- prevent single point faliure
- batch requests
- scaling
- job recovery (retry)
RabbitMQ
- message acknowledgment
- message durability (both for queue and messages)
- dispatch: round robin (
prefetch_count
)
Redis
- suitable for short-live message
Kafka
- large amounts of data for a long time
NGINX
- load balance methods
- round robin
- least connection
- least time (PLUS)
- generic hash
- IP hash
- random
kube-proxy IPVS
rr
: round robinlc
: least connection (smallest number of open connections)dh
: destination hashingsh
: source hashingsed
: shortest expected delaynq
: never queue