This is literally crazy podcast, I do not think i have finished any podcast without a break ever till this one arrived. A good start to the weekend. Thanks Arpit for doing this.
There is a reason behind high expense charges from these big cloud companies . They invent technologies. Docker and Kubernetes was backed by google . Open source is not free. Single command chalake instances spink karna was not possible few years ago. Google invested money in both docker and Kubernetes. So when you pay cloud bills it’s just not billing charges. it also includes R&D for current and future projects.
I liked this perspective lekin ye bhi socho ki why would they invest in open source? It's definitely not because they are so generous and want people to give open source alternatives of their products, if they invest it's because of their interest. They heavily use open source in their "big companies" and their investment is merely "giving back to the community" if not entirely their business interest.
but he has to take care of his own bottomline and his own business. I’m sure nobody biz or consumer including Google would care if he’s making poor business decisions to be socially right
Man this video is insane. I thought I'll watch first 30 mins or so and sleep. But I couldn't skip any single minute. I've literally been doing engineering for last few years but it was all about writing code and code level optimisations but no one ever talks about these optimisations from the perspective of costs, thinking in such a way from first principles is really 'Asli Engineering'. From the insights from Subhash to the insights from Arpit at the end. It was super amazing. Ive never commented on any youtube video before, creating a channel just to comment on this one. Subscribed to your channel as well. Amazing video loved it!
I'm less knowledgeable to this topic. So 10pm k around video start kiya aur 1:11am ho rha hai. 1hr 37 min ka video ko dekhne mai itna time laga... so many thing for me to know. bahut se unknown terms ko google search kr kr ke dekhna pada.learned many things. Plz keep it up !
he single handedly destroyed Devops Engineers/SRE/GCP/AWS (with facts and data of course). But what a nice podcast full of TECHNICAL depth along with real time examples that he have shown. My Key takeaways : 1. Learn KUBERNETES ! {will try to become the 5th SRE/Infra guy at DUKAAN} 2. Any system no matter how reliable , FATEGA ! 3. Cloud vs On-prem (IOPS performance comparison)
@@yash_renaissance_athleteBhai most devops wala people I know can do run bare metal infra as well. Usually cloud ko padhne se pehle system handle krna aana hi chahiye. It's not really a big deal, most companies are moving some services to bare metal. Ye kaam bhi devops wale hi krenge
@@yash_renaissance_athlete not actually destroyed , destroy would be a little harsh word , but the need and use of a big devops/SRE team vs 3-4 guys who actually knows how to handle big infrastructure,(be it On-Prem vs cloud) and get stuff done.
@@animeshsingh4290 toh wahi toh mai bhi maanta hoo, but the main comment poster said “he destroyed devops engineers” and that’s why I asked him how did he destroy them
No.. he didn't.. He is just working with trade offs.. for some companies the traffic is so huge that they cannot afford the downtime with critical transactions.. and that downtime he doesn't care.. and basically managing 10 different other free services instead of using mainstream cloud.
I was like I will watch for 10 mins till my snacks would be finished and now I am having dinner watching this. Man, please do more podcasts like these. Got to learn a different thought process . Thanks Subhash and Arpit !! I hope to see more such podcasts on the channel.
Amazing ! In my recent experience working in a similar company to Dukaan but for the Middle East region. They have struggled alot to reduce the cost of AWS bills, this is the real pain for startups. Eye opener this is called "Asli Engineering" hats off, we need more podcasts with Subhash .. Already a fan of him
Question to Subhash: When you say you are doing read replica of Postgres to cloud and in case your on-prem db services is down, you will still serve it via cloud that means you have allowed writes to that cloud database or your application will work only in read-only mode until you bring back your on-prem services? If you allow write to cloud database, how you bring that delta back on your on-prem services?
One of the best podcast. I can say this is asli engineering and Subhash is very confident and fearless, "Fatega to dekha jayega, Zarurat hi ku h, complex krna hi ni h" 😄and he literally shutdown the actual server for the demo. I learnt, do not over design the architecture, just see the data and build.
Great and insightful podcast. I have couple of doubts - 1. Why do they have servers all over the world when their sellers and users are only in India? 2. For all the servers spread across the world, won't they have to connect with one single db in a particular region (or do they have read-replicas in each region)?
Never heard of this dukaan. Out of the blue you tube showed me :) checked google playstore, just looking into the 1st 4-6 comments close the app and stopped watching the remaining video. Not worth investing time.
There is a flaw in the design to handle peak traffic we cannot move the traffic/proxy to other data centers across the world, there are some regions which have data residency. Not very sure what Subhash was referring to when he said I can move traffic across the world, does he mean that he has compute that can run anywhere in the world? Do you have a globally replicated databases?
CLOUD COMPUTING ISN'T A ONE-SIZE-FITS-ALL SOLUTION. IF YOUR WORKLOAD REQUIRES OVER 1200+ SERVERS, IT MAY BE MORE FINANCIALLY PRUDENT TO MAINTAIN ALL SERVERS ON-PREMISES FOR THE LONG TERM. Otherwise, CLOUD IS THE SOLUTION FOR YOU.....!!!!
Great podcast. It’s real stuff. Liked it a lot. Great if you do these similar ones from experts. May I ask one question, you mentioned why Kafka is fast. Can you please in details or give me some reference.
They told partial story. Lots of open risks. Only DB moved to Bare metal? If no, then going bare metal also means that Dukaan might be facing high latencies in their multi-tenant application when hosted on cloud. 32GB RAM x 5 Server = 1TB RAM Kaise? Also after seeing Arpit turned on while looking at low 250K IOPS it seems he has never touched a bare metal enterprise server
I think the worldwide servers are just for redundancy, not for serving requests. All their requests are served from a single DB (with a read replica maintained by them, and one more backup being replicated to GCP).
This discussion was really a gem, from my experience what i have observed is when people talk about response time, everyone gets so defensive that the first point of check is either code, or query run time, adding multiple indexes to make the the query run faster, but we easily ignore the network calls most of the time. Some great take aways from this discussions for sure.
Clever use of anycast network, but there is one big issue with thats why gslb is preferred over it. In this example, he switched off loadbalancer hence demoing a complete failover. But what if there is issue in one of the python servers ( like slow performance because of gc ), then this solution won't work. In this case gslb, can remove machine intelligently from routing.
Completing it at 2:30am. I got so much to learn that I don’t know if I will get to sleep tonight or not. Amazing one man. Just happily overwhelmed & I am stunned looking at u guys. I’m 20 but now I want to learn & experience all these cool stuffs. Great video Arpit Bro ❤❤❤
Technically, its not on-prem. Its located in a Datacenter, who provide bare metal servers instead of the abstraction provided by Cloud Providers. So, I think the word bare metal is appropriate.
Good stuff. But what is difference between the AWS and the stuff he showed both are on cloud I don’t think so any he just picked cheap hosting nothing else
1. Django is running on devserver in production, it should be runing with gunicorn(wsgi) or uvicorn(asgi) 2. how is k8s deployed+managed (rke2? k3s? rancher?), looks like k3s which is meant to be for edge and not for Baremetal servers. 3. how is the HA (high availability), FT (fault tolerance) of k8s maintained? KubeVip? 4. etcd backup & restore? 5. mock disaster recovery of k8s?
Dukaan doesn't use resource limits & resource requests in the K8s Yamls, any specific reasons for that? It's generally understood to be a good practice to have these specified.
Mehnga mic lene se kuch nahi hota....Correct input stream bhi select karni hoti hai. Dynamic mic ko itna door nahi rakhta us point pe tujhe gain bhi badana padega, aur tere background noise aur static bhi badegi. Us point pe tu USB condenser mic lele, Trend main aake tune SM7B liya... iske saath cloudlifter toh liya hoga???
Awesome podcast. Quick question, when we have servers distributed across so many geographic regions to provide faster response time to the users what happens with the database? I mean all these servers still need to talk to the database and the db roundtrip can become expensive if the server and the db are far apart. Do we create DB replicas for every geographic region were the servers are present? That could be expensive right? Also too many replicas for a RDBMS could result in the DB queries to slow down because the DB now needs to sync the write queries across all the replicas to maintain consistency.
This is where the cache layer will coming into picture. Which is replicated to form a ring across region along with background sync option. You can keep it along with application or separate, upto your business needs and architecture.
What if my all clusters die and my product got outage. Now I plan to switch to AWS cloud. 1. I created Empty brand new Kubernetes Cluster 2. Added some more worker Nodes. 3. Deployed Argo into that new cluster. 4. I would need to create each application and project and configure everything about Argo, into that new cluster too. How to avoid that last 4th step, and to automate that too?
how can I get the backend architecture of Dukaan? In this video, the functionalities and tech stack were explained brilliantly by Subhash sir. But, I wanted to read the diagram explaining the architecture and how the services are working with each other? Is there any source?
What a brilliant podcast! Didn't shy from disclosing the real numbers... no one want's to find out the hard way. You and Subhash are proper engineers and I loved listening to you two talk technology. I've been a generalist for about 20 years, but love to experiment. I've learnt to setup servers etc. only since covid. So while my knowledge is surface level compared to you'll it was interesting to know that my learning is in the right direction. And just like you, i was astonished with the IOPS difference! Anyway, thank you for a wonderful podcast.
Can anyone answer this? Did he pull the plug from just one microservice from the Amsterdam cluster or the Amsterdam cluster itself? If the latter, then who routed the traffic to another region if there's nothing to receive it in the first place? Or the API gateway is somewhere else? If you've idea pls lmk
Man this is dope! Would love to see a follow up podcast where we could see the database architecture. As in, are the databases sharded for a set of shops? Are all the app server regions of the same service hitting a single primary db?
16:16 sums it up. This is great for a talk show and an engineering project. Especially for a retail focused startup, if you want to become a unicorn and then go for IPO, moving out of cloud is not a great strategy. Hybrid is the way to go.
Just for fun, I brought all my old PCs; one of them is running Nginx primarily for load balancing and reverse proxy, while the rest are running Express.js instances, Jellyfin, Nextcloud, and others (In docker of course). In my server, I do serve recorded videos for local coaching centers. But backing up your data is important.
As an SRE I got offended in the beginning but have been worked in both DataCenter and Cloud environments. Definitely understand his requirements and his way of finding a solution for it ( and his comment on SRE ).
I work for a cloud provider so this is counter intuitive to I profess in general. However, the simplicity and ease with which Dukaan manage their infrastructure is amazing. My favorite thing was Subhash's enthusiasm for all this and how he takes pride in his tech work, truly inspiring! I only worry about rapid scaling ability in future, however if the company doesn't project such an exponential growth, then this would work well for a foreseeable future.
This is a magical discussion!!! Man, this is an awesome podcast. I have Subscribed to the channel and turn the notifications ON, and kept shooting such podcasts. Really knowledgable.
I tried ignoring it several times as voice pace is too fast of the "Genius Guy Subhash" but always getting stick to this podcast :p Totally insane, soch raha hu jeevan waste kar diya apna maine faltu ke kaamon mein being a developer
good interview, reminds me of my early days in startup, when our high-end server at datacenter crashed, and all the backup was on my laptop in office, went to datacenter and i had to put the laptop in datacenter and run the business, surprisingiy laptop was in datacenter for 8 months, as we couldnt fix the highend machine as it was assembled and had multiple issues..
How is the latency so low in all regions, doesn't the application connect to the database in the storefront (shown in the KeyCDN test), As I understood the database master is deployed in a single region.
WEB traffic ( or any thing which has session state) on IP anycast is not a good idea , More dangerous is advertising the same IP from 2 different gateways on from same region.( a simple routing change ) can cause lot of hashing change and traffic could drop.
