CredentialUnavailableError using Pytest and Async credentials

[savvaki]

• azure-identity:
• 1.25.1:
• Linux:
• 3.13.7:

Describe the bug
CredentialUnavailableError is raised when I run repeated pytests using the Async DefaultAzureCredential. Note: about 80% of the tests work, but some fail to get the credential.

command_args = ['account', 'get-access-token', '--output', 'json', '--resource', 'https://ai.azure.com'], timeout = 10

    async def _run_command(command_args: List[str], timeout: int) -> str:
        # Ensure executable exists in PATH first. This avoids a subprocess call that would fail anyway.
        if sys.platform.startswith("win"):
            # On Windows, the expected executable is az.cmd, so we check for that first, falling back to 'az' in case.
            az_path = shutil.which(EXECUTABLE_NAME + ".cmd") or shutil.which(EXECUTABLE_NAME)
        else:
            az_path = shutil.which(EXECUTABLE_NAME)
        if not az_path:
            raise CredentialUnavailableError(message=CLI_NOT_FOUND)
    
        args = [az_path] + command_args
        working_directory = get_safe_working_dir()
        try:
            _LOGGER.debug("Executing subprocess with the following arguments %s", args)
            proc = await asyncio.create_subprocess_exec(
                *args,
                stdout=asyncio.subprocess.PIPE,
                stderr=asyncio.subprocess.PIPE,
                stdin=asyncio.subprocess.DEVNULL,
                cwd=working_directory,
                env=dict(os.environ, AZURE_CORE_NO_COLOR="true"),
            )
            stdout_b, stderr_b = await asyncio.wait_for(proc.communicate(), timeout)
            output = stdout_b.decode()
            stderr = stderr_b.decode()
        except asyncio.TimeoutError as ex:
            proc.kill()
>           raise CredentialUnavailableError(message="Timed out waiting for Azure CLI") from ex
E           azure.identity._exceptions.CredentialUnavailableError: Timed out waiting for Azure CLI

[pvaneck]

This might potentially be some sort of resource contention causing flakiness/delays as, depending on your test setup, there could be a lot subprocess calls to az account get-access-token, possibly happening concurrently. If you can't use a mock credential and real tokens are actually needed in your tests, then some things to try are:

• Increase the timeout by passing something like process_timeout=30 to DefaultAzureCredential.
• If multiple test cases are instantiating the same client, then constructing this client once in a pytest fixture, and using the same client across multiple tests will ensure that operations on this client reuse the same access token. This should reduce the number of subprocess calls.