How Write-Around Caching Works - Write-Around Caching | OneNoughtOne

0/319

System Design (HLD)Caching Strategies

Write-Around Caching

LevelIntermediate

Duration55 mins

TopicCaching Strategies

1 / 5

How Write-Around Caching Works

The Cache-Bypass Philosophy

In the landscape of caching strategies, write-around caching represents a deliberate philosophical choice: not all data deserves immediate cache residency. Unlike write-through and write-back strategies that ensure written data lives in the cache, write-around takes a fundamentally different approach—it bypasses the cache entirely for write operations, sending data directly to the persistent storage layer.

This might initially seem counterintuitive. Isn't the whole point of caching to keep frequently accessed data in fast storage? The insight behind write-around is that writes don't predict reads. Just because data was written doesn't mean it will be read soon—or ever. Write-around recognizes this reality and optimizes accordingly.

What You Will Learn

By the end of this page, you will understand the complete mechanics of write-around caching—how data flows through the system, why the cache is bypassed during writes, how reads trigger cache population, and the architectural patterns that emerge from this strategy. You'll develop deep intuition for when write-around is the optimal choice.

Fundamental Mechanics of Write-Around

Write-around caching implements a simple but powerful principle: writes go to the database; reads populate the cache. This creates an asymmetric data flow where the cache serves purely as a read accelerator rather than a write buffer.

The Core Data Flow:

Understanding write-around requires visualizing two distinct paths—the write path and the read path—that operate independently of each other.

Write Path (Cache Bypass)

•Application issues write request — Create, update, or delete operation initiated
•Request goes directly to database — Cache is completely bypassed, no cache interaction
•Database persists the data — Storage layer handles durability
•Confirmation returned to application — Write is complete
•Cache remains unchanged — Existing cache entries untouched (potential staleness)

Read Path (Cache Population)

•Application issues read request — Data retrieval operation initiated
•Cache checked first — Look for existing cached entry (cache hit/miss)
•On cache miss → Database query — Fetch data from persistent storage
•Data loaded into cache — Populate cache with retrieved data
•Return data to application — Subsequent reads served from cache

The Key Insight:

The write path and read path are completely decoupled. A write operation has no effect on the cache—it doesn't add data, doesn't update existing entries, and doesn't invalidate stale data. Only read operations interact with the cache, and only when data isn't already present (cache miss) does the cache get populated.

This decoupling is both the strength and the weakness of write-around caching. It prevents cache pollution from transient writes but creates the potential for stale reads immediately after writes.

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
class WriteAroundCache<T> {
  constructor(
    private cache: CacheStore<T>,
    private database: Database<T>
  ) {}
 
  // Write path: Database only, cache bypassed entirely
  async write(key: string, value: T): Promise<void> {
    await this.database.write(key, value);
    // Note: No cache interaction whatsoever
    // Cache may now contain stale data for this key!
  }
 
  // Read path: Check cache first, populate on miss
  async read(key: string): Promise<T | null> {
    // Step 1: Try cache first
    const cachedValue = await this.cache.get(key);
    if (cachedValue !== null) {
      return cachedValue; // Cache hit - serve from cache
    }
 
    // Step 2: Cache miss - fetch from database
    const dbValue = await this.database.read(key);
    if (dbValue !== null) {
      // Step 3: Populate cache for future reads
      await this.cache.set(key, dbValue);
    }
    
    return dbValue;
  }
 
  // Delete path: Remove from database (optionally invalidate cache)
  async delete(key: string): Promise<void> {
    await this.database.delete(key);
    // Optional: Invalidate cache to prevent stale reads
    await this.cache.delete(key);
  }
}

The Staleness Window

Notice that after a write, the cache may contain stale data. If key 'user:123' is in the cache with value A, and we write value B using write-around, the cache still contains A. Any reads before the cache entry expires will return stale data. This is the fundamental trade-off of write-around—you must decide if this staleness window is acceptable for your use case.

