Troubleshoot Python errors in Azure Functions

This article provides information to help you troubleshoot errors with your Python functions in Azure Functions. This article supports both the v1 and v2 programming models. Choose the model you want to use from the selector at the top of the article. The v2 model is currently in preview. For more information on Python programming models, see the Python developer guide.

Here are the troubleshooting sections for common issues in Python functions:

Troubleshoot ModuleNotFoundError

This section helps you troubleshoot module-related errors in your Python function app. These errors typically result in the following Azure Functions error message:

"Exception: ModuleNotFoundError: No module named 'module_name'."

This error occurs when a Python function app fails to load a Python module. The root cause for this error is one of the following issues:

View project files

To identify the actual cause of your issue, you need to get the Python project files that run on your function app. If you don't have the project files on your local computer, you can get them in one of the following ways:

  • If the function app has a WEBSITE_RUN_FROM_PACKAGE app setting and its value is a URL, download the file by copying and pasting the URL into your browser.
  • If the function app has WEBSITE_RUN_FROM_PACKAGE and it's set to 1, go to https://<app-name>.scm.azurewebsites.net/api/vfs/data/SitePackages and download the file from the latest href URL.
  • If the function app doesn't have either of the preceding app settings, go to https://<app-name>.scm.azurewebsites.net/api/settings and find the URL under SCM_RUN_FROM_PACKAGE. Download the file by copying and pasting the URL into your browser.
  • If none of these suggestions resolves the issue, go to https://<app-name>.scm.azurewebsites.net/DebugConsole and view the content under /home/site/wwwroot.

The rest of this article helps you troubleshoot potential causes of this error by inspecting your function app's content, identifying the root cause, and resolving the specific issue.

Diagnose ModuleNotFoundError

This section details potential root causes of module-related errors. After you figure out which is the likely root cause, you can go to the related mitigation.

The package can't be found

Go to .python_packages/lib/python3.6/site-packages/<package-name> or .python_packages/lib/site-packages/<package-name>. If the file path doesn't exist, this missing path is likely the root cause.

Using third-party or outdated tools during deployment might cause this issue.

To mitigate this issue, see Enable remote build or Build native dependencies.

The package isn't resolved with the proper Linux wheel

Go to .python_packages/lib/python3.6/site-packages/<package-name>-<version>-dist-info or .python_packages/lib/site-packages/<package-name>-<version>-dist-info. Use your favorite text editor to open the wheel file and check the Tag: section. The issue might be that the tag value doesn't contain linux.

Python functions run only on Linux in Azure. The Functions runtime v2.x runs on Debian Stretch, and the v3.x runtime runs on Debian Buster. The artifact is expected to contain the correct Linux binaries. When you use the --build local flag in Core Tools, third-party, or outdated tools, it might cause older binaries to be used.

To mitigate the issue, see Enable remote build or Build native dependencies.

The package is incompatible with the Python interpreter version

Go to .python_packages/lib/python3.6/site-packages/<package-name>-<version>-dist-info or .python_packages/lib/site-packages/<package-name>-<version>-dist-info. In your text editor, open the METADATA file and check the Classifiers: section. If the section doesn't contain Python :: 3, Python :: 3.6, Python :: 3.7, Python :: 3.8, or Python :: 3.9, the package version is either too old or, more likely, it's already out of maintenance.

You can check the Python version of your function app from the Azure portal. Navigate to your function app's Overview resource page to find the runtime version. The runtime version supports Python versions as described in the Azure Functions runtime versions overview.

To mitigate the issue, see Update your package to the latest version or Replace the package with equivalents.

The package conflicts with other packages

If you've verified that the package is resolved correctly with the proper Linux wheels, there might be a conflict with other packages. In certain packages, the PyPi documentation might clarify the incompatible modules. For example, in azure 4.0.0, you'll find the following statement:

"This package isn't compatible with azure-storage. If you installed azure-storage, or if you installed azure 1.x/2.x and didn’t uninstall azure-storage, you must uninstall azure-storage first."

You can find the documentation for your package version in https://pypi.org/project/<package-name>/<package-version>.

To mitigate the issue, see Update your package to the latest version or Replace the package with equivalents.

The package supports only Windows and macOS platforms

Open the requirements.txt with a text editor and check the package in https://pypi.org/project/<package-name>. Some packages run only on Windows and macOS platforms. For example, pywin32 runs on Windows only.

The Module Not Found error might not occur when you're using Windows or macOS for local development. However, the package fails to import on Azure Functions, which uses Linux at runtime. This issue is likely to be caused by using pip freeze to export the virtual environment into requirements.txt from your Windows or macOS machine during project initialization.

To mitigate the issue, see Replace the package with equivalents or Handcraft requirements.txt.

