Sai Charan

@the_saicharan

impressed within expressed without😎

Hyderabad, India

Joined August 2017

947 Following

70 Followers

973 Posts

the_saicharan retweeted

Shreya

@Oblivious9021

3 days ago

Interviewer: You're at Razorpay. A startup just processed their first ever payment. ₹999. 11:59 PM. Your system failed to capture it. Money deducted. ✅ Payment never received. ❌ Founder's first customer. Gone. What went wrong and how do you make sure nobody loses their first ₹999 again?

the_saicharan retweeted

Shreya

@Oblivious9021

4 days ago

Interviewer: You're the engineer at CRED. Your users are India's most impatient people. High income. Zero tolerance. App takes 3 seconds to load. 50,000 users uninstall. Same day. What's the first thing you fix?

the_saicharan retweeted

freeCodeCamp.org

@freeCodeCamp

5 days ago

Large React apps get harder to scale when every feature has to live inside the core codebase. In this tutorial, Jessica shows you how to design a React plugin architecture with TypeScript that is type-safe, lazy-loaded, and secure. You'll also learn about hosting APIs, plugin lifecycles, separate bundling, runtime loading, and more. https://t.co/SsT4uIKAnH

freeCodeCamp's tweet photo. Large React apps get harder to scale when every feature has to live inside the core codebase.

In this tutorial, Jessica shows you how to design a React plugin architecture with TypeScript that is type-safe, lazy-loaded, and secure.

You'll also learn about hosting APIs, plugin lifecycles, separate bundling, runtime loading, and more.

https://t.co/SsT4uIKAnH

284

283

39K

the_saicharan retweeted

Vishwanath Patil

@patilvishi

5 days ago

System Design - EDA and Messaging - Day 50 – Kafka Delivery Guarantees What happens if a message is lost? Or processed twice? Distributed systems must choose between speed, reliability, and duplicates. That's where Kafka's delivery guarantees come in. --- What Are Delivery Guarantees? Delivery guarantees define what Kafka promises when delivering messages from a producer to a consumer. There are three models: - At Most Once - At Least Once - Exactly Once Each makes different trade-offs. --- 1. At Most Once A message is delivered zero or one time. Producer │ ▼ Kafka │ ▼ Consumer ❌ Crash Message Lost If a failure occurs before processing finishes... The message is lost forever. Characteristics ✅ Fastest ❌ No retries ❌ Possible message loss --- 2. At Least Once A message is delivered one or more times. Producer │ ▼ Kafka │ ▼ Consumer Crash Retry Consumer Again If processing fails... Kafka delivers the message again. No data is lost. But duplicates are possible. Characteristics ✅ Reliable ✅ Retries ⚠ Duplicate processing --- 3. Exactly Once Kafka guarantees each message is processed exactly one time. Producer ↓ Kafka ↓ Consumer ↓ Processed Once ✅ - No duplicates. - No message loss. This requires idempotent producers and Kafka transactions. --- Delivery Guarantee Comparison At Most Once: Possible message loss • No duplicates • Fastest At Least Once: No message loss • Possible duplicates • High throughput Exactly Once: No message loss • No duplicates • Lowest throughput --- Real-World Examples At Most Once - Metrics - Application Logs - Monitoring Data Losing one message usually isn't critical. --- At Least Once - Order Processing - Email Notifications - Inventory Updates - Audit Events Missing data is unacceptable. Duplicates can be handled. --- Exactly Once - Banking - Payments - Financial Transactions - Invoice Generation Processing the same event twice can cause serious problems. --- Production Example Customer pays ₹100. Without protection: Payment Event ↓ Processed ↓ Consumer Crash ↓ Retry ↓ ₹100 Charged Again ❌ With Exactly Once: Payment Event ↓ Kafka Transaction ↓ Processed Once ✅ The customer is charged only once. --- Common Mistake Many developers assume Exactly Once is always the best option. It isn't. It adds extra coordination and overhead. For many systems, At Least Once + Idempotent Consumers provides the best balance of reliability and performance. --- Best Practices - Use At Most Once for telemetry and logs. - Use At Least Once for most business events. - Use Exactly Once only for critical financial workflows. --- Key Takeaway At Most Once → Fast, but messages may be lost. At Least Once → Reliable, but duplicates are possible. Exactly Once → No loss, no duplicates, with additional complexity. Choosing the right delivery guarantee depends on your business requirements not just the technology. --- Tomorrow ->Dead Letter Queues (DLQ) Learn how production systems handle messages that repeatedly fail without blocking the rest of the queue.

