Choosing Synchronization… — Data Synchronization… | OneNoughtOne

0/105

System Design (HLD)Distributed Data Patterns

Data Synchronization Patterns

LevelAdvanced

Duration90 mins

TopicDistributed Data Patterns

5 / 5

Choosing Synchronization Approach

The Architecture Decision That Defines Your System

We've explored a spectrum of synchronization approaches: from strongly consistent 2PC to non-blocking 3PC, from consensus protocols to async replication. Each has its strengths and trade-offs. The question now is: how do you choose the right one for your system? This isn't an academic exercise. The synchronization strategy you select fundamentally shapes your system's behavior under load, during failures, and across geographic distances. Choose wrong, and you'll either pay unnecessary latency costs or face data integrity issues in production. This page provides a structured framework for making this critical decision.

What You Will Learn

By the end of this page, you will have a decision framework for evaluating synchronization approaches, understand the key questions to ask about your requirements, and be able to select and justify the appropriate strategy for different scenarios.

The Fundamental Trade-offs

Before diving into specific approaches, understand that every synchronization choice is a trade-off . There is no universally optimal solution. The CAP Theorem Reminder: During a network partition, you must choose between: Consistency — All nodes see the same data at the same time Availability — Every request receives a response The PACELC Extension: Even without partitions (normal operation), there's a trade-off: Latency — Time to complete operations Consistency — Strength of consistency guarantees Strong consistency typically requires coordination, which adds latency.

Synchronization Approaches: Core Trade-offs
Approach	Consistency	Availability	Latency	Durability
2PC	Strong	Low (blocking)	High	High
3PC	Weak (partition)	Medium	Higher	High
Paxos/Raft	Strong (majority)	High (majority)	Medium	Majority
Async Primary-Replica	Eventual	Very High	Very Low	Risk of loss
Semi-Sync	Eventual	High	Low-Medium	Configurable
Multi-Primary	Eventual	Very High	Very Low	Conflict risk

No Silver Bullet

Notice that no approach excels in all dimensions. The goal isn't to find the 'best' approach—it's to find the approach whose trade-offs align with your requirements.

Key Questions to Ask

Before selecting a synchronization approach, systematically answer these questions:

Requirement Analysis Questions

•Q1: What happens if data is temporarily inconsistent? Financial transactions require strong consistency—a balance must never show two different values. Social media likes can tolerate brief inconsistency.
•Q2: What happens if the system is unavailable? E-commerce checkout must keep working; billing reports can wait. Identify which operations are availability-critical.
•Q3: What are your latency requirements? Real-time gaming needs <50ms; batch processing can tolerate seconds. Synchronous replication adds latency proportional to network distance.
•Q4: How is your data geographically distributed? Single datacenter vs. multi-region changes everything. Cross-region synchronous replication adds 50-200ms per round trip.
•Q5: What failure modes must you survive? Single node failure? Datacenter failure? Network partition? Each requires different redundancy strategies.
•Q6: What is your write volume and pattern? High write volume stresses synchronous protocols. Write conflicts in multi-primary increase with write rate.

requirement-scoring.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
// Requirement Scoring Framework
 
interface SystemRequirements {
    // Consistency requirements (1-5 scale)
    consistencyNeeded: 1 | 2 | 3 | 4 | 5;  // 1=eventual, 5=strong
    
    // Availability requirements (1-5 scale)  
    availabilityNeeded: 1 | 2 | 3 | 4 | 5;  // 1=can tolerate downtime, 5=must always work
    
    // Latency requirements
    maxWriteLatencyMs: number;
    
    // Geographic distribution
    deploymentScope: 'single-dc' | 'multi-dc-same-region' | 'multi-region' | 'global';
    
    // Failure tolerance
    tolerateNodeFailure: boolean;
    tolerateDatacenterFailure: boolean;
    tolerateNetworkPartition: boolean;
    
    // Write characteristics
    writesPerSecond: number;
    writeConflictLikelihood: 'low' | 'medium' | 'high';
}
 
function recommendSyncApproach(req: SystemRequirements): string {
    // Strong consistency, single DC → 2PC or consensus
    if (req.consistencyNeeded >= 4 && req.deploymentScope === 'single-dc') {
        if (req.availabilityNeeded >= 4) {
            return 'Consensus (Raft) - Strong consistency with majority availability';
        }
        return '2PC - Strong consistency, acceptable blocking risk';
    }
    
    // Strong consistency, multi-region → Consensus with geographic awareness
    if (req.consistencyNeeded >= 4 && req.deploymentScope !== 'single-dc') {
        return 'Geo-distributed Consensus (Spanner-style) - Accept latency for consistency';
    }
    
    // High availability, eventual consistency OK → Async replication
    if (req.availabilityNeeded >= 4 && req.consistencyNeeded <= 2) {
        if (req.deploymentScope === 'global' && req.writeConflictLikelihood === 'low') {
            return 'Multi-Primary with LWW - Global writes, eventual consistency';
        }
        return 'Async Primary-Replica - Simple, fast, eventually consistent';
    }
    
    // Balanced requirements → Semi-sync
    if (req.consistencyNeeded === 3 && req.availabilityNeeded === 3) {
        return 'Semi-Sync Replication - Balance of durability and latency';
    }
    
    // Default
    return 'Evaluate specific trade-offs - requirements are complex';
}

The Decision Framework

Use this decision tree to narrow down your options: START │ ▼ ┌─────────────────────────────┐ │ Do you need STRONG │ │ consistency (serializable)? │ └─────────────┬───────────────┘ │ ┌───────────┴───────────┐ │ │ YES NO │ │ ▼ ▼ ┌──────────────┐ ┌──────────────────┐ │ Geographic │ │ Write conflicts │ │ distribution? │ │ possible? │ └───────┬──────┘ └────────┬─────────┘ │ │ ┌──────┴──────┐ ┌──────┴──────┐ │ │ │ │ Single Multi YES NO DC Region │ │ │ │ ▼ ▼ │ │ Multi-Primary Async Primary- │ │ + Conflict Replica │ │ Resolution ▼ ▼ 2PC or Spanner- Raft style (Paxos + TrueTime)

When to Use Each Approach
Approach	Best For	Avoid When
2PC	Distributed transactions within single DC	High availability needed, WAN latency
Consensus (Raft)	Replicated state machines, leader election, metadata stores	Very high write throughput (leader bottleneck)
Primary-Replica Async	Read-heavy workloads, analytics replicas	Zero data loss required
Semi-Sync	Balance of durability and latency	Global distribution (latency too high)
Multi-Primary	Global writes, offline-first apps	High conflict rate, complex merge logic

Real-World Architecture Patterns

In practice, systems often combine multiple approaches for different data types:

Common Hybrid Architectures

•E-commerce: Consensus for inventory/payments (strong consistency), async replication for product catalog and reviews (eventual consistency)
•Social Media: Multi-primary for posts/likes (availability), consensus for auth tokens and account data (consistency)
•Gaming: Async for leaderboards and non-critical state, strong consistency for purchases and matchmaking
•Banking: 2PC/consensus for all transactions within regions, async batch reconciliation across regions

hybrid-sync.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
// Hybrid Synchronization Strategy Example
 
interface DataClassification {
    type: 'critical' | 'important' | 'best-effort';
    syncStrategy: SyncStrategy;
}
 
const dataClassifications: Record<string, DataClassification> = {
    // Critical: Strong consistency required
    'account_balance': { type: 'critical', syncStrategy: 'consensus' },
    'payment_transactions': { type: 'critical', syncStrategy: '2pc' },
    'inventory_count': { type: 'critical', syncStrategy: 'consensus' },
    
    // Important: Durability required, eventual consistency OK
    'order_history': { type: 'important', syncStrategy: 'semi-sync' },
    'user_profile': { type: 'important', syncStrategy: 'semi-sync' },
    
