mirror of
https://github.com/GreptimeTeam/greptimedb.git
synced 2026-08-23 22:48:19 +00:00
discord9
8d887ddd00
* fix(query): restore columnar group-by for dictionary-encoded tags Bump the DataFusion fork to be93ffd85 (feat/dict-group-column-53), which backports apache/datafusion #23187: DictionaryGroupValuesColumn lets dictionary-encoded group keys use the columnar GroupValuesColumn fast path (hash distinct dictionary values once per batch, resolve rows by key index) instead of falling back to row-based GroupValuesRows. This fixes the TSBS double-groupby regression introduced by #8541 (preserve dictionary-encoded query labels): v1.2.0-beta.1 scan output changed tag columns to Dictionary(UInt32, Utf8), which DataFusion 53.1.0 did not support in GroupValuesColumn's supported_type allow-list, so GROUP BY queries silently dropped to the ~60% slower row path (time_calculating_group_ids +57%, peak_mem +50%, end-to-end +38%). Adds an end-to-end integration test (dict_groupby_sst) that flushes a flat-format SST with dictionary-encoded hostname, runs the tsbs-style double-groupby query, and asserts correct results with no CastExec inserted before the aggregate. Signed-off-by: discord9 <discord9@greptime.dev> * chore(deps): pin merged dictionary group-by support Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com> --------- Signed-off-by: discord9 <discord9@greptime.dev> Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com> Co-authored-by: discord9 <discord9@greptime.dev>