The following MIB2D error is viewed in the log messages:
VC10 mib2d: MIB2D_KVM_FAILURE: tu_kread: kvm_read returns error: short read
Here is a description of the problem:
The error is generated because SNMP is trying to collect data on the TCP connections of the switch and this data might be incorrect.
What actually is happening is that mib2d is working on 32 bits whereas the kernel is using 64 bits.
When mib2d is requesting data from the kernel, it’s receiving an 8byte pointer, which is truncated by mib2d to 4 byte pointer.
Then the mib2d is requesting that data by providing the 4 byte pointer. If that pointer doesn’t exist, then this message is returned.
The work-around for this is the following steps:
These two commands will restart both the snmp and mib2d process. There will be no traffic impact doing this. Should this not work, you will want to reboot the switch.
This is not a permanent fix however and the messages may reappear in some time.
This bug has been reported and deal in this PR 1007305 [Confidential, Mistaken] – “DAEMON-3-MIB2D_KVM_FAILURE: tu_kread: kvm_read returns error: short read” messages seen in the logs
A more robust approach for mib2d to gather kernel information was implemented in Junos version R15.1
The fix for this situation is in Junos 15.1 version.
Please do an upgarde.