kgyrtkirk commented on PR #5037:
URL: https://github.com/apache/hive/pull/5037#issuecomment-1926644977
> each CTE is compiled separately
got it now - I was thinking its the other way around
I was thinking about this a bit in the last couple days; here is a summary
of my thoughts:
* to have this feature cause trouble the following is needed:
* a failing query
* an incorrectly taken stat entry
* the bad stat entry causes further issues
* ....not very likely :D
* keeping multiple equiv groups which should be merged is not fortunate -
but may not cause trouble
* at the end of the day all a `Map<Signature, RuntimeStat>` will be created
from all these details
* if there is are 2 groups which have the same `Signature` that will mean
there will be multiple candidates for the key in the map => which means those
should have been in a single group...that's why I didn't really like to relax
the check on the signature
* I believe lookup-by-signatue will either return one a representative
group ; even when there are more
from the listed approaches in the doc:
* I liked the best the `RelTreeSignature` related; I think that would be the
best if possioble
* this might be a way to achieve a merge between the 2 groups
* have you tried somehow relax the alias name of `HiveTableScan` in the
`RelTreeSignature` ?
* that class creates the signature of the relnodes 1-by-1 ; so you
could access/customize/etc based on the type
* possibly use the `@Signature` annotation
* use something to filter out poisonous stuff `@IgnoreSig` or
something
* by-hand stuff
* on the other hand after thinking it thru what might be the outcomes - I
find that the idea to not pull the signatures during linking might actually
work correctly...
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