Architectural Data Flow Patterns

Write-around caching creates distinct architectural patterns based on how the system handles the interplay between written data, cached data, and read requests. Understanding these patterns is essential for designing robust systems.

Pattern 1: Pure Write-Around (No Invalidation)

In its purest form, write-around never touches the cache during writes. This creates maximum write performance but relies entirely on TTL (Time-To-Live) for eventual consistency.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
// Pure Write-Around: Maximum write performance
async function pureWriteAround(key: string, data: Data): Promise<void> {
  // Single database write - cache completely ignored
  await database.write(key, data);
  
  // Staleness resolved by:
  // 1. TTL expiration on cache entries (eventually consistent)
  // 2. Natural read patterns bringing fresh data
  // 3. Background refresh processes (if implemented)
}
 
// Implications:
// - Writes: O(1) database operation only
// - Reads after write: May return stale data until TTL expires
// - Best for: Write-heavy workloads with tolerance for eventual consistency

Pattern 2: Write-Around with Invalidation

A more conservative approach adds cache invalidation to the write path. While still bypassing cache population, it removes stale entries to ensure subsequent reads fetch fresh data from the database.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
// Write-Around with Invalidation: Balance of performance and consistency
async function writeAroundWithInvalidation(key: string, data: Data): Promise<void> {
  // Step 1: Write to database
  await database.write(key, data);
  
  // Step 2: Invalidate cache entry (but don't populate)
  await cache.delete(key); // Fast O(1) operation
  
  // Next read will be a cache miss → fetch from DB → populate cache
}
 
// Implications:
// - Writes: O(1) database + O(1) cache delete
// - Reads after write: Cache miss guaranteed, fresh data loaded
// - Slightly slower writes, but immediate consistency on next read

Pattern 3: Write-Around with Background Refresh

Advanced implementations combine write-around with background processes that proactively refresh stale cache entries based on database change events or scheduled refreshes.

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
// Write-Around with Background Refresh
class WriteAroundWithRefresh {
  private changeLog: ChangeEvent[] = [];
  
  async write(key: string, data: Data): Promise<void> {
    await database.write(key, data);
    
    // Emit change event for background processing
    this.changeLog.push({ 
      key, 
      timestamp: Date.now(), 
      action: 'write' 
    });
  }
  
  // Background worker refreshes hot cache entries
  async backgroundRefreshWorker(): Promise<void> {
    while (true) {
      const changes = await this.consumeChanges();
      
      for (const change of changes) {
        const isHotKey = await this.isFrequentlyAccessed(change.key);
        
        if (isHotKey) {
          // Proactively refresh cache for hot keys
          const freshData = await database.read(change.key);
          await cache.set(change.key, freshData);
        }
        // Cold keys can wait for natural read-through
      }
      
      await sleep(REFRESH_INTERVAL);
    }
  }
}

Write-Around Pattern Comparison
Pattern	Write Latency	Read Consistency	Complexity	Best Use Case
Pure Write-Around	Lowest	Eventually consistent (TTL-based)	Minimal	Logging, analytics, metrics
With Invalidation	Low-Medium	Consistent on next read	Low	User profiles, configurations
With Background Refresh	Lowest	Near real-time for hot data	High	Mixed workloads, critical hot data

Write-Around and Cache-Aside: A Unified View

Write-around caching is often implemented as the write policy within a broader cache-aside (also called lazy-loading) architecture. Understanding this relationship clarifies how write-around fits into the larger caching ecosystem.

Cache-Aside Architecture:

In cache-aside, the application is responsible for managing the cache—the cache doesn't automatically handle reads or writes. The application must:

Check the cache before reading from the database
Populate the cache after a cache miss
Decide what to do with the cache during writes

Write-around provides the answer to #3: do nothing (or invalidate).

Cache-Aside Write Policies

•Write-Through — Write to cache AND database synchronously. Cache always has fresh data but writes are slower.
•Write-Back (Write-Behind) — Write to cache first, async write to database. Fastest writes but risk of data loss.
•Write-Around — Write to database only, cache bypassed. Prevents cache pollution from transient writes.

Why Write-Around is Natural for Cache-Aside:

Cache-aside already assumes the application manages cache population lazily—on read misses. Write-around extends this philosophy to writes: if we don't know that data will be read, why eagerly cache it? Let the read path handle cache population when (and if) the data is actually needed.

This creates a demand-driven cache where cache contents precisely reflect actual read patterns rather than write patterns. The cache becomes a true LRU (Least Recently Used) read cache.

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
class CacheAsideWithWriteAround<T> {
  constructor(
    private cache: Cache<T>,
    private database: Database<T>,
    private ttl: number = 3600 // 1 hour default
  ) {}
 
  // Cache-Aside Read Pattern (Lazy Loading)
  async get(key: string): Promise<T | null> {
    // 1. Check cache first
    const cached = await this.cache.get(key);
    if (cached !== null) {
      return cached; // Cache hit
    }
 
    // 2. Cache miss: Load from database
    const data = await this.database.get(key);
    if (data !== null) {
      // 3. Populate cache for future reads
      await this.cache.set(key, data, this.ttl);
    }
    
    return data;
  }
 
  // Write-Around Write Pattern (Database Only)
  async set(key: string, value: T): Promise<void> {
    // Write directly to database
    await this.database.set(key, value);
    
    // Optionally invalidate stale cache entry
    await this.cache.delete(key);
    
    // Note: NOT populating cache
    // If this data is needed, get() will load it
  }
 
  // Bulk write example: Write-around shines here
  async bulkInsert(records: Record<string, T>[]): Promise<void> {
    // All writes go to database
    await this.database.bulkInsert(records);
    
    // Cache is NOT polluted with bulk data
    // Only frequently-read records will enter cache naturally
  }
}

The Lazy Philosophy

Write-around embraces 'lazy' as a virtue. Why do work that might never be needed? By deferring cache population to actual read demand, the system automatically optimizes cache contents for real access patterns. This is analogous to lazy evaluation in functional programming—compute only what's necessary.

Timing and Latency Characteristics

Understanding the timing profile of write-around caching is crucial for capacity planning and setting performance expectations. Let's analyze the latency characteristics for different operations.

Write-Around Latency Profile
Operation	Cache Involvement	Expected Latency	Components
Write (pure)	None	5-20ms	Network → Database write → Response
Write (with invalidation)	Delete only	5-25ms	Network → Database write + Cache delete → Response
Read (cache hit)	Read only	1-5ms	Network → Cache read → Response
Read (cache miss)	Read + Write	10-30ms	Network → Cache read → DB read → Cache write → Response
First read after write	Miss + Populate	10-30ms	Always a cache miss (guaranteed fresh data)

The First-Read Penalty:

A defining characteristic of write-around is the first-read penalty after writes. When data is written, the cache either doesn't have it (pure write-around) or has been invalidated (write-around with invalidation). Either way, the first read operation pays the full cost of a cache miss:

Write → [Database stores data]
Read  → [Cache miss] → [Database read] → [Cache population] → [Return data]

This is the price paid for write-path simplicity. Whether this trade-off is acceptable depends on the read-after-write latency requirements of your application.

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
56
57
58
59
60
interface LatencyMetrics {
  writeLatency: number;
  readHitLatency: number;
  readMissLatency: number;
  cacheHitRate: number;
}
 
class InstrumentedWriteAroundCache<T> {
  private metrics: LatencyMetrics = {
    writeLatency: 0,
    readHitLatency: 0,
    readMissLatency: 0,
    cacheHitRate: 0,
  };
  
  async write(key: string, value: T): Promise<void> {
    const start = performance.now();
    
    await this.database.write(key, value);
    await this.cache.delete(key); // Invalidation
    
    this.metrics.writeLatency = this.ema(
      this.metrics.writeLatency, 
      performance.now() - start
    );
  }
  
  async read(key: string): Promise<T | null> {
    const start = performance.now();
    
    const cached = await this.cache.get(key);
    if (cached !== null) {
      this.metrics.readHitLatency = this.ema(
        this.metrics.readHitLatency,
        performance.now() - start
      );
      this.recordHit();
      return cached;
    }
    
