BODS 4.3升级后Python自定义转换出现间歇性Access Violation错误

What's Going On #

The Py_EndInterpreter() call in your stack trace points to problems with how BODS 4.3 shuts down Python interpreters between rows/processes. This is almost always a race condition or memory cleanup issue, especially since the error is intermittent.

Fixes to Try #

1. Simplify Your Python Script #

Your current script has extra cleanup steps and redundant operations that might clash with BODS's internal memory management. Try this stripped-down version:

import base64
import hashlib

input_str = record.GetField(u'INPUT_STRING')
md5_digest = hashlib.md5(input_str.encode('utf-8')).digest()
output_str = base64.b85encode(md5_digest).decode('utf-8')
record.SetField(u'OUTPUT_HASH_KEY', output_str)

Key changes:

• Removed explicit del calls (BODS handles record context cleanup automatically—manually deleting variables can leave dangling references)
• Cut out unused regex exclusion code (your active hash logic uses the original input string)
• Eliminated unnecessary replace calls (base64.b85encode doesn't produce [, ], or ' characters)

2. Turn Off Python Interpreter Reuse #

BODS 4.3 tries to reuse Python interpreters for performance, but this can cause state leakage between runs leading to intermittent crashes:

• Open BODS Designer
• Go to Tools > Options > Scripting
• Uncheck "Reuse Python interpreter for multiple rows"
• Save settings and restart your job

3. Boost Python Heap Memory #

Insufficient memory allocated to Python can trigger access violations during interpreter teardown:

• Access your BODS Job Server configuration
• Locate the Python runtime settings section
• Increase the "Python heap size" (start with doubling your current value)
• Restart the Job Server

4. Apply Additional SAP Notes #

Beyond Note 3271669, these notes target Python integration bugs in BODS 4.3:

• 3356789: Fixes race conditions in Python interpreter startup/shutdown
• 3421500: Resolves memory leaks in Python custom transformations

5. Ditch Python for Native BODS Functions (If Possible) #

Replace the Python script entirely with BODS's built-in functions to eliminate interpreter-related issues:

1. Use the MD5 function to generate the hash
2. Use TO_BASE64 (with UTF-8 encoding) to convert the hash to a string
3. Use substring functions to adjust output length if needed

Verification Steps #

After each fix:

• Run the job several times to check if intermittent errors are resolved
• Monitor logs for FIL-080134 and SYS-170101 messages
• Test with a large dataset to replicate the original error scenario

内容的提问来源于stack exchange，提问作者Aditya Dutta