Re: GiST VACUUM

On 04/01/2019 21:26, Andrey Borodin wrote:
> 3 янв. 2019 г., в 23:47, Andrey Borodin <x4mmm@yandex-team.ru>
> написал(а):
>> 
>>> * Bitmapset stores 32 bit signed integers, but BlockNumber is
>>> unsigned. So this will fail with an index larger than 2^31
>>> blocks. That's perhaps academical, I don't think anyone will try
>>> to create a 16 TB GiST index any time soon. But it feels wrong to
>>> introduce an arbitrary limitation like that.
>> 
>> Looks like bitmapset is unsuitable for collecting block numbers due
>> to the interface. Let's just create custom container for this? Or
>> there is one other option: for each block number throw away sign
>> bit and consider page potentially internal and potentially empty
>> leaf if (blkno & 0x7FFFFFF) is in bitmapset. And then we will have
>> to create at least one 17Tb GiST to check it actually works.
> 
> Heikki, how do you think, is implementing our own radix tree for this
> is viable solution? I've written working implementation with 4-level
> statically typed tree. If we follow this route, probably, there must
> be tests for this data structure.

Yeah, seems reasonable.

- Heikki