CPU wait consistently high - Postgres process
by carolinahusker from LinuxQuestions.org on (#571NC)
A VM running on a vSphere ESX 6.7 host is showing consistently high CPU wait times. The VM is a postgres database server. Top running from the VM shows that the CPU wait time is usually 40+. Iotop shows that only about 50-100 M/s are reading from storage and (with 90%+ IO) the command is a postgres select statement. Iostat shows that disk utilization is 95%+ on the disk. The VM was using a datastore with an SSD cache layer. I moved it to a different compute cluster that is backed by an all flash array, but I'm seeing the same result. Previous iPerf tests showed that the network is not the bottleneck and that makes sense given that the hosts have 40Gbps NICs and are connected to the same core swtich (less than 1msec of latency).
It seems like this is a storage issue, but I moved it to an all flash array with plenty of capacity. Other VMs on that compute cluster do not have performance issues.
What other testing should I try to sort this out?
It seems like this is a storage issue, but I moved it to an all flash array with plenty of capacity. Other VMs on that compute cluster do not have performance issues.
What other testing should I try to sort this out?