    // Best-effort: Availability over consistency
    'product_reviews': { type: 'best-effort', syncStrategy: 'async' },
    'page_views': { type: 'best-effort', syncStrategy: 'async' },
    'recommendations': { type: 'best-effort', syncStrategy: 'async' },
};
 
class HybridDataStore {
    private consensusStore: RaftCluster;
    private asyncStore: AsyncReplicatedDB;
    private twoPhaseCoordinator: TwoPhaseCommitCoordinator;
 
    async write(table: string, data: Record<string, any>): Promise<void> {
        const classification = dataClassifications[table];
        
        switch (classification?.syncStrategy) {
            case 'consensus':
                await this.consensusStore.write(table, data);
                break;
            case '2pc':
                await this.twoPhaseCoordinator.execute([
                    { table, operation: 'insert', data }
                ]);
                break;
            case 'semi-sync':
                await this.asyncStore.writeSemiSync(table, data);
                break;
            case 'async':
            default:
                await this.asyncStore.writeAsync(table, data);
                break;
        }
    }
}

Classify Your Data

The first step in choosing synchronization is classifying your data. Ask: 'What's the cost of this data being wrong or lost?' Financial data: very high. Cached recommendations: very low. Match the protection level to the cost.

Module Summary: Data Synchronization Patterns

We've covered the complete spectrum of data synchronization patterns in distributed systems:

Module Takeaways

•Two-Phase Commit provides atomic transactions but blocks under coordinator failure
•Three-Phase Commit eliminates blocking but fails under network partitions
•Consensus protocols (Paxos/Raft) enable fault-tolerant agreement with majority liveness
•Async replication trades consistency for availability and low latency
•No single approach is universally best — choose based on your specific requirements
•Hybrid approaches combine strategies based on data classification

Quick Reference: Synchronization Selection
If you need...	Consider...
Distributed transactions, single DC	2PC or Saga pattern
Replicated state machine	Raft consensus
Global strong consistency	Spanner-style (Paxos + TrueTime)
Maximum availability	Async replication + eventual consistency
Global writes without conflicts	CRDTs or careful multi-primary
Balanced durability/latency	Semi-synchronous replication

Module Complete

Congratulations! You've completed the Data Synchronization Patterns module. You now understand the full spectrum of distributed synchronization approaches—from classical protocols to modern consensus and async replication. This knowledge is foundational for designing any distributed system that manages data across multiple nodes.

System Design (HLD)Distributed Data Patterns

Data Synchronization Patterns

LevelAdvanced

Duration90 mins

TopicDistributed Data Patterns

5 / 5

Choosing Synchronization Approach

The Architecture Decision That Defines Your System

We've explored a spectrum of synchronization approaches: from strongly consistent 2PC to non-blocking 3PC, from consensus protocols to async replication. Each has its strengths and trade-offs. The question now is: how do you choose the right one for your system? This isn't an academic exercise. The synchronization strategy you select fundamentally shapes your system's behavior under load, during failures, and across geographic distances. Choose wrong, and you'll either pay unnecessary latency costs or face data integrity issues in production. This page provides a structured framework for making this critical decision.

What You Will Learn

By the end of this page, you will have a decision framework for evaluating synchronization approaches, understand the key questions to ask about your requirements, and be able to select and justify the appropriate strategy for different scenarios.

The Fundamental Trade-offs

Before diving into specific approaches, understand that every synchronization choice is a trade-off . There is no universally optimal solution. The CAP Theorem Reminder: During a network partition, you must choose between: Consistency — All nodes see the same data at the same time Availability — Every request receives a response The PACELC Extension: Even without partitions (normal operation), there's a trade-off: Latency — Time to complete operations Consistency — Strength of consistency guarantees Strong consistency typically requires coordination, which adds latency.

Synchronization Approaches: Core Trade-offs
Approach	Consistency	Availability	Latency	Durability
2PC	Strong	Low (blocking)	High	High
3PC	Weak (partition)	Medium	Higher	High
Paxos/Raft	Strong (majority)	High (majority)	Medium	Majority
Async Primary-Replica	Eventual	Very High	Very Low	Risk of loss
Semi-Sync	Eventual	High	Low-Medium	Configurable
Multi-Primary	Eventual	Very High	Very Low	Conflict risk

No Silver Bullet

Notice that no approach excels in all dimensions. The goal isn't to find the 'best' approach—it's to find the approach whose trade-offs align with your requirements.

Key Questions to Ask

Before selecting a synchronization approach, systematically answer these questions:

Requirement Analysis Questions

•Q1: What happens if data is temporarily inconsistent? Financial transactions require strong consistency—a balance must never show two different values. Social media likes can tolerate brief inconsistency.
•Q2: What happens if the system is unavailable? E-commerce checkout must keep working; billing reports can wait. Identify which operations are availability-critical.
•Q3: What are your latency requirements? Real-time gaming needs <50ms; batch processing can tolerate seconds. Synchronous replication adds latency proportional to network distance.
•Q4: How is your data geographically distributed? Single datacenter vs. multi-region changes everything. Cross-region synchronous replication adds 50-200ms per round trip.
•Q5: What failure modes must you survive? Single node failure? Datacenter failure? Network partition? Each requires different redundancy strategies.
•Q6: What is your write volume and pattern? High write volume stresses synchronous protocols. Write conflicts in multi-primary increase with write rate.

requirement-scoring.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
// Requirement Scoring Framework
 
interface SystemRequirements {
    // Consistency requirements (1-5 scale)
    consistencyNeeded: 1 | 2 | 3 | 4 | 5;  // 1=eventual, 5=strong
    
    // Availability requirements (1-5 scale)  
    availabilityNeeded: 1 | 2 | 3 | 4 | 5;  // 1=can tolerate downtime, 5=must always work
    
    // Latency requirements
    maxWriteLatencyMs: number;
    
    // Geographic distribution
    deploymentScope: 'single-dc' | 'multi-dc-same-region' | 'multi-region' | 'global';
    
    // Failure tolerance
    tolerateNodeFailure: boolean;
    tolerateDatacenterFailure: boolean;
    tolerateNetworkPartition: boolean;
    
    // Write characteristics
    writesPerSecond: number;
    writeConflictLikelihood: 'low' | 'medium' | 'high';
}
 
function recommendSyncApproach(req: SystemRequirements): string {
    // Strong consistency, single DC → 2PC or consensus
    if (req.consistencyNeeded >= 4 && req.deploymentScope === 'single-dc') {
        if (req.availabilityNeeded >= 4) {
            return 'Consensus (Raft) - Strong consistency with majority availability';
        }
        return '2PC - Strong consistency, acceptable blocking risk';
    }
    
    // Strong consistency, multi-region → Consensus with geographic awareness
    if (req.consistencyNeeded >= 4 && req.deploymentScope !== 'single-dc') {
        return 'Geo-distributed Consensus (Spanner-style) - Accept latency for consistency';
    }
    
    // High availability, eventual consistency OK → Async replication
    if (req.availabilityNeeded >= 4 && req.consistencyNeeded <= 2) {
        if (req.deploymentScope === 'global' && req.writeConflictLikelihood === 'low') {
            return 'Multi-Primary with LWW - Global writes, eventual consistency';
        }
        return 'Async Primary-Replica - Simple, fast, eventually consistent';
    }
    
    // Balanced requirements → Semi-sync
    if (req.consistencyNeeded === 3 && req.availabilityNeeded === 3) {
        return 'Semi-Sync Replication - Balance of durability and latency';
    }
    
    // Default
    return 'Evaluate specific trade-offs - requirements are complex';
}

The Decision Framework

Use this decision tree to narrow down your options: START │ ▼ ┌─────────────────────────────┐ │ Do you need STRONG │ │ consistency (serializable)? │ └─────────────┬───────────────┘ │ ┌───────────┴───────────┐ │ │ YES NO │ │ ▼ ▼ ┌──────────────┐ ┌──────────────────┐ │ Geographic │ │ Write conflicts │ │ distribution? │ │ possible? │ └───────┬──────┘ └────────┬─────────┘ │ │ ┌──────┴──────┐ ┌──────┴──────┐ │ │ │ │ Single Multi YES NO DC Region │ │ │ │ ▼ ▼ │ │ Multi-Primary Async Primary- │ │ + Conflict Replica │ │ Resolution ▼ ▼ 2PC or Spanner- Raft style (Paxos + TrueTime)

When to Use Each Approach
Approach	Best For	Avoid When
2PC	Distributed transactions within single DC	High availability needed, WAN latency
Consensus (Raft)	Replicated state machines, leader election, metadata stores	Very high write throughput (leader bottleneck)
Primary-Replica Async	Read-heavy workloads, analytics replicas	Zero data loss required
Semi-Sync	Balance of durability and latency	Global distribution (latency too high)
Multi-Primary	Global writes, offline-first apps	High conflict rate, complex merge logic

Real-World Architecture Patterns

In practice, systems often combine multiple approaches for different data types:

Common Hybrid Architectures

•E-commerce: Consensus for inventory/payments (strong consistency), async replication for product catalog and reviews (eventual consistency)
•Social Media: Multi-primary for posts/likes (availability), consensus for auth tokens and account data (consistency)
•Gaming: Async for leaderboards and non-critical state, strong consistency for purchases and matchmaking
•Banking: 2PC/consensus for all transactions within regions, async batch reconciliation across regions

hybrid-sync.ts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
// Hybrid Synchronization Strategy Example
 
interface DataClassification {
    type: 'critical' | 'important' | 'best-effort';
    syncStrategy: SyncStrategy;
}
 
const dataClassifications: Record<string, DataClassification> = {
    // Critical: Strong consistency required
    'account_balance': { type: 'critical', syncStrategy: 'consensus' },
    'payment_transactions': { type: 'critical', syncStrategy: '2pc' },
    'inventory_count': { type: 'critical', syncStrategy: 'consensus' },
    
    // Important: Durability required, eventual consistency OK
    'order_history': { type: 'important', syncStrategy: 'semi-sync' },
    'user_profile': { type: 'important', syncStrategy: 'semi-sync' },
    
    // Best-effort: Availability over consistency
    'product_reviews': { type: 'best-effort', syncStrategy: 'async' },
    'page_views': { type: 'best-effort', syncStrategy: 'async' },
    'recommendations': { type: 'best-effort', syncStrategy: 'async' },
};
 
class HybridDataStore {
    private consensusStore: RaftCluster;
    private asyncStore: AsyncReplicatedDB;
    private twoPhaseCoordinator: TwoPhaseCommitCoordinator;
 
    async write(table: string, data: Record<string, any>): Promise<void> {
        const classification = dataClassifications[table];
        
        switch (classification?.syncStrategy) {
            case 'consensus':
                await this.consensusStore.write(table, data);
                break;
            case '2pc':
                await this.twoPhaseCoordinator.execute([
                    { table, operation: 'insert', data }
                ]);
                break;
            case 'semi-sync':
                await this.asyncStore.writeSemiSync(table, data);
                break;
            case 'async':
            default:
                await this.asyncStore.writeAsync(table, data);
                break;
        }
    }
}

Classify Your Data

The first step in choosing synchronization is classifying your data. Ask: 'What's the cost of this data being wrong or lost?' Financial data: very high. Cached recommendations: very low. Match the protection level to the cost.

Module Summary: Data Synchronization Patterns

We've covered the complete spectrum of data synchronization patterns in distributed systems:

Module Takeaways

•Two-Phase Commit provides atomic transactions but blocks under coordinator failure
•Three-Phase Commit eliminates blocking but fails under network partitions
•Consensus protocols (Paxos/Raft) enable fault-tolerant agreement with majority liveness
•Async replication trades consistency for availability and low latency
•No single approach is universally best — choose based on your specific requirements
•Hybrid approaches combine strategies based on data classification

Quick Reference: Synchronization Selection
If you need...	Consider...
Distributed transactions, single DC	2PC or Saga pattern
Replicated state machine	Raft consensus
Global strong consistency	Spanner-style (Paxos + TrueTime)
Maximum availability	Async replication + eventual consistency
Global writes without conflicts	CRDTs or careful multi-primary
Balanced durability/latency	Semi-synchronous replication

Module Complete

Congratulations! You've completed the Data Synchronization Patterns module. You now understand the full spectrum of distributed synchronization approaches—from classical protocols to modern consensus and async replication. This knowledge is foundational for designing any distributed system that manages data across multiple nodes.

Choosing Synchronization Approach

The Architecture Decision That Defines Your System

The Fundamental Trade-offs

Key Questions to Ask

The Decision Framework

Real-World Architecture Patterns

Module Summary: Data Synchronization Patterns

There’s more to understand.

Choosing Synchronization Approach

The Architecture Decision That Defines Your System

The Fundamental Trade-offs

Key Questions to Ask

The Decision Framework

Real-World Architecture Patterns

Module Summary: Data Synchronization Patterns

There’s more to understand.