Re: Declarative partitioning - another take

On Wed, Oct 26, 2016 at 6:12 AM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> On Wed, Oct 26, 2016 at 3:04 PM, Amit Langote
> <Langote_Amit_f8(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>> On 2016/10/26 17:57, Amit Kapila wrote:
>>> @@ -123,6 +123,9 @@ typedef struct RelationData
>>> {
>>> ..
>>> MemoryContext rd_partkeycxt; /* private memory cxt for the below */
>>> struct PartitionKeyData *rd_partkey; /* partition key, or NULL */
>>> + MemoryContext rd_pdcxt; /* private context for partdesc */
>>> + struct PartitionDescData *rd_partdesc; /* partitions, or NULL */
>>> + List *rd_partcheck; /* partition CHECK quals */
>>> ..
>>> }
>>>
>>> I think one thing to consider here is the increase in size of relcache
>>> due to PartitionDescData. I think it will be quite useful in some of
>>> the cases like tuple routing. Isn't it feasible to get it in some
>>> other way, may be by using relpartbound from pg_class tuple?
>>
>> Whereas pg_class.relpartbound stores partition bound of the *individual
>> partitions* in Node form, the above relcache struct is associated with
>> parent tables; it contains some efficient to use (and fairly compact)
>> representation of bounds of *all* the partitions of the parent.
>>
>
> Okay, but still it will be proportional to number of partitions and
> the partition keys. Is it feasible to store ranges only for
> partitions that are actively accessed? For example, consider a table
> with 100 partitions and the first access to table requires to access
> 5th partition, then we store ranges for first five partitions or
> something like that. This could be helpful, if we consider cases that
> active partitions are much less as compare to total partitions of a
> table.

I have serious doubt about whether it's a good idea to do that EVER,
but it certainly doesn't need to be in the first version of this
patch.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company