I didnt get few things. How is he serving the sites from different locations ? Does he have servers rented from different locations ? and those servers are basically Kubernetes nodes which are joined to master nodes ? when he tested fail over what did he actually do ? Did he turn off the kubernetes worker node of new york ? He said he only have one gateway ? By gateway he means ISTIO gateway ?
Great insights, how do they replicate realtime data across their regions? It was mentioned it does it on gcp, but how on premise? Does postgres support it out of the box?
35:17 although the server has been proxied to another server in a different location, since their db is in prem, wouldnt it still cause db servers to crash because of the extra load?
How beautifully demonstrated ✨🔥❤️ Although this can help my organisation to fabricate new load balancers for L7 based load balancing as well internally instead of traditional way!!
This is truely an insane podcast!
Subhash by default speaks in 1.5x
His brain works faster than his speech
yep. I wanted to watch in 1.25x or 1.5x but couldn't understand him
i had to listen on .75 to understand
I literally checked if I’m on 1.5x
Same..instinctively i put the video on 1.5x as soon as it started but had to switch back once he started speaking 😂
This is literally crazy podcast, I do not think i have finished any podcast without a break ever till this one arrived. A good start to the weekend. Thanks Arpit for doing this.
Same, 1st podcast I completed.
I hope Arpit bring more content like this (not necessarily in podcast) but practical hands on for "asli engineering"
Bro this cto is dope.... knowlege...curiosity....fearlessness...man insane... kudos to him...Thank you Arpit bhaiya for this podcast
Never saw a world class dev process explained in plain Hindi like this before . Also please see if sensitive info is not being shown. ❤
There is a reason behind high expense charges from these big cloud companies . They invent technologies. Docker and Kubernetes was backed by google . Open source is not free. Single command chalake instances spink karna was not possible few years ago. Google invested money in both docker and Kubernetes. So when you pay cloud bills it’s just not billing charges. it also includes R&D for current and future projects.
I liked this perspective lekin ye bhi socho ki why would they invest in open source? It's definitely not because they are so generous and want people to give open source alternatives of their products, if they invest it's because of their interest. They heavily use open source in their "big companies" and their investment is merely "giving back to the community" if not entirely their business interest.
but he has to take care of his own bottomline and his own business. I’m sure nobody biz or consumer including Google would care if he’s making poor business decisions to be socially right
Man this video is insane. I thought I'll watch first 30 mins or so and sleep. But I couldn't skip any single minute. I've literally been doing engineering for last few years but it was all about writing code and code level optimisations but no one ever talks about these optimisations from the perspective of costs, thinking in such a way from first principles is really 'Asli Engineering'. From the insights from Subhash to the insights from Arpit at the end. It was super amazing. Ive never commented on any youtube video before, creating a channel just to comment on this one. Subscribed to your channel as well. Amazing video loved it!
I recognize this writing
I'm less knowledgeable to this topic. So 10pm k around video start kiya aur 1:11am ho rha hai. 1hr 37 min ka video ko dekhne mai itna time laga...
so many thing for me to know. bahut se unknown terms ko google search kr kr ke dekhna pada.learned many things. Plz keep it up !
Dude got more microservices than twitter. 😂😂😂
he single handedly destroyed Devops Engineers/SRE/GCP/AWS (with facts and data of course).
But what a nice podcast full of TECHNICAL depth along with real time examples that he have shown.
My Key takeaways :
1. Learn KUBERNETES ! {will try to become the 5th SRE/Infra guy at DUKAAN}
2. Any system no matter how reliable , FATEGA !
3. Cloud vs On-prem (IOPS performance comparison)
lol how did he destroy devops engineers
@@yash_renaissance_athleteBhai most devops wala people I know can do run bare metal infra as well. Usually cloud ko padhne se pehle system handle krna aana hi chahiye. It's not really a big deal, most companies are moving some services to bare metal. Ye kaam bhi devops wale hi krenge
@@yash_renaissance_athlete not actually destroyed , destroy would be a little harsh word , but the need and use of a big devops/SRE team vs 3-4 guys who actually knows how to handle big infrastructure,(be it On-Prem vs cloud) and get stuff done.
@@animeshsingh4290 toh wahi toh mai bhi maanta hoo, but the main comment poster said “he destroyed devops engineers” and that’s why I asked him how did he destroy them
No.. he didn't.. He is just working with trade offs.. for some companies the traffic is so huge that they cannot afford the downtime with critical transactions.. and that downtime he doesn't care.. and basically managing 10 different other free services instead of using mainstream cloud.
This CTO is amazing! Very inspiring and logical and keeps things simple! Ankit thanks for bringing this, looking for more
I was like I will watch for 10 mins till my snacks would be finished and now I am having dinner watching this. Man, please do more podcasts like these. Got to learn a different thought process . Thanks Subhash and Arpit !! I hope to see more such podcasts on the channel.
Amazing ! In my recent experience working in a similar company to Dukaan but for the Middle East region. They have struggled alot to reduce the cost of AWS bills, this is the real pain for startups. Eye opener this is called "Asli Engineering" hats off, we need more podcasts with Subhash .. Already a fan of him
It's great to see.
Man he is knowledgeable and confident enough to show the inside workings of his company.
Really respect him
Question to Subhash: When you say you are doing read replica of Postgres to cloud and in case your on-prem db services is down, you will still serve it via cloud that means you have allowed writes to that cloud database or your application will work only in read-only mode until you bring back your on-prem services? If you allow write to cloud database, how you bring that delta back on your on-prem services?
when Arpit looked like a kid
One of the best podcast. I can say this is asli engineering and Subhash is very confident and fearless, "Fatega to dekha jayega, Zarurat hi ku h, complex krna hi ni h" 😄and he literally shutdown the actual server for the demo. I learnt, do not over design the architecture, just see the data and build.
Great and insightful podcast. I have couple of doubts -
1. Why do they have servers all over the world when their sellers and users are only in India?
2. For all the servers spread across the world, won't they have to connect with one single db in a particular region (or do they have read-replicas in each region)?
To maintain redundancy?
That's what I think.Even I'm interested in knowing more.
I think all of those servers are served from a single DB in India.
This is peak of youtube! SO MUCH KNOWLEDGE in a SINGLE VIDEO!
I am DevOps Engineer this guy almost concluded my entire job in just 1 and half hours😅😅
Never heard of this dukaan. Out of the blue you tube showed me :) checked google playstore, just looking into the 1st 4-6 comments close the app and stopped watching the remaining video. Not worth investing time.
What happened dude😂
Ane doh, Ane doh 😂😂, waiting for this video
There is a flaw in the design to handle peak traffic we cannot move the traffic/proxy to other data centers across the world, there are some regions which have data residency. Not very sure what Subhash was referring to when he said I can move traffic across the world, does he mean that he has compute that can run anywhere in the world? Do you have a globally replicated databases?
What's the replacement for S3 that dukkan is using ?
You can make your own S3 and maintain it .
CLOUD COMPUTING ISN'T A ONE-SIZE-FITS-ALL SOLUTION.
IF YOUR WORKLOAD REQUIRES OVER 1200+ SERVERS, IT MAY BE MORE FINANCIALLY PRUDENT TO MAINTAIN ALL SERVERS ON-PREMISES FOR THE LONG TERM.
Otherwise, CLOUD IS THE SOLUTION FOR YOU.....!!!!
Great work Dukaan team, one of the best podcast out there. Keep posting 💥💥💥
ok pranav
He fired whole dukaan team now saying its good
Subhash Choudhary, sir, looks like Viraj Ghelani, an Instagram influencer
Desi CTO 💌
Great podcast. It’s real stuff. Liked it a lot. Great if you do these similar ones from experts. May I ask one question, you mentioned why Kafka is fast. Can you please in details or give me some reference.
