Error message when hold more than 30000 sockets about dpdk-ans HOT 11 CLOSED

Do you invoke anssock_close() or anssock_shutdown to close the fd?

You run so many lcore, so you shall update ans_main.c to increase the mbuf pool size.

from dpdk-ans.

If can't malloc mbuf from pool, will print error code 12 (NOMEM).
Please increase the mbuf pool size.
And share your detail test steps.

from dpdk-ans.

After I increase the mbuf pool size, It works fine. Thanks.
Maybe you treat the close of connection as a EPOLLIN event and would malloc a mbuf to store the data,
It confused me at first.

from dpdk-ans.

Yes, if remote send 'fin' to ANS, ANS will notify application via EPOLLIN event.

from dpdk-ans.

Could you please share the test result?

from dpdk-ans.

I want to verify whether I could use it to hold 1M connection, the result is OK.
Then I would test if it works OK when read and write on such a mount of connections, especially If it would block when do connect, read or write action.

from dpdk-ans.

Good.
What is your ENV? CPU? NIC?
How many lcores are reserved for ANS?
How many hugepage are reserved?

from dpdk-ans.

I use Cenos6.5, kernel 3.18
CPU:Intel(R) Xeon(R) CPU E5-2620 0 @ 2.00GHz
NIC: Intel Corporation 82599ES 10-Gigabit SFI/SFP+ Network Connection (rev 01)
alloc 4 cores for ANS
4096 * 2MB hugepages, if increase the mbuf pool size, more hugepages needed.
Create 1M sockets and connect to another directly (port to port without any router and switcher), cost about 8 minutes.

from dpdk-ans.

Alse there is a serious problem, the connect action would blocking, It must considered in high load environment.

from dpdk-ans.

Yes, anssock_connect() API is always block API, it block there to wait remote server reply.

from dpdk-ans.

Any result for your reading/writing case?

from dpdk-ans.