Add time zone config to set non-UTC [databricks]

integration_tests/run_pyspark_from_build.sh

@@ -223,11 +223,14 @@ else
         export PYSP_TEST_spark_jars="${ALL_JARS//:/,}"
     fi
 
+    # time zone will be tested; use export TZ=time_zone_name before run this script
+    TZ=${TZ:-UTC}
+
     # Set the Delta log cache size to prevent the driver from caching every Delta log indefinitely
-    export PYSP_TEST_spark_driver_extraJavaOptions="-ea -Duser.timezone=UTC -Ddelta.log.cacheSize=10 $COVERAGE_SUBMIT_FLAGS"
-    export PYSP_TEST_spark_executor_extraJavaOptions='-ea -Duser.timezone=UTC'
+    export PYSP_TEST_spark_driver_extraJavaOptions="-ea -Duser.timezone=$TZ -Ddelta.log.cacheSize=10 $COVERAGE_SUBMIT_FLAGS"
+    export PYSP_TEST_spark_executor_extraJavaOptions="-ea -Duser.timezone=$TZ"
     export PYSP_TEST_spark_ui_showConsoleProgress='false'
-    export PYSP_TEST_spark_sql_session_timeZone='UTC'
+    export PYSP_TEST_spark_sql_session_timeZone=$TZ
     export PYSP_TEST_spark_sql_shuffle_partitions='4'
     # prevent cluster shape to change
     export PYSP_TEST_spark_dynamicAllocation_enabled='false'

integration_tests/src/main/python/conftest.py

@@ -77,6 +77,15 @@ def is_emr_runtime():
 def is_dataproc_runtime():
     return runtime_env() == "dataproc"
 
+def get_test_tz():
+    return os.environ.get('TZ', 'UTC')
+
+def is_utc():
+    return get_test_tz() == "UTC"
+
+def is_not_utc():
+    return not is_utc()
+
 _is_nightly_run = False
 _is_precommit_run = False
 

integration_tests/src/main/python/data_gen.py

@@ -21,7 +21,7 @@
 from pyspark.sql.types import *
 import pyspark.sql.functions as f
 import random
-from spark_session import is_tz_utc, is_before_spark_340, with_cpu_session
+from spark_session import is_before_spark_340, with_cpu_session
 import sre_yield
 import struct
 from conftest import skip_unless_precommit_tests,get_datagen_seed
@@ -30,11 +30,6 @@
 from functools import lru_cache
 import hashlib
 
-# set time zone to UTC for timestamp test cases to avoid `datetime` out-of-range error:
-# refer to: https://github.com/NVIDIA/spark-rapids/issues/7535
-os.environ['TZ'] = 'UTC'
-time.tzset()
-
 class DataGen:
     """Base class for data generation"""
 
@@ -584,12 +579,18 @@ def __init__(self, start=None, end=None, nullable=True, tzinfo=timezone.utc):
         elif not isinstance(start, datetime):
             raise RuntimeError('Unsupported type passed in for start {}'.format(start))
 
+        # Spark supports time through: "9999-12-31 23:59:59.999999"
+        # but in order to avoid out-of-range error in non-UTC time zone, here use 9999-12-30 instead of 12-31 as max end
+        # for details, refer to https://github.com/NVIDIA/spark-rapids/issues/7535
+        max_end = datetime(9999, 12, 30, 23, 59, 59, 999999, tzinfo=tzinfo)
         if end is None:
-            # Spark supports time through
-            # "9999-12-31 23:59:59.999999"
-            end = datetime(9999, 12, 31, 23, 59, 59, 999999, tzinfo=tzinfo)
+            end = max_end
         elif isinstance(end, timedelta):
-            end = start + end
+            max_timedelta = max_end - start
+            if ( end >= max_timedelta):
+                end = max_end
+            else:
+                end = start + end
         elif not isinstance(start, date):
             raise RuntimeError('Unsupported type passed in for end {}'.format(end))
 
@@ -749,10 +750,6 @@ def gen_bytes():
             return bytes([ rand.randint(0, 255) for _ in range(length) ])
         self._start(rand, gen_bytes)
 
-def skip_if_not_utc():
-    if (not is_tz_utc()):
-        skip_unless_precommit_tests('The java system time zone is not set to UTC')
-
 # Note: Current(2023/06/06) maxmium IT data size is 7282688 bytes, so LRU cache with maxsize 128
 # will lead to 7282688 * 128 = 932 MB additional memory usage in edge case, which is acceptable.
 @lru_cache(maxsize=128, typed=True)
@@ -776,10 +773,6 @@ def gen_df(spark, data_gen, length=2048, seed=None, num_slices=None):
         # we cannot create a data frame from a nullable struct
         assert not data_gen.nullable
 
-    # Before we get too far we need to verify that we can run with timestamps
-    if src.contains_ts():
-        skip_if_not_utc()
-
     data = gen_df_help(src, length, seed_value)
 
     # We use `numSlices` to create an RDD with the specific number of partitions,
@@ -832,10 +825,6 @@ def _gen_scalars_common(data_gen, count, seed=None):
     else:
         seed_value = seed
 
-    # Before we get too far we need to verify that we can run with timestamps
-    if src.contains_ts():
-        skip_if_not_utc()
-
     rand = random.Random(seed_value)
     src.start(rand)
     return src

integration_tests/src/main/python/spark_session.py

@@ -55,16 +55,6 @@ def _from_scala_map(scala_map):
     'spark.sql.legacy.allowNegativeScaleOfDecimal': 'true',
 }
 
-def is_tz_utc(spark=_spark):
-    """
-    true if the tz is UTC else false
-    """
-    # Now we have to do some kind of ugly internal java stuff
-    jvm = spark.sparkContext._jvm
-    utc = jvm.java.time.ZoneId.of('UTC').normalized()
-    sys_tz = jvm.java.time.ZoneId.systemDefault().normalized()
-    return utc == sys_tz
-
 def _set_all_confs(conf):
     newconf = _default_conf.copy()
     if (should_inject_oom()):