Mitigate ModuleNotFoundError

The following are potential mitigations for module-related issues. Use the previously mentioned diagnoses to determine which of these mitigations to try.

Enable remote build

Make sure that remote build is enabled. The way that you make sure depends on your deployment method.

Make sure that the latest version of the Azure Functions extension for Visual Studio Code is installed. Verify that the .vscode/settings.json file exists and it contains the setting "azureFunctions.scmDoBuildDuringDeployment": true. If it doesn't, create the file with the azureFunctions.scmDoBuildDuringDeployment setting enabled, and then redeploy the project.

Build native dependencies

Make sure that the latest versions of both Docker and Azure Functions Core Tools are installed. Go to your local function project folder, and use func azure functionapp publish <app-name> --build-native-deps for deployment.

Update your package to the latest version

In the latest package version of https://pypi.org/project/<package-name>, check the Classifiers: section. The package should be OS Independent, or compatible with POSIX or POSIX :: Linux in Operating System. Also, the programming language should contain: Python :: 3, Python :: 3.6, Python :: 3.7, Python :: 3.8, or Python :: 3.9.

If these package items are correct, you can update the package to the latest version by changing the line <package-name>~=<latest-version> in requirements.txt.

Handcraft requirements.txt

Some developers use pip freeze > requirements.txt to generate the list of Python packages for their developing environments. Although this convenience should work in most cases, there can be issues in cross-platform deployment scenarios, such as developing functions locally on Windows or macOS, but publishing to a function app, which runs on Linux. In this scenario, pip freeze can introduce unexpected operating system-specific dependencies or dependencies for your local development environment. These dependencies can break the Python function app when it's running on Linux.

The best practice is to check the import statement from each .py file in your project source code and then check in only the modules in the requirements.txt file. This practice guarantees that the resolution of packages can be handled properly on different operating systems.

Replace the package with equivalents

First, take a look into the latest version of the package in https://pypi.org/project/<package-name>. This package usually has its own GitHub page. Go to the Issues section on GitHub and search to see whether your issue has been fixed. If it has been fixed, update the package to the latest version.

Sometimes, the package might have been integrated into Python Standard Library (such as pathlib). If so, because we provide a certain Python distribution in Azure Functions (Python 3.6, Python 3.7, Python 3.8, and Python 3.9), the package in your requirements.txt file should be removed.

However, if you're finding that the issue hasn't been fixed, and you're on a deadline, we encourage you to do some research to find a similar package for your project. Usually, the Python community will provide you with a wide variety of similar libraries that you can use.


Troubleshoot cannot import 'cygrpc'

This section helps you troubleshoot 'cygrpc'-related errors in your Python function app. These errors typically result in the following Azure Functions error message:

"Cannot import name 'cygrpc' from 'grpc._cython'"

This error occurs when a Python function app fails to start with a proper Python interpreter. The root cause for this error is one of the following issues:

Diagnose the 'cygrpc' reference error

The Python interpreter mismatches OS architecture

This mismatch is most likely caused by a 32-bit Python interpreter being installed on your 64-bit operating system.

If you're running on an x64 operating system, ensure that your Python version 3.6, 3.7, 3.8, or 3.9 interpreter is also on a 64-bit version.

You can check your Python interpreter bitness by running the following commands:

On Windows in PowerShell, run py -c 'import platform; print(platform.architecture()[0])'.

On a Unix-like shell, run python3 -c 'import platform; print(platform.architecture()[0])'.

If there's a mismatch between Python interpreter bitness and the operating system architecture, download a proper Python interpreter from Python Software Foundation.

The Python interpreter isn't supported by Azure Functions Python Worker

The Azure Functions Python Worker supports only Python versions 3.6, 3.7, 3.8, and 3.9.

Check to see whether your Python interpreter matches your expected version by py --version in Windows or python3 --version in Unix-like systems. Ensure that the return result is Python 3.6.x, Python 3.7.x, Python 3.8.x, or Python 3.9.x.

If your Python interpreter version doesn't meet the requirements for Azure Functions, instead download the Python version 3.6, 3.7, 3.8, or 3.9 interpreter from Python Software Foundation.


Troubleshoot "python exited with code 137"

Code 137 errors are typically caused by out-of-memory issues in your Python function app. As a result, you get the following Azure Functions error message:

"Microsoft.Azure.WebJobs.Script.Workers.WorkerProcessExitException : python exited with code 137"

This error occurs when a Python function app is forced to terminate by the operating system with a SIGKILL signal. This signal usually indicates an out-of-memory error in your Python process. The Azure Functions platform has a service limitation that terminates any function apps that exceed this limit.

Visit the tutorial section in memory profiling on Python functions to analyze the memory bottleneck in your function app.


Troubleshoot "python exited with code 139"

This section helps you troubleshoot segmentation fault errors in your Python function app. These errors typically result in the following Azure Functions error message:

"Microsoft.Azure.WebJobs.Script.Workers.WorkerProcessExitException : python exited with code 139"

This error occurs when a Python function app is forced to terminate by the operating system with a SIGSEGV signal. This signal indicates violation of the memory segmentation, which can result from an unexpected reading from or writing into a restricted memory region. In the following sections, we provide a list of common root causes.

A regression from third-party packages

In your function app's requirements.txt file, an unpinned package will be upgraded to the latest version in every Azure Functions deployment. Vendors of these packages might introduce regressions in their latest release. To recover from this issue, try commenting out the import statements, disabling the package references, or pinning the package to a previous version in requirements.txt.

Unpickling from a malformed .pkl file

If your function app is using the Python pickle library to load a Python object from a .pkl file, it's possible that the file contains a malformed bytes string or an invalid address reference. To recover from this issue, try commenting out the pickle.load() function.

Pyodbc connection collision

If your function app is using the popular ODBC database driver pyodbc, it's possible that multiple connections are open within a single function app. To avoid this issue, use the singleton pattern, and ensure that only one pyodbc connection is used across the function app.


Troubleshoot errors with Protocol Buffers

Version 4.x.x of the Protocol Buffers (Protobuf) package introduces breaking changes. Because the Python worker process for Azure Functions relies on v3.x.x of this package, pinning your function app to use v4.x.x can break your app. At this time, you should also avoid using any libraries that require Protobuf v4.x.x.

Example error logs:

 [Information] File "/azure-functions-host/workers/python/3.8/LINUX/X64/azure_functions_worker/protos/shared/NullableTypes_pb2.py", line 38, in <module>
 [Information] _descriptor.FieldDescriptor(
 [Information] File "/home/site/wwwroot/.python_packages/lib/site-packages/google/protobuf/descriptor.py", line 560, in __new__
 [Information] _message.Message._CheckCalledFromGeneratedFile()
 [Error] TypeError: Descriptors cannot be created directly.
 [Information] If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
 [Information] If you cannot immediately regenerate your protos, some other possible workarounds are:
 [Information] 1. Downgrade the protobuf package to 3.20.x or lower.
 [Information] 2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).
 [Information] More information: https://developers.google.com/protocol-buffers/docs/news/2022-05-06#python-updates

You can mitigate this issue in either of two ways:

  • Set the application setting PYTHON_ISOLATE_WORKER_DEPENDENCIES to a value of 1.

  • Pin Protobuf to a non-4.x.x. version, as in the following example:

    protobuf >= 3.19.3, == 3.*
    

Multiple Python workers not supported

The multiple Python workers setting isn't supported in the v2 programming model at this time. More specifically, enabling intelligent concurrency by setting FUNCTIONS_WORKER_PROCESS_COUNT to greater than 1 isn't supported for functions that are developed by using the v2 model.

Troubleshoot "could not load file or assembly"

If you receive this error, it might be because you're using the v2 programming model. This error results from a known issue that will be resolved in an upcoming release.

This specific error might read:

"DurableTask.Netherite.AzureFunctions: Could not load file or assembly 'Microsoft.Azure.WebJobs.Extensions.DurableTask, Version=2.0.0.0, Culture=neutral, PublicKeyToken=014045d636e89289'.
The system cannot find the file specified."

This error might occur because of an issue with how the extension bundle was cached. To troubleshoot this issue, you can run the following command with --verbose to see more details:

func host start --verbose

After you run the command, if you notice that Loading startup extension <> isn't followed by Loaded extension <> for each extension, it's likely that you have a caching issue.

To resolve this issue:

  1. Find the .azure-functions-core-tools path by running:

    func GetExtensionBundlePath
    
  2. Delete the .azure-functions-core-tools directory.

    rm -r <insert path>/.azure-functions-core-tools
    

Troubleshoot "unable to resolve the Azure Storage connection"

You might see this error in your local output as the following message:

"Microsoft.Azure.WebJobs.Extensions.DurableTask: Unable to resolve the Azure Storage connection named 'Storage'.
Value cannot be null. (Parameter 'provider')"

This error is a result of how extensions are loaded from the bundle locally. To resolve this error, take one of the following actions:

  • Use a storage emulator such as Azurite. This option is a good one when you aren't planning to use a storage account in your function application.

  • Create a storage account and add a connection string to the AzureWebJobsStorage environment variable in the localsettings.json file. Use this option when you're using a storage account trigger or binding with your application, or if you have an existing storage account. To get started, see Create a storage account.

Issue with deployment

In the Azure portal, select Settings > Configuration, and then ensure that the AzureWebJobsFeatureFlags application setting has a value of EnableWorkerIndexing. If it's not found, add this setting to the function app.

Next steps

If you're unable to resolve your issue, contact the Azure Functions team: