500 RPS per client limit?

[yorek]

Description of issue

It seems that using latest distributed version 0.11.0 on Debian Stretch or Ubuntu I cannot do more than around 500 RPS per locust slave. Slaves are 8GB 4CPU linux machines

Expected behavior

Scale as much as slave resources allows

Actual behavior

Each slaves stops at something around 500 RPS

Environment settings

• OS: Linux (Debian or Ubuntu)
• Python version: 3.6.8
• Locust version: 0.11

Steps to reproduce (for bug reports)

Create a container based on python:3.6.8-stretchj and run a distributed locust. Try to send more then 500 HTTP request per second to an HTTP endpoint

[yorek]

Forgot to mention that using v 8.1 on Alpine Linux I don't have that problem (using exactly the same locust file)

[cgoldberg]

add more local slaves so you can use all CPU cores

[yorek]

This doesn't answer the question completely. Don't understand why you closed the issue. Is there any know limit the prevents locust to have more the 500 RPS? With locust 8.1 I can easily do more, with locust 11, I can't. Any changes that may have introduced this limit?

[cgoldberg]

This doesn't answer the question completely.
Don't understand why you closed the issue.

This is not a forum for questions... We only keep actionable bugs/issues open. For general questions, please join the Slack channel.

Is there any know limit the prevents locust to have more the 500 RPS?

no limits besides resources. But remember, locust only runs in a single process, so it won't make use of multiple cpu cores unless you use multiple slaves.

[yorek]

Thanks for the answers :)

[aldenpeterson-wf]

This was in the body of the issue you opened:

For general questions about how to use Locust, use either the Slack link provided in the Readme or ask a question on Stack Overflow tagged Locust.

[yorek]

Yeah, but mine was not a general "how to use" question. It was: with version 8.1 I can do 550 RPS per locust while, running exactly the same locust file, with any version after that I can do only 350. It's a huge drop and it may be caused by a bug, a change in the used HTTP library or anything else. It looks like an issue to me, not an "how to" question. I though you may also be interested in knowing this and figuring out why....