Hash partion

[subrat]

Hi

Can anyone help me to understand how Hash partioning is happenening in parallel Job. How the key is generating?

Subrat

[ray.wurlod]

You identify the key column(s). The characters from those columns are used as input to a hashing function, which always returns the same uint32 value (the "hashvalue") for any given set of characters.

This "hashvalue" is divided by the number of partitions and the remainder is the partition number to which that row is allocated.

[subrat]

Could u please briefly explain whats mean by 'which always returns the same uint32 value'.
Moreover can a processing node contain more then one partion?

You identify the key column(s). The characters from those columns are used as input to a hashing function, which always returns the same uint32 value (the "hashvalue") for any given set of characters ...

[ray.wurlod]

I don't believe that U's technical expertise is up to providing the explanation you seek so I will attempt your edification.

Code: Select all

hashvalue = f(keyvalue) 

hashvalue is the uint32 result
f() is the partitioning algorithm

Code: Select all

partition_number_for_row = Mod(hashvalue, partition_count)

There is a maximum of one partition per processing node. Use of node pools may mean that there are fewer partitions than processing nodes.

[subrat]

Thanks Ray for this valuable info...

Can u please also suggest me if we are doing hash partioning, is it always better to take all table keys as partion keys also? If yes then can i do same thing for other type of partion as well...

Moreover in case of join, lookup etc... are the data match within the partion or across partions also.

Thanks
Subrat

I don't believe that U's technical expertise is up to providing the explanation you seek so I will attempt your edification.

Code: Select all

hashvalue = f(keyvalue) 

hashvalue is the uint32 r ...