[
https://issues.apache.org/jira/browse/DRILL-6259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16402120#comment-16402120
]
Paul Rogers commented on DRILL-6259:
------------------------------------
What is meant when we say "complex type"? Drill has multiple "complex" types:
* Arrays
* Nested tuples (AKA "maps")
* Arrays of nested tuples
* Multi-dimensional arrays (AKA "repeated lists")
* Hetrogenous values (AKA "unions")
* Hetrogenous lists (AKA "non-repeated lists")
* Combinations of the above (a repeated map that contains a union that
contains a 2D list of maps)
Then there are the "complex" scalar types (complex because they are not simply
bit values like an int or a float):
* Decimal
* Date/time
* Date
* Time
* Period
The write-up mentions arrays. Is this only for arrays? Also for maps? For map
arrays?
Please identify which complex types are now supported.
> Support parquet filter push down for complex types
> --------------------------------------------------
>
> Key: DRILL-6259
> URL: https://issues.apache.org/jira/browse/DRILL-6259
> Project: Apache Drill
> Issue Type: Improvement
> Affects Versions: 1.13.0
> Reporter: Arina Ielchiieva
> Assignee: Arina Ielchiieva
> Priority: Major
> Fix For: 1.14.0
>
>
> Currently parquet filter push down is not working for complex types
> (including arrays).
> This Jira aims to implement filter push down for complex types which
> underneath type is among supported simple types for filter push down. For
> instance, currently Drill does not support filter push down for varchars,
> decimals etc. Though once Drill will start support, this support will be
> applied for complex type automatically.
> Complex fields will be pushed down the same way regular fields are, except
> for one case with arrays.
> Query with predicate {{where users.hobbies_ids[2] is null}} won't be able to
> push down because we are not able to determine exact number of nulls in
> arrays fields.
> {{Consider [1, 2, 3]}} vs {{[1, 2]}} if these arrays are in different files.
> Statistics for the second case won't show any nulls but when querying from
> two files, in terms of data the third value in array is null.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
