In traditional replication terminology, what are replicas also called?

Older terminology uses 'slave' for replica and 'master' for primary, though modern systems prefer inclusive language.

How does a replica stay synchronized with the primary?

The primary continuously sends a stream of changes (often from a write-ahead log) to keep replicas up-to-date.

Why should all writes go to the primary instead of replicas?

Routing all writes through a single primary prevents conflicting updates and maintains data consistency.

In a meal logging app, a user logs a meal (write) and immediately views their history (read). What consistency issue might occur?

Replication lag means the replica may not yet have the newly written meal, so a read to the replica returns stale data.

In a URL shortener, users create short URLs (rare) and click them (frequent). What is the approximate read-to-write ratio?

A viral link can get millions of clicks from a single creation, making reads vastly more frequent than writes.

What is replication lag?

Replication lag is the asynchronous delay in propagating writes from primary to replicas.

What does 'eventual consistency' mean in the context of replicas?

Eventual consistency guarantees convergence over time, but allows temporary divergence due to lag.

A user refreshes a page immediately after posting a new recipe. Why might they not see their new recipe?

If the read request hits a replica before replication lag has completed, the replica still has stale data.

Typical replication lag in a modern system is usually in the range of:

Replication lag is typically very short (ms-scale) for nearby replicas but can reach seconds for distant ones.

What is the key difference between asynchronous and synchronous replication?

Asynchronous replication is faster but risks lag; synchronous replication ensures consistency at higher latency.

In synchronous replication, what is the performance trade-off?

Synchronous replication adds latency to every write because the primary must wait for a replica response.

What is a write-ahead log (WAL)?

The WAL is the source of truth that replicas consume to stay in sync with the primary.

Semi-synchronous replication is a compromise between async and sync. What does it guarantee?

Semi-sync reduces latency vs. full sync by not waiting for replicas to fully apply changes, just receive them.

What is the primary storage medium for a cache like Redis?

Caches are in-memory stores, providing microsecond latency compared to milliseconds for disk-based replicas.

When should you use a cache instead of a read replica?

Caches excel at storing pre-computed results for predictable, repetitive queries; replicas handle arbitrary queries.

What is cache invalidation?

Cache invalidation ensures stale data is removed so users don't see outdated information.

A cache hit means the requested data is already in the cache, avoiding a slower database query.

What does TTL (time-to-live) mean in caching?

TTL controls how long cached data is considered fresh before it must be refreshed.

Why might caching a user's personalized data be problematic?

User-specific cache keys are critical; a mistake could leak private data across sessions.

In Stage 1 of scaling with replicas, what is the bottleneck?

A single database becomes a bottleneck because it must serve all traffic simultaneously.

What does a load balancer do in a multi-replica setup?

A load balancer ensures reads are spread evenly across all available replicas to prevent any single replica overload.

In Stage 3 scaling (multiple replicas), where do write operations still go?

Writes always route to the primary; replicas remain read-only even in a multi-replica setup.

What is database failover?

Failover involves both automated detection and promotion of a replica to take over write responsibility.

In a URL shortener, why is caching critical for handling the redirect flow?

Caching redirects (reads) is far faster than querying replicas, critical for high-frequency redirect requests.

After creating a new short URL in a URL shortener, what should you do immediately?

Immediate cache population ensures the creator's next redirect is served from cache, avoiding lag from the replica.

If a URL shortener only used replicas (no cache) and a new short code was created, what could go wrong?

Replication lag means a replica might not yet have received the new mapping, causing a redirect to fail.

What is the 'read-your-own-writes' problem in a meal logging app?

A user logs a meal to primary, then immediately queries a lagging replica that doesn't yet have it.

How can you solve the read-your-own-writes problem in a meal logging app?

Session-aware routing ensures a user's recent writes are visible by directing their reads to the source of truth.

In a meal logging app, where should nutritional summary queries (complex aggregations) ideally run?

Complex aggregations are slower but replicas have the full data set; slight lag is acceptable for summaries.

What cache invalidation strategy makes sense for a user's meal history cache?

Invalidating on write ensures the next read gets fresh data; implicit expiration via TTL is a fallback.

In a recipe app, why are reads (browsing, searching) far more frequent than writes (posting recipes)?

Content consumption naturally exceeds content creation, making read replicas highly valuable for recipe apps.

What data from a recipe app could be effectively cached?

Shared, frequently-accessed data like top recipes and categories benefits from caching; personal data should not.

If a recipe app stores recipes in a primary and replicas, and a user immediately searches for a newly posted recipe, what might happen?

Replication lag means a newly posted recipe may not appear in replica-backed search results immediately.

How can a recipe app avoid showing an immediate blank page after a user posts a recipe?

Optimistic UI, immediate caching, or primary-routed reads prevent user-visible lag.

Read Replicas and Write Scaling — System Design Concept

Q: What is the primary purpose of a database replica?

Replicas are copies of the primary database that serve read queries. They do not accept writes; the primary remains the sole writer.

🍪 TechCookies

Free

📚Read Replicas and Write ScalingFree

11 sections

~28 min total

35 quick quizzes

3 SD challenges linked

0 of 11 done·~30 min left

Concepts›Read Replicas and Write Scaling›What Is a Database Replica?

0 / 11

11 sections~28 min

What Is a Database Replica?

A replica is an identical copy of the primary database that serves read requests.

ReadQuizCode

~3 min

⋯

Primary for Writes, Replicas for Reads

All writes route to the primary; reads distribute across replicas to scale read throughput.

ReadQuizCode

~3 min

⋯

Replication Lag and Eventual Consistency

Changes propagate asynchronously to replicas, creating a brief window of stale data.

ReadQuizCode

~3 min

⋯

How Replication Works Internally

Asynchronous replication is fast but risks lag; synchronous replication ensures consistency at higher latency cost.

ReadQuizCode

~3 min

⋯

When to Use a Read Replica vs. a Cache

Replicas store full queryable data on disk; caches store only pre-computed results in RAM.

ReadQuizCode

~3 min

⋯

Scaling with Read Replicas: Step-by-Step

Evolve from single database to one replica to multiple replicas with load balancing.

ReadQuizCode

~3 min

⋯

Relevant System Design Questions

Read replicas apply across URL shorteners, meal logging, recipe apps, and other read-heavy systems.

Read

~3 min

⋯

URL Shortener

Redirect clicks are reads; creation is rare writes; caching prevents replication lag issues.

ReadQuizCode

~3 min

⋯

Meal Logging Scaling

User-owned data requires read-your-own-writes; coaches tolerate slight lag; summaries are cacheable.

ReadQuizCode

~3 min

⋯

Recipe App Scaling

Browse and search are read-heavy; user writes are rare; caching and replicas handle scale.

ReadQuiz

~3 min

⋯

Practice test

35 questions

~12 min

Section 1 of 11ReadQuick quiz

What Is a Database Replica?

A replica is an identical copy of the primary database that serves read requests.

~3 min read

3 quick quizzes

A replica is simply a copy of your database. The original database is called the primary (sometimes called the master), and the copies are called replicas (sometimes called secondaries or slaves).

Think of it like a textbook:

The primary is the original manuscript — only the author (your application's write operations) can edit it.
The replicas are printed copies that many students (read requests) can read at the same time.

Whenever the primary gets updated, those changes are sent to the replicas so they stay up-to-date.

☑ Quick check 1/3

What is the primary purpose of a database replica?

ATo serve read requests and reduce load on the primary database

BTo accept write operations when the primary is unavailable

CTo replace the primary as the source of truth

DTo encrypt data for security

Answer the quiz to continue

Notes

🔍

Loading…