This article explains issues with VMs going unresponsive when changes are made to the MaxAddressableSpaceTB setting on the ESXi host
pbElem->shadowInfo.addrs is allocated from VMFS Heap and it has a copy of the PB data.
This is used to tack the changes done as part of the memTxns and to consolidate the changes.
These are used as shadow PBs for Redo logging for Hiperf Transactions.
This is allocated here:
1) Fil6_ResolveAddress->PB6_LockOffset
2) Fil6_ResolveAddress->PB6_UnmapIOLock
These are freed when the corresponding PB is evicted.
PB3EvictLRU->PB3EvictElem->PB3_FreePBElem
Once these Shadow PBs are allocated as part of any memTxn, it is staying there until the
corresponding PB gets evicted. While the PBs are allocated from the PBGC slab, these
shadow PBs are allocated from VMFS heap and they are taking up huge space in the heap
causing VMFS Heap to get full, eventually causing memory failures for any operation
that needs VMFS Heap allocation.
Issue is currently set to be corrected in future release.
Current recommendation is to not make any changes to the configuration.