administrator command")
+                        ).await?;
+                        return Ok(())
+                    }
+
+                    // Admin clients ignore shutdown.
+                    else {
+                        read_message(&mut self.read).await?
+                    }
                 },
                 message_result = read_message(&mut self.read) => message_result?
             };
 
-            // Avoid taking a server if the client just wants to disconnect.
-            if message[0] as char == 'X' {
-                debug!("Client disconnecting");
-                return Ok(());
+            match message[0] as char {
+                // Buffer extended protocol messages even if we do not have
+                // a server connection yet. Hopefully, when we get the S message
+                // we'll be able to allocate a connection. Also, clients do not expect
+                // the server to respond to these messages so even if we were not able to
+                // allocate a connection, we wouldn't be able to send back an error message
+                // to the client so we buffer them and defer the decision to error out or not
+                // to when we get the S message
+                'P' | 'B' | 'D' | 'E' => {
+                    self.buffer.put(&message[..]);
+                    continue;
+                }
+                'X' => {
+                    debug!("Client disconnecting");
+                    return Ok(());
+                }
+                _ => (),
             }
 
             // Handle admin database queries.
@@ -540,15 +645,14 @@ where
             // Get a pool instance referenced by the most up-to-date
             // pointer. This ensures we always read the latest config
             // when starting a query.
-            let pool = match get_pool(self.target_pool_name.clone(), self.target_user_name.clone())
-            {
+            let pool = match get_pool(self.pool_name.clone(), self.username.clone()) {
                 Some(pool) => pool,
                 None => {
                     error_response(
                         &mut self.write,
                         &format!(
                             "No pool configured for database: {:?}, user: {:?}",
-                            self.target_pool_name, self.target_user_name
+                            self.pool_name, self.username
                         ),
                     )
                     .await?;
@@ -625,18 +729,20 @@ where
                     show_response(&mut self.write, "primary reads", &value).await?;
                     continue;
                 }
+
+                // COPY .. SHARDING_KEY_COLUMN ..
+                Some((Command::StartShardedCopy, value)) => {
+                    custom_protocol_response_ok(&mut self.write, "SHARDED_COPY").await?;
+                    self.sharding_key_column = Some(value.parse::<usize>().unwrap());
+                    continue;
+                }
             };
 
             debug!("Waiting for connection from pool");
 
             // Grab a server from the pool.
             let connection = match pool
-                .get(
-                    query_router.shard(),
-                    query_router.role(),
-                    self.process_id,
-                    round_robin,
-                )
+                .get(query_router.shard(), query_router.role(), self.process_id)
                 .await
             {
                 Ok(conn) => {
@@ -644,9 +750,19 @@ where
                     conn
                 }
                 Err(err) => {
-                    error!("Could not get connection from pool: {:?}", err);
+                    // Client is attempting to get results from the server,
+                    // but we were unable to grab a connection from the pool
+                    // We'll send back an error message and clean the extended
+                    // protocol buffer
+                    if message[0] as char == 'S' {
+                        error!("Got Sync message but failed to get a connection from the pool");
+                        self.buffer.clear();
+                    }
                     error_response(&mut self.write, "could not get connection from the pool")
                         .await?;
+
+                    error!("Could not get connection from pool: {{ pool_name: {:?}, username: {:?}, shard: {:?}, role: \"{:?}\", error: \"{:?}\" }}",
+                    self.pool_name.clone(), self.username.clone(), query_router.shard(), query_router.role(), err);
                     continue;
                 }
             };
@@ -655,11 +771,10 @@ where
             let address = connection.1;
             let server = &mut *reference;
 
-            round_robin += 1;
-
             // Server is assigned to the client in case the client wants to
             // cancel a query later.
             server.claim(self.process_id, self.secret_key);
+            self.connected_to_server = true;
 
             // Update statistics.
             if let Some(last_address_id) = self.last_address_id {
@@ -677,13 +792,10 @@ where
                 server.address()
             );
 
-            // Set application_name if any.
             // TODO: investigate other parameters and set them too.
-            if self.parameters.contains_key("application_name") {
-                server
-                    .set_name(&self.parameters["application_name"])
-                    .await?;
-            }
+
+            // Set application_name.
+            server.set_name(&self.application_name).await?;
 
             // Transaction loop. Multiple queries can be issued by the client here.
             // The connection belongs to the client until the transaction is over,
@@ -692,7 +804,7 @@ where
             // If the client is in session mode, no more custom protocol
             // commands will be accepted.
             loop {
-                let mut message = if message.len() == 0 {
+                let message = if message.len() == 0 {
                     trace!("Waiting for message inside transaction or in session mode");
 
                     match read_message(&mut self.read).await {
@@ -700,12 +812,7 @@ where
                         Err(err) => {
                             // Client disconnected inside a transaction.
                             // Clean up the server and re-use it.
-                            // This prevents connection thrashing by bad clients.
-                            if server.in_transaction() {
-                                server.query("ROLLBACK").await?;
-                                server.query("DISCARD ALL").await?;
-                                server.set_name("pgcat").await?;
-                            }
+                            server.checkin_cleanup().await?;
 
                             return Err(err);
                         }
@@ -716,178 +823,174 @@ where
                     msg
                 };
 
-                // The message will be forwarded to the server intact. We still would like to
-                // parse it below to figure out what to do with it.
-                let original = message.clone();
+                match self.handle_message(&pool, server, &address, message).await? {
+                    Some(done) => if done { break; },
+                    None => return Ok(()),
+                };
 
-                let code = message.get_u8() as char;
-                let _len = message.get_i32() as usize;
-
-                trace!("Message: {}", code);
-
-                match code {
-                    // ReadyForQuery
-                    'Q' => {
-                        debug!("Sending query to server");
-
-                        self.send_and_receive_loop(
-                            code,
-                            original,
-                            server,
-                            &address,
-                            query_router.shard(),
-                            &pool,
-                        )
-                        .await?;
-
-                        if !server.in_transaction() {
-                            // Report transaction executed statistics.
-                            self.stats.transaction(self.process_id, address.id);
-
-                            // Release server back to the pool if we are in transaction mode.
-                            // If we are in session mode, we keep the server until the client disconnects.
-                            if self.transaction_mode {
-                                break;
-                            }
-                        }
-                    }
-
-                    // Terminate
-                    'X' => {
-                        // Client closing. Rollback and clean up
-                        // connection before releasing into the pool.
-                        // Pgbouncer closes the connection which leads to
-                        // connection thrashing when clients misbehave.
-                        if server.in_transaction() {
-                            server.query("ROLLBACK").await?;
-                            server.query("DISCARD ALL").await?;
-                            server.set_name("pgcat").await?;
-                        }
-
-                        self.release();
-
-                        return Ok(());
-                    }
-
-                    // Parse
-                    // The query with placeholders is here, e.g. `SELECT * FROM users WHERE email = $1 AND active = $2`.
-                    'P' => {
-                        self.buffer.put(&original[..]);
-                    }
-
-                    // Bind
-                    // The placeholder's replacements are here, e.g. 'user@email.com' and 'true'
-                    'B' => {
-                        self.buffer.put(&original[..]);
-                    }
-
-                    // Describe
-                    // Command a client can issue to describe a previously prepared named statement.
-                    'D' => {
-                        self.buffer.put(&original[..]);
-                    }
-
-                    // Execute
-                    // Execute a prepared statement prepared in `P` and bound in `B`.
-                    'E' => {
-                        self.buffer.put(&original[..]);
-                    }
-
-                    // Sync
-                    // Frontend (client) is asking for the query result now.
-                    'S' => {
-                        debug!("Sending query to server");
-
-                        self.buffer.put(&original[..]);
-
-                        self.send_and_receive_loop(
-                            code,
-                            self.buffer.clone(),
-                            server,
-                            &address,
-                            query_router.shard(),
-                            &pool,
-                        )
-                        .await?;
-
-                        self.buffer.clear();
-
-                        if !server.in_transaction() {
-                            self.stats.transaction(self.process_id, address.id);
-
-                            // Release server back to the pool if we are in transaction mode.
-                            // If we are in session mode, we keep the server until the client disconnects.
-                            if self.transaction_mode {
-                                break;
-                            }
-                        }
-                    }
-
-                    // CopyData
-                    'd' => {
-                        // Forward the data to the server,
-                        // don't buffer it since it can be rather large.
-                        self.send_server_message(
-                            server,
-                            original,
-                            &address,
-                            query_router.shard(),
-                            &pool,
-                        )
-                        .await?;
-                    }
-
-                    // CopyDone or CopyFail
-                    // Copy is done, successfully or not.
-                    'c' | 'f' => {
-                        self.send_server_message(
-                            server,
-                            original,
-                            &address,
-                            query_router.shard(),
-                            &pool,
-                        )
-                        .await?;
-
-                        let response = self
-                            .receive_server_message(server, &address, query_router.shard(), &pool)
-                            .await?;
-
-                        match write_all_half(&mut self.write, response).await {
-                            Ok(_) => (),
-                            Err(err) => {
-                                server.mark_bad();
-                                return Err(err);
-                            }
-                        };
-
-                        if !server.in_transaction() {
-                            self.stats.transaction(self.process_id, address.id);
-
-                            // Release server back to the pool if we are in transaction mode.
-                            // If we are in session mode, we keep the server until the client disconnects.
-                            if self.transaction_mode {
-                                break;
-                            }
-                        }
-                    }
-
-                    // Some unexpected message. We either did not implement the protocol correctly
-                    // or this is not a Postgres client we're talking to.
-                    _ => {
-                        error!("Unexpected code: {}", code);
-                    }
-                }
             }
 
             // The server is no longer bound to us, we can't cancel it's queries anymore.
             debug!("Releasing server back into the pool");
+            server.checkin_cleanup().await?;
             self.stats.server_idle(server.process_id(), address.id);
+            self.connected_to_server = false;
+
             self.release();
             self.stats.client_idle(self.process_id, address.id);
         }
     }
 
+    async fn handle_message(&mut self, pool: &ConnectionPool, server: &mut Server, address: &Address, mut message: BytesMut) -> Result<Option<bool>, Error> {
+        let original = message.clone();
+        let code = message.get_u8() as char;
+        let _len = message.get_i32() as usize;
+
+        trace!("Message: {}", code);
+
+        match code {
+            // ReadyForQuery
+            'Q' => {
+                debug!("Sending query to server");
+
+                self.send_and_receive_loop(code, original, server, &address, &pool)
+                    .await?;
+
+                if !server.in_transaction() {
+                    // Report transaction executed statistics.
+                    self.stats.transaction(self.process_id, address.id);
+
+                    // Release server back to the pool if we are in transaction mode.
+                    // If we are in session mode, we keep the server until the client disconnects.
+                    if self.transaction_mode {
+                        return Ok(Some(true));
+                    }
+                }
+            }
+
+            // Terminate
+            'X' => {
+                server.checkin_cleanup().await?;
+                self.release();
+
+                return Ok(None);
+            }
+
+            // Parse
+            // The query with placeholders is here, e.g. `SELECT * FROM users WHERE email = $1 AND active = $2`.
+            'P' => {
+                self.buffer.put(&original[..]);
+            }
+
+            // Bind
+            // The placeholder's replacements are here, e.g. 'user@email.com' and 'true'
+            'B' => {
+                self.buffer.put(&original[..]);
+            }
+
+            // Describe
+            // Command a client can issue to describe a previously prepared named statement.
+            'D' => {
+                self.buffer.put(&original[..]);
+            }
+
+            // Execute
+            // Execute a prepared statement prepared in `P` and bound in `B`.
+            'E' => {
+                self.buffer.put(&original[..]);
+            }
+
+            // Sync
+            // Frontend (client) is asking for the query result now.
+            'S' => {
+                debug!("Sending query to server");
+
+                self.buffer.put(&original[..]);
+
+                // Clone after freeze does not allocate
+                let first_message_code = (*self.buffer.get(0).unwrap_or(&0)) as char;
+
+                // Almost certainly true
+                if first_message_code == 'P' {
+                    // Message layout
+                    // P followed by 32 int followed by null-terminated statement name
+                    // So message code should be in offset 0 of the buffer, first character
+                    // in prepared statement name would be index 5
+                    let first_char_in_name = *self.buffer.get(5).unwrap_or(&0);
+                    if first_char_in_name != 0 {
+                        // This is a named prepared statement
+                        // Server connection state will need to be cleared at checkin
+                        server.mark_dirty();
+                    }
+                }
+
+                self.send_and_receive_loop(
+                    code,
+                    self.buffer.clone(),
+                    server,
+                    &address,
+                    &pool,
+                )
+                .await?;
+
+                self.buffer.clear();
+
+                if !server.in_transaction() {
+                    self.stats.transaction(self.process_id, address.id);
+
+                    // Release server back to the pool if we are in transaction mode.
+                    // If we are in session mode, we keep the server until the client disconnects.
+                    if self.transaction_mode {
+                        return Ok(Some(true));
+                    }
+                }
+            }
+
+            // CopyData
+            'd' => {
+                // Forward the data to the server,
+                // don't buffer it since it can be rather large.
+                self.send_server_message(server, original, &address, &pool)
+                    .await?;
+            }
+
+            // CopyDone or CopyFail
+            // Copy is done, successfully or not.
+            'c' | 'f' => {
+                self.send_server_message(server, original, &address, &pool)
+                    .await?;
+
+                let response = self.receive_server_message(server, &address, &pool).await?;
+
+                match write_all_half(&mut self.write, response).await {
+                    Ok(_) => (),
+                    Err(err) => {
+                        server.mark_bad();
+                        return Err(err);
+                    }
+                };
+
+                if !server.in_transaction() {
+                    self.stats.transaction(self.process_id, address.id);
+
+                    // Release server back to the pool if we are in transaction mode.
+                    // If we are in session mode, we keep the server until the client disconnects.
+                    if self.transaction_mode {
+                        return Ok(Some(true));
+                    }
+                }
+            }
+
+            // Some unexpected message. We either did not implement the protocol correctly
+            // or this is not a Postgres client we're talking to.
+            _ => {
+                error!("Unexpected code: {}", code);
+            }
+        };
+
+        Ok(Some(false))
+    }
+
     /// Release the server from the client: it can't cancel its queries anymore.
     pub fn release(&self) {
         let mut guard = self.client_server_map.lock();
@@ -900,20 +1003,17 @@ where
         message: BytesMut,
         server: &mut Server,
         address: &Address,
-        shard: usize,
         pool: &ConnectionPool,
     ) -> Result<(), Error> {
         debug!("Sending {} to server", code);
 
-        self.send_server_message(server, message, &address, shard, &pool)
+        self.send_server_message(server, message, &address, &pool)
             .await?;
 
         // Read all data the server has to offer, which can be multiple messages
         // buffered in 8196 bytes chunks.
         loop {
-            let response = self
-                .receive_server_message(server, &address, shard, &pool)
-                .await?;
+            let response = self.receive_server_message(server, &address, &pool).await?;
 
             match write_all_half(&mut self.write, response).await {
                 Ok(_) => (),
@@ -939,13 +1039,12 @@ where
         server: &mut Server,
         message: BytesMut,
         address: &Address,
-        shard: usize,
         pool: &ConnectionPool,
     ) -> Result<(), Error> {
         match server.send(message).await {
             Ok(_) => Ok(()),
             Err(err) => {
-                pool.ban(address, shard, self.process_id);
+                pool.ban(address, self.process_id);
                 Err(err)
             }
         }
@@ -955,7 +1054,6 @@ where
         &mut self,
         server: &mut Server,
         address: &Address,
-        shard: usize,
         pool: &ConnectionPool,
     ) -> Result<BytesMut, Error> {
         if pool.settings.user.statement_timeout > 0 {
@@ -968,7 +1066,7 @@ where
                 Ok(result) => match result {
                     Ok(message) => Ok(message),
                     Err(err) => {
-                        pool.ban(address, shard, self.process_id);
+                        pool.ban(address, self.process_id);
                         error_response_terminal(
                             &mut self.write,
                             &format!("error receiving data from server: {:?}", err),
@@ -983,7 +1081,7 @@ where
                         address, pool.settings.user.username
                     );
                     server.mark_bad();
-                    pool.ban(address, shard, self.process_id);
+                    pool.ban(address, self.process_id);
                     error_response_terminal(&mut self.write, "pool statement timeout").await?;
                     Err(Error::StatementTimeout)
                 }
@@ -992,7 +1090,7 @@ where
             match server.recv().await {
                 Ok(message) => Ok(message),
                 Err(err) => {
-                    pool.ban(address, shard, self.process_id);
+                    pool.ban(address, self.process_id);
                     error_response_terminal(
                         &mut self.write,
                         &format!("error receiving data from server: {:?}", err),
@@ -1010,15 +1108,16 @@ impl<S, T> Drop for Client<S, T> {
         let mut guard = self.client_server_map.lock();
         guard.remove(&(self.process_id, self.secret_key));
 
-        // Update statistics.
+        // Dirty shutdown
+        // TODO: refactor, this is not the best way to handle state management.
         if let Some(address_id) = self.last_address_id {
             self.stats.client_disconnecting(self.process_id, address_id);
 
-            if let Some(process_id) = self.last_server_id {
-                self.stats.server_idle(process_id, address_id);
+            if self.connected_to_server {
+                if let Some(process_id) = self.last_server_id {
+                    self.stats.server_idle(process_id, address_id);
+                }
             }
         }
-
-        // self.release();
     }
 }

src/config.rs

@@ -57,15 +57,35 @@ impl PartialEq<Role> for Option<Role> {
 /// Address identifying a PostgreSQL server uniquely.
 #[derive(Clone, PartialEq, Hash, std::cmp::Eq, Debug)]
 pub struct Address {
+    /// Unique ID per addressable Postgres server.
     pub id: usize,
+
+    /// Server host.
     pub host: String,
-    pub port: String,
+
+    /// Server port.
+    pub port: u16,
+
+    /// Shard number of this Postgres server.
     pub shard: usize,
+
+    /// The name of the Postgres database.
     pub database: String,
+
+    /// Server role: replica, primary.
     pub role: Role,
+
+    /// If it's a replica, number it for reference and failover.
     pub replica_number: usize,
+
+    /// Position of the server in the pool for failover.
+    pub address_index: usize,
+
+    /// The name of the user configured to use this pool.
     pub username: String,
-    pub poolname: String,
+
+    /// The name of this pool (i.e. database name visible to the client).
+    pub pool_name: String,
 }
 
 impl Default for Address {
@@ -73,13 +93,14 @@ impl Default for Address {
         Address {
             id: 0,
             host: String::from("127.0.0.1"),
-            port: String::from("5432"),
+            port: 5432,
             shard: 0,
+            address_index: 0,
             replica_number: 0,
             database: String::from("database"),
             role: Role::Replica,
             username: String::from("username"),
-            poolname: String::from("poolname"),
+            pool_name: String::from("pool_name"),
         }
     }
 }
@@ -88,11 +109,11 @@ impl Address {
     /// Address name (aka database) used in `SHOW STATS`, `SHOW DATABASES`, and `SHOW POOLS`.
     pub fn name(&self) -> String {
         match self.role {
-            Role::Primary => format!("{}_shard_{}_primary", self.poolname, self.shard),
+            Role::Primary => format!("{}_shard_{}_primary", self.pool_name, self.shard),
 
             Role::Replica => format!(
                 "{}_shard_{}_replica_{}",
-                self.poolname, self.shard, self.replica_number
+                self.pool_name, self.shard, self.replica_number
             ),
         }
     }
@@ -339,9 +360,9 @@ impl Config {
 
         for (pool_name, pool_config) in &self.pools {
             // TODO: Make this output prettier (maybe a table?)
-            info!("--- Settings for pool {} ---", pool_name);
             info!(
-                "Pool size from all users: {}",
+                "[pool: {}] Maximum user connections: {}",
+                pool_name,
                 pool_config
                     .users
                     .iter()
@@ -349,20 +370,39 @@ impl Config {
                     .sum::<u32>()
                     .to_string()
             );
-            info!("Pool mode: {}", pool_config.pool_mode);
-            info!("Sharding function: {}", pool_config.sharding_function);
-            info!("Primary reads: {}", pool_config.primary_reads_enabled);
-            info!("Query router: {}", pool_config.query_parser_enabled);
-
-            // TODO: Make this prettier.
-            info!("Number of shards: {}", pool_config.shards.len());
-            info!("Number of users: {}", pool_config.users.len());
+            info!("[pool: {}] Pool mode: {}", pool_name, pool_config.pool_mode);
+            info!(
+                "[pool: {}] Sharding function: {}",
+                pool_name, pool_config.sharding_function
+            );
+            info!(
+                "[pool: {}] Primary reads: {}",
+                pool_name, pool_config.primary_reads_enabled
+            );
+            info!(
+                "[pool: {}] Query router: {}",
+                pool_name, pool_config.query_parser_enabled
+            );
+            info!(
+                "[pool: {}] Number of shards: {}",
+                pool_name,
+                pool_config.shards.len()
+            );
+            info!(
+                "[pool: {}] Number of users: {}",
+                pool_name,
+                pool_config.users.len()
+            );
 
             for user in &pool_config.users {
                 info!(
-                    "{} pool size: {}, statement timeout: {}",
-                    user.1.username, user.1.pool_size, user.1.statement_timeout
+                    "[pool: {}][user: {}] Pool size: {}",
+                    pool_name, user.1.username, user.1.pool_size,
                 );
+                info!(
+                    "[pool: {}][user: {}] Statement timeout: {}",
+                    pool_name, user.1.username, user.1.statement_timeout
+                )
             }
         }
     }
@@ -460,6 +500,18 @@ pub async fn parse(path: &str) -> Result<(), Error> {
             }
         };
 
+        match pool.pool_mode.as_ref() {
+            "transaction" => (),
+            "session" => (),
+            other => {
+                error!(
+                    "pool_mode can be 'session' or 'transaction', got: '{}'",
+                    other
+                );
+                return Err(Error::BadConfig);
+            }
+        };
+
         for shard in &pool.shards {
             // We use addresses as unique identifiers,
             // let's make sure they are unique in the config as well.

src/errors.rs

@@ -12,4 +12,5 @@ pub enum Error {
     ClientError,
     TlsError,
     StatementTimeout,
+    ShuttingDown,
 }

src/main.rs

@@ -24,6 +24,7 @@ extern crate async_trait;
 extern crate bb8;
 extern crate bytes;
 extern crate env_logger;
+extern crate exitcode;
 extern crate log;
 extern crate md5;
 extern crate num_cpus;
@@ -66,18 +67,20 @@ mod stats;
 mod tls;
 
 use crate::config::{get_config, reload_config, VERSION};
+use crate::errors::Error;
 use crate::pool::{ClientServerMap, ConnectionPool};
 use crate::prometheus::start_metric_server;
 use crate::stats::{Collector, Reporter, REPORTER};
 
 #[tokio::main(worker_threads = 4)]
 async fn main() {
-    env_logger::init();
+    env_logger::builder().format_timestamp_micros().init();
+
     info!("Welcome to PgCat! Meow. (Version {})", VERSION);
 
     if !query_router::QueryRouter::setup() {
         error!("Could not setup query router");
-        return;
+        std::process::exit(exitcode::CONFIG);
     }
 
     let args = std::env::args().collect::<Vec<String>>();
@@ -92,7 +95,7 @@ async fn main() {
         Ok(_) => (),
         Err(err) => {
             error!("Config parse error: {:?}", err);
-            return;
+            std::process::exit(exitcode::CONFIG);
         }
     };
 
@@ -107,7 +110,7 @@ async fn main() {
             Ok(addr) => addr,
             Err(err) => {
                 error!("Invalid http address: {}", err);
-                return;
+                std::process::exit(exitcode::CONFIG);
             }
         };
         tokio::task::spawn(async move {
@@ -121,7 +124,7 @@ async fn main() {
         Ok(sock) => sock,
         Err(err) => {
             error!("Listener socket error: {:?}", err);
-            return;
+            std::process::exit(exitcode::CONFIG);
         }
     };
 
@@ -133,125 +136,33 @@ async fn main() {
     let client_server_map: ClientServerMap = Arc::new(Mutex::new(HashMap::new()));
 
     // Statistics reporting.
-    let (tx, rx) = mpsc::channel(100_000);
-    REPORTER.store(Arc::new(Reporter::new(tx.clone())));
+    let (stats_tx, stats_rx) = mpsc::channel(100_000);
+    REPORTER.store(Arc::new(Reporter::new(stats_tx.clone())));
 
     // Connection pool that allows to query all shards and replicas.
     match ConnectionPool::from_config(client_server_map.clone()).await {
         Ok(_) => (),
         Err(err) => {
             error!("Pool error: {:?}", err);
-            return;
+            std::process::exit(exitcode::CONFIG);
         }
     };
 
-    // Statistics collector task.
-    let collector_tx = tx.clone();
-
-    // Save these for reloading
-    let reload_client_server_map = client_server_map.clone();
-    let autoreload_client_server_map = client_server_map.clone();
-
     tokio::task::spawn(async move {
-        let mut stats_collector = Collector::new(rx, collector_tx);
+        let mut stats_collector = Collector::new(stats_rx, stats_tx.clone());
         stats_collector.collect().await;
     });
 
-    info!("Waiting for clients");
+    info!("Config autoreloader: {}", config.general.autoreload);
 
-    let (shutdown_event_tx, mut shutdown_event_rx) = broadcast::channel::<()>(1);
-
-    let shutdown_event_tx_clone = shutdown_event_tx.clone();
-
-    // Client connection loop.
+    let mut autoreload_interval = tokio::time::interval(tokio::time::Duration::from_millis(15_000));
+    let autoreload_client_server_map = client_server_map.clone();
     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();
+            autoreload_interval.tick().await;
+            if config.general.autoreload {
+                info!("Automatically reloading config");
 
-            // 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,
-                    client_shutdown_handler_rx,
-                )
-                .await
-                {
-                    Ok(_) => {
-                        let duration = chrono::offset::Utc::now().naive_utc() - start;
-
-                        info!(
-                            "Client {:?} disconnected, session duration: {}",
-                            addr,
-                            format_duration(&duration)
-                        );
-                    }
-
-                    Err(err) => {
-                        debug!("Client disconnected with error {:?}", err);
-                    }
-                };
-                // Drop this transmitter so receiver knows that the task is completed
-                drop(dummy_tx);
-            });
-        }
-    });
-
-    // Reload config:
-    // kill -SIGHUP $(pgrep pgcat)
-    tokio::task::spawn(async move {
-        let mut stream = unix_signal(SignalKind::hangup()).unwrap();
-
-        loop {
-            stream.recv().await;
-
-            info!("Reloading config");
-
-            match reload_config(reload_client_server_map.clone()).await {
-                Ok(_) => (),
-                Err(_) => continue,
-            };
-
-            get_config().show();
-        }
-    });
-
-    if config.general.autoreload {
-        let mut interval = tokio::time::interval(tokio::time::Duration::from_millis(15_000));
-
-        tokio::task::spawn(async move {
-            info!("Config autoreloader started");
-
-            loop {
-                interval.tick().await;
                 match reload_config(autoreload_client_server_map.clone()).await {
                     Ok(changed) => {
                         if changed {
@@ -261,43 +172,131 @@ async fn main() {
                     Err(_) => (),
                 };
             }
-        });
-    }
+        }
+    });
 
     let mut term_signal = unix_signal(SignalKind::terminate()).unwrap();
     let mut interrupt_signal = unix_signal(SignalKind::interrupt()).unwrap();
+    let mut sighup_signal = unix_signal(SignalKind::hangup()).unwrap();
+    let (shutdown_tx, _) = broadcast::channel::<()>(1);
+    let (drain_tx, mut drain_rx) = mpsc::channel::<i32>(2048);
+    let (exit_tx, mut exit_rx) = mpsc::channel::<()>(1);
 
-    tokio::select! {
-        // Initiate graceful shutdown sequence on sig int
-        _ = interrupt_signal.recv() => {
-            info!("Got SIGINT, waiting for client connection drain now");
+    info!("Waiting for clients");
 
-            // Broadcast that client tasks need to finish
-            shutdown_event_tx.send(()).unwrap();
-            // Closes transmitter
-            drop(shutdown_event_tx);
+    let mut admin_only = false;
+    let mut total_clients = 0;
 
-            // 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;
+    loop {
+        tokio::select! {
+            // Reload config:
+            // kill -SIGHUP $(pgrep pgcat)
+            _ = sighup_signal.recv() => {
+                info!("Reloading config");
+
+                match reload_config(client_server_map.clone()).await {
+                    Ok(_) => (),
+                    Err(_) => (),
+                };
+
+                get_config().show();
+            },
+
+            // Initiate graceful shutdown sequence on sig int
+            _ = interrupt_signal.recv() => {
+                info!("Got SIGINT, waiting for client connection drain now");
+                admin_only = true;
+
+                // Broadcast that client tasks need to finish
+                let _ = shutdown_tx.send(());
+                let exit_tx = exit_tx.clone();
+                let _ = drain_tx.send(0).await;
+
+                tokio::task::spawn(async move {
+                    let mut interval = tokio::time::interval(tokio::time::Duration::from_millis(config.general.shutdown_timeout));
+
+                    // First tick fires immediately.
+                    interval.tick().await;
+
+                    // Second one in the interval time.
+                    interval.tick().await;
+
+                    // We're done waiting.
+                    error!("Graceful shutdown timed out. {} active clients being closed", total_clients);
+
+                    let _ = exit_tx.send(()).await;
+                });
+            },
+
+            _ = term_signal.recv() => {
+                info!("Got SIGTERM, closing with {} clients active", total_clients);
+                break;
+            },
+
+            new_client = listener.accept() => {
+                let (socket, addr) = match new_client {
+                    Ok((socket, addr)) => (socket, addr),
+                    Err(err) => {
+                        error!("{:?}", err);
+                        continue;
                     }
+                };
+
+                let shutdown_rx = shutdown_tx.subscribe();
+                let drain_tx = drain_tx.clone();
+                let client_server_map = client_server_map.clone();
+
+                tokio::task::spawn(async move {
+                    let start = chrono::offset::Utc::now().naive_utc();
+
+                    match client::client_entrypoint(
+                        socket,
+                        client_server_map,
+                        shutdown_rx,
+                        drain_tx,
+                        admin_only,
+                    )
+                    .await
+                    {
+                        Ok(()) => {
+
+                            let duration = chrono::offset::Utc::now().naive_utc() - start;
+
+                            info!(
+                                "Client {:?} disconnected, session duration: {}",
+                                addr,
+                                format_duration(&duration)
+                            );
+                        }
+
+                        Err(err) => {
+                            match err {
+                                // Don't count the clients we rejected.
+                                Error::ShuttingDown => (),
+                                _ => {
+                                    // drain_tx.send(-1).await.unwrap();
+                                }
+                            }
+
+                            debug!("Client disconnected with error {:?}", err);
+                        }
+                    };
+                });
+            }
+
+            _ = exit_rx.recv() => {
+                break;
+            }
+
+            client_ping = drain_rx.recv() => {
+                let client_ping = client_ping.unwrap();
+                total_clients += client_ping;
+
+                if total_clients == 0 && admin_only {
+                    let _ = exit_tx.send(()).await;
                 }
             }
-        },
-        _ = term_signal.recv() => (),
+        }
     }
 
     info!("Shutting down...");
@@ -309,34 +308,18 @@ async fn main() {
 ///
 /// * `duration` - A duration of time
 fn format_duration(duration: &chrono::Duration) -> String {
-    let seconds = {
-        let seconds = duration.num_seconds() % 60;
-        if seconds < 10 {
-            format!("0{}", seconds)
-        } else {
-            format!("{}", seconds)
-        }
-    };
+    let milliseconds = format!("{:0>3}", duration.num_milliseconds() % 1000);
 
-    let minutes = {
-        let minutes = duration.num_minutes() % 60;
-        if minutes < 10 {
-            format!("0{}", minutes)
-        } else {
-            format!("{}", minutes)
-        }
-    };
+    let seconds = format!("{:0>2}", duration.num_seconds() % 60);
 
-    let hours = {
-        let hours = duration.num_hours() % 24;
-        if hours < 10 {
-            format!("0{}", hours)
-        } else {
-            format!("{}", hours)
-        }
-    };
+    let minutes = format!("{:0>2}", duration.num_minutes() % 60);
+
+    let hours = format!("{:0>2}", duration.num_hours() % 24);
 
     let days = duration.num_days().to_string();
 
-    format!("{}d {}:{}:{}", days, hours, minutes, seconds)
+    format!(
+        "{}d {}:{}:{}.{}",
+        days, hours, minutes, seconds, milliseconds
+    )
 }

src/pool.rs

@@ -6,44 +6,80 @@ use chrono::naive::NaiveDateTime;
 use log::{debug, error, info, warn};
 use once_cell::sync::Lazy;
 use parking_lot::{Mutex, RwLock};
+use rand::seq::SliceRandom;
+use rand::thread_rng;
 use std::collections::HashMap;
 use std::sync::Arc;
 use std::time::Instant;
 
-use crate::config::{get_config, Address, Role, Shard, User};
+use crate::config::{get_config, Address, Role, User};
 use crate::errors::Error;
 
 use crate::server::Server;
+use crate::sharding::ShardingFunction;
 use crate::stats::{get_reporter, Reporter};
 
 pub type BanList = Arc<RwLock<Vec<HashMap<Address, NaiveDateTime>>>>;
-pub type ClientServerMap = Arc<Mutex<HashMap<(i32, i32), (i32, i32, String, String)>>>;
+pub type ClientServerMap = Arc<Mutex<HashMap<(i32, i32), (i32, i32, String, u16)>>>;
 pub type PoolMap = HashMap<(String, String), ConnectionPool>;
 /// The connection pool, globally available.
 /// This is atomic and safe and read-optimized.
 /// The pool is recreated dynamically when the config is reloaded.
 pub static POOLS: Lazy<ArcSwap<PoolMap>> = Lazy::new(|| ArcSwap::from_pointee(HashMap::default()));
 
+/// Pool mode:
+/// - transaction: server serves one transaction,
+/// - session: server is attached to the client.
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub enum PoolMode {
+    Session,
+    Transaction,
+}
+
+impl std::fmt::Display for PoolMode {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match *self {
+            PoolMode::Session => write!(f, "session"),
+            PoolMode::Transaction => write!(f, "transaction"),
+        }
+    }
+}
+
+/// Pool settings.
 #[derive(Clone, Debug)]
 pub struct PoolSettings {
-    pub pool_mode: String,
-    pub shards: HashMap<String, Shard>,
+    /// Transaction or Session.
+    pub pool_mode: PoolMode,
+
+    // Number of shards.
+    pub shards: usize,
+
+    // Connecting user.
     pub user: User,
-    pub default_role: String,
+
+    // Default server role to connect to.
+    pub default_role: Option<Role>,
+
+    // Enable/disable query parser.
     pub query_parser_enabled: bool,
+
+    // Read from the primary as well or not.
     pub primary_reads_enabled: bool,
-    pub sharding_function: String,
+
+    // Sharding function.
+    pub sharding_function: ShardingFunction,
 }
+
 impl Default for PoolSettings {
     fn default() -> PoolSettings {
         PoolSettings {
-            pool_mode: String::from("transaction"),
-            shards: HashMap::from([(String::from("1"), Shard::default())]),
+            pool_mode: PoolMode::Transaction,
+            shards: 1,
             user: User::default(),
-            default_role: String::from("any"),
+            default_role: None,
             query_parser_enabled: false,
             primary_reads_enabled: true,
-            sharding_function: "pg_bigint_hash".to_string(),
+            sharding_function: ShardingFunction::PgBigintHash,
         }
     }
 }
@@ -71,6 +107,7 @@ pub struct ConnectionPool {
     /// on pool creation and save the K messages here.
     server_info: BytesMut,
 
+    /// Pool configuration.
     pub settings: PoolSettings,
 }
 
@@ -78,11 +115,13 @@ impl ConnectionPool {
     /// Construct the connection pool from the configuration.
     pub async fn from_config(client_server_map: ClientServerMap) -> Result<(), Error> {
         let config = get_config();
-        let mut new_pools = PoolMap::default();
 
+        let mut new_pools = HashMap::new();
         let mut address_id = 0;
+
         for (pool_name, pool_config) in &config.pools {
-            for (_user_index, user_info) in &pool_config.users {
+            // There is one pool per database/user pair.
+            for (_, user) in &pool_config.users {
                 let mut shards = Vec::new();
                 let mut addresses = Vec::new();
                 let mut banlist = Vec::new();
@@ -96,10 +135,11 @@ impl ConnectionPool {
                 // Sort by shard number to ensure consistency.
                 shard_ids.sort_by_key(|k| k.parse::<i64>().unwrap());
 
-                for shard_idx in shard_ids {
-                    let shard = &pool_config.shards[&shard_idx];
+                for shard_idx in &shard_ids {
+                    let shard = &pool_config.shards[shard_idx];
                     let mut pools = Vec::new();
                     let mut servers = Vec::new();
+                    let mut address_index = 0;
                     let mut replica_number = 0;
 
                     for server in shard.servers.iter() {
@@ -116,15 +156,17 @@ impl ConnectionPool {
                             id: address_id,
                             database: shard.database.clone(),
                             host: server.0.clone(),
-                            port: server.1.to_string(),
+                            port: server.1 as u16,
                             role: role,
+                            address_index,
                             replica_number,
                             shard: shard_idx.parse::<usize>().unwrap(),
-                            username: user_info.username.clone(),
-                            poolname: pool_name.clone(),
+                            username: user.username.clone(),
+                            pool_name: pool_name.clone(),
                         };
 
                         address_id += 1;
+                        address_index += 1;
 
                         if role == Role::Replica {
                             replica_number += 1;
@@ -132,14 +174,14 @@ impl ConnectionPool {
 
                         let manager = ServerPool::new(
                             address.clone(),
-                            user_info.clone(),
+                            user.clone(),
                             &shard.database,
                             client_server_map.clone(),
                             get_reporter(),
                         );
 
                         let pool = Pool::builder()
-                            .max_size(user_info.pool_size)
+                            .max_size(user.pool_size)
                             .connection_timeout(std::time::Duration::from_millis(
                                 config.general.connect_timeout,
                             ))
@@ -166,13 +208,27 @@ impl ConnectionPool {
                     stats: get_reporter(),
                     server_info: BytesMut::new(),
                     settings: PoolSettings {
-                        pool_mode: pool_config.pool_mode.clone(),
-                        shards: pool_config.shards.clone(),
-                        user: user_info.clone(),
-                        default_role: pool_config.default_role.clone(),
+                        pool_mode: match pool_config.pool_mode.as_str() {
+                            "transaction" => PoolMode::Transaction,
+                            "session" => PoolMode::Session,
+                            _ => unreachable!(),
+                        },
+                        // shards: pool_config.shards.clone(),
+                        shards: shard_ids.len(),
+                        user: user.clone(),
+                        default_role: match pool_config.default_role.as_str() {
+                            "any" => None,
+                            "replica" => Some(Role::Replica),
+                            "primary" => Some(Role::Primary),
+                            _ => unreachable!(),
+                        },
                         query_parser_enabled: pool_config.query_parser_enabled.clone(),
                         primary_reads_enabled: pool_config.primary_reads_enabled,
-                        sharding_function: pool_config.sharding_function.clone(),
+                        sharding_function: match pool_config.sharding_function.as_str() {
+                            "pg_bigint_hash" => ShardingFunction::PgBigintHash,
+                            "sha1" => ShardingFunction::Sha1,
+                            _ => unreachable!(),
+                        },
                     },
                 };
 
@@ -185,7 +241,9 @@ impl ConnectionPool {
                         return Err(err);
                     }
                 };
-                new_pools.insert((pool_name.clone(), user_info.username.clone()), pool);
+
+                // There is one pool per database/user pair.
+                new_pools.insert((pool_name.clone(), user.username.clone()), pool);
             }
         }
 
@@ -201,16 +259,9 @@ impl ConnectionPool {
     /// the pooler starts up.
     async fn validate(&mut self) -> Result<(), Error> {
         let mut server_infos = Vec::new();
-        let stats = self.stats.clone();
-
         for shard in 0..self.shards() {
-            let mut round_robin = 0;
-
-            for _ in 0..self.servers(shard) {
-                // To keep stats consistent.
-                let fake_process_id = 0;
-
-                let connection = match self.get(shard, None, fake_process_id, round_robin).await {
+            for server in 0..self.servers(shard) {
+                let connection = match self.databases[shard][server].get().await {
                     Ok(conn) => conn,
                     Err(err) => {
                         error!("Shard {} down or misconfigured: {:?}", shard, err);
@@ -218,25 +269,21 @@ impl ConnectionPool {
                     }
                 };
 
-                let proxy = connection.0;
-                let address = connection.1;
+                let proxy = connection;
                 let server = &*proxy;
                 let server_info = server.server_info();
 
-                stats.client_disconnecting(fake_process_id, address.id);
-
                 if server_infos.len() > 0 {
                     // Compare against the last server checked.
                     if server_info != server_infos[server_infos.len() - 1] {
                         warn!(
                             "{:?} has different server configuration than the last server",
-                            address
+                            proxy.address()
                         );
                     }
                 }
 
                 server_infos.push(server_info);
-                round_robin += 1;
             }
         }
 
@@ -246,6 +293,8 @@ impl ConnectionPool {
             return Err(Error::AllServersDown);
         }
 
+        // We're assuming all servers are identical.
+        // TODO: not true.
         self.server_info = server_infos[0].clone();
 
         Ok(())
@@ -254,58 +303,31 @@ impl ConnectionPool {
     /// Get a connection from the pool.
     pub async fn get(
         &self,
-        shard: usize,           // shard number
-        role: Option<Role>,     // primary or replica
-        process_id: i32,        // client id
-        mut round_robin: usize, // round robin offset
+        shard: usize,       // shard number
+        role: Option<Role>, // primary or replica
+        process_id: i32,    // client id
     ) -> Result<(PooledConnection<'_, ServerPool>, Address), Error> {
         let now = Instant::now();
-        let addresses = &self.addresses[shard];
+        let mut candidates: Vec<&Address> = self.addresses[shard]
+            .iter()
+            .filter(|address| address.role == role)
+            .collect();
 
-        let mut allowed_attempts = match role {
-            // Primary-specific queries get one attempt, if the primary is down,
-            // nothing we should do about it I think. It's dangerous to retry
-            // write queries.
-            Some(Role::Primary) => 1,
-
-            // Replicas get to try as many times as there are replicas
-            // and connections in the pool.
-            _ => addresses.len(),
-        };
-
-        debug!("Allowed attempts for {:?}: {}", role, allowed_attempts);
-
-        let exists = match role {
-            Some(role) => addresses.iter().filter(|addr| addr.role == role).count() > 0,
-            None => true,
-        };
-
-        if !exists {
-            error!("Requested role {:?}, but none are configured", role);
-            return Err(Error::BadConfig);
-        }
+        // Random load balancing
+        candidates.shuffle(&mut thread_rng());
 
         let healthcheck_timeout = get_config().general.healthcheck_timeout;
         let healthcheck_delay = get_config().general.healthcheck_delay as u128;
 
-        while allowed_attempts > 0 {
-            // Round-robin replicas.
-            round_robin += 1;
+        while !candidates.is_empty() {
+            // Get the next candidate
+            let address = match candidates.pop() {
+                Some(address) => address,
+                None => break,
+            };
 
-            let index = round_robin % addresses.len();
-            let address = &addresses[index];
-
-            // Make sure you're getting a primary or a replica
-            // as per request. If no specific role is requested, the first
-            // available will be chosen.
-            if address.role != role {
-                continue;
-            }
-
-            allowed_attempts -= 1;
-
-            // Don't attempt to connect to banned servers.
-            if self.is_banned(address, shard, role) {
+            if self.is_banned(&address, role) {
+                debug!("Address {:?} is banned", address);
                 continue;
             }
 
@@ -313,12 +335,14 @@ impl ConnectionPool {
             self.stats.client_waiting(process_id, address.id);
 
             // Check if we can connect
-            let mut conn = match self.databases[shard][index].get().await {
+            let mut conn = match self.databases[address.shard][address.address_index]
+                .get()
+                .await
+            {
                 Ok(conn) => conn,
                 Err(err) => {
-                    error!("Banning replica {}, error: {:?}", index, err);
-                    self.ban(address, shard, process_id);
-                    self.stats.client_disconnecting(process_id, address.id);
+                    error!("Banning instance {:?}, error: {:?}", address, err);
+                    self.ban(&address, process_id);
                     self.stats
                         .checkout_time(now.elapsed().as_micros(), process_id, address.id);
                     continue;
@@ -332,6 +356,9 @@ impl ConnectionPool {
             let require_healthcheck =
                 server.last_activity().elapsed().unwrap().as_millis() > healthcheck_delay;
 
+            // Do not issue a health check unless it's been a little while
+            // since we last checked the server is ok.
+            // Health checks are pretty expensive.
             if !require_healthcheck {
                 self.stats
                     .checkout_time(now.elapsed().as_micros(), process_id, address.id);
@@ -345,7 +372,7 @@ impl ConnectionPool {
 
             match tokio::time::timeout(
                 tokio::time::Duration::from_millis(healthcheck_timeout),
-                server.query(";"),
+                server.query(";"), // Cheap query (query parser not used in PG)
             )
             .await
             {
@@ -359,62 +386,67 @@ impl ConnectionPool {
                     }
 
                     // Health check failed.
-                    Err(_) => {
-                        error!("Banning replica {} because of failed health check", index);
+                    Err(err) => {
+                        error!(
+                            "Banning instance {:?} because of failed health check, {:?}",
+                            address, err
+                        );
 
                         // Don't leave a bad connection in the pool.
                         server.mark_bad();
 
-                        self.ban(address, shard, process_id);
+                        self.ban(&address, process_id);
                         continue;
                     }
                 },
 
                 // Health check timed out.
-                Err(_) => {
-                    error!("Banning replica {} because of health check timeout", index);
+                Err(err) => {
+                    error!(
+                        "Banning instance {:?} because of health check timeout, {:?}",
+                        address, err
+                    );
                     // Don't leave a bad connection in the pool.
                     server.mark_bad();
 
-                    self.ban(address, shard, process_id);
+                    self.ban(&address, process_id);
                     continue;
                 }
             }
         }
 
-        return Err(Error::AllServersDown);
+        Err(Error::AllServersDown)
     }
 
     /// Ban an address (i.e. replica). It no longer will serve
     /// traffic for any new transactions. Existing transactions on that replica
     /// will finish successfully or error out to the clients.
-    pub fn ban(&self, address: &Address, shard: usize, process_id: i32) {
+    pub fn ban(&self, address: &Address, process_id: i32) {
         self.stats.client_disconnecting(process_id, address.id);
-        self.stats
-            .checkout_time(Instant::now().elapsed().as_micros(), process_id, address.id);
 
         error!("Banning {:?}", address);
+
         let now = chrono::offset::Utc::now().naive_utc();
         let mut guard = self.banlist.write();
-        guard[shard].insert(address.clone(), now);
+        guard[address.shard].insert(address.clone(), now);
     }
 
     /// Clear the replica to receive traffic again. Takes effect immediately
     /// for all new transactions.
-    pub fn _unban(&self, address: &Address, shard: usize) {
+    pub fn _unban(&self, address: &Address) {
         let mut guard = self.banlist.write();
-        guard[shard].remove(address);
+        guard[address.shard].remove(address);
     }
 
     /// Check if a replica can serve traffic. If all replicas are banned,
     /// we unban all of them. Better to try then not to.
-    pub fn is_banned(&self, address: &Address, shard: usize, role: Option<Role>) -> bool {
+    pub fn is_banned(&self, address: &Address, role: Option<Role>) -> bool {
         let replicas_available = match role {
-            Some(Role::Replica) => self.addresses[shard]
+            Some(Role::Replica) => self.addresses[address.shard]
                 .iter()
                 .filter(|addr| addr.role == Role::Replica)
                 .count(),
-            None => self.addresses[shard].len(),
+            None => self.addresses[address.shard].len(),
             Some(Role::Primary) => return false, // Primary cannot be banned.
         };
 
@@ -423,17 +455,17 @@ impl ConnectionPool {
         let guard = self.banlist.read();
 
         // Everything is banned = nothing is banned.
-        if guard[shard].len() == replicas_available {
+        if guard[address.shard].len() == replicas_available {
             drop(guard);
             let mut guard = self.banlist.write();
-            guard[shard].clear();
+            guard[address.shard].clear();
             drop(guard);
             warn!("Unbanning all replicas.");
             return false;
         }
 
         // I expect this to miss 99.9999% of the time.
-        match guard[shard].get(address) {
+        match guard[address.shard].get(address) {
             Some(timestamp) => {
                 let now = chrono::offset::Utc::now().naive_utc();
                 let config = get_config();
@@ -443,7 +475,7 @@ impl ConnectionPool {
                     drop(guard);
                     warn!("Unbanning {:?}", address);
                     let mut guard = self.banlist.write();
-                    guard[shard].remove(address);
+                    guard[address.shard].remove(address);
                     false
                 } else {
                     debug!("{:?} is banned", address);
@@ -527,11 +559,7 @@ impl ManageConnection for ServerPool {
 
     /// Attempts to create a new connection.
     async fn connect(&self) -> Result<Self::Connection, Self::Error> {
-        info!(
-            "Creating a new connection to {:?} using user {:?}",
-            self.address.name(),
-            self.user.username
-        );
+        info!("Creating a new server connection {:?}", self.address);
 
         // Put a temporary process_id into the stats
         // for server login.
@@ -580,6 +608,7 @@ pub fn get_pool(db: String, user: String) -> Option<ConnectionPool> {
     }
 }
 
+/// How many total servers we have in the config.
 pub fn get_number_of_addresses() -> usize {
     get_all_pools()
         .iter()
@@ -587,6 +616,7 @@ pub fn get_number_of_addresses() -> usize {
         .sum()
 }
 
+/// Get a pointer to all configured pools.
 pub fn get_all_pools() -> HashMap<(String, String), ConnectionPool> {
     return (*(*POOLS.load())).clone();
 }

src/query_router.rs

@@ -10,10 +10,10 @@ use sqlparser::parser::Parser;
 
 use crate::config::Role;
 use crate::pool::PoolSettings;
-use crate::sharding::{Sharder, ShardingFunction};
+use crate::sharding::Sharder;
 
 /// Regexes used to parse custom commands.
-const CUSTOM_SQL_REGEXES: [&str; 7] = [
+const CUSTOM_SQL_REGEXES: [&str; 8] = [
     r"(?i)^ *SET SHARDING KEY TO '?([0-9]+)'? *;? *$",
     r"(?i)^ *SET SHARD TO '?([0-9]+|ANY)'? *;? *$",
     r"(?i)^ *SHOW SHARD *;? *$",
@@ -21,6 +21,7 @@ const CUSTOM_SQL_REGEXES: [&str; 7] = [
     r"(?i)^ *SHOW SERVER ROLE *;? *$",
     r"(?i)^ *SET PRIMARY READS TO '?(on|off|default)'? *;? *$",
     r"(?i)^ *SHOW PRIMARY READS *;? *$",
+    r"(?i)^ *SHARDED_COPY '?([0-9]+)'? *;? *$",
 ];
 
 /// Custom commands.
@@ -33,6 +34,7 @@ pub enum Command {
     ShowServerRole,
     SetPrimaryReads,
     ShowPrimaryReads,
+    StartShardedCopy,
 }
 
 /// Quickly test for match when a query is received.
@@ -55,11 +57,16 @@ pub struct QueryRouter {
     /// Include the primary into the replica pool for reads.
     primary_reads_enabled: bool,
 
+    /// Pool configuration.
     pool_settings: PoolSettings,
+
+    // Sharding key column
+    sharding_key_column: Option<usize>,
 }
 
 impl QueryRouter {
-    /// One-time initialization of regexes.
+    /// One-time initialization of regexes
+    /// that parse our custom SQL protocol.
     pub fn setup() -> bool {
         let set = match RegexSet::new(&CUSTOM_SQL_REGEXES) {
             Ok(rgx) => rgx,
@@ -74,10 +81,7 @@ impl QueryRouter {
             .map(|rgx| Regex::new(rgx).unwrap())
             .collect();
 
-        // Impossible
-        if list.len() != set.len() {
-            return false;
-        }
+        assert_eq!(list.len(), set.len());
 
         match CUSTOM_SQL_REGEX_LIST.set(list) {
             Ok(_) => true,
@@ -90,7 +94,8 @@ impl QueryRouter {
         }
     }
 
-    /// Create a new instance of the query router. Each client gets its own.
+    /// Create a new instance of the query router.
+    /// Each client gets its own.
     pub fn new() -> QueryRouter {
         QueryRouter {
             active_shard: None,
@@ -98,9 +103,11 @@ impl QueryRouter {
             query_parser_enabled: false,
             primary_reads_enabled: false,
             pool_settings: PoolSettings::default(),
+            sharding_key_column: None,
         }
     }
 
+    /// Pool settings can change because of a config reload.
     pub fn update_pool_settings(&mut self, pool_settings: PoolSettings) {
         self.pool_settings = pool_settings;
     }
@@ -136,19 +143,6 @@ impl QueryRouter {
             return None;
         }
 
-        let sharding_function = match self.pool_settings.sharding_function.as_ref() {
-            "pg_bigint_hash" => ShardingFunction::PgBigintHash,
-            "sha1" => ShardingFunction::Sha1,
-            _ => unreachable!(),
-        };
-
-        let default_server_role = match self.pool_settings.default_role.as_ref() {
-            "any" => None,
-            "primary" => Some(Role::Primary),
-            "replica" => Some(Role::Replica),
-            _ => unreachable!(),
-        };
-
         let command = match matches[0] {
             0 => Command::SetShardingKey,
             1 => Command::SetShard,
@@ -157,6 +151,7 @@ impl QueryRouter {
             4 => Command::ShowServerRole,
             5 => Command::SetPrimaryReads,
             6 => Command::ShowPrimaryReads,
+            7 => Command::StartShardedCopy,
             _ => unreachable!(),
         };
 
@@ -164,7 +159,8 @@ impl QueryRouter {
             Command::SetShardingKey
             | Command::SetShard
             | Command::SetServerRole
-            | Command::SetPrimaryReads => {
+            | Command::SetPrimaryReads
+            | Command::StartShardedCopy => {
                 // Capture value. I know this re-runs the regex engine, but I haven't
                 // figured out a better way just yet. I think I can write a single Regex
                 // that matches all 5 custom SQL patterns, but maybe that's not very legible?
@@ -200,7 +196,10 @@ impl QueryRouter {
 
         match command {
             Command::SetShardingKey => {
-                let sharder = Sharder::new(self.pool_settings.shards.len(), sharding_function);
+                let sharder = Sharder::new(
+                    self.pool_settings.shards,
+                    self.pool_settings.sharding_function,
+                );
                 let shard = sharder.shard(value.parse::<i64>().unwrap());
                 self.active_shard = Some(shard);
                 value = shard.to_string();
@@ -208,11 +207,18 @@ impl QueryRouter {
 
             Command::SetShard => {
                 self.active_shard = match value.to_ascii_uppercase().as_ref() {
-                    "ANY" => Some(rand::random::<usize>() % self.pool_settings.shards.len()),
+                    "ANY" => Some(rand::random::<usize>() % self.pool_settings.shards),
                     _ => Some(value.parse::<usize>().unwrap()),
                 };
             }
 
+            Command::StartShardedCopy => {
+                self.sharding_key_column = match value.parse::<usize>() {
+                    Ok(value) => Some(value),
+                    Err(_) => return None,
+                }
+            }
+
             Command::SetServerRole => {
                 self.active_role = match value.to_ascii_lowercase().as_ref() {
                     "primary" => {
@@ -236,7 +242,7 @@ impl QueryRouter {
                     }
 
                     "default" => {
-                        self.active_role = default_server_role;
+                        self.active_role = self.pool_settings.default_role;
                         self.query_parser_enabled = self.query_parser_enabled;
                         self.active_role
                     }
@@ -367,10 +373,10 @@ impl QueryRouter {
 
 #[cfg(test)]
 mod test {
-    use std::collections::HashMap;
-
     use super::*;
     use crate::messages::simple_query;
+    use crate::pool::PoolMode;
+    use crate::sharding::ShardingFunction;
     use bytes::BufMut;
 
     #[test]
@@ -633,13 +639,13 @@ mod test {
         QueryRouter::setup();
 
         let pool_settings = PoolSettings {
-            pool_mode: "transaction".to_string(),
-            shards: HashMap::default(),
+            pool_mode: PoolMode::Transaction,
+            shards: 0,
             user: crate::config::User::default(),
-            default_role: Role::Replica.to_string(),
+            default_role: Some(Role::Replica),
             query_parser_enabled: true,
             primary_reads_enabled: false,
-            sharding_function: "pg_bigint_hash".to_string(),
+            sharding_function: ShardingFunction::PgBigintHash,
         };
         let mut qr = QueryRouter::new();
         assert_eq!(qr.active_role, None);
@@ -661,9 +667,6 @@ mod test {
 
         let q2 = simple_query("SET SERVER ROLE TO 'default'");
         assert!(qr.try_execute_command(q2) != None);
-        assert_eq!(
-            qr.active_role.unwrap().to_string(),
-            pool_settings.clone().default_role
-        );
+        assert_eq!(qr.active_role.unwrap(), pool_settings.clone().default_role);
     }
 }

src/server.rs

@@ -1,7 +1,8 @@
 /// Implementation of the PostgreSQL server (database) protocol.
 /// Here we are pretending to the a Postgres client.
 use bytes::{Buf, BufMut, BytesMut};
-use log::{debug, error, info, trace};
+use log::{debug, error, info, trace, warn};
+use std::io::Read;
 use std::time::SystemTime;
 use tokio::io::{AsyncReadExt, BufReader};
 use tokio::net::{
@@ -48,6 +49,9 @@ pub struct Server {
     /// Is the server broken? We'll remote it from the pool if so.
     bad: bool,
 
+    /// If server connection requires a DISCARD ALL before checkin
+    needs_cleanup: bool,
+
     /// Mapping of clients and servers used for query cancellation.
     client_server_map: ClientServerMap,
 
@@ -75,7 +79,7 @@ impl Server {
         stats: Reporter,
     ) -> Result<Server, Error> {
         let mut stream =
-            match TcpStream::connect(&format!("{}:{}", &address.host, &address.port)).await {
+            match TcpStream::connect(&format!("{}:{}", &address.host, address.port)).await {
                 Ok(stream) => stream,
                 Err(err) => {
                     error!("Could not connect to server: {}", err);
@@ -316,6 +320,7 @@ impl Server {
                         in_transaction: false,
                         data_available: false,
                         bad: false,
+                        needs_cleanup: false,
                         client_server_map: client_server_map,
                         connected_at: chrono::offset::Utc::now().naive_utc(),
                         stats: stats,
@@ -342,7 +347,7 @@ impl Server {
     /// Uses a separate connection that's not part of the connection pool.
     pub async fn cancel(
         host: &str,
-        port: &str,
+        port: u16,
         process_id: i32,
         secret_key: i32,
     ) -> Result<(), Error> {
@@ -440,6 +445,29 @@ impl Server {
                     break;
                 }
 
+                // CommandComplete
+                'C' => {
+                    let mut command_tag = String::new();
+                    match message.reader().read_to_string(&mut command_tag) {
+                        Ok(_) => {
+                            // Non-exhaustive list of commands that are likely to change session variables/resources
+                            // which can leak between clients. This is a best effort to block bad clients
+                            // from poisoning a transaction-mode pool by setting inappropriate session variables
+                            match command_tag.as_str() {
+                                "SET\0" | "PREPARE\0" => {
+                                    debug!("Server connection marked for clean up");
+                                    self.needs_cleanup = true;
+                                }
+                                _ => (),
+                            }
+                        }
+
+                        Err(err) => {
+                            warn!("Encountered an error while parsing CommandTag {}", err);
+                        }
+                    }
+                }
+
                 // DataRow
                 'D' => {
                     // More data is available after this message, this is not the end of the reply.
@@ -529,7 +557,7 @@ impl Server {
                 self.process_id,
                 self.secret_key,
                 self.address.host.clone(),
-                self.address.port.clone(),
+                self.address.port,
             ),
         );
     }
@@ -553,14 +581,43 @@ impl Server {
         Ok(())
     }
 
+    /// Perform any necessary cleanup before putting the server
+    /// connection back in the pool
+    pub async fn checkin_cleanup(&mut self) -> Result<(), Error> {
+        // Client disconnected with an open transaction on the server connection.
+        // Pgbouncer behavior is to close the server connection but that can cause
+        // server connection thrashing if clients repeatedly do this.
+        // Instead, we ROLLBACK that transaction before putting the connection back in the pool
+        if self.in_transaction() {
+            self.query("ROLLBACK").await?;
+        }
+
+        // Client disconnected but it perfromed session-altering operations such as
+        // SET statement_timeout to 1 or create a prepared statement. We clear that
+        // to avoid leaking state between clients. For performance reasons we only
+        // send `DISCARD ALL` if we think the session is altered instead of just sending
+        // it before each checkin.
+        if self.needs_cleanup {
+            self.query("DISCARD ALL").await?;
+            self.needs_cleanup = false;
+        }
+
+        return Ok(());
+    }
+
     /// A shorthand for `SET application_name = $1`.
-    #[allow(dead_code)]
     pub async fn set_name(&mut self, name: &str) -> Result<(), Error> {
         if self.application_name != name {
             self.application_name = name.to_string();
-            Ok(self
+            // We don't want `SET application_name` to mark the server connection
+            // as needing cleanup
+            let needs_cleanup_before = self.needs_cleanup;
+
+            let result = Ok(self
                 .query(&format!("SET application_name = '{}'", name))
-                .await?)
+                .await?);
+            self.needs_cleanup = needs_cleanup_before;
+            return result;
         } else {
             Ok(())
         }
@@ -581,6 +638,11 @@ impl Server {
     pub fn last_activity(&self) -> SystemTime {
         self.last_activity
     }
+
+    // Marks a connection as needing DISCARD ALL at checkin
+    pub fn mark_dirty(&mut self) {
+        self.needs_cleanup = true;
+    }
 }
 
 impl Drop for Server {
@@ -607,7 +669,8 @@ impl Drop for Server {
         let duration = now - self.connected_at;
 
         info!(
-            "Server connection closed, session duration: {}",
+            "Server connection closed {:?}, session duration: {}",
+            self.address,
             crate::format_duration(&duration)
         );
     }

tests/docker/Dockerfile

@@ -0,0 +1,5 @@
+FROM rust:bullseye
+
+RUN apt-get update && apt-get install llvm-11 psmisc postgresql-contrib postgresql-client ruby ruby-dev libpq-dev python3 python3-pip lcov sudo curl -y
+RUN cargo install cargo-binutils rustfilt
+RUN rustup component add llvm-tools-preview

