Re: squid, DNS and no buffer space

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Fri, 6 Dec 2024, Michael van Elst wrote:
> On Thu, Dec 05, 2024 at 11:34:56AM +0000, Stephen Borrill wrote:
>
> > Here's a netstat -mssv at the time when squid is complaining:
> >
> > 18043 mbufs in use:
> >         3502 mbufs allocated to data
> >         14538 mbufs allocated to packet headers
> >         3 mbufs allocated to socket names and addresses
> > 0 calls to protocol drain routines
>
>
> >              tcp               inuse         14538          0          0
> >                                claims      6036881          0          0
> >                                releases    6022343          0          0
>
> Looks fine for a busy system.
>
> Please check the kern.somaxkva value, it is probably at
> the 16MB default, if yes this would be the limit that you hit
> with about 9MB mbufs and 7MB mbuf clusters allocated.

Seems plausible. I already had kern.sbmax=67108864

2024/12/06 09:21:24 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:04:43 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:18:45 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:21:31 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:33:29 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:33:54 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 10:38:30 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available

At 10:42, I increased from 16MB to 64MB (and to match kern.sbmax):
10:42   sysctl -w kern.somaxkva=67108864

But the warnings continued:

2024/12/06 12:11:39 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 12:11:45 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 12:16:24 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 12:39:12 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 12:39:13 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 12:39:57 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 13:00:25 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 13:15:41 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available

At 13:17 I doubled the values again:
13:17   sysctl -w kern.somaxkva=134217728
13:17   sysctl -w kern.sbmax=134217728

The warnings continued:

2024/12/06 13:24:01 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 13:26:35 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 14:23:01 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 14:30:52 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available
2024/12/06 14:31:09 kid1| idnsSendQuery FD 9: sendto: (55) No buffer space available

I wrote a script to output a massaged version of netstat -m every minute 
and final 3 warnings above correspond to:

Date       Time     data  packets names 
2024/12/06 14:22:39 16084 12186 3
2024/12/06 14:23:39 23610 13078 3
2024/12/06 14:30:40 7915 12214 3
2024/12/06 14:31:40 9023 13282 3
2024/12/06 14:32:40 7677 13278 3

These numbers don't seem particularly high

--
Stephen