patilvishi's tweet photo. System Design - EDA and Messaging - Day 50 – Kafka Delivery Guarantees

What happens if a message is lost?
Or processed twice?

Distributed systems must choose between speed, reliability, and duplicates.

That's where Kafka's delivery guarantees come in.

---

What Are Delivery Guarantees?

Delivery guarantees define what Kafka promises when delivering messages from a producer to a consumer.

There are three models:

- At Most Once
- At Least Once
- Exactly Once

Each makes different trade-offs.

---

1. At Most Once

A message is delivered zero or one time.

Producer
│
▼
Kafka
│
▼
Consumer

❌ Crash
Message Lost

If a failure occurs before processing finishes...

The message is lost forever.

Characteristics

✅ Fastest
❌ No retries
❌ Possible message loss

---

2. At Least Once

A message is delivered one or more times.

Producer
│
▼
Kafka
│
▼
Consumer

Crash
Retry
Consumer Again

If processing fails...

Kafka delivers the message again.

No data is lost.

But duplicates are possible.

Characteristics
✅ Reliable
✅ Retries
⚠ Duplicate processing

---

3. Exactly Once

Kafka guarantees each message is processed exactly one time.

Producer

↓

Kafka

↓

Consumer

↓

Processed Once ✅

- No duplicates.
- No message loss.

This requires idempotent producers and Kafka transactions.

---

Delivery Guarantee Comparison

At Most Once: Possible message loss • No duplicates • Fastest
At Least Once: No message loss • Possible duplicates • High throughput
Exactly Once: No message loss • No duplicates • Lowest throughput

---

Real-World Examples

At Most Once
- Metrics
- Application Logs
- Monitoring Data

Losing one message usually isn't critical.

---

At Least Once
- Order Processing
- Email Notifications
- Inventory Updates
- Audit Events

Missing data is unacceptable.

Duplicates can be handled.

---

Exactly Once
- Banking
- Payments
- Financial Transactions
- Invoice Generation

Processing the same event twice can cause serious problems.

---

Production Example

Customer pays ₹100.

Without protection:

Payment Event

↓

Processed

↓

Consumer Crash

↓

Retry

↓

₹100 Charged Again ❌

With Exactly Once:

Payment Event

↓

Kafka Transaction

↓

Processed Once ✅

The customer is charged only once.

---

Common Mistake

Many developers assume Exactly Once is always the best option.

It isn't.

It adds extra coordination and overhead.

For many systems, At Least Once + Idempotent Consumers provides the best balance of reliability and performance.

---

Best Practices
- Use At Most Once for telemetry and logs.
- Use At Least Once for most business events.
- Use Exactly Once only for critical financial workflows.

---

Key Takeaway

At Most Once → Fast, but messages may be lost.

At Least Once → Reliable, but duplicates are possible.

Exactly Once → No loss, no duplicates, with additional complexity.

Choosing the right delivery guarantee depends on your business requirements not just the technology.

---

Tomorrow ->Dead Letter Queues (DLQ)

Learn how production systems handle messages that repeatedly fail without blocking the rest of the queue.

Who to follow

Lukman Nakib

@mr_mihi

Tech Lead @fluent_forms @WPManageninja aka @AuthLab #WCEU

Elad Gasner 🇮🇱

@GasnerElad

Thoughts of a team leader 😎. Freelance Laravel , Vuejs , Livewire Developer.

