Run #2
trips · submitted 11 Oct 2026, 18:57
SQL
successSELECT COUNT(*) AS total FROM trips WHERE region = 'EU'
Dataset trips · ./data/trips
| Row | total |
|---|---|
| 1 | 100,000 |
The plan after optimization, which is what actually executed — projection and predicate pushdown have already been applied, so the Scan node shows the columns and filters it really used.
Text plan
Aggregate group=[-] aggs=[count(*)] Scan trips cols=[passengers] part_filter
Where the bytes went
The middle bars are bytes a scan-everything engine would have read and this one did not. They add up to the full scan exactly — it is one number decomposed, not four separate measurements.
Row-group map (40 of 40 read)
Scan metrics as recorded by the engine
- total_files
- 15
- files_scanned
- 5
- files_pruned
- 10
- total_row_groups
- 40
- row_groups_scanned
- 40
- row_groups_pruned
- 0
- rows_scanned
- 100000
- columns_projected
- 1
- columns_total
- 7
- bytes_scanned
- 29041
- bytes_total
- 6259257
- bytes_pruned_files
- 4196966
- bytes_pruned_row_groups
- 0
- bytes_unprojected
- 2033250
Byte counts are compressed on-disk bytes on both sides of the ratio.
bytes_total covers every row-group including the pruned ones, plus the
file size of files that were never opened.
- 10 of 15 files were never opened — the partition filter ruled them out from the directory path alone, costing zero I/O.
- Only 1 of 7 columns came off disk — the other 6 were never referenced, and Parquet is columnar.
The middle bars are bytes a scan-everything engine would have read and this one did not. They add up to the full scan exactly — it is one number decomposed, not four separate measurements.