Hashing won't stop but there may be a scenario of hash collisions. Where two different primary index rows would get same hash value.
To resolve this there exist one more 32 bit values generated called uniqueness values (added to hash value)
I agree with Sri, if the there is huge number of rows with small hash map, hashing would not stop, rather hash collsion will occure. A 32 bit uniqueness values is used to differentiate between the hash collisions. in TD 14, a larger hash map is introduced to avoid hash collisions.
On what basis number of Hashmap buckets will be decided? Let's say I have a Teradata system with 100 AMPS installing on 64bit SUSE linux, then howmany buckets will be created in Hashmap to hold AMP numbers?
Thanks in advance.
The number of hash buckets is always the same, 2**20 = 1,048,576
There's a process to fill the hash buckets for a n-AMP systems with numbers between 0 and n-1 during sysinit.
See the Database Design manual: