That is true. FWIW, as an experiment I built a 50-node ECS cluster and launched 4000 tasks on it. I couldn't build a single service with 4000 tasks due to soft limits, so I built 4 services with 1000 tasks each (nginx, 256MB, 128 CPU units). No load balancer.
Overall task replacement time was about 15 minutes -- which is pretty respectable, in my view. Obviously not as fast as doing a hot code swap, though. Tradeoffs abound.
Comments
That is true. FWIW, as an experiment I built a 50-node ECS cluster and launched 4000 tasks on it. I couldn't build a single service with 4000 tasks due to soft limits, so I built 4 services with 1000 tasks each (nginx, 256MB, 128 CPU units). No load balancer.
Overall task replacement time was about 15 minutes -- which is pretty respectable, in my view. Obviously not as fast as doing a hot code swap, though. Tradeoffs abound.