Suggestion/Clarifcation on Sentinel --down-after-milliseconds settings

[Dann S Washko]

What is the sweet spot for setting the --down-after-milliseconds setting for Sentinel? I see the default is 30 seconds and that seems awful high. From my understanding that would mean the Redis server could be unresponsive for at least 30 seconds before the sentinels would attempt a failover. That seems pretty high.

We are using Redis as a cache and initially figured that we should set this value as low as possible since response and uptime is critical. We started with 2000 milliseconds but we are seeing failovers on a regular basis, but no consistent pattern on what might be causing the failover. I am beginning to wonder if we have set the value to aggressively?

Are there any suggestions on this? My research has not found anything specific other than an indication that setting the value to a second seems to cause a lot of chatterring and pre-mature failover's.

Thanks for any advice.

[The Real Bill]

In my testing I've found that depending on networking three seconds is the lower bounds. But really it is a mix of your hard requirements and what your network will let you get away with. Oh and if you frequently run any long commands such as flushing millions of keys.

Cheers,
Bill

[Dann S Washko]

Thanks a lot The Real Bill, I appreciate the feedback.