tests/docker/docker-compose.yml

@@ -0,0 +1,47 @@
+version: "3"
+services:
+  pg1:
+    image: postgres:14
+    network_mode: "service:main"
+    environment:
+      POSTGRES_USER: postgres
+      POSTGRES_DB: postgres
+      POSTGRES_PASSWORD: postgres
+      POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+    command: ["postgres", "-c", "shared_preload_libraries=pg_stat_statements", "-c", "pg_stat_statements.track=all", "-p", "5432"]
+  pg2:
+    image: postgres:14
+    network_mode: "service:main"
+    environment:
+      POSTGRES_USER: postgres
+      POSTGRES_DB: postgres
+      POSTGRES_PASSWORD: postgres
+      POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+    command: ["postgres", "-c", "shared_preload_libraries=pg_stat_statements", "-c", "pg_stat_statements.track=all", "-p", "7432"]
+  pg3:
+    image: postgres:14
+    network_mode: "service:main"
+    environment:
+      POSTGRES_USER: postgres
+      POSTGRES_DB: postgres
+      POSTGRES_PASSWORD: postgres
+      POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+    command: ["postgres", "-c", "shared_preload_libraries=pg_stat_statements", "-c", "pg_stat_statements.track=all", "-p", "8432"]
+  pg4:
+    image: postgres:14
+    network_mode: "service:main"
+    environment:
+      POSTGRES_USER: postgres
+      POSTGRES_DB: postgres
+      POSTGRES_PASSWORD: postgres
+      POSTGRES_HOST_AUTH_METHOD: scram-sha-256
+    command: ["postgres", "-c", "shared_preload_libraries=pg_stat_statements", "-c", "pg_stat_statements.track=all", "-p", "9432"]
+  main:
+    build: .
+    command: ["bash", "/app/tests/docker/run.sh"]
+    environment:
+      RUSTFLAGS: "-C instrument-coverage"
+      LLVM_PROFILE_FILE: "pgcat-%m.profraw"
+    volumes:
+      - ../../:/app/
+      - /app/target/

tests/docker/run.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+rm /app/*.profraw || true
+rm /app/pgcat.profdata || true
+rm -rf /app/cov || true
+
+cd /app/
+
+cargo build
+cargo test --tests
+
+bash .circleci/run_tests.sh
+
+rust-profdata merge -sparse pgcat-*.profraw -o pgcat.profdata
+
+rust-cov export -ignore-filename-regex="rustc|registry" -Xdemangler=rustfilt -instr-profile=pgcat.profdata --object ./target/debug/pgcat --format lcov > ./lcov.info
+
+genhtml lcov.info --output-directory cov --prefix $(pwd)
+
+rm /app/*.profraw
+rm /app/pgcat.profdata

tests/python/tests.py

@@ -14,12 +14,18 @@ PGCAT_PORT = "6432"
 def pgcat_start():
     pg_cat_send_signal(signal.SIGTERM)
     os.system("./target/debug/pgcat .circleci/pgcat.toml &")
+    time.sleep(2)
 
 
 def pg_cat_send_signal(signal: signal.Signals):
-    for proc in psutil.process_iter(["pid", "name"]):
-        if "pgcat" == proc.name():
-            os.kill(proc.pid, signal)
+    try:
+        for proc in psutil.process_iter(["pid", "name"]):
+            if "pgcat" == proc.name():
+                os.kill(proc.pid, signal)
+    except Exception as e:
+        # The process can be gone when we send this signal
+        print(e)
+
     if signal == signal.SIGTERM:
         # Returns 0 if pgcat process exists
         time.sleep(2)
@@ -27,11 +33,23 @@ def pg_cat_send_signal(signal: signal.Signals):
             raise Exception("pgcat not closed after SIGTERM")
 
 
-def connect_normal_db(
-    autocommit: bool = False,
+def connect_db(
+    autocommit: bool = True,
+    admin: bool = False,
 ) -> Tuple[psycopg2.extensions.connection, psycopg2.extensions.cursor]:
+
+    if admin:
+        user = "admin_user"
+        password = "admin_pass"
+        db = "pgcat"
+    else:
+        user = "sharding_user"
+        password = "sharding_user"
+        db = "sharded_db"
+
     conn = psycopg2.connect(
-        f"postgres://sharding_user:sharding_user@{PGCAT_HOST}:{PGCAT_PORT}/sharded_db?application_name=testing_pgcat"
+        f"postgres://{user}:{password}@{PGCAT_HOST}:{PGCAT_PORT}/{db}?application_name=testing_pgcat",
+        connect_timeout=2,
     )
     conn.autocommit = autocommit
     cur = conn.cursor()
@@ -45,7 +63,7 @@ def cleanup_conn(conn: psycopg2.extensions.connection, cur: psycopg2.extensions.
 
 
 def test_normal_db_access():
-    conn, cur = connect_normal_db()
+    conn, cur = connect_db(autocommit=False)
     cur.execute("SELECT 1")
     res = cur.fetchall()
     print(res)
@@ -53,11 +71,7 @@ def test_normal_db_access():
 
 
 def test_admin_db_access():
-    conn = psycopg2.connect(
-        f"postgres://admin_user:admin_pass@{PGCAT_HOST}:{PGCAT_PORT}/pgcat"
-    )
-    conn.autocommit = True  # BEGIN/COMMIT is not supported by admin db
-    cur = conn.cursor()
+    conn, cur = connect_db(admin=True)
 
     cur.execute("SHOW POOLS")
     res = cur.fetchall()
@@ -67,15 +81,14 @@ def test_admin_db_access():
 
 def test_shutdown_logic():
 
-    ##### NO ACTIVE QUERIES SIGINT HANDLING #####
+    # - - - - - - - - - - - - - - - - - -
+    # NO ACTIVE QUERIES SIGINT HANDLING
+
     # Start pgcat
     pgcat_start()
 
-    # Wait for server to fully start up
-    time.sleep(2)
-
     # Create client connection and send query (not in transaction)
-    conn, cur = connect_normal_db(True)
+    conn, cur = connect_db()
 
     cur.execute("BEGIN;")
     cur.execute("SELECT 1;")
@@ -97,17 +110,14 @@ def test_shutdown_logic():
     cleanup_conn(conn, cur)
     pg_cat_send_signal(signal.SIGTERM)
 
-    ##### END #####
+    # - - - - - - - - - - - - - - - - - -
+    # HANDLE TRANSACTION WITH SIGINT
 
-    ##### HANDLE TRANSACTION WITH SIGINT #####
     # Start pgcat
     pgcat_start()
 
-    # Wait for server to fully start up
-    time.sleep(2)
-
     # Create client connection and begin transaction
-    conn, cur = connect_normal_db(True)
+    conn, cur = connect_db()
 
     cur.execute("BEGIN;")
     cur.execute("SELECT 1;")
@@ -126,17 +136,97 @@ def test_shutdown_logic():
     cleanup_conn(conn, cur)
     pg_cat_send_signal(signal.SIGTERM)
 
-    ##### END #####
-
-    ##### HANDLE SHUTDOWN TIMEOUT WITH SIGINT #####
+    # - - - - - - - - - - - - - - - - - -
+    # NO NEW NON-ADMIN CONNECTIONS DURING SHUTDOWN
     # Start pgcat
     pgcat_start()
 
-    # Wait for server to fully start up
-    time.sleep(3)
+    # Create client connection and begin transaction
+    transaction_conn, transaction_cur = connect_db()
+
+    transaction_cur.execute("BEGIN;")
+    transaction_cur.execute("SELECT 1;")
+
+    # Send sigint to pgcat while still in transaction
+    pg_cat_send_signal(signal.SIGINT)
+    time.sleep(1)
+
+    start = time.perf_counter()
+    try:
+        conn, cur = connect_db()
+        cur.execute("SELECT 1;")
+        cleanup_conn(conn, cur)
+    except psycopg2.OperationalError as e:
+        time_taken = time.perf_counter() - start
+        if time_taken > 0.1:
+            raise Exception(
+                "Failed to reject connection within 0.1 seconds, got", time_taken, "seconds")
+        pass
+    else:
+        raise Exception("Able connect to database during shutdown")
+
+    cleanup_conn(transaction_conn, transaction_cur)
+    pg_cat_send_signal(signal.SIGTERM)
+
+    # - - - - - - - - - - - - - - - - - -
+    # ALLOW NEW ADMIN CONNECTIONS DURING SHUTDOWN
+    # Start pgcat
+    pgcat_start()
+
+    # Create client connection and begin transaction
+    transaction_conn, transaction_cur = connect_db()
+
+    transaction_cur.execute("BEGIN;")
+    transaction_cur.execute("SELECT 1;")
+
+    # Send sigint to pgcat while still in transaction
+    pg_cat_send_signal(signal.SIGINT)
+    time.sleep(1)
+
+    try:
+        conn, cur = connect_db(admin=True)
+        cur.execute("SHOW DATABASES;")
+        cleanup_conn(conn, cur)
+    except psycopg2.OperationalError as e:
+        raise Exception(e)
+
+    cleanup_conn(transaction_conn, transaction_cur)
+    pg_cat_send_signal(signal.SIGTERM)
+
+    # - - - - - - - - - - - - - - - - - -
+    # ADMIN CONNECTIONS CONTINUING TO WORK AFTER SHUTDOWN
+    # Start pgcat
+    pgcat_start()
+
+    # Create client connection and begin transaction
+    transaction_conn, transaction_cur = connect_db()
+    transaction_cur.execute("BEGIN;")
+    transaction_cur.execute("SELECT 1;")
+
+    admin_conn, admin_cur = connect_db(admin=True)
+    admin_cur.execute("SHOW DATABASES;")
+
+    # Send sigint to pgcat while still in transaction
+    pg_cat_send_signal(signal.SIGINT)
+    time.sleep(1)
+
+    try:
+        admin_cur.execute("SHOW DATABASES;")
+    except psycopg2.OperationalError as e:
+        raise Exception("Could not execute admin command:", e)
+
+    cleanup_conn(transaction_conn, transaction_cur)
+    cleanup_conn(admin_conn, admin_cur)
+    pg_cat_send_signal(signal.SIGTERM)
+
+    # - - - - - - - - - - - - - - - - - -
+    # HANDLE SHUTDOWN TIMEOUT WITH SIGINT
+
+    # Start pgcat
+    pgcat_start()
 
     # Create client connection and begin transaction, which should prevent server shutdown unless shutdown timeout is reached
-    conn, cur = connect_normal_db(True)
+    conn, cur = connect_db()
 
     cur.execute("BEGIN;")
     cur.execute("SELECT 1;")
@@ -159,7 +249,7 @@ def test_shutdown_logic():
     cleanup_conn(conn, cur)
     pg_cat_send_signal(signal.SIGTERM)
 
-    ##### END #####
+    # - - - - - - - - - - - - - - - - - -
 
 
 test_normal_db_access()

tests/ruby/Gemfile

@@ -1,6 +1,8 @@
 source "https://rubygems.org"
 
 gem "pg"
-gem "activerecord"
+gem "toml"
+gem "rspec"
 gem "rubocop"
-gem "toml", "~> 0.3.0"
+gem "toxiproxy"
+gem "activerecord"

tests/ruby/Gemfile.lock

@@ -13,6 +13,7 @@ GEM
       tzinfo (~> 2.0)
     ast (2.4.2)
     concurrent-ruby (1.1.10)
+    diff-lcs (1.5.0)
     i18n (1.11.0)
       concurrent-ruby (~> 1.0)
     minitest (5.16.2)
@@ -24,6 +25,19 @@ GEM
     rainbow (3.1.1)
     regexp_parser (2.3.1)
     rexml (3.2.5)
+    rspec (3.11.0)
+      rspec-core (~> 3.11.0)
+      rspec-expectations (~> 3.11.0)
+      rspec-mocks (~> 3.11.0)
+    rspec-core (3.11.0)
+      rspec-support (~> 3.11.0)
+    rspec-expectations (3.11.0)
+      diff-lcs (>= 1.2.0, < 2.0)
+      rspec-support (~> 3.11.0)
+    rspec-mocks (3.11.1)
+      diff-lcs (>= 1.2.0, < 2.0)
+      rspec-support (~> 3.11.0)
+    rspec-support (3.11.0)
     rubocop (1.29.0)
       parallel (~> 1.10)
       parser (>= 3.1.0.0)
@@ -38,19 +52,23 @@ GEM
     ruby-progressbar (1.11.0)
     toml (0.3.0)
       parslet (>= 1.8.0, < 3.0.0)
+    toxiproxy (2.0.1)
     tzinfo (2.0.4)
       concurrent-ruby (~> 1.0)
     unicode-display_width (2.1.0)
 
 PLATFORMS
+  aarch64-linux
   arm64-darwin-21
   x86_64-linux
 
 DEPENDENCIES
   activerecord
   pg
+  rspec
   rubocop
-  toml (~> 0.3.0)
+  toml
+  toxiproxy
 
 BUNDLED WITH
-   2.3.7
+   2.3.21

tests/ruby/helpers/pg_instance.rb

@@ -0,0 +1,82 @@
+require 'pg'
+require 'toxiproxy'
+
+class PgInstance
+  attr_reader :port
+  attr_reader :username
+  attr_reader :password
+  attr_reader :database_name
+
+  def initialize(port, username, password, database_name)
+    @original_port = port
+    @toxiproxy_port = 10000 + port.to_i
+    @port = @toxiproxy_port
+
+    @username = username
+    @password = password
+    @database_name = database_name
+    @toxiproxy_name = "database_#{@original_port}"
+    Toxiproxy.populate([{
+      name: @toxiproxy_name,
+      listen: "0.0.0.0:#{@toxiproxy_port}",
+      upstream: "localhost:#{@original_port}",
+    }])
+
+    # Toxiproxy server will outlive our PgInstance objects
+    # so we want to destroy our proxies before exiting
+    # Ruby finalizer is ideal for doing this
+    ObjectSpace.define_finalizer(@toxiproxy_name, proc { Toxiproxy[@toxiproxy_name].destroy })
+  end
+
+  def with_connection
+    conn = PG.connect("postgres://#{@username}:#{@password}@localhost:#{port}/#{database_name}")
+    yield conn
+  ensure
+    conn&.close
+  end
+
+  def reset
+    reset_toxics
+    reset_stats
+  end
+
+  def toxiproxy
+    Toxiproxy[@toxiproxy_name]
+  end
+
+  def take_down
+    if block_given?
+      Toxiproxy[@toxiproxy_name].toxic(:limit_data, bytes: 5).apply { yield }
+    else
+      Toxiproxy[@toxiproxy_name].toxic(:limit_data, bytes: 5).toxics.each(&:save)
+    end
+  end
+
+  def add_latency(latency)
+    if block_given?
+      Toxiproxy[@toxiproxy_name].toxic(:latency, latency: latency).apply { yield }
+    else
+      Toxiproxy[@toxiproxy_name].toxic(:latency, latency: latency).toxics.each(&:save)
+    end
+  end
+
+  def delete_proxy
+    Toxiproxy[@toxiproxy_name].delete
+  end
+
+  def reset_toxics
+    Toxiproxy[@toxiproxy_name].toxics.each(&:destroy)
+  end
+
+  def reset_stats
+    with_connection { |c| c.async_exec("SELECT pg_stat_statements_reset()") }
+  end
+
+  def count_query(query)
+    with_connection { |c| c.async_exec("SELECT SUM(calls) FROM pg_stat_statements WHERE query = '#{query}'")[0]["sum"].to_i }
+  end
+
+  def count_select_1_plus_2
+    with_connection { |c| c.async_exec("SELECT SUM(calls) FROM pg_stat_statements WHERE query = 'SELECT $1 + $2'")[0]["sum"].to_i }
+  end
+end

tests/ruby/helpers/pgcat_helper.rb

@@ -0,0 +1,100 @@
+require 'json'
+require 'ostruct'
+require_relative 'pgcat_process'
+require_relative 'pg_instance'
+
+module Helpers
+  module Pgcat
+    def self.three_shard_setup(pool_name, pool_size, pool_mode="transaction")
+      user = {
+        "password" => "sharding_user",
+        "pool_size" => pool_size,
+        "statement_timeout" => 0,
+        "username" => "sharding_user"
+      }
+
+      pgcat    = PgcatProcess.new("info")
+      primary0 = PgInstance.new(5432, user["username"], user["password"], "shard0")
+      primary1 = PgInstance.new(7432, user["username"], user["password"], "shard1")
+      primary2 = PgInstance.new(8432, user["username"], user["password"], "shard2")
+
+      pgcat_cfg = pgcat.current_config
+      pgcat_cfg["pools"] = {
+        "#{pool_name}" => {
+          "default_role" => "any",
+          "pool_mode" => pool_mode,
+          "primary_reads_enabled" => false,
+          "query_parser_enabled" => false,
+          "sharding_function" => "pg_bigint_hash",
+          "shards" => {
+            "0" => { "database" => "shard0", "servers" => [["localhost", primary0.port.to_s, "primary"]] },
+            "1" => { "database" => "shard1", "servers" => [["localhost", primary1.port.to_s, "primary"]] },
+            "2" => { "database" => "shard2", "servers" => [["localhost", primary2.port.to_s, "primary"]] },
+          },
+          "users" => { "0" => user }
+        }
+      }
+      pgcat.update_config(pgcat_cfg)
+
+      pgcat.start
+      pgcat.wait_until_ready
+
+      OpenStruct.new.tap do |struct|
+        struct.pgcat = pgcat
+        struct.shards = [primary0, primary1, primary2]
+        struct.all_databases = [primary0, primary1, primary2]
+      end
+    end
+
+    def self.single_shard_setup(pool_name, pool_size, pool_mode="transaction")
+      user = {
+        "password" => "sharding_user",
+        "pool_size" => pool_size,
+        "statement_timeout" => 0,
+        "username" => "sharding_user"
+      }
+
+      pgcat = PgcatProcess.new("info")
+      pgcat_cfg = pgcat.current_config
+
+      primary  = PgInstance.new(5432, user["username"], user["password"], "shard0")
+      replica0 = PgInstance.new(7432, user["username"], user["password"], "shard0")
+      replica1 = PgInstance.new(8432, user["username"], user["password"], "shard0")
+      replica2 = PgInstance.new(9432, user["username"], user["password"], "shard0")
+
+      # Main proxy configs
+      pgcat_cfg["pools"] = {
+        "#{pool_name}" => {
+          "default_role" => "any",
+          "pool_mode" => pool_mode,
+          "primary_reads_enabled" => false,
+          "query_parser_enabled" => false,
+          "sharding_function" => "pg_bigint_hash",
+          "shards" => {
+            "0" => {
+              "database" => "shard0",
+              "servers" => [
+                ["localhost", primary.port.to_s, "primary"],
+                ["localhost", replica0.port.to_s, "replica"],
+                ["localhost", replica1.port.to_s, "replica"],
+                ["localhost", replica2.port.to_s, "replica"]
+              ]
+            },
+          },
+          "users" => { "0" => user }
+        }
+      }
+      pgcat_cfg["general"]["port"] = pgcat.port
+      pgcat.update_config(pgcat_cfg)
+      pgcat.start
+      pgcat.wait_until_ready
+
+      OpenStruct.new.tap do |struct|
+        struct.pgcat = pgcat
+        struct.primary = primary
+        struct.replicas = [replica0, replica1, replica2]
+        struct.all_databases = [primary, replica0, replica1, replica2]
+      end
+    end
+  end
+end

tests/ruby/helpers/pgcat_process.rb

@@ -0,0 +1,116 @@
+require 'pg'
+require 'toml'
+require 'fileutils'
+require 'securerandom'
+
+class PgcatProcess
+  attr_reader :port
+  attr_reader :pid
+
+  def self.finalize(pid, log_filename, config_filename)
+    `kill #{pid}`
+    File.delete(config_filename) if File.exist?(config_filename)
+    File.delete(log_filename) if File.exist?(log_filename)
+  end
+
+  def initialize(log_level)
+    @env = {"RUST_LOG" => log_level}
+    @port = rand(20000..32760)
+    @log_level = log_level
+    @log_filename = "/tmp/pgcat_log_#{SecureRandom.urlsafe_base64}.log"
+    @config_filename = "/tmp/pgcat_cfg_#{SecureRandom.urlsafe_base64}.toml"
+
+    @command = "../../target/debug/pgcat #{@config_filename}"
+
+    FileUtils.cp("../../pgcat.toml", @config_filename)
+    cfg = current_config
+    cfg["general"]["port"] = @port.to_i
+    cfg["general"]["enable_prometheus_exporter"] = false
+
+    update_config(cfg)
+  end
+
+  def logs
+    File.read(@log_filename)
+  end
+
+  def update_config(config_hash)
+    @original_config = current_config
+    output_to_write = TOML::Generator.new(config_hash).body
+    output_to_write = output_to_write.gsub(/,\s*["|'](\d+)["|']\s*,/, ',\1,')
+    File.write(@config_filename, output_to_write)
+  end
+
+  def current_config
+    old_cfg = File.read(@config_filename)
+    loadable_string = old_cfg.gsub(/,\s*(\d+)\s*,/, ', "\1",')
+    TOML.load(loadable_string)
+  end
+
+  def reload_config
+    `kill -s HUP #{@pid}`
+    sleep 0.1
+  end
+
+  def start
+    raise StandardError, "Process is already started" unless @pid.nil?
+    @pid = Process.spawn(@env, @command, err: @log_filename, out: @log_filename)
+    ObjectSpace.define_finalizer(@log_filename, proc { PgcatProcess.finalize(@pid, @log_filename, @config_filename) })
+
+    return self
+  end
+
+  def wait_until_ready
+    exc = nil
+    10.times do
+      PG::connect(example_connection_string).close
+
+      return self
+    rescue => e
+      exc = e
+      sleep(0.5)
+    end
+    puts exc
+    raise StandardError, "Process #{@pid} never became ready. Logs #{logs}"
+  end
+
+  def stop
+    `kill #{@pid}`
+    sleep 0.1
+  end
+
+  def shutdown
+    stop
+    File.delete(@config_filename) if File.exist?(@config_filename)
+    File.delete(@log_filename) if File.exist?(@log_filename)
+  end
+
+  def admin_connection_string
+    cfg = current_config
+    username = cfg["general"]["admin_username"]
+    password = cfg["general"]["admin_password"]
+
+    "postgresql://#{username}:#{password}@0.0.0.0:#{@port}/pgcat"
+  end
+
+  def connection_string(pool_name, username)
+    cfg = current_config
+
+    user_idx, user_obj = cfg["pools"][pool_name]["users"].detect { |k, user| user["username"] == username }
+    password = user_obj["password"]
+
+    "postgresql://#{username}:#{password}@0.0.0.0:#{@port}/#{pool_name}"
+  end
+
+  def example_connection_string
+    cfg = current_config
+    first_pool_name = cfg["pools"].keys[0]
+
+    db_name = first_pool_name
+
+    username = cfg["pools"][first_pool_name]["users"]["0"]["username"]
+    password = cfg["pools"][first_pool_name]["users"]["0"]["password"]
+
+    "postgresql://#{username}:#{password}@0.0.0.0:#{@port}/#{db_name}"
+  end
+end

tests/ruby/load_balancing_spec.rb

@@ -0,0 +1,61 @@
+# frozen_string_literal: true
+require_relative 'spec_helper'
+
+describe "Load Balancing" do
+  let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5) }
+  after do
+    processes.all_databases.map(&:reset)
+    processes.pgcat.shutdown
+  end
+
+  context "under regular circumstances" do
+    it "balances query volume between all instances" do
+      conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+
+      query_count = QUERY_COUNT
+      expected_share = query_count / processes.all_databases.count
+      failed_count = 0
+
+      query_count.times do
+        conn.async_exec("SELECT 1 + 2")
+      rescue
+        failed_count += 1
+      end
+
+      expect(failed_count).to eq(0)
+      processes.all_databases.map(&:count_select_1_plus_2).each do |instance_share|
+        expect(instance_share).to be_within(expected_share * MARGIN_OF_ERROR).of(expected_share)
+      end
+    end
+  end
+
+  context "when some replicas are down" do
+    it "balances query volume between working instances" do
+      conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      expected_share = QUERY_COUNT / (processes.all_databases.count - 2)
+      failed_count = 0
+
+      processes[:replicas][0].take_down do
+        processes[:replicas][1].take_down do
+          QUERY_COUNT.times do
+            conn.async_exec("SELECT 1 + 2")
+          rescue
+            conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+            failed_count += 1
+          end
+        end
+      end
+
+      expect(failed_count).to eq(2)
+      processes.all_databases.each do |instance|
+        queries_routed = instance.count_select_1_plus_2
+        if processes.replicas[0..1].include?(instance)
+          expect(queries_routed).to eq(0)
+        else
+          expect(queries_routed).to be_within(expected_share * MARGIN_OF_ERROR).of(expected_share)
+        end
+      end
+    end
+  end
+end
+

