Re: parse partition strategy string in gram.y

It will often happen that some hash keys are more frequently referenced than others. Consider a scenario where customer_id is the hash key, and one customer is very large in terms of their activity, like IBM, and other keys have much less activity. This asymmetry creates a noisy neighbor problem. Some partitions may have more than one noisy neighbor, and in general it would be more flexible to be able to divide the work evenly in terms of activity instead of evenly with respect to the encoding of the keys.

On 10/24/22, 8:50 PM, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.

Alvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> writes:
> On 2022-Oct-24, Finnerty, Jim wrote:
>> The advantage of hash partition bounds is that they are not
>> domain-specific, as they are for ordinary RANGE partitions, but they
>> are more flexible than MODULUS/REMAINDER partition bounds.

I'm more than a bit skeptical of that claim. Under what
circumstances (other than a really awful hash function,
perhaps) would it make sense to not use equi-sized hash
partitions?

<snip>

regards, tom lane