Locust: I had increase open file limit but still exception "Device or resource busy"?

[theanhvo]

With Logs

Mar  3 09:32:32 loadtest-v2 locust: thread_result.set(func(*args, **kwargs))
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,558] loadtest-v2/ERROR/stderr:
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,558] loadtest-v2/ERROR/stderr: OSError: [Errno 16] Device or resource busy
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,558] loadtest-v2/ERROR/stderr:
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,558] loadtest-v2/ERROR/stderr: 2020-03-03T02:32:32Z
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,559] loadtest-v2/ERROR/stderr:
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,559] loadtest-v2/ERROR/stderr: (<ThreadPoolWorker at 0x7f6f9aa7ccc0 thread_ident=0x7f6f8dbe5700 hub=<Hub at 0x7f6f93947588 thread_ident=0x7f6fa2693740>>, <built-in function getaddrinfo>) failed with OSError
Mar  3 09:32:32 loadtest-v2 locust: [2020-03-03 09:32:32,559] loadtest-v2/ERROR/stderr: Traceback (most recent call last):
Mar  3 09:33:11 loadtest-v2 locust: [2020-03-03 09:33:11,683] loadtest-v2/WARNING/urllib3.connectionpool: Connection pool is full, discarding connection: ...

[cyberw]

I did a quick search and found that this is most likely related to gevent's DNS resolving.

Are you using a very high number of clients/users (above maybe 5000 per process) or high rampup (above maybe 100/s per locust process)?

Would it be possible for you to use ip-adresses instead of names?

Maybe you can try one of gevent's other name resolving strategies http://www.gevent.org/dns.html

Or try using the new FastHttpLocust instead of the regular one.