0

I am experiencing diskless clients losing connection to their nfsroot server within 24 hours of booting. Initially I thought it was hardware related as i simultaneously upgraded 16 blades from Centos6 to Centos7 (diskless/pxe boot with nfsroot) and they all lose connection at the same time after booting ok and running 12 hours+. When they do they all print to the console "task blocked for more than 120 seconds". I setup one of the blades to boot from local disk and when reproducing the problem the 15 diskless blades fail as described and the blade with boot disk continues as before. The nfs server continues serving other clients fine.

I've concluded that my nfsroot connection is getting lost on these diskless blades (Dell M620s in M1000e chassis). Nothing interesting is getting logged in messages file either end. I do not think it is hardware because the all that's changed is upgrade from Centos6 to 7, although there could be compatibilty issue i suppose. The hardware does claim to support Centos7.

Can anyone advise good way to debug why the nfsroot conenction is getting lost ? kernel = 3.10.0-1160.59.1.el7.x86_64

4
  • About the ton of possibilities: booting using DHCP without keeping a DHCP client to renew the lease? As the subject is broad, I can't tell about the One method to debug this. You have to try it for all possible aspects: network, restart of critical tools etc.
    – A.B
    Commented Jan 14, 2023 at 16:15
  • "without keeping a dhcp client to renew the release ?" I had not thought of that but according to my logs both the pxe interface and the booted client interface are on the same mac/ip so i dont think that's an issue (ie dhcpoffer is recorded twice in messages log - once for pxe and once for the booted os ).
    – richm1000
    Commented Jan 14, 2023 at 16:56
  • the nfs server prints "kernel: nf_conntrack: falling back to vmalloc" in /var/log/messages - shortly before the diskless clients go off line. could that explain clients loss of their nfsroot ?
    – richm1000
    Commented Jan 15, 2023 at 0:18
  • found this bugzilla.redhat.com/show_bug.cgi?id=1132396 and trying the recommendation to set lease time to forever for diskless clients . .
    – richm1000
    Commented Jan 15, 2023 at 10:23

1 Answer 1

0

Confirmed solution/workaround : adding the following lines to dhcpd.conf prevented dropping the connection to the nfs server :

default-lease-time infinite;
max-lease-time infinite;

as suggested in bug report https://bugzilla.redhat.com/show_bug.cgi?id=1132396

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .