apache/pinot

Unify some behavior in TableDataManager

Chiusa

#9752 aperta il 7 nov 2022

 (2 commenti) (2 reazioni) (0 assegnatari)Java (1234 fork)batch import
bugcleanupgood first issue

Metriche repository

Star
 (4937 stelle)
Metriche merge PR
 (Metriche PR in attesa)

Descrizione

In the TableDataManager:

  • void addSegment(File indexDir, IndexLoadingConfig indexLoadingConfig) is never used in production code (the only non-testing caller addOfflineSegment() is not wired up)
  • void addSegment(String segmentName, TableConfig tableConfig, IndexLoadingConfig indexLoadingConfig) can be simplified because TableConfig is included in the IndexLoadingConfig
  • void addOrReplaceSegment(String segmentName, IndexLoadingConfig indexLoadingConfig, SegmentZKMetadata zkMetadata, @Nullable SegmentMetadata localMetadata) should be broken into 2 methods: addSegment() and refreshSegment()
  • Loading committed segment into real-time table should share the same logic as loading segment into offline table. Currently the logic is diverged, which causes moving tier not happening for real-time table

Ideally we want to keep the following methods:

  • addSegment
  • addConsumingSegment (real-time only)
  • refreshSegment
  • reloadSegment
  • removeSegment

Guida contributor