Connect to `etcd` server having minimal latency

[yunkon-kim]

Connect to `etcd` server having minimal latency

[yunkon-kim]

In this experiment, the etcd cluster has 3 member nodes as follows:

• Member 1: Located in Singapore (AWS)
• Member 2: Located in Seoul (GCP)
• Member 3: Located in Seoul (MS Azure)

Case 1: individual connection

A test program established the connection to each member node respectively.
I was able to get a response to the Status() call right away, but not for other calls.
I think it takes time to perform Raft (consensus algorithm) in a cluster.
As a result, it has an average response time.

Test result

Test case	Member 1	Member 2	Member 3
Status() call (like ping)	232ms	18ms	15ms
Health-check REST call	154ms	172ms	159ms
Get() call	149ms	169ms	154ms
Put() call	169ms	175ms	154ms

Case 2: multiple connections

A test program established the connection by multiple endpoints to utilize client load balancer.
There was not much difference in response time even when using multiple endpoints for the client load balancer.

Test result

Test case	Member 1	Member 2	Member 3
Get() call	149ms	172ms	153ms
Put() call	172ms	153ms	152ms

Currently, it seems that the effect of the latency is less than expected.