Harald Schmalzbauer
2014-Jul-24 09:03 UTC
nfsd server cache flooded, try to increase nfsrc_floodlevel
Bez?glich Rick Macklem's Nachricht vom 08.08.2013 14:20 (localtime):> Lars Eggert wrote: >> Hi, >> >> every few days or so, my -STABLE NFS server (v3 and v4) gets wedged >> with a ton of messages about "nfsd server cache flooded, try to >> increase nfsrc_floodlevel" in the log, and nfsstat shows TCPPeak at >> 16385. It requires a reboot to unwedge, restarting the server does >> not help. >> >> The clients are (mostly) six -CURRENT nfsv4 boxes that netboot from >> the server and mount all drives from there. >> >> I googled around and saw that others have hit this issue, but I >> haven't seen any resolution posted. I guess I can increase >> NFSRVCACHE_FLOODLEVEL in the source, but I wonder if I wouldn't >> simply hit the increase value after a little while longer... >> >> Lars >> > You can either try this patch (which dynamically adjusts nfsrc_floodlevel > along with handling a variety of overhead issues for the DRC under heavy load): > http://people.freebsd.org/~rmacklem/drc4.patch > > or just bump it up a bunch. The default value was safe for a server with 256Mbytes > of ram and a default mbuf cluster limit. The only thing you might have to do > along with bumping NFSRC_FLOODLEVEL up is increasing kern.ipc.mbclusters. > > The variant of the above patch will make it into head someday, once I merge > in changes from ivoras@'s similar patch and confer with him about it.Dear all, regarding the conversation from last year - quoted above, I think I found the mentioned patch (it's variants) MFCd in r255532 (from http://svnweb.freebsd.org/base?view=revision&revision=254337), so it's included in 9.3-RELEASE. Unfortunately I'm still having the nfsrc_floodlevel problem with OpenOwner=16385, CacheSize=16385 (in nfsstat -e -s) in my production environment under 9.3-RELEASE-amd64. Extremely light load on the server (2 (FreeBSD8/9) clients), but the building client (nfsv4) locks up frequently. It mounts 'home' and 'ports/ports' via NFSv4 (this time, 'make index' in nfs-mounted /usr/ports killed the nfsv4server). I found another interesting 3 years old patch/thread, which seems never beeing comitted: http://lists.freebsd.org/pipermail/freebsd-fs/2011-July/012016.html I don't really understand all these details of nfs(v4), but I observe problems with regular usage, so I wanted to ask if there are new findings regarding the "nfsd server cache flooded, try to increase nfsrc_floodlevel" messages (while 'nfsrc_floodlevel' doesn't seem to be tunable in 9.3). To my understanding, it's a problem on the server side, right? Is the fix from 3 years back still adequate (does apply with view offsets only to 9.3)? I'm currently testing 9.3-RELEASE+noopen.patch, but it usually took two or three days until the client locked up (hadn't looked for the reason before the last issue, nfs(v4) was brand new reintroduced here) Thanks, -Harry -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 196 bytes Desc: OpenPGP digital signature URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20140724/f10e69d3/attachment.sig>
Rick Macklem
2014-Jul-25 00:14 UTC
nfsd server cache flooded, try to increase nfsrc_floodlevel
Harald Schmalzbauer wrote:> Bez?glich Rick Macklem's Nachricht vom 08.08.2013 14:20 (localtime): > > Lars Eggert wrote: > >> Hi, > >> > >> every few days or so, my -STABLE NFS server (v3 and v4) gets > >> wedged > >> with a ton of messages about "nfsd server cache flooded, try to > >> increase nfsrc_floodlevel" in the log, and nfsstat shows TCPPeak > >> at > >> 16385. It requires a reboot to unwedge, restarting the server does > >> not help. > >> > >> The clients are (mostly) six -CURRENT nfsv4 boxes that netboot > >> from > >> the server and mount all drives from there. > >>Have you tried increasing vfs.nfsd.tcphighwater? This needs to be increased to increase the flood level above 16384. Garrett Wollman sets: vfs.nfsd.tcphighwater=100000 vfs.nfsd.tcpcachetimeo=300 or something like that, if I recall correctly. rick> >> I googled around and saw that others have hit this issue, but I > >> haven't seen any resolution posted. I guess I can increase > >> NFSRVCACHE_FLOODLEVEL in the source, but I wonder if I wouldn't > >> simply hit the increase value after a little while longer... > >> > >> Lars > >> > > You can either try this patch (which dynamically adjusts > > nfsrc_floodlevel > > along with handling a variety of overhead issues for the DRC under > > heavy load): > > http://people.freebsd.org/~rmacklem/drc4.patch > > > > or just bump it up a bunch. The default value was safe for a server > > with 256Mbytes > > of ram and a default mbuf cluster limit. The only thing you might > > have to do > > along with bumping NFSRC_FLOODLEVEL up is increasing > > kern.ipc.mbclusters. > > > > The variant of the above patch will make it into head someday, once > > I merge > > in changes from ivoras@'s similar patch and confer with him about > > it. > > Dear all, > > regarding the conversation from last year - quoted above, > I think I found the mentioned patch (it's variants) MFCd in r255532 > (from > http://svnweb.freebsd.org/base?view=revision&revision=254337), > so it's included in 9.3-RELEASE. > > Unfortunately I'm still having the nfsrc_floodlevel problem with > OpenOwner=16385, CacheSize=16385 (in nfsstat -e -s) in my production > environment under 9.3-RELEASE-amd64. > Extremely light load on the server (2 (FreeBSD8/9) clients), but the > building client (nfsv4) locks up frequently. It mounts 'home' and > 'ports/ports' via NFSv4 (this time, 'make index' in nfs-mounted > /usr/ports killed the nfsv4server). > > > I found another interesting 3 years old patch/thread, which seems > never beeing comitted: > http://lists.freebsd.org/pipermail/freebsd-fs/2011-July/012016.html > > I don't really understand all these details of nfs(v4), but I observe > problems with regular usage, so I wanted to ask if there are new > findings regarding the "nfsd server cache flooded, try to increase > nfsrc_floodlevel" messages (while 'nfsrc_floodlevel' doesn't seem to > be > tunable in 9.3). > To my understanding, it's a problem on the server side, right? > > Is the fix from 3 years back still adequate (does apply with view > offsets only to 9.3)? > > I'm currently testing 9.3-RELEASE+noopen.patch, but it usually took > two > or three days until the client locked up (hadn't looked for the > reason > before the last issue, nfs(v4) was brand new reintroduced here) > > Thanks, > > -Harry > >