# Proposed changes
[Parquet v1.11+ supports page skipping](https://github.com/apache/parquet-format/blob/master/PageIndex.md),
which helps the scanner reduce the amount of data scanned, decompressed, decoded, and insertion.
According to the performance FlameGraph, decompression takes up 20% cpu time.
If a page can be filtered as a whole, the page can not be decompressed.
However, the row numbers between pages are not aligned. Columns containing predicates can be filtered by page granularity,
but other columns need to be skipped within pages, so non predicate columns can only save the decoding and insertion time.
Array column needs the repetition level to align with other columns, so the array column can only save the decoding and insertion time.
## Explore
`OffsetIndex` in the column metadata can locate the page position.
Theoretically, a page can be completely skipped, including the time of reading from HDFS.
However, the average size of a page is around 500KB. Skipping a page requires calling the `skip`.
The performance of `skip` is low when it is called frequently,
and may not be better than continuous reading of large blocks of data (such as 4MB).
If multiple consecutive pages are filtered, `skip` reading can be performed according to`OffsetIndex`.
However, for the convenience of programming and readability, the data of all pages are loaded and filtered in turn.
Add some utils and provide the candidate row range (generated with skipped row range of each column)
to read for page index filter
this version support binary operator filter
todo:
- use context instead of structures in close()
- process complex type filter
- use this instead of row group minmax filter
- refactor _eval_binary() for row group filter and page index filter
Refactor the scanners for hms external catalog, work in progress.
Use VFileScanner, will remove NewFileParquetScanner, NewFileOrcScanner and NewFileTextScanner after fully tested.
Query for parquet file has been tested, still need to add readers for orc file, text file and load logic as well.
The predicate column type for char, varchar and string is PredicateColumnType<TYPE_STRING>, so _base_evaluate method should convert the input column to PredicateColumnType<TYPE_STRING> always.
After the consume mem tracker exceeds the mem limit in the mem hook, the boost stacktrace will be printed. A query/load will only be printed once, and the process tracker will only be printed once per second.
After the process memory reaches the upper limit, the boost stacktrace will be printed every second. The observed phenomena are as follows:
After query/load is canceled, the memory increases instantly;
tcmalloc profile total physical memory is less than perf process memory;
The process mem tracker is smaller than the perf process memory;
tcmalloc/jemalloc allocator cache does not participate in the mem check as part of the process physical memory.
because new/malloc will trigger mem hook when using tcmalloc/jemalloc allocator cache, but it may not actually alloc physical memory, which is not expected in mem hook fail.
in addition:
The value of tcmalloc/jemalloc allocator cache is used as a mem tracker, the parent is the process mem tracker, which is updated every 1s.
Modify the process default mem_limit to 90%. expect mem tracker to effectively limit the memory usage of the process.
Fix _delete_sign_idx and _seq_col_idx when append_column or build_schema when load.
Tablet schema cache support recycle when schema sptr use count equals 1.
Add a http interface for flink-connector to sync ddl.
Improve tablet->tablet_schema() by max_version_schema.
When the flush is triggered when the load channel exceeds the mem limit, if the flush fails, an error message is returned and the load is terminated.
Usually flush failure is -238 error code. Because the memtable is frequently flushed after the load channel exceeds the mem limit, the number of segments exceeds the max value.
actual result
select cast("0.0000031417" as date);
+------------------------------+
| CAST('0.0000031417' AS DATE) |
+------------------------------+
| 2000-00-00 |
+------------------------------+
expect result
select cast("0.0000031417" as date);
+------------------------------+
| CAST('0.0000031417' AS DATE) |
+------------------------------+
| NULL |
+------------------------------+
Reuse compression ctx and buffer.
Use a global instance for every compression algorithm, and use a
thread saft buffer pool to reuse compression buffer, pool size is equal
to max parallel thread num in compression, and this will not be too large.
Test shows this feature increase 5% of data import and compaction.
Co-authored-by: yixiutt <yixiu@selectdb.com>
Store the offset rather than the length in file for the data with array type. The new file format can improve the seek performance. Please refer to #12246 to get the performance report.
Co-authored-by: xy720 <22125576+xy720@users.noreply.github.com>
Currently, Doris has a variety of readers for different file formats,
such as parquet reader, orc reader, csv reader, json reader and so on.
The interfaces of these readers are not unified, which makes it impossible to call them through a unified method.
In this PR, I added a `GenericReader` interface class, and other Readers will implement this interface class
to use the `get_next_block()` method.
This PR currently only modifies `arrow_reader` and `parquet reader`.
Other readers will be modified one by one in subsequent PRs.
In compute level, CHAR type will shrink suffix zeros.
To keep the logic the same as CHAR type, we also shrink for ARRAY or ARRAY<ARRAY> types.
Co-authored-by: cambyzju <zhuxiaoli01@baidu.com>
