soham.tsx

20+1 | ex-engineering @/Euron | upskilling and chasing sunsets | sharing what i find interesting

[soham.tsx @Itysoham]

oh i hate the way ive to post it like threads to bypass that character limit TvT. might just create a blog next time

[soham.tsx @Itysoham]

years back i watched this video (https://t.co/Bdqbe39uRy) and it randomly popped up on my feed again, so thought of sharing it. it's about how Hotstar handled massive cricket traffic.
- 25.3M concurrent viewers.
- 1M requests/sec.
- 10 TB/s of video bandwidth.

[soham.tsx @Itysoham]

when things get close to the limit, panic mode + graceful degradation kicks in. non-critical stuff gets reduced or throttled while critical services keep running.
the mindset i liked most: don't wait for millions of users to break your system. break it yourself first.

[soham.tsx @Itysoham]

they also used chaos engineering to simulate things breaking.
AZ failures, network issues, database problems, latency spikes, service failures.
the goal was simple:
find the bottleneck before the actual match finds it for you.

[soham.tsx @Itysoham]

reactive autoscaling wasn't enough. when millions of users suddenly join, waiting 5-6 minutes for new instances isn't exactly useful. so Hotstar used expected concurrency + request rate to pre-scale before the traffic actually arrived.

[soham.tsx @Itysoham]

the part that caught my attention was Project HULK.
Hotstar built their own massive load testing setup to basically attack their infrastructure before the actual match.
3,000+ machines across 8 AWS regions.
they recreated real traffic, sudden spikes, latency and failures.

[soham.tsx @Itysoham]

oh i hate the way ive to post it like threads to bypass that limit TvT. might just create a blog next time

[soham.tsx @Itysoham]

the part that caught my attention was Project HULK.
Hotstar built their own massive load testing setup to basically attack their infrastructure before the actual match.
3,000+ machines across 8 AWS regions.
they recreated real traffic, sudden spikes, latency and failures.

[soham.tsx @Itysoham]

reactive autoscaling wasn't enough.
when millions of users suddenly join, waiting 5-6 minutes for new instances isn't exactly useful.
so Hotstar used expected concurrency + request rate to pre-scale before the traffic actually arrived.

[soham.tsx @Itysoham]

they also used chaos engineering to simulate things breaking.
AZ failures, network issues, database problems, latency spikes, service failures.
the goal was simple:
find the bottleneck before the actual match finds it for you.

[soham.tsx @Itysoham]

when things get close to the limit, panic mode + graceful degradation kicks in.
non-critical stuff gets reduced or throttled while critical services keep running.
the mindset I liked most:
dont wait for millions of users to break your system. break it yourself first.

[soham.tsx @Itysoham]

@KarthikIO haha thats so creative!

[soham.tsx @Itysoham]

@Samaytwt been working on building something similar, if anybody wanna help, dm me!

[soham.tsx @Itysoham]

An idiot in motion goes further than a genius at rest.

- anonymous

[Mrinal @Hi_Mrinal]

READ ARTICLES
READ BLOGS
READ RESEARCH PAPERS
LEARN NEW STUFFS
MAKE PROJECTS AROUND THEM
SHOW EVERYONE

[soham.tsx @Itysoham]

@DarthKermi72747 nice find buddy but i would suggest reporting it to CERT and wait for them to fix it.

[soham.tsx @Itysoham]

watched this video expecting a normal layoff story and ended up listening to one of the most knowledgeable engineers ive seen explain infra and systems so casually. crazy.

[soham.tsx @Itysoham]

@Rushu_Tushu because SMTP was designed as a decentralized protocol. but it doesnt send the mail right away theres a 5 sec(which you can increase till 30 sec if i remember well) holding state during which you can undo send

[soham.tsx @Itysoham]

@arpit_bhayani Congrats sir!!

[soham.tsx @Itysoham]

@adi_1298 @subhashchy definitely adding this to my reading list!