Hi, We are running redis on a virtual machine. If we migrate the VM when the press is high, like running redis benchmark locally, the redis ping from another VM may be delayed much, even longer than 30s, and this may trigger master/backup swap if the timeout is set to 30s. During VM migration, we may use CPU or memory throttle to make the dirty data converge. This will certainly decrease the performance of the VM. However, is it the expected result that the ping delay becomes so long? Is it because redis server is single thread, and the performance degrade of every request accumulates to the delay of the ping request if it happens to be queued at the tail? I'm not an expert of redis. Looking forward to your real expert advice. Thanks, Heyi
Comment From: yossigo
Hi @iwishguo, reports of latency spikes when doing live VM migration are generally common and are not specific to Redis. This of course depends on many factors like the VM migration technology used, workload, dataset size and access pattern, etc.
I would suggest to consult your VM for possible ways to optimize this, or avoid live migration of Redis (which is something users often do for databases and other data-intensive apps).
Closing this ticket, please feel free to re-open if you have more details that indicate this is a specific issue with Redis.