[DRILL-7418] MetadataDirectGroupScan improvements

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.16.0
• Fix Version/s: 1.17.0
• Labels: ready-to-commit
• Assignee: Arina Ielchiieva
• Reporter: Arina Ielchiieva
• Reviewer: Vova Vysotskyi
• Created: 22/Oct/19 14:32
• Updated: 30/Oct/19 15:09
• Resolved: 30/Oct/19 15:09

Description

When count is converted to direct scan (case when statistics or table metadata are available and there is no need to perform count operation), MetadataDirectGroupScan is used. Proposed MetadataDirectGroupScan enhancements:
1. Show table selection root instead listing all table files. If table has lots of files, query plan gets polluted with all files enumeration. Since files are not used for calculation (only metadata), they are not relevant and can be excluded from the plan.

Before:

| 00-00    Screen
00-01      Project(EXPR$0=[$0], EXPR$1=[$1], EXPR$2=[$2], EXPR$3=[$3])
00-02        DirectScan(groupscan=[files = [/drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_0.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_5.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_4.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_9.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_3.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_6.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_7.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_10.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_2.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_1.parquet, /drill/testdata/metadata_cache/store_sales_null_blocks_all/0_0_8.parquet], numFiles = 11, usedMetadataSummaryFile = false, DynamicPojoRecordReader{records = [[1560060, 2880404, 2880404, 0]]}])

After:

| 00-00    Screen
00-01      Project(EXPR$0=[$0], EXPR$1=[$1], EXPR$2=[$2], EXPR$3=[$3])
00-02        DirectScan(groupscan=[selectionRoot = /drill/testdata/metadata_cache/store_sales_null_blocks_all, numFiles = 11, 
usedMetadataSummaryFile = false, DynamicPojoRecordReader{records = [[1560060, 2880404, 2880404, 0]]}])

For Hive tables which were scanned directly, selection root is not available thus will be omitted.

2. Submission of physical plan which contains MetadataDirectGroupScan fails with deserialization errors, proper ser / de should be implemented.

Issue Links

• links to: GitHub Pull Request #1883