Closed sflow closed 1 month ago
Would raising the netlink socket write buffer be helpful?
In linux-cp
, we specifically set larger read buffer so that the kernel can burst as we retrieve netlink messages.
See LIBNL's nl_socket_set_buffer_size()
in src/plugins/linux-cp/lcp_nl.c L803 and L918.
It turns out this is related to issue #6
Under high load conditions we see EMSGSIZ returned from sendmsg() call to send sample to PSAMPLE, although some samples appear to be getting through. If the netlink socket were full we might expect to see EAGAIN or EWOULDBLOCK. Not sure why we get EMSGSIZ.