10.0-RC1: bad mbuf leak?

[Mark Felder]

On Wed, Jan 8, 2014, at 14:57, Adam McDougall wrote:
> On 01/08/2014 15:45, Mark Felder wrote:
> > On Wed, Jan 8, 2014, at 14:32, Adam McDougall wrote:
> >> On 01/06/2014 13:32, Mark Felder wrote:
> >>> It's not looking promising. mbuf usage is really high again. I haven't
> >>> hit the point where the system is unavailable on the network but it
> >>> appears to be approaching.
> >>>
> >>> root at skeletor:/usr/home/feld # netstat -m
> >>> 4093391/3109/4096500 mbufs in use (current/cache/total)
> >>> 1025/1725/2750/1017354 mbuf clusters in use (current/cache/total/max)
> >>> 1025/1725 mbuf+clusters out of packet secondary zone in use
> >>> (current/cache)
> >>> 0/492/492/508677 4k (page size) jumbo clusters in use
> >>> (current/cache/total/max)
> >>> 0/0/0/150719 9k jumbo clusters in use (current/cache/total/max)
> >>> 0/0/0/84779 16k jumbo clusters in use (current/cache/total/max)
> >>> 1025397K/6195K/1031593K bytes allocated to network (current/cache/total)
> >>> 0/0/0 requests for mbufs denied (mbufs/clusters/mbuf+clusters)
> >>> 0/0/0 requests for mbufs delayed (mbufs/clusters/mbuf+clusters)
> >>> 0/0/0 requests for jumbo clusters delayed (4k/9k/16k)
> >>> 0/0/0 requests for jumbo clusters denied (4k/9k/16k)
> >>> 0 requests for sfbufs denied
> >>> 0 requests for sfbufs delayed
> >>> 0 requests for I/O initiated by sendfile
> >>>
> >>> root at skeletor:/usr/home/feld # vmstat -z | grep mbuf
> >>> mbuf_packet:            256, 6511065,    1025,    1725, 9153363,   0,  
> >>> 0
> >>> mbuf:                   256, 6511065, 4092367,    1383,74246554,   0,  
> >>> 0
> >>> mbuf_cluster:          2048, 1017354,    2750,       0,    2750,   0,  
> >>> 0
> >>> mbuf_jumbo_page:       4096, 508677,       0,     492, 2655317,   0,   0
> >>> mbuf_jumbo_9k:         9216, 150719,       0,       0,       0,   0,   0
> >>> mbuf_jumbo_16k:       16384,  84779,       0,       0,       0,   0,   0
> >>> mbuf_ext_refcnt:          4,      0,       0,       0,       0,   0,   0
> >>>
> >>> root at skeletor:/usr/home/feld # uptime
> >>> 12:30PM  up 15:05, 1 user, load averages: 0.24, 0.23, 0.27
> >>>
> >>> root at skeletor:/usr/home/feld # uname -a
> >>> FreeBSD skeletor.feld.me 10.0-PRERELEASE FreeBSD 10.0-PRERELEASE #17
> >>> r260339M: Sun Jan  5 21:23:10 CST 2014
> >>> _______________________________________________
> >>> freebsd-stable at freebsd.org mailing list
> >>> http://lists.freebsd.org/mailman/listinfo/freebsd-stable
> >>> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at freebsd.org"
> >>>
> >>
> >> Can you try your NFS mounts from directly within the jails, or stop one
> >> or more jails for a night and see if it becomes stable?  Anything else
> >> unusual besides the jails/nullfs such as pf, ipfw, nat, vimages?  My
> >> systems running 10 seem fine including the one running poudriere builds
> >> which uses jails and I think nullfs, but not nfs.  Do mbufs go up when
> >> you cause nfs traffic?
> >>
> > 
> > You can't do NFS mounts from within a jail, which is why I have to do it
> > this way.
> > 
> > Nothing else unusual. Very few services running. The box sits mostly
> > idle and the traffic is light -- watching some TV shows (the jail runs
> > Plex Media Server). I haven't been able to locate a reason for the mbufs
> > to go up, but often a wake up in the morning after it has been doing
> > nothing all night and see it made a large jump in mbufs used. When I'm
> > running an 11-CURRENT kernel these problems do not exist.
> 
> Can you have a script run some stats like netstat -m every few minutes
> during the night to see if it happens at a particular time?  I'm
> wondering if the system scripts are crawling the mountpoints to cause
> this.  Alternately, as far as NFS mounts and jails, with a reasonable
> amount of work could you replace the nullfs/nfs usage with temporary NFS
> mounts outside of the jails but mounted in the jail root fs?
> 

Yes, that won't be difficult. I only nullfs because it's convenient to
be able to access the mounts without having to be in the jail :-)

I also just realized that there will be some filesystem activity when
I'm not using the system: when a download completes my NAS signals the
Plex program to rescan for new media files, so it does walk the entire
NFS mount occasionally.