2:50 Today we run node that are 10s of Terrabytes, holds 100's of Billions of Financial Records 4:12 we had 5 thousand users and today I think we have 11 Million users 4:32 & every morning when stock markets open we have like 2+ Million user who Login in within a span of let's say 10 or 15 seconds, so we handle huge amount of concurrency & real time traffic 4:56 & DataBases that serve 100's of 1000's of request every second 8:08 It Run's on the EC2 System which cost peanuts 14:07 we Built our earlier systems in python & we soon ran into Big stumbling blocks and bottle necks with python, so we picked up GO 14:30 Even today GO all our High concurrency High Traffic systems are written in GO & when you trade on our trading platform ( kite ) we have a baseline most request finish in Less than 15ms 16:44 K8 was a huge black box that we struggled to understand, we Discarded it. so today we use Nomad... I think now the code that we write for infrastructure is **really simple**, it's something that a new developer can sitdown read and understand lets say 20 Lines of configuration, & how ur application id deployed within in ur stack is also really really visible, it's not hidden behind Blackboxes, so it's a principle that we value greatly **simplicity** 18:45 so that again gave us the Privelage & Laxury of doing things really really.. really slowly 20:53 the desition to master a database before using it 20:30 the small decisions dt ( compounded ) 30:06 ListMonk, we use it at Zerodha to send out 200 Million emails every month, & ListMonk is self hosted on an EC2 instance, on a tiny EC2 instance. the emails that hv to go out has to hit an SMTP server, v don't trust an external SMTP server providers so v self host Haraca ( FOSS ) & d cost is peanuts.
Is there any guarantee from the linux kernel that it will deliver the message from A2 to A1 when they're both listening on the same port and A2 forwards the message back to that port? What about when there are 3 or more pgbouncer instances listening on the same port? It seems to me like you could just end up thrashing between A2 and A3 when the cancel request needs to go to A1.