JingsongLi opened a new pull request, #9351:
URL: https://github.com/apache/paimon/pull/9351

   ### Purpose
   
   Catalog-managed format table partitions can carry exact row counts, but 
FormatDataSplit cannot safely expose a partition count because one partition 
may produce multiple splits. This change attaches the pruned, scan-level row 
count to FormatTableScan.Plan and feeds it into Spark PaimonStatistics.
   
   Split generation and row-count aggregation reuse one catalog partition 
listing per plan. Unknown partition counts keep the scan row count unknown, and 
runtime filtering creates a fresh plan.
   
   The scan statistics allow Spark to make better cost-based decisions. A 
TPC-DS-style store_sales/date_dim test verifies that an analyzed dimension 
partition changes the physical plan from SortMergeJoin to BroadcastHashJoin and 
avoids fact-side shuffle.
   
   ### Tests
   
   - CatalogManagedPartitionScanTest (19 tests)
   - CatalogManagedPartitionAnalyzeTest (21 tests)
   - FormatTableTest on Spark 3.5 (21 tests)
   - Spotless checks for paimon-core, paimon-spark-common, and paimon-spark-ut
   - git diff --check


-- 
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]

Reply via email to