here is what a typical file group instantiation of MDT partition looks like
t10: create a dummy commit w/ base commit time "0000000".
So this will create a log file w/ dummy delete block.
Immediately following this, we take the bulk_insert which will create a new file slice but w/ same commit time.
base_file_00000.parquet.
Theoretically, these both belong to diff file slices and when latest snapshot is read, only latest base file should be read. but as of now, we consider the log file also as latest and read it. Since its dummy delete log block, there is no correctness issue here.
Just some code clean up is required.
this is an issue only w/ a fresh table.
JIRA info
here is what a typical file group instantiation of MDT partition looks like
t10: create a dummy commit w/ base commit time "0000000".
So this will create a log file w/ dummy delete block.
Immediately following this, we take the bulk_insert which will create a new file slice but w/ same commit time.
base_file_00000.parquet.
Theoretically, these both belong to diff file slices and when latest snapshot is read, only latest base file should be read. but as of now, we consider the log file also as latest and read it. Since its dummy delete log block, there is no correctness issue here.
Just some code clean up is required.
this is an issue only w/ a fresh table.
JIRA info