Error Message

WARNING: msgcnt 4 mesg 014: V-2-14: vx_iget - inode table overflow 

 

WARNING: msgcnt 1 mesg 069: V-2-69: memory usage specified by the vxfs:vxfs_ninode and vxfs:vx_bc_bufhwm parameters exceeds available memory; the system may hang under heavy load.

VxFS uses the difference between the ekernelheap and kernelheap OS kernel variables for calculating the value of vx_virtmem. Both these OS kernel variables are 64bit data types whereas vx_virtmem was only a 32bit data type and in systems with large amounts of RAM, it was seen that that value of vx_virtmem might not be correctly stored in a 32bit data type.

The workaround of manually setting the following value in the /etc/system file avoided these low values:

set vxfs:vxfs_ninode=400000

 

vx_virtmem has been updated to be a 64bit data type and this has been fixed in patches for VxFS 6.1.1 and 7.3.1 (as well as other VxFS versions) and is included in 7.4.1.

 

 

 

 

Low default values were being automatically set for the vxfs_ninode and vx_bc_bufhwm tunables on Solaris 11.3 systems with large amounts of RAM and running VxFS 6.1.1 and 7.3.1. For example on a system with 8TB RAM, the default values for the vxfs_ninode and vx_bc_bufhwm tunables were set to: # echo "vx_bc_bufhwm/D" | adb -k /dev/ksyms /dev/mem
physmem 3ae7cca3
vx_bc_bufhwm:
vx_bc_bufhwm: 20000
# echo "vxfs_ninode /D" | adb -k /dev/ksyms /dev/mem
physmem 3ae7cca3
vxfs_ninode:
vxfs_ninode: 6000 In the event that the system was under memory pressure, the low values of these two VxFS tunables could lead to a hang.

ETrack: 3947648