Releases: TileDB-Inc/TileDB
Releases · TileDB-Inc/TileDB
TileDB v2.2.1-rc1
Pre-release for upcoming 2.2.1. This version is not guaranteed to have a stable API or format, only the final 2.2.1 release will be covered by the backwards compatible guarantees.
Please see HISTORY.md for current list of changes.
TileDB v2.2.1
TileDB v2.2.1 Release Notes
Breaking behavior
- The tile extent can now be set to null, in which case internally TileDB sets the extent to the dimension domain range. #1880
- The C++ API
std::pair<uint64_t, uint64_t> Query::est_result_size_var
has been changed to 1) a return type ofstd::array<uint64_t, 2>
and 2) returns the offsets as a size in bytes rather than elements. #1946
New features
- Support for nullable attributes. #1895 #1938 #1948 #1945
- Support for Hilbert order sorting for sparse arrays. #1880
- Support for AWS S3 "AssumeRole" temporary credentials #1882
- Support for zero-copy import/export with the Apache Arrow adapter #2001
- Experimental support for an in-memory backend used with bootstrap option "--enable-memfs" #1873
- Support for element offsets when reading var-sized attributes. #1897
- Support for an extra offset indicating the size of the returned data when reading var-sized attributes. #1932
- Support for 32-bit offsets when reading var-sized attributes. #1950
Improvements
- Optimized string dimension performance.
- Added functionality to get fragment information from an array. #1900
- Prevented unnecessary sorting when (1) there is a single fragment and (i) either the query layout is global order, or (ii) the number of dimensions is 1, and (2) when there is a single range for which the result coordinates have already been sorted. #1880
- Added extra stats for consolidation. #1880
- Disabled checking if cells are written in global order when consolidating, as it was redundant (the cells are already being read in global order during consolidation). #1880
- Optimize consolidated fragment metadata loading #1975
Bug fixes
- Fix tiledb_dimension_alloc returning a non-null pointer after error #1959
- Fixed issue with string dimensions and non-set subarray (which implies spanning the whole domain). There was an assertion being triggered. Now it works properly.
- Fixed bug when checking the dimension domain for infinity or NaN values. #1880
- Fixed bug with string dimension partitioning. #1880
API additions
C API
- Added functions for getting fragment information. #1900
- Added APIs for getting and setting ranges of queries using a dimension name. #1920
C++ API
- Added class
FragmentInfo
and functions for getting fragment information. #1900 - Added function
Dimension::create
that allows not setting a space tile extent. #1880 - Added APIs for getting and setting ranges of queries using a dimension name. #1920
- Changed
std::pair<uint64_t, uint64_t> Query::est_result_size_var
tostd::array<uint64_t, 2> Query::est_result_size_var
. Additionally, the size estimate for the offsets have been changed from elements to bytes. #1946
TileDB v2.1.6
TileDB v2.1.5
TileDB v2.1.5 Release Notes
Improvements
- Optimize consolidated fragment metadata loading #1975
- Optimize
Reader::load_tile_offsets
for loading only relevant fragments #1976 #1983 - Optimize locking in
FragmentMetadata::load_tile_offsets
andFragmentMetadata::load_tile_var_offsets
#1979 - Exit early in
Reader::copy_coordinates
/Reader::copy_attribute_values
when no results #1984
Bug fixes
TileDB v2.1.4
TileDB v2.1.4 Release Notes
Improvements
- Optimize
ResultTile::compute_results_sparse<char>
resulting in significant performance increases in certain cases with string dimensions #1963
TileDB v2.1.3
TileDB v2.1.3 Release Notes
Improvements
- Optimized string dimension performance. #1922
Bug fixes
- Updated the AWS SDK to v1.8.84 to fix an uncaught exception when using S3 #1899TileDB-Py #409
- Fixed bug where a read on a sparse array may return duplicate values. #1905
- Fixed bug where an array could not be opened if created with an array schema from an older version #1889