[
https://issues.apache.org/jira/browse/HIVE-9188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14304661#comment-14304661
]
Gopal V commented on HIVE-9188:
-------------------------------
The predicate evaluation should always use min/max comparisons.
The min-max pruning is turned off for a column which has a bloom filter. This
inadvertantly turns off the the fastest check in favour of a slower check.
{code}
+ // if bloom filter exists, check in bloom filter else min/max stats
+ if (bloomFilter == null) {
+ loc = compareToRange((Comparable) predObj, minValue, maxValue);
+ if (loc == Location.MIN) {
+ return hasNull ? TruthValue.YES_NULL : TruthValue.YES;
+ }
{code}
I ran L_ORDERKEY filters with bloom filters and with min-max pruning. The
rows-read were surprising at the 1Tb scale.
{code}
With bloom filters:
VERTICES TOTAL_TASKS DURATION_SECONDS CPU_TIME_MILLIS
GC_TIME_MILLIS INPUT_RECORDS
Map 1 198 7.88 1,162,490
16,270 2,960,000 8
Without bloom filters:
Map 1 194 6.28 1,422,550
33,483 410,000 4
{code}
Without PPD, that actually reads 5,999,989,709 records in ~10s.
> BloomFilter support in ORC
> --------------------------
>
> Key: HIVE-9188
> URL: https://issues.apache.org/jira/browse/HIVE-9188
> Project: Hive
> Issue Type: New Feature
> Components: File Formats
> Affects Versions: 0.15.0
> Reporter: Prasanth Jayachandran
> Assignee: Prasanth Jayachandran
> Labels: orcfile
> Attachments: HIVE-9188.1.patch, HIVE-9188.2.patch, HIVE-9188.3.patch,
> HIVE-9188.4.patch, HIVE-9188.5.patch, HIVE-9188.6.patch, HIVE-9188.7.patch,
> HIVE-9188.8.patch, HIVE-9188.9.patch
>
>
> BloomFilters are well known probabilistic data structure for set membership
> checking. We can use bloom filters in ORC index for better row group pruning.
> Currently, ORC row group index uses min/max statistics to eliminate row
> groups (stripes as well) that do not satisfy predicate condition specified in
> the query. But in some cases, the efficiency of min/max based elimination is
> not optimal (unsorted columns with wide range of entries). Bloom filters can
> be an effective and efficient alternative for row group/split elimination for
> point queries or queries with IN clause.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
