performance problems after OLB

Document ID : KB000117155
Last Modified Date : 09/10/2018
Show Technical Document Details
Issue:
After every Spectrum Online Backup (OLB), a lot of devices show "Management agent lost" alerts.

When trying to contact the devices using their management protocol, the contact is successful. Is striking that the phenomenon is limited to the devices, which are connected trough an Secure Domain Connector (SDC). It is noticed that the SDC process eats permanently 2,7% CPU. 
Environment:
Spectrum 10.2.3.BMP_301
Resolution:
Fix provided with Spectrum_10.02.03.D431 (D431a - extra debug):
------------------------------------------------------
Secondary SpectroSERVER restart flushing out the SNMPv3 profiles at SDC during OnLineBackup and creating "Management agent lost" alarms.
(DE345502, 00788951)
------------------------------------------------------

D431(a) makes a very subtle change to how Spectrum manages SNMPv3 profiles, especially from a FaultTolerant (FT) SpectroServer pair. SDC should behave much like the SpectroSERVER does for direct SNMPv3 communication, so Spectrum does a full ‘flush/reset’ of the SDC SNMPv3 profiles when the SpectroServer starts up.
The problem comes into play in a FT environment, as the secondary SpectroSERVER also triggers this SDC-side SNMPv3 ‘flush/reset’ when it starts up.  This was disrupting pending SNMPv3 device communication that was coming from the primary.
The solution was just to disregard the SNMPv3 ‘flush/reset’, if the incoming set of SNMPv3 profile data matches the current set.  In a FT environment, this should always be the case, and will just be dropped by the SDC.
Additional Information:
Planned to be included in Spectrum 10.3.1