Features

Bugfixes

Fill partition value for tunnel records
Use PERCENTILE_APPROX for doubles under ODPS 2.0
Convert all requirement files to UNIX format for pyodps-pack
Fix error when reloading volume tunnel session
Fix logview setting not working in options
Dump SQL statement when encountered ParseError
Remove misplaced warnings when pickling user functions
Fix errors of to_pandas for InSessionInstance readers
Fix position of tablesample clause for sample
Fix compatibility for SQLAlchemy 2.0
Fix results of value_counts when values are None
Remove empty equal mark for url actions
Stop copying and caching for DataFrame(pd).persist if possible to reduce memory usage
Fix missing quotaName in full lifecycle of tunnel requests
Fill partition value for tunnel records
Fix starting of Mars notebook and Mars import in some case
Delete deflate Content-Encoding header for halo in storage api

Supports scanning dependencies for pkg_resources
Add PEP517 args for pyodps-pack
Persist pandas dataframes in batches
Use date in response headers to replace fields in Schemas
Add detailed logs for sign server on errors
Make option context as thread locals
Adapt to extended types for ODPS arrow format
Supports schema API along with SQL implementations
Add support for MaxFieldSize passed by server end
Add options to allow keeping resources for DataFrame
Add support for timestamp_ntz type
Refine error message for malfunctioning create instance response
Allow adding custom log handlers to support displaying logs in notebook kernels
Allow using run_sql to execute merge smallfiles or compact commands
Allow specifying transactional table property
Unify verbose_log into standard Python logging and dump progress when waiting for instances
Return struct values as namedtuples by default and fix DataFrame customized functions on complex types
Add retry for BufferedRecordWriter when writing blocks
Reuse task utilities to simplify MCQA submission

PyODPS now returns struct values as namedtuples for tunnels to keep consistency with UDFs. For most of the cases your code might still work. If it doesn't, try configuring options.struct_as_dict = True.
From v0.11.5 nullable property of columns is added for transactional tables, and default value for partition columns is False. If you use these column instances in some scenario, for instance, using them as common columns to create tables, non-nullable columns could be created and insertion of null values will result in errors. To ignore nullable flags in columns, try configuring sql.ignore_fields_not_null = True.