JioCinema

stream live IPL ~20 million concurrent devices
Tons of audits on each hop of request. Frontend, load balancer, etc.
Calculate the limit of scaling for each of the service
War rooms, dashboard and metrics
Frontend is code freezed much much before
Use feature flags, so that features which are causing lot of issues can be turned off
- In case of fire, switch off less critical components
Perform feature flags simulation, performing permutations/combinations
config service returns feature flags
autoscaling takes time to kick in, couple of minutes to an hour
autoscaling is not useful since game would have ended by then
scale up before the game, and perform back of the envelop calculation
- how many requests (for example 20 million)
- how many database calls per request
- need to ask cloud providers before to give the capacity (nodes)
- Sometimes it could be 75% of India’s bandwidth
If Database goes down:
- There are some APIs which basically are put in static storage with fixed response
- APIs are routed to static storage instead of origin server
- This is known as Panic mode, User does not feel that the system is down
Scale down:
- Never scale down between match, since Dhoni might come back for bowling
- After the match scale down on ladders based on number of users
- Do not scale down immediately since There are a lot of users sticking around for some time after the match
Kafka:
- Producer rate consumer rate
- Plan partitions based on above rates

More Notes