Lonskie

@Mr_Asncn

RDG🔥

the_saicharan retweeted

freeCodeCamp.org

@freeCodeCamp

7 days ago

OpenStreetMap is a free, open alternative to Google Maps you can use for many location-based apps. In this tutorial, @joshuaaabraham explains how OpenStreetMap works and how to integrate it into a React app with Leaflet. You'll also learn about geocoding with Nominatim, when to choose OpenStreetMap over Google Maps, and more. https://t.co/l3UbxEa7jw

freeCodeCamp's tweet photo. OpenStreetMap is a free, open alternative to Google Maps you can use for many location-based apps.

In this tutorial, @joshuaaabraham explains how OpenStreetMap works and how to integrate it into a React app with Leaflet.

You'll also learn about geocoding with Nominatim, when to choose OpenStreetMap over Google Maps, and more.

https://t.co/l3UbxEa7jw

356

314

40K

the_saicharan retweeted

Shubh Jain

@shubh19

7 days ago

I’ve seen people go from: - Infosys (4 LPA) → Google (40+ LPA) - Tech Mahindra (4.2 LPA) → Adobe (25+ LPA) - Cognizant (4.5 LPA) → Microsoft (30+ LPA) What made the difference? It wasn't luck. It wasn't having an IIT or NIT tag. It wasn't waiting for the "perfect opportunity." It was a few consistent habits: 1. They mastered DSA topic wise and pattern wise, instead of just solving random questions. 2. They built strong fundamentals in one tech stack, instead of chasing every new trend. 3. They kept upskilling - system design, AI, a new framework, even when they weren't actively looking for a job. 4. They learned from every rejection and kept improving, instead of giving up. 5. They stayed consistent for months, not just a few weekends before interviews. Your first salary doesn't define your career. Your learning mindset does.

397

260

26K

the_saicharan retweeted

Abhishek Singh

@0xlelouch_

7 days ago

Most distributed systems work in 2026 is still 10 concepts, done under failure and pager noise: 1) Timeouts + retries. Cap them, add jitter, and avoid retry storms when the dependency is already burning. 2) Idempotency. Every write endpoint needs a key or dedupe window, or at-least-once delivery turns into double charges. 3) Backpressure. Bounded queues and load shedding beat infinite buffers that just convert spikes into OOMs. 4) Consistency model. Pick what can be stale, for how long, and where; document it or it becomes an incident later. 5) Partitioning + hotspots. One bad key (tenant_id, user_id) can pin a shard; add salting and watch skew. 6) Concurrency control. Use optimistic locking, version checks, or leases; last-write-wins quietly loses money. 7) Deployment safety. Canary, feature flags, and rollbacks; schema changes need forward and backward compatibility. 8) Observability you can use. Correlation IDs, RED metrics, traces with sampling that still catches rare errors. 9) Debugging in prod. Packet captures, thread dumps, pprof, slow query logs; know what to grab before you SSH. 10) Security basics. mTLS or at least service auth, secrets rotation, least-privilege IAM; most breaches start as internal misuse. [generated using my AI agent, shared so you can learn from it]

164

177

the_saicharan retweeted

Vishwanath Patil

@patilvishi

7 days ago

toString() Best Practices in Java - Day 36 Have you ever printed an object and seen something like this? User@6d06d69c Not very helpful, right? That is the default implementation of toString(). A well-written toString() can save hours of debugging and make logs much easier to understand. Let's see why. --- What is toString()? Every Java object inherits the toString() method from the Object class. public String toString() Its purpose is simple: Return a human-readable representation of an object. Whenever you do this: System.out.println(user); Java automatically calls: user.toString(); --- Default Implementation Suppose we have: class User { int id; String name; } Now: User user = new User(); System.out.println(user); Output: User@6d06d69c What does that mean? User → Class name 6d06d69c → Hexadecimal hash code Useful for the JVM. Not useful for developers. --- Override toString() Instead, provide meaningful information. @ Override public String toString() { return "User{id=" + id + ", name='" + name + "'}"; } Now: System.out.println(user); Output: User{id=1, name='Alice'} Much easier to read. --- Where is toString() Used? You use it more often than you think. ✔ Logging ✔ Debugging ✔ IDE object inspection ✔ Exception messages ✔ String concatenation System.out.println("User = " + user); Internally Java calls: user.toString(); Automatically. --- Best Practices A good toString() should: - Include meaningful fields - Keep the output concise - Be easy to read - Handle null values gracefully Avoid including: - Passwords - Access tokens - API keys - Sensitive personal information Remember: Logs often end up in monitoring systems. Never expose secrets through toString(). Use IDEs to Generate It You don't need to write it manually. Most IDEs can generate it for you. IntelliJ IDEA: Right Click → Generate → toString() This saves time and reduces mistakes. --- Interview Takeaway Should every field be included in toString()? Not necessarily. Include fields that help identify and understand the object's state. Exclude: - Passwords - Authentication tokens - Sensitive identifiers - Large collections if they make logs unreadable A clean toString() should help developers not overwhelm them. --- Easy Way to Remember equals() → Are these two objects logically equal? hashCode() → Which bucket should this object go into? toString() → How should this object look when printed? These three methods from the Object class are among the most commonly overridden methods in Java and are essential for writing clean, maintainable code. That completes Phase 2: Object-Oriented Programming (OOP) Next, we will move into the next phase of the roadmap - Memory & JVM --- Question: Do you usually write toString() manually, generate it using your IDE, or rely on Lombok's @ ToString annotation?

patilvishi's tweet photo. toString() Best Practices in Java - Day 36

Have you ever printed an object and seen something like this?
User@6d06d69c

Not very helpful, right?
That is the default implementation of toString().

A well-written toString() can save hours of debugging and make logs much easier to understand.

Let's see why.

---
What is toString()?

Every Java object inherits the toString() method from the Object class.

public String toString()

Its purpose is simple:
Return a human-readable representation of an object.

Whenever you do this:
System.out.println(user);

Java automatically calls:
user.toString();

---
Default Implementation
Suppose we have:

class User {
int id;
String name;
}

Now:
User user = new User();
System.out.println(user);

Output:
User@6d06d69c

What does that mean?
User → Class name
6d06d69c → Hexadecimal hash code

Useful for the JVM.

Not useful for developers.

---
Override toString()
Instead, provide meaningful information.
@ Override
public String toString() {
return "User{id=" + id +
", name='" + name + "'}";
}
Now:
System.out.println(user);

Output:
User{id=1, name='Alice'}

Much easier to read.

---
Where is toString() Used?
You use it more often than you think.
✔ Logging
✔ Debugging
✔ IDE object inspection
✔ Exception messages
✔ String concatenation

System.out.println("User = " + user);

Internally Java calls:
user.toString();

Automatically.

---
Best Practices
A good toString() should:
- Include meaningful fields
- Keep the output concise
- Be easy to read
- Handle null values gracefully

Avoid including:
- Passwords
- Access tokens
- API keys
- Sensitive personal information

Remember:
Logs often end up in monitoring systems.
Never expose secrets through toString().

Use IDEs to Generate It
You don't need to write it manually.

Most IDEs can generate it for you.

IntelliJ IDEA:
Right Click → Generate → toString()

This saves time and reduces mistakes.
---
Interview Takeaway
Should every field be included in toString()?
Not necessarily.

Include fields that help identify and understand the object's state.

Exclude:
- Passwords
- Authentication tokens
- Sensitive identifiers
- Large collections if they make logs unreadable

A clean toString() should help developers not overwhelm them.

---
Easy Way to Remember
equals() → Are these two objects logically equal?
hashCode() → Which bucket should this object go into?
toString() → How should this object look when printed?

