Too many vCPUs for the underlying hardware?

[djh47 (Daren James Hasenkamp)]

I have a 64-core dedicated machine which in order to use, I must allocate a virtualized instance.

I have the option to virtualize with anything from 1 to 240 vCPUs. Perhaps naively, I decide to allocate the maximum of 240. No other virtualized instances are running on this hardware.

I set up elasticsearch on this instance and move traffic over from an older, high-traffic server which we are replacing.

Elasticsearch works normal at first but CPU usage and system load gradually increase until the entire instance becomes unresponsive. (Not just elasticsearch--the entire machine goes unresponsive.) The machine regains responsiveness after a minute or so, works for a few minutes, goes unresponsive again, and continues to cycle back and forth.

I've investigated the usual stuff, swapping, file descriptors, etc. Garbage collection looks normal. So I'm running out of ideas.

Is it possible -- am I using too many vCPU?

[stephenb (Stephen Brown)]

Hi @djh47 welcome to the community

Thin provisioning in other words overallocating vCPUs is absolutely not recommended for elasticsearch. Best practice also it's recommended to pin the vCPUs and the RAM for virtualized machines running elasticsearch.

I'm not sure if this is what's causing your problems but 4:1 over allocation of CPU is probably not a good idea.