Add better pooling controls

server/db/pg/driver.ts

@@ -1,7 +1,7 @@
 import { drizzle as DrizzlePostgres } from "drizzle-orm/node-postgres";
-import { Pool } from "pg";
 import { readConfigFile } from "@server/lib/readConfigFile";
 import { withReplicas } from "drizzle-orm/pg-core";
+import { createPool } from "./poolConfig";
 
 function createDb() {
     const config = readConfigFile();
@@ -39,12 +39,17 @@ function createDb() {
 
     // Create connection pools instead of individual connections
     const poolConfig = config.postgres.pool;
-    const primaryPool = new Pool({
+    const maxConnections = poolConfig?.max_connections || 20;
+    const idleTimeoutMs = poolConfig?.idle_timeout_ms || 30000;
+    const connectionTimeoutMs = poolConfig?.connection_timeout_ms || 5000;
+
+    const primaryPool = createPool(
         connectionString,
-        max: poolConfig?.max_connections || 20,
-        idleTimeoutMillis: poolConfig?.idle_timeout_ms || 30000,
-        connectionTimeoutMillis: poolConfig?.connection_timeout_ms || 5000
-    });
+        maxConnections,
+        idleTimeoutMs,
+        connectionTimeoutMs,
+        "primary"
+    );
 
     const replicas = [];
 
@@ -55,14 +60,16 @@ function createDb() {
             })
         );
     } else {
+        const maxReplicaConnections =
+            poolConfig?.max_replica_connections || 20;
         for (const conn of replicaConnections) {
-            const replicaPool = new Pool({
-                connectionString: conn.connection_string,
-                max: poolConfig?.max_replica_connections || 20,
-                idleTimeoutMillis: poolConfig?.idle_timeout_ms || 30000,
-                connectionTimeoutMillis:
-                    poolConfig?.connection_timeout_ms || 5000
-            });
+            const replicaPool = createPool(
+                conn.connection_string,
+                maxReplicaConnections,
+                idleTimeoutMs,
+                connectionTimeoutMs,
+                "replica"
+            );
             replicas.push(
                 DrizzlePostgres(replicaPool, {
                     logger: process.env.QUERY_LOGGING == "true"
@@ -84,4 +91,4 @@ export default db;
 export const primaryDb = db.$primary;
 export type Transaction = Parameters<
     Parameters<(typeof db)["transaction"]>[0]
->[0];
+>[0];

server/db/pg/logsDriver.ts

@@ -1,9 +1,9 @@
 import { drizzle as DrizzlePostgres } from "drizzle-orm/node-postgres";
-import { Pool } from "pg";
 import { readConfigFile } from "@server/lib/readConfigFile";
 import { withReplicas } from "drizzle-orm/pg-core";
 import { build } from "@server/build";
 import { db as mainDb, primaryDb as mainPrimaryDb } from "./driver";
+import { createPool } from "./poolConfig";
 
 function createLogsDb() {
     // Only use separate logs database in SaaS builds
@@ -42,12 +42,17 @@ function createLogsDb() {
 
     // Create separate connection pool for logs database
     const poolConfig = logsConfig?.pool || config.postgres?.pool;
-    const primaryPool = new Pool({
+    const maxConnections = poolConfig?.max_connections || 20;
+    const idleTimeoutMs = poolConfig?.idle_timeout_ms || 30000;
+    const connectionTimeoutMs = poolConfig?.connection_timeout_ms || 5000;
+
+    const primaryPool = createPool(
         connectionString,
-        max: poolConfig?.max_connections || 20,
-        idleTimeoutMillis: poolConfig?.idle_timeout_ms || 30000,
-        connectionTimeoutMillis: poolConfig?.connection_timeout_ms || 5000
-    });
+        maxConnections,
+        idleTimeoutMs,
+        connectionTimeoutMs,
+        "logs-primary"
+    );
 
     const replicas = [];
 
@@ -58,14 +63,16 @@ function createLogsDb() {
             })
         );
     } else {
+        const maxReplicaConnections =
+            poolConfig?.max_replica_connections || 20;
         for (const conn of replicaConnections) {
-            const replicaPool = new Pool({
-                connectionString: conn.connection_string,
-                max: poolConfig?.max_replica_connections || 20,
-                idleTimeoutMillis: poolConfig?.idle_timeout_ms || 30000,
-                connectionTimeoutMillis:
-                    poolConfig?.connection_timeout_ms || 5000
-            });
+            const replicaPool = createPool(
+                conn.connection_string,
+                maxReplicaConnections,
+                idleTimeoutMs,
+                connectionTimeoutMs,
+                "logs-replica"
+            );
             replicas.push(
                 DrizzlePostgres(replicaPool, {
                     logger: process.env.QUERY_LOGGING == "true"
@@ -84,4 +91,4 @@ function createLogsDb() {
 
 export const logsDb = createLogsDb();
 export default logsDb;
-export const primaryLogsDb = logsDb.$primary;
+export const primaryLogsDb = logsDb.$primary;

server/db/pg/poolConfig.ts

@@ -0,0 +1,63 @@
+import { Pool, PoolConfig } from "pg";
+import logger from "@server/logger";
+
+export function createPoolConfig(
+    connectionString: string,
+    maxConnections: number,
+    idleTimeoutMs: number,
+    connectionTimeoutMs: number
+): PoolConfig {
+    return {
+        connectionString,
+        max: maxConnections,
+        idleTimeoutMillis: idleTimeoutMs,
+        connectionTimeoutMillis: connectionTimeoutMs,
+        // TCP keepalive to prevent silent connection drops by NAT gateways,
+        // load balancers, and other intermediate network devices (e.g. AWS
+        // NAT Gateway drops idle TCP connections after ~350s)
+        keepAlive: true,
+        keepAliveInitialDelayMillis: 10000, // send first keepalive after 10s of idle
+        // Allow connections to be released and recreated more aggressively
+        // to avoid stale connections building up
+        allowExitOnIdle: false
+    };
+}
+
+export function attachPoolErrorHandlers(pool: Pool, label: string): void {
+    pool.on("error", (err) => {
+        // This catches errors on idle clients in the pool. Without this
+        // handler an unexpected disconnect would crash the process.
+        logger.error(
+            `Unexpected error on idle ${label} database client: ${err.message}`
+        );
+    });
+
+    pool.on("connect", (client) => {
+        // Set a statement timeout on every new connection so a single slow
+        // query can't block the pool forever
+        client.query("SET statement_timeout = '30s'").catch((err: Error) => {
+            logger.warn(
+                `Failed to set statement_timeout on ${label} client: ${err.message}`
+            );
+        });
+    });
+}
+
+export function createPool(
+    connectionString: string,
+    maxConnections: number,
+    idleTimeoutMs: number,
+    connectionTimeoutMs: number,
+    label: string
+): Pool {
+    const pool = new Pool(
+        createPoolConfig(
+            connectionString,
+            maxConnections,
+            idleTimeoutMs,
+            connectionTimeoutMs
+        )
+    );
+    attachPoolErrorHandlers(pool, label);
+    return pool;
+}