These three methods from the Object class are among the most commonly overridden methods in Java and are essential for writing clean, maintainable code.

That completes Phase 2: Object-Oriented Programming (OOP)

Next, we will move into the next phase of the roadmap - Memory & JVM

---
Question:
Do you usually write toString() manually, generate it using your IDE, or rely on Lombok's @ ToString annotation?

417

the_saicharan retweeted

Ben Dicken

@BenjDicken

8 days ago

Fast backups at-scale are made possible by sharding. These require an amazing amount of engineering to nail the infrastructure, orchestration, and reliability. I wrote all about it... with the help of some PlanetScale database geniuses, of course!

446

220

36K

the_saicharan retweeted

SumitM

@SumitM_X

8 days ago

Need 30 LPA ? Need to prepare like that... And trust me its not that tough..

410

309

52K

the_saicharan retweeted

alex shapalov

@alxshp

11 days ago

Postgres 19 can query relationships like a GRAPH NOT with a long manual join chain. With a path: customer → bought → product product ← bought ← similar customer similar customer → follows → brand That is a recommendation system in one readable pattern. SQL/PGQ finds the relationship. COLUMNS decides what comes back. Postgres keeps eating the backend stack. https://t.co/3LUxPoZWaN

56K

the_saicharan retweeted

alex shapalov

@alxshp

9 days ago

PostgreSQL adds OAuth 2.0 support 🚀 Instead of managing separate database passwords, users can authenticate through an external identity provider using SSO and OAuth tokens. PostgreSQL can integrate with providers such as: → Google → Auth0 → Keycloak → Microsoft Entra ID Centralized access control. Short-lived credentials. Fewer database passwords to manage. Modern authentication, directly in PostgreSQL. https://t.co/hq0R5tIMPM

452

260

71K

the_saicharan retweeted

SumitM

@SumitM_X

9 days ago

As a Frontend Developer, Please slap yourself if you cannot clearly explain at least 10 of the following : Hydration Partial hydration Islands architecture Streaming SSR Concurrent rendering Time slicing Reconciliation algorithm Fiber architecture Virtual DOM diffing complexity Stale closure problem Event loop (macro vs microtasks) Task starvation Layout thrashing Critical rendering path Render blocking resources Tree shaking internals Code splitting strategies Dynamic import chunking Module federation Shadow DOM Web Components interoperability Web Workers vs Service Workers SharedArrayBuffer Transferable objects OffscreenCanvas WebAssembly integration Browser compositing layers Paint vs composite vs layout GPU acceleration in CSS CSS containment Subpixel rendering IntersectionObserver internals ResizeObserver loop limits MutationObserver cost IndexedDB Service Worker lifecycle traps Cache invalidation strategies Stale-while-revalidate ETag vs Cache-Control HTTP/3 and QUIC Priority hints Preload vs Prefetch vs Preconnect CORS preflight SameSite cookie modes CSRF vs XSS mitigation Content Security Policy (CSP) Trusted Types DOM clobbering Prototype pollution Race conditions in UI state Suspense boundaries Selective hydration Server components Edge rendering Micro-frontend orchestration Finite state modeling Event sourcing in frontend Optimistic UI rollback strategy WebRTC Backpressure in streams API AbortController Streaming fetch response handling Browser memory leak detection Detached DOM nodes Garbage collection timing PerformanceObserver API Long tasks API First Input Delay (FID) Interaction to Next Paint (INP) Cumulative Layout Shift (CLS) Largest Contentful Paint (LCP) Speculative prerendering Priority inversion in async code

118

158

the_saicharan retweeted

SumitM

@SumitM_X

9 days ago

As a Java developer, Please learn: 1. Core Java Mastery - OOP principles (SOLID, DRY, KISS) - Generics, Lambda expressions, Functional interfaces - Java Streams API (map/reduce, collectors) - Java Collections framework - Java Reflection API - Exception handling 2. Multithreading & Concurrency - Thread synchronization, Executors, Locks - Fork/Join framework - Understanding of race conditions, deadlocks, and thread pools - Concurrency utilities (java.util.concurrent) 3. Design Patterns & Architecture - Common design patterns (Singleton, Factory, Builder) - Architectural patterns (MVC, - Microservices, Event-Driven Architecture) - Dependency Injection (DI), Inversion of Control (IoC) 4. Java Memory Management - Garbage Collection (G1, CMS, ZGC) - JVM heap and stack management - Profiling tools (JProfiler, VisualVM) - Analyzing memory leaks, thread dumps, and heap dumps 5. Classloaders and Reflection - Custom class loaders - Dynamic class loading - Reflection for runtime behavior manipulation 6. Spring Framework & Spring Boot - Spring Core (Dependency Injection, AOP) - Spring Boot (Auto-configuration, Microservices support) - Spring Security (OAuth2, JWT) - Spring Data (JPA, Hibernate integration) - Spring Cloud (Netflix OSS, Circuit Breakers) 7. Microservices Architecture - Service discovery (Eureka, Consul) - Load balancing, distributed tracing, and circuit breaking - API Gateway (Zuul, NGINX) - Asynchronous communication with Kafka, RabbitMQ 8. RESTful Web Services - REST principles, building APIs - JSON/XML handling - API versioning, OpenAPI/Swagger documentation 9. Java I/O and NIO - Blocking vs non-blocking I/O (NIO) - Asynchronous I/O, channels, selectors - File handling, serialization and deserialization 10. Reactive Programming - Project Reactor, RxJava - Event-driven architecture, backpressure - Reactive streams, non-blocking IO 11. JPA/Hibernate - ORM principles, entity relationships - Lazy vs eager loading - Caching strategies, query optimization 12. Database Optimization - SQL optimization, indexing, and transactions - NoSQL databases (MongoDB, Cassandra) - ACID principles, CAP theorem 13. Distributed Systems - Consistency, availability, partitioning (CAP) - Event sourcing, CQRS (Command Query Responsibility Segregation) - Distributed caching (Redis, Hazelcast) - Tools: Apache ZooKeeper, Consul, etcd 14. Testing & TDD/BDD - Unit testing (JUnit, Mockito) - Integration and functional testing - Behavior-driven development (Cucumber) 15. CI/CD & DevOps - Continuous integration (Jenkins, CircleCI) - Containerization with Docker - Orchestration with Kubernetes - Source control (Git), versioning, branching strategies 16. Performance Tuning & Optimization - JVM tuning, garbage collection optimization - Caching strategies, load balancing -Tools for profiling and monitoring (Prometheus, Grafana) 17. Security - Authentication/authorization (OAuth2, JWT) - Secure coding (OWASP guidelines) - Encryption (TLS/SSL), CSRF/XSS prevention 18. Cloud Platforms & API Gateways - AWS, Azure, GCP deployment -API Gateway (Zuul, NGINX) for rate-limiting and routing - Cloud-native microservices tools 19. Logging & Monitoring - Centralized logging (ELK Stack, Graylog) - Distributed tracing (Jaeger, Zipkin) - Application Performance Monitoring (APM) 20. Caching Mechanisms - In-memory caching (Ehcache, Caffeine) - Distributed caching (Redis, Memcached) - Cache invalidation strategies (LRU, LFU) 21. Build Tools - Maven and Gradle for dependency management - Multi-module project handling - Continuous delivery pipelines 22. Message Queuing Systems - Kafka, RabbitMQ for event-driven microservices - Pub/Sub models, distributed message brokers

576

37K

the_saicharan retweeted

KrunalSinh Sisodia

@krunalbuilds

10 days ago

Senior Backend Interview Question: Two engineers cache product data. Engineer A — Cache-Aside 👇 Read: Check cache → miss → query DB → store in cache → return Write: Update DB only. Cache expires on TTL. Engineer B — Write-Through 👇 Read: Check cache → always there → return instantly Write: Update DB AND cache simultaneously. Always in sync. Engineer A's users see stale product prices for up to 1 hour. A flash sale starts. Old prices show for 60 minutes. Engineer B's cache is always fresh. But every write is slower because it updates two places. At Shopify, flash sales affect millions of users simultaneously. Which strategy would you choose — and what's the tradeoff you're accepting? 👇

krunalbuilds's tweet photo. Senior Backend Interview Question:

Two engineers cache product data.

Engineer A — Cache-Aside 👇

Read: Check cache → miss → query DB → store in cache → return
Write: Update DB only. Cache expires on TTL.

Engineer B — Write-Through 👇

Read: Check cache → always there → return instantly
Write: Update DB AND cache simultaneously. Always in sync.

Engineer A's users see stale product prices for up to 1 hour.

A flash sale starts. Old prices show for 60 minutes.

Engineer B's cache is always fresh. But every write is slower because it updates two places.

At Shopify, flash sales affect millions of users simultaneously.

Which strategy would you choose — and what's the tradeoff you're accepting? 👇

the_saicharan retweeted

yashaswi.

@pixperk

10 days ago

(another dbi glaze post) database internals is the best book i've ever read. it can be a bit dense in its approach, but it has taught me a lot. this book really means a lot to me.

758

352

21K

the_saicharan retweeted

Vishwanath Patil

@patilvishi

10 days ago

Idempotency ≠ Deduplication ≠ Exactly-Once Processing These three concepts are often used together in distributed systems... But they solve completely different problems. Confusing them can lead to duplicate payments, repeated emails, or inconsistent order processing. Here is the easiest way to remember them: Idempotency ➜ The same request can be executed multiple times, but the final result remains the same. Deduplication ➜ Detects and ignores duplicate requests or messages before processing. Exactly-Once Processing ➜ Ensures a message's effect is applied only once across the entire processing pipeline. Quick memory trick - Idempotency = Safe Retries - Deduplication = Remove Duplicates - Exactly-Once = One Effect ---- Imagine a Payment API A customer clicks "Pay Now"... The network times out. The client retries the request. What happens next? Idempotency The request includes: Idempotency-Key: abc123 Even if the request is sent 5 times... Only one payment is created. Every retry returns the same result. Deduplication Suppose Kafka delivers the same event twice: PaymentCompleted PaymentCompleted The consumer checks the Message ID. If it's already processed... The duplicate is skipped. Only one copy is handled. Exactly-Once Processing Imagine: Producer → Kafka → Consumer → Database The consumer crashes after writing to the database but before acknowledging the message. Without proper guarantees... The message may be processed again. Exactly-once processing ensures that even if retries happen, the business effect is applied only once. Examples include transactional messaging systems such as Kafka's Exactly-Once Semantics (EOS) combined with idempotent consumers or transactions where appropriate. --- When should you use each? Idempotency - Payment APIs - REST APIs - Retryable operations - Order creation - Money transfers Purpose: 👉 Make retries safe. Deduplication - Kafka consumers - RabbitMQ consumers - Email notifications - Webhooks - Event processing Purpose: 👉 Filter duplicate messages. Exactly-Once Processing - Financial systems - Inventory management - Banking - Event-driven microservices - Critical business workflows Purpose: 👉 Guarantee a single business effect. --- The biggest misconception Many developers think: Idempotency gives Exactly-Once Processing. Not necessarily. Idempotency makes repeated requests safe. Deduplication filters repeated messages. Exactly-once processing is an end-to-end guarantee that typically requires coordination across producers, brokers, consumers, and storage. In production systems, you will often use all three together. --- One sentence to remember forever Idempotency = Same request, same outcome. Deduplication = Ignore duplicates. Exactly-Once = One business effect. Reliable distributed systems aren't built by hoping messages arrive only once... They are built by assuming retries, duplicates, and failures will happen and designing the system to handle them correctly.