Transfers optimize

[aalan3]

This PR should bring transfers time filtering to the same speed as the rest of the tables.

The reason why it doesn't work atm is because there's a date_trunc in the where. We don't need to truncate the date there, we can just pass the value.

Thanks to @jkylling for pointing this out and solving this!

[jkylling]

Just a note about why the date_trunc breaks the query: Spark does not know much about the date_trunc('hour', -) function, and for the planner it is likely equivalent to writing random(). Hence, the query planner needs to evaluate random() > now() - interval 2 days for each row in the table, while when the query is rewritten to block_time > now() - interval 2 days, it knows that most files which make up the table does not contain any row which could match this predicate.

[soispoke]

Oh interesting, thanks @aalan3 and @jkylling !

[soispoke]

Can this be applied to erc721 and erc1155 transfers as well ?

[aalan3]

Can this be applied to erc721 and erc1155 transfers as well ?

It can but those tables are not yet z-ordered but we can still change the queries. I can follow up with PR after this one.