- 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
redis-server --port 6666
redis-server --port 6667
redis-server --port 6668
redis-cli -p 6667 SLAVEOF 127.0.0.1 6666
redis-cli -p 6668 SLAVEOF 127.0.0.1 6666
redis-server sentinel.conf --sentinel
haproxy -f haproxy.cfg -db
Open http://localhost:8080/ and try kill some redis
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
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent