Salesforce Architecture Variants
Overview
Three validated architecture blueprints for Salesforce integrations: Direct API (simple), Event-Driven (scalable), and Middleware/iPaaS (enterprise). Each pattern addresses different scale, latency, and complexity requirements.
Prerequisites
- Understanding of your data volume and sync frequency requirements
- Decision on unidirectional vs bidirectional data flow
- Knowledge of Salesforce edition (affects available features like CDC)
Variant A: Direct API Integration (Simple)
Best for: MVPs, < 50K records/day, single-direction sync
┌─────────────┐ jsforce ┌─────────────┐
│ Your App │ ──── REST API ──▶ │ Salesforce │
│ (Node.js) │ ◀── SOQL/SOSL ── │ Org │
└─────────────┘ └─────────────┘
Data flow:
- App queries SF via SOQL (polling or on-demand)
- App writes to SF via sObject CRUD
- Scheduled cron for periodic sync
Key Characteristics
- Single jsforce connection per process
- Polling-based reads (cron schedule)
- Direct REST writes
- In-memory or Redis caching for describe/metadata
- Suitable for: internal tools, admin dashboards, simple data sync
Code Pattern
// Cron-based sync — runs every 15 minutes
import cron from 'node-cron';
cron.schedule('*/15 * * * *', async () => {
const conn = await getConnection();
// Fetch recently modified accounts
const accounts = await conn.query(`
SELECT Id, Name, Industry, AnnualRevenue
FROM Account
WHERE LastModifiedDate >= ${fifteenMinutesAgo}
`);
// Sync to local database
for (const account of accounts.records) {
await localDb.upsert('accounts', mapFromSalesforce(account));
}
});
Variant B: Event-Driven Integration (Scalable)
Best for: Real-time sync, 50K-5M records/day, bidirectional flow
┌─────────────┐ ┌─────────────┐
│ Your App │ ◀─── CDC Events ─── │ Salesforce │
│ (listener) │ Change Data Capture │ Org │
│ │ │ │
│ │ ── Bulk API 2.0 ──▶ │ │
│ │ (write-back) │ │
└──────┬──────┘ └─────────────┘
│
┌────▼────┐
│ Queue │ (Redis/SQS/Pub-Sub)
│ (async │
│ writes)│
└─────────┘
Key Characteristics
- CDC for real-time change notifications (no polling waste)
- Bulk API 2.0 for high-volume writes
- Queue-based async processing for write-back
- ReplayId tracking for event resumption
- Suitable for: CRM sync, data warehouse ETL, real-time dashboards
Code Pattern
// Event-driven — near-real-time sync
import { getConnection } from './salesforce/connection';
const conn = await getConnection();
// Subscribe to Account changes via CDC
conn.streaming.topic('/data/AccountChangeEvent').subscribe(async (event) => {
const { changeType, recordIds, changedFields } = event.payload.ChangeEventHeader;
switch (changeType) {
case 'CREATE':
await localDb.insert('accounts', mapFromSalesforce(event.payload));
break;
case 'UPDATE':
await localDb.update('accounts', recordIds[0], mapChangedFields(event.payload, changedFields));
break;
case 'DELETE':
await localDb.delete('accounts', recordIds[0]);
break;
}
});
// Write-back via queue (async, decoupled)
queue.process('sync-to-salesforce', async (job) => {
const conn = await getConnection();
await conn.sobject(job.data.objectType).upsert(
job.data.records,
'External_ID__c'
);
});
Required Salesforce Features
- Change Data Capture (Enterprise Edition+)
- Platform Events (all editions)
- Bulk API 2.0 (all editions with API access)
Variant C: Middleware/iPaaS Integration (Enterprise)
Best for: Multi-system integration, 5M+ records/day, complex transformations
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ Your App │ ── API ──────▶ │ Middleware │ ── REST ────▶ │ Salesforce │
│ │ │ (MuleSoft/ │ │ Org │
│ │ │ Workato/ │ ◀─ CDC ──────│ │
│ │ ◀─ Webhooks ── │ Zapier) │ │ │
└─────────────┘ └──────┬───────┘ └─────────────┘
│
┌─────▼──────┐
│ Other │
│ Systems │
│ (ERP, DW, │
│ Marketing) │
└─────────────┘
Key Characteristics
- Middleware handles transformation, routing, error handling
- Pre-built Salesforce connectors (no custom code for basic sync)
- Visual flow builders for business users
- Enterprise monitoring and audit trails
- Suitable for: multi-system integration, complex business rules, compliance-heavy
iPaaS Options
| Platform | Salesforce Integration | Best For | |----------|----------------------|----------| | MuleSoft | Native (Salesforce owns it) | Enterprise, complex flows | | Heroku Connect | Bi-directional auto-sync | Simple sync to Postgres | | Workato | Pre-built recipes | Business user automation | | Zapier | Trigger-based | Simple 1:1 integrations | | Dell Boomi | Enterprise iPaaS | Legacy system integration |
Decision Matrix
| Factor | Direct API (A) | Event-Driven (B) | Middleware (C) | |--------|---------------|-------------------|----------------| | Records/day | < 50K | 50K-5M | 5M+ | | Latency | Minutes (polling) | Seconds (CDC) | Depends on config | | Direction | Usually one-way | Bidirectional | Multi-directional | | API call efficiency | Medium | High (no polling) | High (batched) | | Complexity | Low | Medium | Low (config-driven) | | Cost | jsforce only | jsforce + queue infra | iPaaS license ($$$) | | SF Edition required | Any with API | Enterprise+ (CDC) | Any | | Team skills | Node.js developers | Node.js + streaming | Business analysts OK |
Migration Path
Variant A (Direct API)
│
│ Growing data volume / need real-time sync
▼
Variant B (Event-Driven)
│
│ Multiple systems / compliance / no-code needed
▼
Variant C (Middleware/iPaaS)
Output
- Architecture variant selected based on requirements
- Integration pattern implemented
- Data flow documented
- Scaling path planned
Error Handling
| Issue | Cause | Solution | |-------|-------|----------| | Polling misses changes | Interval too long | Switch to CDC (Variant B) | | CDC events lost | No replay tracking | Persist last replayId | | API limits exhausted | Polling too frequently | Batch with Collections/Bulk API | | Middleware cost too high | Over-engineered | Start with Variant A or B |
Resources
Next Steps
For common anti-patterns, see salesforce-known-pitfalls.