Thread Pool max Size and Number of Processor

Hi, In latest elasticSearch Version 7 series, How to decide the maximum Search Thread Pool Size, will it depends on number of vCPUs , and decided by the formula (vCPU*3/2)+1 OR the max size still is based on 32 vCPUs as was earlier .Thanx

Welcome to our community! :smiley:

It's based on the number of processors allocated to the node, see Thread pools | Elasticsearch Guide [7.12] | Elastic and Thread pools | Elasticsearch Guide [7.12] | Elastic

Thanx for reply. I had also gone through the Thread Pool Guide but somewhere I have read that there is max limit of 32 Thread irrespective of allocated vCPUs (even more than 32). So now it's confirmed that , if I increase the vCPUs to cater my heavier load , Thred Pool size will be more. For Ex, If I assign 48 vCPUs, then effecctive search Thread Pool size will be 48*3/2 + 1 = 73
Pl acknowledge

For the search thread pool, yes.

Thanks

But there has to be limit for maximum number of vCpu allocation for a node because vertical scaling is not supported by elastic search...what is limit for vCpu allocation ?

Welcome to our community! :smiley:

Why do you say that?

Once i had a meeting with elasticsearch team In that meeting they mentioned that vertical scaling is possible upto some limit (due to heap memory and garbage collection ), beyond that we need to do horizontal scaling,So i want to know about the limit of that vertical scaling . is there any benchmark for that?

I think you should create a separate topic on that question :slight_smile:

1 Like

What is your use case? Have you identified what is limiting performance for you? Elasticsearch is often more limited by storage performance that CPU or memory. The exception to this is e.g. if all your data is cached in the operating system page cache and is rarely changed. How much data do you have? What is the hardware specified fixation of the node/cluster? Is it a query heavy load?

It'd be better to move any further discussions into a new thread, to save taking this off topic :slight_smile: