[
https://issues.apache.org/jira/browse/PARQUET-2237?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17691949#comment-17691949
]
ASF GitHub Bot commented on PARQUET-2237:
-----------------------------------------
gszadovszky commented on code in PR #1023:
URL: https://github.com/apache/parquet-mr/pull/1023#discussion_r1113886761
##########
parquet-hadoop/src/main/java/org/apache/parquet/filter2/statisticslevel/StatisticsFilter.java:
##########
@@ -289,8 +320,14 @@ public <T extends Comparable<T>> Boolean visit(Lt<T> lt) {
T value = lt.getValue();
- // drop if value <= min
- return stats.compareMinToValue(value) >= 0;
+ // we are looking for records where v < someValue
+ if (stats.compareMinToValue(value) >= 0) {
+ // drop if value <= min
+ return BLOCK_CANNOT_MATCH;
+ } else {
+ // if value > min, we must take it
+ return BLOCK_MUST_MATCH;
Review Comment:
FYI: for column index we specified it explicitly to allow shortening the
min/max values (e.g. string are truncated to 64bytes by default).
The older min/max statistics for the row groups are trickier since it is not
properly specified. The issue is if only one implementation of parquet in the
wild would do the similar truncate we may loose data when filtering if we use
this "aggressive optimization".
Dictionary is a completely different thing. It contains the very values from
the pages themselves. It is not designed to be used for filtering (but we use
anyway) but to save space.
> Improve performance when filters in RowGroupFilter can match exactly
> --------------------------------------------------------------------
>
> Key: PARQUET-2237
> URL: https://issues.apache.org/jira/browse/PARQUET-2237
> Project: Parquet
> Issue Type: Improvement
> Reporter: Mars
> Priority: Major
>
> If we can accurately judge by the minMax status, we don’t need to load the
> dictionary from filesystem and compare one by one anymore.
> Similarly , Bloomfilter needs to load from filesystem, it may costs time and
> memory. If we can exactly determine the existence/nonexistence of the value
> from minMax or dictionary filters , then we can avoid using Bloomfilter to
> Improve performance.
> For example,
> # read data greater than {{x1}} in the block, if minMax in status is all
> greater than {{{}x1{}}}, then we don't need to read dictionary and compare
> one by one.
> # If we already have page dictionaries and have compared one by one, we
> don't need to read BloomFilter and compare.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
