Redis2.8.8, how does sentinel handle other sentinel failure

[nfwlpw]

How does redis sentinel handle the detection of another sentinel going down while determining if it should fail over

Here's our set up, 2 VMs, A and B, each runs redis-server and redis-sentinel, and redis-serverB is the slaveof redis-serverA.  The quorum is set to 2.

So in our set up, from my testing, if something catastrophic happened to VM A and turned off both redis-serverA and sentinelA, sentinelB will never initiate failover because the master will not be considered ODOWN due to quorum being set to 2, even though it seems to know it's the only sentinel left according to sentinelB's log:

[15170] 18 Apr 16:06:32.555 # +sdown master mymaster
[15170] 18 Apr 16:06:59.266 # +sdown sentinel

Is there a way to get around this without setting quorum to 1, or putting sentinel on its own VMs?

[Adi Chiru]

In my test in a very similar scenarios I found the same and ended up having more sentinels. So just add a sentinel on another VM that may any other role, not necessarily a redis-server running on it. So you don't need to have VMs specifically for sentinel processes since they can run on any other host (that makes sens from a networking perspective) and help with voting, authorizing, promoting etc.

I would also definitely run sentinels on eac redis-server instance too.

Hope it helps.

Adi