Connection Pooling Guide

How to efficiently manage connections to Kimberlite clusters across different SDKs.

Overview

Connection pooling improves performance by reusing connections instead of creating new ones for each operation. This is especially important for:

• Web applications with concurrent requests
• Microservices with high throughput
• Long-running batch processors

Python SDK

Python SDK handles connection pooling internally through the FFI layer.

Basic Pattern

from kimberlite import Client

# Create a single client instance for your application
client = Client.connect(
    addresses=["localhost:5432", "localhost:5433"],
    tenant_id=1,
    auth_token="token"
)

# Reuse this client across multiple operations
def handle_request(data):
    stream_id = client.create_stream("events", DataClass.NON_PHI)
    client.append(stream_id, [data])

With Flask

from flask import Flask, request
from kimberlite import Client, DataClass

app = Flask(__name__)

# Initialize client once at startup
client = None

@app.before_first_request
def init_client():
    global client
    client = Client.connect(
        addresses=["localhost:5432"],
        tenant_id=1
    )

@app.route('/events', methods=['POST'])
def post_event():
    data = request.get_json()
    stream_id = int(data['stream_id'])
    events = [e.encode('utf-8') for e in data['events']]

    offset = client.append(stream_id, events)
    return {'offset': int(offset)}

@app.teardown_appcontext
def cleanup(exc):
    if client:
        client.disconnect()

With Django

# settings.py
from kimberlite import Client

KIMBERLITE_CLIENT = None

def get_kimberlite_client():
    global KIMBERLITE_CLIENT
    if KIMBERLITE_CLIENT is None:
        KIMBERLITE_CLIENT = Client.connect(
            addresses=["localhost:5432"],
            tenant_id=1
        )
    return KIMBERLITE_CLIENT

# views.py
from django.conf import settings

def my_view(request):
    client = settings.get_kimberlite_client()
    # Use client...

TypeScript SDK

TypeScript SDK manages connections at the client level.

Basic Pattern

import { Client, DataClass } from '@kimberlite/client';

// Create client once
const client = await Client.connect({
  addresses: ['localhost:5432', 'localhost:5433'],
  tenantId: 1n,
  authToken: 'token'
});

// Reuse across operations
async function handleRequest(data: Buffer) {
  const streamId = await client.createStream('events', DataClass.NonPHI);
  await client.append(streamId, [data]);
}

With Express.js

import express from 'express';
import { Client } from '@kimberlite/client';

const app = express();
let client: Client;

// Initialize on startup
async function init() {
  client = await Client.connect({
    addresses: ['localhost:5432'],
    tenantId: 1n
  });
}

app.post('/events', async (req, res) => {
  try {
    const streamId = BigInt(req.body.stream_id);
    const events = req.body.events.map((e: string) => Buffer.from(e));
    const offset = await client.append(streamId, events);
    res.json({ offset: offset.toString() });
  } catch (error) {
    res.status(500).json({ error: (error as Error).message });
  }
});

// Graceful shutdown
process.on('SIGTERM', async () => {
  await client.disconnect();
  process.exit(0);
});

init().then(() => {
  app.listen(3000);  // Express.js app port (NOT Kimberlite server - that's :5432)
});

With NestJS

import { Injectable, OnModuleInit, OnModuleDestroy } from '@nestjs/common';
import { Client } from '@kimberlite/client';

@Injectable()
export class KimberliteService implements OnModuleInit, OnModuleDestroy {
  private client: Client;

  async onModuleInit() {
    this.client = await Client.connect({
      addresses: ['localhost:5432'],
      tenantId: 1n
    });
  }

  async onModuleDestroy() {
    await this.client.disconnect();
  }

  async append(streamId: bigint, events: Buffer[]) {
    return await this.client.append(streamId, events);
  }
}

Rust SDK

Rust SDK uses synchronous connections with Send + Sync for thread safety.

Basic Pattern

use kimberlite::Kimberlite;
use kimberlite_types::TenantId;
use std::sync::Arc;

// Create once and share via Arc
let db = Arc::new(Kimberlite::open("./data")?);
let tenant = db.tenant(TenantId::new(1));

// Clone Arc for each thread
let tenant_clone = tenant.clone();
std::thread::spawn(move || {
    tenant_clone.append(stream_id, events)?;
});

With Actix Web

use actix_web::{web, App, HttpServer};
use kimberlite::Kimberlite;
use std::sync::Arc;

#[actix_web::main]
async fn main() -> std::io::Result<()> {
    // Initialize once
    let db = Arc::new(Kimberlite::open("./data").unwrap());

    HttpServer::new(move || {
        App::new()
            .app_data(web::Data::new(db.clone()))
            .route("/events", web::post().to(post_event))
    })
    .bind(("127.0.0.1", 8080))?
    .run()
    .await
}

async fn post_event(
    db: web::Data<Arc<Kimberlite>>,
    req: web::Json<EventRequest>,
) -> impl Responder {
    let tenant = db.tenant(TenantId::new(1));
    let offset = tenant.append(req.stream_id, req.events.clone())?;
    web::Json(json!({ "offset": offset }))
}

Best Practices

1. Single Client per Application

Create one client instance and reuse it:

# ✅ GOOD
client = Client.connect(...)
for i in range(1000):
    client.append(stream_id, [data])

# ❌ BAD
for i in range(1000):
    client = Client.connect(...)
    client.append(stream_id, [data])
    client.disconnect()

2. Graceful Shutdown

Always disconnect on application shutdown:

process.on('SIGTERM', async () => {
  await client.disconnect();
  process.exit(0);
});

3. Health Checks

Implement periodic health checks:

import threading
import time

def health_check():
    while True:
        try:
            # Ping operation
            client.read(health_stream_id, from_offset=0, max_bytes=1)
        except Exception as e:
            logger.error(f"Health check failed: {e}")
            # Reconnect logic here
        time.sleep(30)

threading.Thread(target=health_check, daemon=True).start()

4. Error Recovery

Implement retry logic with exponential backoff:

async function withRetry<T>(
  fn: () => Promise<T>,
  maxRetries = 3
): Promise<T> {
  let lastError;
  for (let i = 0; i < maxRetries; i++) {
    try {
      return await fn();
    } catch (error) {
      lastError = error;
      await new Promise(resolve => setTimeout(resolve, Math.pow(2, i) * 100));
    }
  }
  throw lastError;
}

const offset = await withRetry(() => client.append(streamId, events));

Multi-Cluster Setup

For high availability, connect to multiple cluster addresses:

client = Client.connect(
    addresses=[
        "cluster1.example.com:5432",
        "cluster2.example.com:5432",
        "cluster3.example.com:5432",
    ],
    tenant_id=1
)

The client will automatically:

• Discover the cluster leader
• Failover to a new leader if needed
• Retry on transient failures

Performance Tips

1. Batch Operations: Group multiple appends into single calls
2. Concurrent Reads: Multiple read operations can run in parallel
3. Connection Limits: Don’t create more clients than needed
4. Keep-Alive: FFI layer maintains persistent connections
5. Resource Cleanup: Always disconnect when done

Monitoring

Track connection metrics:

import time

class MonitoredClient:
    def __init__(self, client):
        self.client = client
        self.operations = 0
        self.errors = 0
        self.start_time = time.time()

    def append(self, stream_id, events):
        try:
            result = self.client.append(stream_id, events)
            self.operations += 1
            return result
        except Exception as e:
            self.errors += 1
            raise

    def stats(self):
        elapsed = time.time() - self.start_time
        return {
            'operations': self.operations,
            'errors': self.errors,
            'ops_per_sec': self.operations / elapsed if elapsed > 0 else 0,
            'error_rate': self.errors / self.operations if self.operations > 0 else 0
        }

See Also

• Quickstart Guide - Python
• Quickstart Guide - TypeScript
• Protocol Specification