Hyukjin Kwon created SPARK-58071:
------------------------------------
Summary: Do not import numpy at 'import pyspark' time in
stateful_processor_api_client
Key: SPARK-58071
URL: https://issues.apache.org/jira/browse/SPARK-58071
Project: Spark
Issue Type: Bug
Components: Structured Streaming, PySpark
Affects Versions: 4.2.0, 5.0.0
Reporter: Hyukjin Kwon
SPARK-57760 moved `import numpy as np` to the module top level (inside a
try/except ImportError) in
`python/pyspark/sql/streaming/stateful_processor_api_client.py`. This module is
transitively imported by `import pyspark` (pyspark -> pyspark.sql ->
pyspark.sql.streaming -> stateful_processor_api_client), so numpy is now
imported during `import pyspark`.
This breaks the fast-import guarantee enforced by
`python/pyspark/tests/test_import_spark.py::ImportSparkTest::test_import_spark_libraries`,
which fails with:
{code}
AssertionError: Unexpected 3rd party package 'numpy' imported during 'import
pyspark'
{code}
The scheduled Non-ANSI builds on master and branch-4.x have been failing
because of this (e.g. build_non_ansi.yml).
Before SPARK-57760, numpy was imported lazily inside `_serialize_to_bytes`. The
fix is to keep numpy out of the module top level: resolve it lazily via a
cached `has_numpy()` helper and import numpy inside `_normalize_state_value`
only when normalizing values. Behavior is unchanged.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
