ParquetReader does not respect time range provided

[v0y4g3r]

What type of enhancement is this?

Performance

What does the enhancement do?

ParquetReaderBuilder provides a time_range option to filter the timestamps of rows to read. But currently it is not used anywhere when reading parquet files. We need to respect this option which will boost the scan performance when an exact time range is provided.

greptimedb/src/mito2/src/sst/parquet/reader.rs

Line 69 in 5a0629e

time_range: Option<TimestampRange>,

Implementation challenges

There's a workaround: we can transform the time range into predicates and inserts them into ParquetReaderBuilder::predicate. These predicates will be applied to the batches read from parquet files.

But this workaround does not skip pages and row groups, which may waste extra IO overhead. We need to use the time range to prune row groups and data pages according the the file metadata.