While there are many reasons SCSI reservation conflicts can occur in a shared LUN environment, sometimes vendors write agents without knowing the full impact they could have in various environments they do not fully test. In HP’s case, there are 2 agents that are part of the Insight Manager suite that are responsible for excessive SCSI reservations.
VMware has written a KB article that addresses this issue:
HP has also released a customer advisory on the possibility of SCSI “anomalies” if one of these agents (cmafcad) are running:
The “cmahostd” agent has only affected HDS based arrays (HP XP arrays are included since they are HDS based) as these arrays are optimized for SCSI-3 reservations however ESX uses SCSI-2 reservations and as a result are slower to lock/unlock these reservations.