kern/146792: [flowtable] flowcleaner 100% cpu's core load

[Brandon Gooch]

On Sun, May 23, 2010 at 5:06 PM, Kurt Jaeger <pi at opsec.eu> wrote:
> The following reply was made to PR kern/146792; it has been noted by GNATS.
>
> From: Kurt Jaeger <pi at opsec.eu>
> To: bug-followup at FreeBSD.org, niko at gtelecom.ru
> Cc:
> Subject: Re: kern/146792: [flowtable] flowcleaner 100% cpu's core load
> Date: Sun, 23 May 2010 16:19:25 +0200
>
>  Hi!
>
>  I observe a similar behaviour on a 8.0-RELEASE-p2 i386 GENERIC
>  kernel.
>
>  System receives 2 BGP4 fullfeeds (approx. 310K routes each).
>
>  The system is still running, a few processes are unkillable or
>  die only after a long amount (1-2h) of time.
>
>  Here's the list of unkillable processes:
>
>  80871  ??  R      0:00.00 /bin/sh /etc/periodic/daily/470.status-named
>  76499  ??  Rs     0:00.01 sshd: [accepted] (sshd)
>  76922  ??  Rs     0:00.01 sshd: [accepted] (sshd)
>
>  flowcleaner looks pretty busy (for an uptime of approx. 40h):
>
>    22  ??  RL   1209:50.98 [flowcleaner]
>
>  4:17PM  up 1 day, 22:22, 2 users, load averages: 7.20, 6.53, 5.81
>
>  quagga is running on the system, bgpd mgmt cli is no longer reachable:
>
>  # telnet 0 2605
>  Trying 0.0.0.0...
>  Connected to 0.
>  Escape character is '^]'.
>
>  ^]
>  telnet> close
>  Connection closed.
>  #
>
>  What can I do to help to debug this ?
>  No console access available right now, but can probably made available.
>
>  This is a production host, but not yet super-critical, so...

I know absolutely nothing about quagga, and very, very little about
the flowcleaner process (or flowtable, no man page), but I DO KNOW
that Kip Macy suggested disabling:

options FLOWTABLE

from the kernel config of the machine experiencing the issue. This was
back in December 2009, so I'm not sure about a resolution to the
actual problem (or if it is just inherent in the design of the per-cpu
routing cache).

Perhaps Kip may have more insight?

-Brandon