Meta system design interview: Design Instagram (with ex-Meta data engineer)

[@IGotAnOffer-Engineering]

Get 1-to-1 coaching and ace your tech interviews: igotanoffer.com/en/interview-coaching/type/tech-interview?KZbin&

[@CWaldman]

Useful video but IMO the guy spent way too much time on metrics. The interviewer should have moved him forward to focus on the architecture.

[@jinhuili3181]

This one is pretty high-quality compared to previous mock system designs in the channel, surprisingly it only got 3k views so far. Thank you, keep up the good work.

[@IGotAnOffer-Engineering]

Glad you like it!

[@voipgrl6735]

Doing my first System Design interview in many years. This was a great refresher on main points to hit and Karthik did a great job of explaining them. Props for the quality 'on the fly' diagraming!

[@alfishanaqeel3178]

5TB/day i.e. 5*365 = 1825TB = 1.8 PB/day correction required here

[@rock2050]

Yup thanks ! agreed 1.8 PB / year 👍

[@yas5673]

Thank you so much Karthik. Previously, even though I knew the technical concepts, I had absolutely no idea how to present them. Because of that, I think I missed a few good opportunities in the past. This video helped me a lot to organize my 40-minute presentation very clearly along with a lot of technical knowledge.

[@mertkahyaoglu48]

Thanks for the video. But it wasn't really focusing on the essential parts. 30 mins passed and then he starts doing high level design... In a real world interview, I think this was a reject.

[@PoRBvG]

Thanks for the effort on this channel! But I really didn't like either the design or the scheduling of the interviewee. His priorities of discussion was terrible and focused a lot on not-very fun and important parts of the design instead of timeline generation and celebrity issues for a system with 1B users!. Was not clear why we have an extra service right before the Kafka queue ( why not just directly go from LB to Kafka) and then the uploader reads from Kafka and uploads/updates the storages and many more other issues for a design like that.

[@ajaydhingra1982]

sorry to say if I give an interview like this then I fail.

[@DevSoni-yy7th]

How about using Read and write replica as more users will read few will upload?

[@rahulsbhatt]

Hey, first of all this is a great video! I have one question in the metrics section, why is the TPS calculated based on how many users? The transactions should include no of users and their activities like uploading a photos, generating newsfeed and following other account, right?

[@ksdekamil]

What is the name of this drawing/diagramming tool which is used by Karthik?

[@jameshizon4861]

Hey I know this guy. I haven't had much experience with having to do System Design interview questions as a Data Engineer. However, I still think that it can be highly beneficial to start prepping now just in case. Not every DE role requires System & API Design questions. My challenge may also be s.t. I don't have any real-world experience building APIs so I am unsure if that would diminish my ability to land role despite being able to pass the API design round.

[@sukhdeepkharbanda1743]

why media service is interacting with CDN?

[@sastryjumpani6969]

How are you handling hot sharding? What’s the use of oltp with retriever service when no sql would be faster to retrieve the data with caching ?

[@mohammadtoficmohammad3594]

thank you very useful

[@IGotAnOffer-Engineering]

Glad it was helpful!

[@rohit-ld6fc]

I will reject someone who spends 15 mins on BOE. It is useless and provides no insight about the candidate. Off-course the system should be scalable, that is the whole purpose of designing a system.

[@harshnagpal4212]

Is it not Imp to be correct on these calculations I think the calculations around BW were in correct instead of 10% it has been 1% i think that needs to be called out. Also I think assumption needs to be a little more realistic the assumption of 200Kb for photo was absurd.

[@zodradana]

Yes, it was pretty unrealistic to assume that both read and write operations will have same weight in terms of payload data. 200kb is possible for photo with right compression techniques, but 200kb for read sounds like a ultra heavy query.