Hi Bob, > On Feb 23, 2017, at 22:04, Bob Proulx <[email protected]> wrote: > > Assaf Gordon wrote: >> I'm leaning towards the possibility that it is related to >> secondary-groups/NFS issues, >> such as: >> >> http://unix.stackexchange.com/questions/206062/nfs-permission-problem-with-secondary-groups > > It does look like the same problem. Too bad there isn't a solution > there. > > We are using the -g manage-groups option with nfs. And it has been > working. So it is strange that it would now be failed after a > restart. > > This won't be encouraging but I had this exact same problem on another > system a couple of years ago and never figured it out. Eventually the > problem evaporated around me and I never knew what caused it.
I have the solution! (i think). This guy deserves a beer! http://www.xkyle.com/solving-the-nfs-16-group-limit-problem/ 1. Because we DO use "--manage-gids" in download:/etc/default/nfs-kernel-server , it means that rpc.mountd on download IGNORES the GIDs sent from download0 and fetches the group-ids by itself. 2. Which groups are fetched for each user can be checked here: # cat /proc/net/rpc/auth.unix.gid/content #uid cnt: gids... 0 1: 0 131035 0: 33 1: 33 65534 1: 65534 I'm user "131035(agn)", and it found no groups for me. Where do groups come from? MySQL on internal0 ! which was just rebooted! Then, checking download:/var/log/daemon.log shows clearly: === Feb 24 03:03:25 download mountd[4812]: libnss-mysql: Connection to server '208.118.235.75' failed: Lost connection to MySQL server at 'reading initial communication packet', system error: 113 === and 208.118.235.75 is old 'internal' which was shutdown! so somehow, "rpc.mountd" running on old download didn't get the memo :) And while other user-space programs on 'download' did use the correct mysql on 'internal0', rpc.mountd did not! or at least, someone the reboot messed things up. I would guess that perhaps just restarting rpc/nfs on download would fix it ?? But 'download' is too delicate for me to touch. Also: should the NFS on download0 be unmounted first? Fingers crossed! - assaf
