Mehran Tarif

Fullstack Developer | Machine learning

[Mehran Tarif @Mehran_Tarif]

@badass_molook مگه بیمه دندون رو پوشش میده؟

[Mehran Tarif @Mehran_Tarif]

@Michaelpfiffer @s00ad00 لیس عمیق

[Mehran Tarif @Mehran_Tarif]

@Mrtz_Fz @SayehRosan تا به حال شده بلایند تست سعی کنی حدس بزنی کدوما رو ai زده

[Mehran Tarif @Mehran_Tarif]

@redheadfoxie تو گوه خوردی سرت تو کون بقیه است عقده ای بدبخت مگه از تو نظر پرسیدن؟

[Mehran Tarif @Mehran_Tarif]

@DaraaCodes یه چیزی. تو هر کشور اروپایی باشی تقریبا همون سه روز در ماهت رو می تونی سیو کنی. اون سه روز در ماه سوییس میشه ایفون ولی سه روز در ماه یه کشور دیگه میشه میان رده شیائومی.

[Mehran Tarif @Mehran_Tarif]

@fordprifect @DaraaCodes جمله ات تناقض داره. اگر فاند داشته باشی میشه پول بلیت، وثیقه، ترجمه مدارک و یه ماه اولت. یه چیز حدود سه هزارتا. تمکن نمیخواد دیگه.

[Mehran Tarif @Mehran_Tarif]

Most load balancers don’t actually “balance” load. 🚀

In production, traffic is never evenly distributed.

That’s why scaling becomes hard.

Real-world problems:

- heavy requests overload specific nodes
- retry storms create cascading failures
- slow instances destroy P99 latency
- “healthy” servers may still be overloaded internally

Modern load balancing is less about routing traffic and more about:

- failure isolation
- latency control
- congestion management

The hardest part of distributed systems isn’t handling traffic.

It’s handling uneven traffic.

#LoadBalancing #DistributedSystems #Backend #SystemDesign

[Mehran Tarif @Mehran_Tarif]

@SahilExec The pattern is consistent, great engineers optimize for long-term system health, not short-term output.

[Mehran Tarif @Mehran_Tarif]

@emad__elsaid Feels like the friction of building was part of the motivation. When you remove it with agents, you also remove the reward loop.

[Mehran Tarif @Mehran_Tarif]

@ash_twtz Storage vendors use decimal (GB = 10^9), while operating systems report binary (GiB = 2^30). So 32 GB shows up as ~29.8 GiB.

[Mehran Tarif @Mehran_Tarif]

@droidbuilds Hype without a clear long-term value usually dies fast. Curious what problem it actually solved better than existing tools?

[Mehran Tarif @Mehran_Tarif]

@theo Slow dashboards are usually a symptom, not the problem. Either inefficient queries or missing aggregation layers. Curious if it was pulling raw data instead of precomputed views?

[Mehran Tarif @Mehran_Tarif]

@adahstwt Exactly. Writing code is the easy part. Operating it under real traffic, failures, and evolving requirements is where engineering actually starts.

[Mehran Tarif @Mehran_Tarif]

@devcansado404 The real bug is not the code, it’s shipping TODOs into production. Every system has at least one of these landmines waiting to explode.

[Mehran Tarif @Mehran_Tarif]

@Its_Nova1012 Classic idempotency issue. Retries + no deduplication usually cause this. Curious if there’s a message queue or webhook retry mechanism involved?

[Mehran Tarif @Mehran_Tarif]

@CaptainInsightX Sounds like a resource leak or something tied to a TTL/timeout cycle. What’s happening around the 30–40 min mark, cache expiry, connection pool exhaustion, or scheduled jobs?

[Mehran Tarif @Mehran_Tarif]

@devcansado404 Rewrites solve technical dissatisfaction, not business problems. Most of the time, you just reset your bug backlog.

[Mehran Tarif @Mehran_Tarif]

@techxutkarsh These are great for onboarding, but real learning starts when you build something that breaks under real-world constraints.

[Mehran Tarif @Mehran_Tarif]

@flaviocopes That works until you need cross-project context. Isolation improves focus, but real workflows often depend on shared state and references.

[Mehran Tarif @Mehran_Tarif]

@asaio87 Using AI is an advantage, but replacing fundamentals with tools is risky. The best engineers use AI to accelerate thinking, not replace it.