Increase tail latencies when migrated service to arm64 architecture

Ask questions at https://stackoverflow.com/questions/ask?tags=go+redis

Our service uses redigo as redis client library. Our application sees an rps of 4k. So for p99 requests the latencies are under the sla but for tail latencies we saw redis Get calls were taking more that 1second. We checked server for slow queries but there were no slow queries since all calls wew just GET calls. We also saw cpu usage and memory usage less than 50%. Number of go routines were also normal. The only anomaly was that when we migrated our application from aws ec2 to graviton then after few days tail latencies started increasing. Once we reverted the deployments from arm64 to amd64 instances the latencies became normal.

Can anybody help us understanding the root cause?

gomodule / redigo

Increase tail latencies when migrated service to arm64 architecture #651