Completed 3 years ⁦@Google⁩ 🎉

RT @asmah2107: A user likes a photo. A friend across the world reloads their feed a second later, and the like count hasn't updated. Is thi….

Subscribe to my daily newsletter where I dissect the world of software engineering :

I'm a curious software engineer who loves exploring how things work. I break down complex topics. Subscribe to YT :

Follow @asmah2107, I keep pushing interesting software engineering tweets !.

So, the real question isn't technical. It's: For this piece of data, what's the maximum time we can tolerate it being stale before it causes a real problem?.

For a bank balance, the cost of being wrong is catastrophic. You must use Strong Consistency. It's slower, but it guarantees correctness.

For a "like" count, the cost is near zero. Choosing Eventual Consistency makes the system faster and cheaper. A small delay is an acceptable trade-off for a better user experience.

The magic keyword is the Consistency Model. This isn't a technical choice; it's a business one. Engineers think data must always be perfectly up-to-date. This is expensive and often unnecessary. Your job is to ask the business: "What's the real cost of a user seeing data that.

A user likes a photo. A friend across the world reloads their feed a second later, and the like count hasn't updated. Is this a bug?. A user transfers $500. Their partner reloads the balance a second later, and it hasn't updated. Is this a bug?. "No" to the first. A catastrophic.

Thread for wrong answers only 🔽.

RT @asmah2107: Your company has dozens of AI models in production with no central monitoring. A data science team recently deployed a bugg….

Your company has dozens of AI models in production with no central monitoring. A data science team recently deployed a buggy recommendation model, which went unnoticed for a week. …causing a 20% drop in user engagement for a key product feature. How do you fix this?.

RT @asmah2107: The CAP theorem only tells you what happens during a network partition. The PACELC theorem is more complete. It states that….

RT @asmah2107: You built a SaaS platform like Figma, but the third-party plugin ecosystem is failing. Plugins are insecure, consume too m….

Mandatory thread for wrong answers only 🧵 Go ⤵️.

The CAP theorem only tells you what happens during a network partition. The PACELC theorem is more complete. It states that even if there is no partition (P), a system must still trade between Latency (L) and Consistency (C). Every time you read from a replica, you are making a.

You built a SaaS platform like Figma, but the third-party plugin ecosystem is failing. Plugins are insecure, consume too many resources, and frequently crash the main application for users. The platform is now unstable, and developer trust is eroding. How do you fix this?.

RT @asmah2107: The most dangerous server in your distributed system isn't the one that's dead. It's the one that's a zombie 🏴‍☠️ https://t….

The most dangerous server in your distributed system isn't the one that's dead. It's the one that's a zombie 🏴‍☠️

RT @asmah2107: A profound and often overlooked truth in distributed systems is that …. you can never truly know the current state of anothe….