All pull requests

Pull requests

• Pack and split Delta server side planned GPU scan files into partitions [databricks]

  performance
  #16649·sdrp713 opened Oct 28, 2026Collaborator·3 of 8 tasks·

  ·
  1

  2

  

• Preserve Delta 4.3 partitioning on saveAsTable overwrite [databricks]

  bugdelta
  #16647·liurenjie1024 opened Oct 28, 2026Collaborator·3 of 8 tasks·

  ·
  4

• Fix Delta 4.x DV MERGE metadata assertions [databricks]

  test
  #16646·liurenjie1024 opened Oct 28, 2026Collaborator·3 of 8 tasks·

  ·
  5

• [WIP] [AutoSparkUT] Restore ORC nested-column COUNT pushdown test [databricks]

  test
  #16645·wjxiz1992 opened Oct 28, 2026Collaborator·3 of 11 tasks·

  ·
  1

• Fix DBR 18.3 GPU MERGE type mismatch with schema evolution [databricks]

  bug
  #16642·sdrp713 opened Oct 27, 2026Collaborator·3 of 8 tasks·

  ·
  1

  5

  

• Add Delta 4.3 DELETE zero metrics and MERGE error equivalence tests [databricks]

  test
  #16637·sdrp713 opened Oct 27, 2026Collaborator·3 of 8 tasks·

  ·
  1

  2

  

• Reduce Parquet cache heap pressure with smaller row-group records [databricks]

  #16633·revans2 opened Oct 27, 2026Collaborator·3 of 8 tasks·

  ·
  1

• Copy each Arrow batch into host columns when HostColumnarToGpu receives it [databricks]

  bugperformance
  #16632·thirtiseven opened Oct 27, 2026Collaborator·3 of 8 tasks·

  ·
  2

  4

  

• Accelerate Delta Lake 4.3 selective replacement with replaceOn and replaceUsing [databricks]

  feature request
  #16619·sdrp713 opened Oct 26, 2026Collaborator·3 of 8 tasks·

  ·
  1

  2

  

• Test Delta 4.3 dynamic partition overwrite partition value normalization on GPU [databricks]

  test
  #16611·sdrp713 opened Oct 26, 2026Collaborator·3 of 8 tasks·

  ·
  1

  4

  

• Shorten unbounded aggregate window resource lifetimes [databricks]

  task
  #16608·wjxiz1992 opened Oct 26, 2026Collaborator·3 of 8 tasks·

  ·
  4

• Audit Spark package-private access and root-load Parquet cache readers [databricks]

  #16602·gerashegalov opened Oct 26, 2026Collaborator·2 of 10 tasks·

  ·
  2

  5

  

• Add bounded ready-handle admission and continuation tracking

  #16600·liurenjie1024 opened Oct 26, 2026Collaborator·5/5·3 of 8 tasks·

  ·
  1

  3

• Add owned asynchronous work scopes for unified readers

  #16599·liurenjie1024 opened Oct 26, 2026Collaborator·4/5·3 of 8 tasks·

  ·
  1

  4

• Add the format-neutral unified reader execution loop

  #16598·liurenjie1024 opened Oct 26, 2026Collaborator·3/5·3 of 8 tasks·

  ·
  1

  1

• Add Java resource and task context adapters for unified readers

  #16597·liurenjie1024 opened Oct 26, 2026Collaborator·2/5·3 of 8 tasks·

  ·
  1

  2

• Add unified reader ownership contracts in Java [reduced-it]

  #16596·liurenjie1024 opened Oct 26, 2026Collaborator·1/5·3 of 8 tasks·

  ·
  1

  3

• Use shared shim source inheritance in public builds [databricks]

  #16595·gerashegalov opened Oct 26, 2026Collaborator·4 of 11 tasks·

  ·
  6

• Pin shared Spark installation dependency generator [databricks]

  #16549·gerashegalov opened Oct 25, 2026Collaborator·3 of 10 tasks·

  ·

• Promote runtime control and shuffle protocol values to root-safe API [databricks]

  #16539·gerashegalov opened Oct 24, 2026Collaborator·3 of 8 tasks·

  ·
  2

  

• Expose data transfer metric during GPU partitioning

  improve
  #16536·igorpeshansky opened Oct 23, 2026Collaborator·3 of 8 tasks·

  ·
  12

• Integrate cuDF hybrid parquet reader for PERFILE mode

  task
  #16534·pmattione-nvidia opened Oct 23, 2026Collaborator·3 of 8 tasks·

  ·
  65

  

• Optimize Iceberg v3 row ID generation

  performance
  #16533·res-life opened Oct 23, 2026Collaborator·3 of 8 tasks·

  ·
  1

  1

• Prevent unsafe GPU round-robin shuffle retries [databricks]

  bugshuffle
  #16530·wjxiz1992 opened Oct 23, 2026Collaborator·3 of 8 tasks·

  ·
  1

  14

  

• Handle rows-only sides and count-only joins in nested loop joins [databricks]

  reliability
  #16518·amahussein opened Oct 22, 2026Collaborator·3 of 8 tasks·

  ·
  1

  3