They told partial story. Lots of open risks.
Only DB moved to Bare metal? If no, then going bare metal also means that Dukaan might be facing high latencies in their multi-tenant application when hosted on cloud.
32GB RAM x 5 Server = 1TB RAM Kaise?
Also after seeing Arpit turned on while looking at low 250K IOPS it seems he has never touched a bare metal enterprise server
I think the worldwide servers are just for redundancy, not for serving requests. All their requests are served from a single DB (with a read replica maintained by them, and one more backup being replicated to GCP).
Listening in 0.75x speed😂
Dukaan is a cringe company, didn't they just lay off their whole support team?
He is dissing Google cloud at 1:23:57 infront of a Google Senior Staff employee 🤣🤣🤣
Backed by data :)
@@AsliEngineering I think this aspect is important to mention, and it's how conversations in engineering should be. Amazing talk!
This discussion was really a gem, from my experience what i have observed is when people talk about response time, everyone gets so defensive that the first point of check is either code, or query run time, adding multiple indexes to make the the query run faster, but we easily ignore the network calls most of the time. Some great take aways from this discussions for sure.
Clever use of anycast network, but there is one big issue with thats why gslb is preferred over it. In this example, he switched off loadbalancer hence demoing a complete failover. But what if there is issue in one of the python servers ( like slow performance because of gc ), then this solution won't work. In this case gslb, can remove machine intelligently from routing.
Completing it at 2:30am. I got so much to learn that I don’t know if I will get to sleep tonight or not. Amazing one man. Just happily overwhelmed & I am stunned looking at u guys. I’m 20 but now I want to learn & experience all these cool stuffs.
Great video Arpit Bro ❤❤❤
I liked the person more than the pod cast, just after watching 5 minutes
"Jindegi hai Software hai fatega aaj nahi kal"😂 😂
OnPrem is a better word than Bare Metal. BM instantly gave me a feel that they have done something with the hardware
Yeah. Correct. "On Premise" is terminology yes!
Technically, its not on-prem. Its located in a Datacenter, who provide bare metal servers instead of the abstraction provided by Cloud Providers. So, I think the word bare metal is appropriate.
Good stuff.
But what is difference between the AWS and the stuff he showed both are on cloud I don’t think so any he just picked cheap hosting nothing else
Cheap cloud . That we can get on aws as well
1. Django is running on devserver in production, it should be runing with gunicorn(wsgi) or uvicorn(asgi)
2. how is k8s deployed+managed (rke2? k3s? rancher?), looks like k3s which is meant to be for edge and not for Baremetal servers.
3. how is the HA (high availability), FT (fault tolerance) of k8s maintained? KubeVip?
4. etcd backup & restore?
5. mock disaster recovery of k8s?
They blocked anyone who question them they shared info about their customers.
Dukaan doesn't use resource limits & resource requests in the K8s Yamls, any specific reasons for that? It's generally understood to be a good practice to have these specified.
This is the answer to over engineering of things which don't even go live.
What an amazing podcast!!! Subhash sir is just on another level. Loved it throughout.
Mehnga mic lene se kuch nahi hota....Correct input stream bhi select karni hoti hai. Dynamic mic ko itna door nahi rakhta us point pe tujhe gain bhi badana padega, aur tere background noise aur static bhi badegi. Us point pe tu USB condenser mic lele, Trend main aake tune SM7B liya... iske saath cloudlifter toh liya hoga???
This was pure awesome. Loved every second of it ❤.
Ye to new hires ke liye KT video ban gayi Dukaan me work karne ke liye 😆
Everything from Infra, CI/CD, Git Workflow, sab bata diya 😆
Awesome podcast.
Quick question, when we have servers distributed across so many geographic regions to provide faster response time to the users what happens with the database? I mean all these servers still need to talk to the database and the db roundtrip can become expensive if the server and the db are far apart.
Do we create DB replicas for every geographic region were the servers are present? That could be expensive right? Also too many replicas for a RDBMS could result in the DB queries to slow down because the DB now needs to sync the write queries across all the replicas to maintain consistency.
This is where the cache layer will coming into picture. Which is replicated to form a ring across region along with background sync option.
You can keep it along with application or separate, upto your business needs and architecture.
I am also having the same question.
Depends on the size of your db
What if my all clusters die and my product got outage.
Now I plan to switch to AWS cloud.
1. I created Empty brand new Kubernetes Cluster
2. Added some more worker Nodes.
3. Deployed Argo into that new cluster.
4. I would need to create each application and project and configure everything about Argo, into that new cluster too.
How to avoid that last 4th step, and to automate that too?
how can I get the backend architecture of Dukaan? In this video, the functionalities and tech stack were explained brilliantly by Subhash sir. But, I wanted to read the diagram explaining the architecture and how the services are working with each other?
Is there any source?
Even though I didn't want to I had to pause the video and then like it before being mesmerized again 🔥🔥🔥
Benchmark on bare metal, it's possible just ran a test read_throughput:
read: IOPS=257k, BW=1004MiB/s (1053MB/s)(58.8GiB/60001msec)
What a brilliant podcast! Didn't shy from disclosing the real numbers... no one want's to find out the hard way. You and Subhash are proper engineers and I loved listening to you two talk technology. I've been a generalist for about 20 years, but love to experiment. I've learnt to setup servers etc. only since covid. So while my knowledge is surface level compared to you'll it was interesting to know that my learning is in the right direction. And just like you, i was astonished with the IOPS difference! Anyway, thank you for a wonderful podcast.
India is going to eat the competitionin coming future.
We need more such engineers! 🙏
Dilemma between changing the speed from 0.75 x and 1.25x
Can anyone answer this?
Did he pull the plug from just one microservice from the Amsterdam cluster or the Amsterdam cluster itself? If the latter, then who routed the traffic to another region if there's nothing to receive it in the first place? Or the API gateway is somewhere else? If you've idea pls lmk
he said himself that he is advertising same ip for every cluster. DNS request will will route to the shortest/closest route available
This is truly “Asli Engineering” ❤
Man this is dope! Would love to see a follow up podcast where we could see the database architecture. As in, are the databases sharded for a set of shops? Are all the app server regions of the same service hitting a single primary db?
16:16 sums it up. This is great for a talk show and an engineering project. Especially for a retail focused startup, if you want to become a unicorn and then go for IPO, moving out of cloud is not a great strategy. Hybrid is the way to go.
I hope they make a similar episode when they go back into the cloud.
Just for fun, I brought all my old PCs; one of them is running Nginx primarily for load balancing and reverse proxy, while the rest are running Express.js instances, Jellyfin, Nextcloud, and others (In docker of course). In my server, I do serve recorded videos for local coaching centers. But backing up your data is important.
Did you provide ABR streaming locally?
peak level engineering, amazing Subhash sir . Loved it amazing knowledge sir, Big fan 🙌
As an SRE I got offended in the beginning but have been worked in both DataCenter and Cloud environments. Definitely understand his requirements and his way of finding a solution for it ( and his comment on SRE ).
I work for a cloud provider so this is counter intuitive to I profess in general. However, the simplicity and ease with which Dukaan manage their infrastructure is amazing. My favorite thing was Subhash's enthusiasm for all this and how he takes pride in his tech work, truly inspiring!
I only worry about rapid scaling ability in future, however if the company doesn't project such an exponential growth, then this would work well for a foreseeable future.
This is a magical discussion!!! Man, this is an awesome podcast. I have Subscribed to the channel and turn the notifications ON, and kept shooting such podcasts. Really knowledgable.
I tried ignoring it several times as voice pace is too fast of the "Genius Guy Subhash" but always getting stick to this podcast :p
Totally insane, soch raha hu jeevan waste kar diya apna maine faltu ke kaamon mein being a developer
good interview, reminds me of my early days in startup, when our high-end server at datacenter crashed, and all the backup was on my laptop in office,
went to datacenter and i had to put the laptop in datacenter and run the business, surprisingiy laptop was in datacenter for 8 months, as we couldnt fix the highend machine as it was assembled and had multiple issues..
Subhash bhai is Asli Engineer, forget about system design jargons.. he is living system design😇
That’s what we call engineering, dekhte hai kya hota hai, kya fatega
Should have more such podcasts
How is the latency so low in all regions, doesn't the application connect to the database in the storefront (shown in the KeyCDN test), As I understood the database master is deployed in a single region.
Hats off you guys, literally I loved them this is what our Indian Software developer community needs ❤❤
WEB traffic ( or any thing which has session state) on IP anycast is not a good idea , More dangerous is advertising the same IP from 2 different gateways on from same region.( a simple routing change ) can cause lot of hashing change and traffic could drop.
I didnt get few things. How is he serving the sites from different locations ? Does he have servers rented from different locations ? and those servers are basically Kubernetes nodes which are joined to master nodes ? when he tested fail over what did he actually do ? Did he turn off the kubernetes worker node of new york ? He said he only have one gateway ? By gateway he means ISTIO gateway ?
What kind of disk were they using ua-cam.com/video/vFxQyZX84Ro/v-deo.html on gcp? 9000 bhut kam hai
Why 40% of Dukaan employees are open for job ? Anything bad about this org ?
Software hai fatega hi - Subhash
Around minute 50, y not use something like duckdns and not worry about single ip advertising by same instances
It's never a good content on the channel, it's always the best 😅🤣. Great insight 👍👍
This is Tanenbaum's book done bottoms up. Great talk, insight, guts and engineering.
Does that book also cover these kinds of topic? I didn't read that much. But IIRC, it was about the computer stack bottom up
Very informative video man, would love more of such videos/podcasts. Thanks 👍
There are more than 160 such videos on my channel. Do go through them. I have covered deep engineering concepts and real world systems.
@@AsliEngineering Oh, I will go through, Thanks 👍
Great Podcast, very enlightening even to someone who has very less knowledge about cloud Kudos to both 👏
Startups ka fail hone ka jyada chances hai server fail hone se
This is true man 😂😂
watching this at 0.75x while subhash speaks and 1.5x when arpit speaks
All was going good till this boy spoke about a db he is building ...sab db ek system call par khelta hai....ab apun bhi ek system call khojega 😂
Shubahsh Chowdhury is love yr he inspired me a lot for tech or he is a reason of my tech passion
Man power is still cheap in India. I would say still it’s cheap to run on premises. Using cloud as a backup was master stroke.
Craziness on another level!
great knowledge source , both of you guys are amaziing , as a backend senior dev , learned a lot , thanks for this session!
Great insights, how do they replicate realtime data across their regions? It was mentioned it does it on gcp, but how on premise? Does postgres support it out of the box?
It's been great working with Subhash. Really glad to be a part of his impressive operation.
35:17 although the server has been proxied to another server in a different location, since their db is in prem, wouldnt it still cause db servers to crash because of the extra load?
:) kuch log ..cisco routers hi nahi dekhe bhai...gateway ka concept kaise clear ho ga?! :)
Keeping the data on prem is a common thing. Uncommon is everything on cloud, thats kiddish.
dukaan ceo if u have no cybersec team beleive me take this video down, u shared way too much info in this.
How beautifully demonstrated ✨🔥❤️
Although this can help my organisation to fabricate new load balancers for L7 based load balancing as well internally instead of traditional way!!
This was a really good podcast, Arpit. As an old IT guy, I always knew self hosting is cheaper. But these insights from founder were phenomenal.
This podcast will increase the revenue for DELL, HPE, IBM and AWS, GCP, AZURE revenue will hit
Things dont work like that bro 😂
when subhash started speaking, i checked my playback speed 😅
Literally pagal hogaye! Had me there on the 19 year old kid, who goes to hypervisors and says karke dekhte hai 😂