For tiered tables, the block manager maintains an array, WT_BM.handle_array, of pointers to the block handles it has opened for the table.

When reading a block, the block manager cracks the address cookie to get the objectid for the block, and then walks this array looking for a handle (i.e., a *WT_BLOCK) for reading data from that object. 

In practice, this array could get fairly large  It is easy to imagine a workload that routinely adds new records to a tiered table and performs lookups across the entire table. If we flush once per hour, this table will get 720 new objects per month, and more than 8,000 in a year. That's large enough that we don't want to pay the cost of linearly scanning all open handles on every read.

We can address this by making the handle array a hash table instead.