a word of caution about the new default of so-reuseport on FreeBSD

Unbound
1

Hi,

this is a heads-up for people that are about to upgrade to Unbound 1.8.0
on FreeBSD.

When upgrading (and before restarting unbound) I explicitly set
the new defaults in the conf file to make their value explicit.
( so-reuseport, harden-below-nxdomain, minimal-responses)

After I upgraded and restarted unbound all queries were handled by only a single thread
even though I have multiple threads enabled in the config, as a result of that the number of
queries in the queue skyrocketed.

I expected this to be related to so-reuseport and after setting that to 'no',
things were back to normal (all threads handled queries again, queue size back to normal).

kind regards,
nusenu

2

Thank you for bringing this to our attention. so-reuseport behaves
indeed different for Linux and FreeBSD. We will schedule a different
default so-reuseport config setting for FreeBSD in Unbound 1.8.2. (It
will not appear in 1.8.1 which will released this week.)

Best regards,

-- Benno

3

The upcoming FreeBSD 12 has a new SO_REUSEPORT_LB setsockopt that does
the same thing as Linux's SO_REUSEPORT.

4

Hi,

5

Just had a chance to test it. It works great, thanks!

# unbound-control stats_noreset | grep 'num.queries='
thread0.num.queries=17662
thread1.num.queries=17191
thread2.num.queries=17269
thread3.num.queries=17419
total.num.queries=69541

6

Wouter Wijngaards via Unbound-users:

Hi,

The upcoming FreeBSD 12 has a new SO_REUSEPORT_LB setsockopt that does
the same thing as Linux's SO_REUSEPORT.

Thanks for telling that, I have implemented support! It tries to set
the SO_REUSEPORT_LB option when the so-reuseport config option is used.

after enabling so-reuseport on FreeBSD 12.0 (unbound 1.8.2)
the number of handled qps dropped to 0 on half of the threads.
Is this expected?