Redis HA - sentinel + haproxy

• sentinel: autoswitch master/slave
• haproxy: active check for only master node
• haproxy api: disable failed nodes to prevent multimaster

without haproxy maintenance mode

  A is master, B is slave
  A crashes, sentinel convert B to master
  A is recovered (still master)
  haproxy balancing between A and B, until sentinel convert A to slave
  data written to A are lost

witch haproxy maintenance mode via notification script

  A is master, B is slave
  A crashes, sentinel convert B to master
  haproxy disable A
  A is recovered (still master) but disabled in haproxy
  haproxy points only to B

Run redis cluster

redis-server --port 6666
redis-server --port 6667
redis-server --port 6668

Set slaves

redis-cli -p 6667 SLAVEOF 127.0.0.1 6666
redis-cli -p 6668 SLAVEOF 127.0.0.1 6666

Run sentinel

redis-server sentinel.conf  --sentinel

Run haproxy

haproxy -f haproxy.cfg -db

Open http://localhost:8080/ and try kill some redis

Notice

tested on

• redis 2.8.6
• haproxy 1.5-dev21

[!] in production on single host you must specify different data dir before SLAVEOF command otherwise you loose data on master