[Lustre-discuss] servicenode/failnode and comma separated values
Andrus, Brian Contractor
bdandrus at nps.edu
Mon Oct 20 10:04:04 PDT 2014
All,
I notice there seems to be a difference between putting all your nids as comma separated values for servicenode/failnode and putting them as separate entries for each one.
Here is an output of a dryrun against an OST that was created using multiple "--servicenode=" options. I ran tune2fs with the same nids, but as comma separated on a single "--servicenode" option:
============================snip==========================
#tunefs.lustre --dryrun --servicenode=10.100.1.10 at o2ib,10.100.1.11 at o2ib,10.100.1.12 at o2ib,10.100.1.13 at o2ib,10.100.1.14 at o2ib,10.100.1.15 at o2ib --verbose /dev/VG_Hamming/WORK_OST17checking for existing Lustre data: found
Reading CONFIGS/mountdata
Read previous values:
Target: WORK-OST0011
Index: 17
Lustre FS: WORK
Mount type: ldiskfs
Flags: 0x1062
(OST first_time update no_primnode )
Persistent mount opts: errors=remount-ro
Parameters: mgsnode=10.100.1.11 at o2ib:10.100.1.10 at o2ib failover.node=10.100.1.10 at o2ib failover.node=10.100.1.11 at o2ib failover.node=10.100.1.12 at o2ib failover.node=10.100.1.13 at o2ib failover.node=10.100.1.14 at o2ib failover.node=10.100.1.15 at o2ib
Permanent disk data:
Target: WORK:OST0011
Index: 17
Lustre FS: WORK
Mount type: ldiskfs
Flags: 0x1062
(OST first_time update no_primnode )
Persistent mount opts: errors=remount-ro
Parameters: mgsnode=10.100.1.11 at o2ib:10.100.1.10 at o2ib failover.node=10.100.1.10 at o2ib failover.node=10.100.1.11 at o2ib failover.node=10.100.1.12 at o2ib failover.node=10.100.1.13 at o2ib failover.node=10.100.1.14 at o2ib failover.node=10.100.1.15 at o2ib failover.node=10.100.1.10 at o2ib,10.100.1.11 at o2ib,10.100.1.12 at o2ib,10.100.1.13 at o2ib,10.100.1.14 at o2ib,10.100.1.15 at o2ib
exiting before disk write.
============================snip==========================
So is there a difference in how lustre will use those failover.node settings? If so, what is it and if not, shouldn't there be some consistency in how that is stored/displayed?
Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238
More information about the lustre-discuss
mailing list