Devesh Parmar

The fixes: → Return 202, queue logs asynchronously → Isolate log ingestion from critical API routes → Apply per-client rate limits at the edge → Break retry amplification loops with backoff Full incident breakdown backend-bytes.hashnode.dev/how-a-logging-… (3/n)

POST /logs → Gateway → Backend → Elasticsearch ES writes were blocking. As ES slowed, connections piled up. Failed uploads triggered more uploads an amplification loop. Gateway capacity exhausted. Downstream services never received the requests. (2/n)

A logging endpoint took down our API Gateway in production. Not the logging backend. Not the downstream services. A spike in log volume from client applications triggered a bottleneck cascade that ended in widespread 504 Gateway Timeouts. Here's what happened (1/n)

@niveditjain Peace was never an option :(

ambition and stress are symbiotic, one cannot live without the other

@this_is_swati_ Great list! Suggesting two more to add to the mix: ​Granola (Productivity): AI notepad that turns meetings into clean notes. ​Warp (Development): AI terminal that lets you look up commands and debug errors using natural language.

People using AI effectively are quietly gaining unfair advantages right now. And most people still don’t realize how big this gap is becoming. If you’re still learning and working the exact same way you did 2 years ago… you’re already falling behind. The people using AI daily are: • learning faster • building faster • researching faster • preparing for interviews faster • communicating better A few tools quietly creating this shift: ✅️ Perplexity AI Replaces hours of random Googling with instant, researched answers. ✅️ NotebookLM Turns messy PDFs, notes, and documents into summaries, explanations, and study guides. ✅️ Cursor AI One of the biggest leverage tools for developers right now. ✅️ GitHub Copilot A lot of people still don’t realize it’s free for students. ✅️ ChatGPT Basically becoming a personal tutor if you use it correctly. The scary part? Most people are still using none of these tools. And the gap between AI-native professionals and everyone else is only going to grow. I’ve been compiling the best AI tools I’ve found for: • coding • productivity • automation • research • interviews • career growth I’ll drop the full list for anyone who wants it 👇 docs.google.com/document/d/1Cw…

@krunalbuilds Will Prefer UUIDv7 - It prevents enumeration and business data leakage while offering better index locality than UUIDv4. Or Alternatively we can use an internal BIGINT auto-increment key for performance and expose a UUIDv7 externally via APIs to avoid leaking information.

Senior Backend Interview Question: Two engineers design the primary key for a users table. Engineer A 👇 id INT AUTO_INCREMENT → 1, 2, 3, 4, 5... Engineer B 👇 id UUID → "a3f8c2d1-9b4e-4f7a-8c3d-2e1f9b4a7c8d" Engineer A's API ships. A competitor signs up. Their user ID is 4. They now know you have 3 other customers. Which one would you approve — and why?

You can't write to DB and Kafka atomically. One crashes → ledger never updates → balance wrong forever. This is the dual-write problem. The fix? Outbox Pattern write once to DB, guaranteed delivery. Full breakdown - backend-bytes.hashnode.dev/diving-deep-in… #systemdesign #backend #microservices

Fintech systems process millions of transactions daily. Every payment needs two things to happen: → Save transaction to DB → Notify downstream services (ledger, notifications) Looks simple. But in microservices, this is one of the hardest problems to get right.

Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors. Available today at the same price.

@aryanistweeting Vlog soon? 🙃

getting in my morning caffeine to shake off the weekend and turn into a lethal corporate weapon.

Imagine being this early

Wrote detailed technical blog breaking this down e2e inspired by real scenarios like Big Billion Days. Covers Kafka partitioning, dedup, Redis caching, CDN strategy. (3/n) [backend-bytes.hashnode.dev/your-kafka-pip…] #kafka #systemdesign #backend #microservices #redis

Kafka wasn’t the bottleneck. The issue was consumer design: → Partition by ProductID → In-memory dedup before DB writes → Redis write-through + short CDN TTL → Avoid mass cache invalidation (2/n)

Scenario: It's 12 AM. Flash sale goes live. 50,000 price updates hit Kafka in 30 seconds. Consumer lag builds. Customers see stale prices. Carts get abandoned. 4 patterns that prevent this (1/n)

@krunalbuilds Agreed :)

Coding in 2026 is: - 1% actually coding - 25% debugging AI-generated code - 20% rewriting prompts because the AI misunderstood - 15% coffee breaks - 20% asking AI to explain errors you already caused - 10% staring at the screen while the AI confidently hallucinates - 5% copy-pasting from old Stack Overflow answers “just in case” - 4% pretending you totally wrote it yourself 😄

@aryanistweeting @skyjoshii I don't see a query for this, but I'm willing to bypass code review just this once 😂

@skyjoshii karde kya @asyncDevesh? 😂

and just like that my last working day has arrived. everything that felt like routine until now is happening for the very last time.

Even at the scale, JSON finds a way to say hello! Got a delivery confirmation from Flipkart with a raw {"" : ""} JSON object rendered in the message. Looks like a serialization / templating variable issue. At scale, edge cases always find a way:) #SoftwareEngineering #backend

@aryanistweeting Trees & it's traversals

which dsa topic has the most boilerplate? i'll go first. - graphs 🙂

Building on last week’s announcement of practice SATs in Gemini, you can now also take full-length mock JEE Main tests in Gemini at no cost.

applied leave at work with abslutely 0 plans today. woke up at 12pm. no stand up calls, no production issues, no builds in testing, no CI/CD. life feels worth living.