    // Cache miss path
    const data = await this.database.read(key);
    if (data !== null) {
      await this.cache.set(key, data);
    }
    
    this.metrics.readMissLatency = this.ema(
      this.metrics.readMissLatency,
      performance.now() - start
    );
    this.recordMiss();
    
    return data;
  }
  
  // Exponential moving average for smooth metrics
  private ema(current: number, sample: number, alpha = 0.1): number {
    return alpha * sample + (1 - alpha) * current;
  }
}

Amortized Performance

While individual operations may vary, the amortized performance of write-around depends heavily on your read/write ratio. In read-heavy workloads (90%+ reads), the cache hit rate dominates and overall latency is excellent. In write-heavy workloads, the cache provides less benefit but also causes less overhead. This balance is why write-around often outperforms alternatives in aggregate metrics.

Consistency Model Analysis

Write-around caching implements an eventually consistent model with configurable staleness windows. Understanding the consistency guarantees—and limitations—is essential for choosing this strategy appropriately.

Consistency Guarantees

•Write Durability — Writes go directly to the database, ensuring data is persisted before acknowledgment. No data loss risk from cache failures.
•Eventual Consistency — Reads eventually reflect all writes, bounded by TTL or explicit invalidation timing.
•Read-Your-Writes (Conditional) — Only guaranteed if invalidation is used. Pure write-around may return stale data to the same session.
•Monotonic Reads (Not Guaranteed) — A reader may see newer data, then older data if the cache serves a stale entry.
•Causal Consistency (Not Guaranteed) — No ordering guarantees between writes and reads from different sessions.

Staleness Window Analysis:

The staleness window is the period during which cache reads may return outdated data after a write. Understanding this window is critical for determining if write-around is appropriate.

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
// Staleness window depends on implementation variant
 
// Pure Write-Around: Staleness = TTL - elapsed time
function pureWriteAroundStaleness(ttl: number, timeSinceWrite: number): number {
  // Maximum staleness equals remaining TTL
  return Math.max(0, ttl - timeSinceWrite);
  
  // Example: TTL=1hour, write happened 10min ago
  // Staleness window = 50 minutes (worst case)
}
 
// Write-Around with Invalidation: Staleness = 0
function invalidationStaleness(): number {
  // Invalidation removes stale entry
  // Next read fetches from database
  return 0; // No staleness after invalidation
}
 
// Practical staleness scenarios
const scenarios = {
  // User updates profile, reads immediately
  profileUpdate: {
    pattern: "write-then-read",
    pureWriteAround: "May see old profile for up to TTL",
    withInvalidation: "Cache miss, fresh data guaranteed",
    recommendation: "Use invalidation for user-facing updates",
  },
  
  // Analytics event written, never read by same user
  analyticsWrite: {
    pattern: "write-only",
    pureWriteAround: "No staleness concern (no read)",
    withInvalidation: "Unnecessary overhead",
    recommendation: "Pure write-around, skip invalidation",
  },
  
  // Inventory update, read by many other users
  inventoryUpdate: {
    pattern: "write-then-read-by-others",
    pureWriteAround: "Other users may see stale inventory",
    withInvalidation: "All users see fresh data on next read",
    recommendation: "Use invalidation, consider short TTL",
  },
};

Consistency Requirements by Use Case
Use Case	Staleness Tolerance	Recommended Variant	TTL Strategy
User session data	Zero (security)	Write-Through or Invalidation	Short TTL + invalidation
Product catalog	Minutes acceptable	Pure Write-Around	5-15 minute TTL
Inventory counts	Seconds	Invalidation	30 second TTL + invalidation
Analytics/logs	Hours/Never read	Pure Write-Around	Long TTL or no caching
Social feed	Seconds to minutes	Invalidation + Background	1-5 minute TTL
Configuration	Zero	Write-Through	Long TTL + version-based invalidation

The Stale Read Trap

Be careful with pure write-around for user-facing updates. If a user changes their email address and immediately sees their old email displayed (from cache), they'll assume the update failed. This UX degradation may outweigh the performance benefits. For user-initiated updates with immediate visibility requirements, prefer write-around with invalidation.

Comparison with Other Caching Strategies

To truly understand write-around, we must compare it with its siblings: write-through and write-back caching. Each strategy makes different trade-offs along the dimensions of latency, consistency, and failure resilience.

Caching Strategy Comparison Matrix
Dimension	Write-Through	Write-Back	Write-Around
Write latency	High (cache + DB sync)	Low (cache only)	Medium (DB only)
Cache pollution	High (all writes cached)	High (all writes cached)	Low (reads only)
Data loss risk	None (DB confirmed)	High (cache loss = data loss)	None (DB confirmed)
Read-after-write consistency	Guaranteed	Guaranteed	Eventual (or invalidate)
Cache hit rate (write-heavy)	High (but wasted)	High (but wasted)	Low (but efficient)
Implementation complexity	Low	High (async + recovery)	Low

When Write-Around Wins:

Write-around excels in scenarios that penalize the other strategies:

High write volume with sparse reads — If you're writing 100,000 records but only 1% will ever be read, write-through wastes cache space on 99,000 entries that will expire unused. Write-around keeps the cache lean.
Cost-sensitive cache infrastructure — Cache memory is expensive. If your workload would fill the cache with write-through but only reads a fraction, write-around provides better cost efficiency.
Write bursts and bulk operations — Batch imports, ETL processes, and event logging can overwhelm caches. Write-around lets these operations complete without cache interference.
Strict durability requirements — Unlike write-back, write-around confirms database persistence before acknowledging writes. There's no window for data loss.

When Write-Around is Wrong

•Predictable read-after-write patterns — If data is always read immediately after writing (e.g., checkout confirmation pages), write-through eliminates the read miss penalty.
•Real-time requirements — Systems requiring immediate consistency cannot tolerate any staleness window.
•Cache-as-primary-storage — Some architectures use cache as the write target with async DB sync. Write-around doesn't fit this model.
•Session state management — User sessions require read-your-writes guarantees that pure write-around cannot provide.

Hybrid Approaches

Many production systems use hybrid strategies—write-through for hot, frequently-read data and write-around for cold, rarely-read data. The key is understanding your access patterns and applying the appropriate strategy to each data category.

Summary: Write-Around Mechanics

Write-around caching implements a simple but powerful philosophy: let read patterns drive cache contents. By bypassing the cache during writes, this strategy prevents cache pollution and optimizes for workloads where writes don't predict reads.

Key Takeaways

•Writes bypass the cache — Data goes directly to the database with no cache interaction (or only invalidation).
•Reads populate the cache — Cache entries are created lazily, only when data is actually requested.
•Staleness is the trade-off — Without invalidation, cached data may become stale after writes until TTL expires.
•Three variants exist — Pure (no invalidation), with invalidation, and with background refresh—each with different consistency/performance profiles.
•Natural fit for cache-aside — Write-around extends the lazy-loading philosophy to the write path.
•Prevents cache pollution — Write-heavy workloads don't fill the cache with data that may never be read.

What's Next:

Now that you understand how write-around works at a mechanical level, the next page explores why writes go directly to the database—examining the architectural motivations, durability guarantees, and failure mode analysis that make database-first writes the right choice for write-around systems.

Page Complete

You now understand the fundamental mechanics of write-around caching—the data flow patterns, timing characteristics, consistency model, and how it compares to other strategies. You can visualize how writes bypass the cache while reads populate it on demand, and you understand when this trade-off is beneficial.