tests/ruby/misc_spec.rb

@@ -0,0 +1,193 @@
+# frozen_string_literal: true
+require_relative 'spec_helper'
+
+describe "Miscellaneous" do
+  let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5) }
+  after do
+    processes.all_databases.map(&:reset)
+    processes.pgcat.shutdown
+  end
+
+  describe "Extended Protocol handling" do
+    it "does not send packets that client does not expect during extended protocol sequence" do
+      new_configs = processes.pgcat.current_config
+
+      new_configs["general"]["connect_timeout"] = 500
+      new_configs["general"]["ban_time"] = 1
+      new_configs["general"]["shutdown_timeout"] = 1
+      new_configs["pools"]["sharded_db"]["users"]["0"]["pool_size"] = 1
+
+      processes.pgcat.update_config(new_configs)
+      processes.pgcat.reload_config
+
+      25.times do
+        Thread.new do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("SELECT pg_sleep(5)") rescue PG::SystemError
+        ensure
+          conn&.close
+        end
+      end
+
+      sleep(0.5)
+      conn_under_test = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      stdout, stderr = with_captured_stdout_stderr do
+        15.times do |i|
+          conn_under_test.async_exec("SELECT 1") rescue PG::SystemError
+          conn_under_test.exec_params("SELECT #{i} + $1", [i]) rescue PG::SystemError
+          sleep 1
+        end
+      end
+
+      raise StandardError, "Libpq got unexpected messages while idle" if stderr.include?("arrived from server while idle")
+    end
+  end
+
+  describe "Pool recycling after config reload" do
+    let(:processes) { Helpers::Pgcat.three_shard_setup("sharded_db", 5) }
+
+    it "should update pools for new clients and clients that are no longer in transaction" do
+      server_conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      server_conn.async_exec("BEGIN")
+
+      # No config change yet, client should set old configs
+      current_datebase_from_pg = server_conn.async_exec("SELECT current_database();")[0]["current_database"]
+      expect(current_datebase_from_pg).to eq('shard0')
+
+      # Swap shards
+      new_config = processes.pgcat.current_config
+      shard0 = new_config["pools"]["sharded_db"]["shards"]["0"]
+      shard1 = new_config["pools"]["sharded_db"]["shards"]["1"]
+      new_config["pools"]["sharded_db"]["shards"]["0"] = shard1
+      new_config["pools"]["sharded_db"]["shards"]["1"] = shard0
+
+      # Reload config
+      processes.pgcat.update_config(new_config)
+      processes.pgcat.reload_config
+      sleep 0.5
+
+      # Config changed but transaction is in progress, client should set old configs
+      current_datebase_from_pg = server_conn.async_exec("SELECT current_database();")[0]["current_database"]
+      expect(current_datebase_from_pg).to eq('shard0')
+      server_conn.async_exec("COMMIT")
+
+      # Transaction finished, client should get new configs
+      current_datebase_from_pg = server_conn.async_exec("SELECT current_database();")[0]["current_database"]
+      expect(current_datebase_from_pg).to eq('shard1')
+
+      # New connection should get new configs
+      server_conn.close()
+      server_conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      current_datebase_from_pg = server_conn.async_exec("SELECT current_database();")[0]["current_database"]
+      expect(current_datebase_from_pg).to eq('shard1')
+    end
+  end
+
+  describe "Clients closing connection in the middle of transaction" do
+    it "sends a rollback to the server" do
+      conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      conn.async_exec("SET SERVER ROLE to 'primary'")
+      conn.async_exec("BEGIN")
+      conn.close
+
+      expect(processes.primary.count_query("ROLLBACK")).to eq(1)
+    end
+  end
+
+  describe "Server version reporting" do
+    it "reports correct version for normal and admin databases" do
+      server_conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+      expect(server_conn.server_version).not_to eq(0)
+      server_conn.close
+
+      admin_conn = PG::connect(processes.pgcat.admin_connection_string)
+      expect(admin_conn.server_version).not_to eq(0)
+      admin_conn.close
+    end
+  end
+
+  describe "State clearance" do
+    context "session mode" do
+      let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5, "session") }
+
+      it "Clears state before connection checkin" do
+        # Both modes of operation should not raise
+        # ERROR:  prepared statement "prepared_q" already exists
+        15.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("PREPARE prepared_q (int) AS SELECT $1")
+          conn.close
+        end
+
+        conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        initial_value = conn.async_exec("SHOW statement_timeout")[0]["statement_timeout"]
+        conn.async_exec("SET statement_timeout to 1000")
+        current_value = conn.async_exec("SHOW statement_timeout")[0]["statement_timeout"]
+        expect(conn.async_exec("SHOW statement_timeout")[0]["statement_timeout"]).to eq("1s")
+        conn.close
+      end
+
+      it "Does not send DISCARD ALL unless necessary" do
+        10.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("SET SERVER ROLE to 'primary'")
+          conn.async_exec("SELECT 1")
+          conn.close
+        end
+
+        expect(processes.primary.count_query("DISCARD ALL")).to eq(0)
+
+        10.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("SET SERVER ROLE to 'primary'")
+          conn.async_exec("SELECT 1")
+          conn.async_exec("SET statement_timeout to 5000")
+          conn.close
+        end
+
+        expect(processes.primary.count_query("DISCARD ALL")).to eq(10)
+      end
+    end
+
+    context "transaction mode" do
+      let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5, "transaction") }
+      it "Clears state before connection checkin" do
+        # Both modes of operation should not raise
+        # ERROR:  prepared statement "prepared_q" already exists
+        15.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("PREPARE prepared_q (int) AS SELECT $1")
+          conn.close
+        end
+
+        15.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.prepare("prepared_q", "SELECT $1")
+          conn.close
+        end
+      end
+
+      it "Does not send DISCARD ALL unless necessary" do
+        10.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("SET SERVER ROLE to 'primary'")
+          conn.async_exec("SELECT 1")
+          conn.exec_params("SELECT $1", [1])
+          conn.close
+        end
+
+        expect(processes.primary.count_query("DISCARD ALL")).to eq(0)
+
+        10.times do
+          conn = PG::connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+          conn.async_exec("SET SERVER ROLE to 'primary'")
+          conn.async_exec("SELECT 1")
+          conn.async_exec("SET statement_timeout to 5000")
+          conn.close
+        end
+
+        expect(processes.primary.count_query("DISCARD ALL")).to eq(10)
+      end
+    end
+  end
+end

tests/ruby/routing_spec.rb

@@ -0,0 +1,81 @@
+# frozen_string_literal: true
+require_relative 'spec_helper'
+
+
+describe "Routing" do
+  let(:processes) { Helpers::Pgcat.single_shard_setup("sharded_db", 5) }
+  after do
+    processes.all_databases.map(&:reset)
+    processes.pgcat.shutdown
+  end
+
+  describe "SET ROLE" do
+    context "primary" do
+      it "routes queries only to primary" do
+        conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        conn.async_exec("SET SERVER ROLE to 'primary'")
+
+        query_count = 30
+        failed_count = 0
+
+        query_count.times do
+          conn.async_exec("SELECT 1 + 2")
+        rescue
+          failed_count += 1
+        end
+
+        expect(failed_count).to eq(0)
+        processes.replicas.map(&:count_select_1_plus_2).each do |instance_share|
+          expect(instance_share).to eq(0)
+        end
+
+        expect(processes.primary.count_select_1_plus_2).to eq(query_count)
+      end
+    end
+    context "replica" do
+      it "routes queries only to replicas" do
+        conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        conn.async_exec("SET SERVER ROLE to 'replica'")
+
+        expected_share = QUERY_COUNT / processes.replicas.count
+        failed_count = 0
+
+        QUERY_COUNT.times do
+          conn.async_exec("SELECT 1 + 2")
+        rescue
+          failed_count += 1
+        end
+
+        expect(failed_count).to eq(0)
+
+        processes.replicas.map(&:count_select_1_plus_2).each do |instance_share|
+          expect(instance_share).to be_within(expected_share * MARGIN_OF_ERROR).of(expected_share)
+        end
+
+        expect(processes.primary.count_select_1_plus_2).to eq(0)
+      end
+    end
+
+    context "any" do
+      it "routes queries to all instances" do
+        conn = PG.connect(processes.pgcat.connection_string("sharded_db", "sharding_user"))
+        conn.async_exec("SET SERVER ROLE to 'any'")
+
+        expected_share = QUERY_COUNT / processes.all_databases.count
+        failed_count = 0
+
+        QUERY_COUNT.times do
+          conn.async_exec("SELECT 1 + 2")
+        rescue
+          failed_count += 1
+        end
+
+        expect(failed_count).to eq(0)
+
+        processes.all_databases.map(&:count_select_1_plus_2).each do |instance_share|
+          expect(instance_share).to be_within(expected_share * MARGIN_OF_ERROR).of(expected_share)
+        end
+      end
+    end
+  end
+end

tests/ruby/spec_helper.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+require 'pg'
+require_relative 'helpers/pgcat_helper'
+
+QUERY_COUNT = 300
+MARGIN_OF_ERROR = 0.30
+
+def with_captured_stdout_stderr
+  sout = STDOUT.clone
+  serr = STDERR.clone
+  STDOUT.reopen("/tmp/out.txt", "w+")
+  STDERR.reopen("/tmp/err.txt", "w+")
+  STDOUT.sync = true
+  STDERR.sync = true
+  yield
+  return File.read('/tmp/out.txt'), File.read('/tmp/err.txt')
+ensure
+  STDOUT.reopen(sout)
+  STDERR.reopen(serr)
+end

tests/ruby/tests.rb

@@ -1,10 +1,6 @@
 # frozen_string_literal: true
-
-require 'active_record'
 require 'pg'
-require 'toml'
-
-$stdout.sync = true
+require 'active_record'
 
 # Uncomment these two to see all queries.
 # ActiveRecord.verbose_query_logs = true
@@ -115,89 +111,3 @@ begin
 rescue ActiveRecord::StatementInvalid
   puts 'OK'
 end
-
-# Test evil clients
-def poorly_behaved_client
-  conn = PG::connect("postgres://sharding_user:sharding_user@127.0.0.1:6432/sharded_db?application_name=testing_pgcat")
-  conn.async_exec 'BEGIN'
-  conn.async_exec 'SELECT 1'
-
-  conn.close
-  puts 'Bad client ok'
-end
-
-25.times do
-  poorly_behaved_client
-end
-
-
-def test_server_parameters
-  server_conn = PG::connect("postgres://sharding_user:sharding_user@127.0.0.1:6432/sharded_db?application_name=testing_pgcat")
-  raise StandardError, "Bad server version" if server_conn.server_version == 0
-  server_conn.close
-
-  admin_conn = PG::connect("postgres://admin_user:admin_pass@127.0.0.1:6432/pgcat")
-  raise StandardError, "Bad server version" if admin_conn.server_version == 0
-  admin_conn.close
-
-  puts 'Server parameters ok'
-end
-
-
-class ConfigEditor
-  def initialize
-    @original_config_text = File.read('../../.circleci/pgcat.toml')
-    text_to_load = @original_config_text.gsub("5432", "\"5432\"")
-
-    @original_configs = TOML.load(text_to_load)
-  end
-
-  def original_configs
-    TOML.load(TOML::Generator.new(@original_configs).body)
-  end
-
-  def with_modified_configs(new_configs)
-    text_to_write = TOML::Generator.new(new_configs).body
-    text_to_write = text_to_write.gsub("\"5432\"", "5432")
-    File.write('../../.circleci/pgcat.toml', text_to_write)
-    yield
-  ensure
-    File.write('../../.circleci/pgcat.toml', @original_config_text)
-  end
-
-end
-
-
-def test_reload_pool_recycling
-  admin_conn = PG::connect("postgres://admin_user:admin_pass@127.0.0.1:6432/pgcat")
-  server_conn = PG::connect("postgres://sharding_user:sharding_user@127.0.0.1:6432/sharded_db?application_name=testing_pgcat")
-
-  server_conn.async_exec("BEGIN")
-  conf_editor = ConfigEditor.new
-  new_configs = conf_editor.original_configs
-
-  # swap shards
-  new_configs["pools"]["sharded_db"]["shards"]["0"]["database"] = "shard1"
-  new_configs["pools"]["sharded_db"]["shards"]["1"]["database"] = "shard0"
-
-  raise StandardError if server_conn.async_exec("SELECT current_database();")[0]["current_database"] != 'shard0'
-  conf_editor.with_modified_configs(new_configs) { admin_conn.async_exec("RELOAD") }
-  raise StandardError if server_conn.async_exec("SELECT current_database();")[0]["current_database"] != 'shard0'
-  server_conn.async_exec("COMMIT;")
-
-  # Transaction finished, client should get new configs
-  raise StandardError if server_conn.async_exec("SELECT current_database();")[0]["current_database"] != 'shard1'
-  server_conn.close()
-
-  # New connection should get new configs
-  server_conn = PG::connect("postgres://sharding_user:sharding_user@127.0.0.1:6432/sharded_db?application_name=testing_pgcat")
-  raise StandardError if server_conn.async_exec("SELECT current_database();")[0]["current_database"] != 'shard1'
-
-ensure
-  admin_conn.async_exec("RELOAD") # Go back to old state
-  admin_conn.close
-  server_conn.close
-  puts "Pool Recycling okay!"
-end
-
-test_reload_pool_recycling

tests/sharding/query_routing_setup.sql

@@ -70,23 +70,35 @@ GRANT CONNECT ON DATABASE shard2  TO other_user;
 GRANT CONNECT ON DATABASE some_db TO simple_user;
 
 \c shard0
+CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO sharding_user;
 GRANT ALL ON SCHEMA public TO sharding_user;
 GRANT ALL ON TABLE data TO sharding_user;
 GRANT ALL ON SCHEMA public TO other_user;
 GRANT ALL ON TABLE data TO other_user;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO other_user;
 
 \c shard1
+CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO sharding_user;
 GRANT ALL ON SCHEMA public TO sharding_user;
 GRANT ALL ON TABLE data TO sharding_user;
 GRANT ALL ON SCHEMA public TO other_user;
 GRANT ALL ON TABLE data TO other_user;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO other_user;
+
 
 \c shard2
+CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO sharding_user;
 GRANT ALL ON SCHEMA public TO sharding_user;
 GRANT ALL ON TABLE data TO sharding_user;
 GRANT ALL ON SCHEMA public TO other_user;
 GRANT ALL ON TABLE data TO other_user;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO other_user;
 
 \c some_db
+CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+GRANT EXECUTE ON FUNCTION pg_stat_statements_reset TO simple_user;
 GRANT ALL ON SCHEMA public TO simple_user;
 GRANT ALL ON TABLE data TO simple_user;