Data access can be successful even if no topic is available in the data bus (reason: It is possible to subscribe to a non-existent topic).
All steps can retrieve a value from the data bus via AI Inference Server without taking the pipeline sequence into account if the data of the steps originates from different AI Runtime Family apps (asynchronous mode), which can lead to problems in some scenarios.
When you download the logs from Industrial Edge Management System, the file is saved with the ".tar" extension by default, even though the contents are a ZIP folder. Rename the file from .tar to .zip if the archive reader you are using cannot recognize the archive type from the file contents.
Maximum size of the configuration package that can be imported either locally (for example, from a USB flash drive) or remotely via the AI Model Manager app: 2200 MB.
Timeout of the configuration package import: 5 minutes.
Maximum number of input/output topics on IPC227E: 600 variables.
Maximum number of input/output topics on IPC427E: 800 variables.
The app could crash if the pipeline contains 1000 variables (due to the memory limit).
Maximum payload quantity with a message frequency of 1 per second and 200 MB dedicated memory for Orchestrator:
With a static payload quantity (as a large character string), no memory fragmentation occurs:
with SIMATIC S7-Connector v1.6.0-4: 6 MB,
with IE MQTT Connector v1.5.1: 10 MB.
With dynamic payload quantity, memory fragmentation may occur:
with SIMATIC S7-Connector v1.6.0-4: 6 MB,
with IE MQTT Connector v1.5.1: 6 MB.
If there is a problem (for example, restart of the AI Inference Server or restart of the Edge Box), the logging configuration must be activated again (switching off and on).
Side effect in the Runtime Manager when handling large ZIP files (>1.1 GB): The new commands can be lost (e.g. Start pipeline, Stop pipeline) because the Runtime Manager is still working on the previous command.
The internal Python Runtime has a reserved memory of 1.5 GB for MLFB 6AV2170-0LA10-0AA0 (single pipeline version of AI Inference Server) and MLFB 6AV2170-0LA11-0AA0 (AI Inference Server - GPU accelerated - 1 Pipeline). If this memory limit is reached during the runtime of a pipeline, the pipeline execution crashes without warning. The AI Inference Server continues working. An error is displayed in the user interface of AI Inference Server and the process goes to the error state, but there is no message to be seen about the extent of the memory limit.
The internal Python Runtime has a reserved memory of 4 GB for MLFB 6AV2170-0LA10-1AA0 (multiple pipeline versions of AI Inference Server). If this memory limit is reached during the runtime of a pipeline, the pipeline execution crashes without warning. The AI Inference Server continues working. An error is displayed in the user interface of AI Inference Server and the process goes to the error state, but there is no message to be seen about the extent of the memory limit.
When a specific component of the AI Inference Server app is restarted, a RESET message is sent and triggers the stop of the running pipeline. This message is not systematically received by the runtime, so the pipeline may not be automatically terminated (runtime is ready to process the RESET, but the RESET is sporadically lost beforehand). The current workaround is to restart the AI Inference Server app.
Parallel import on multiple Internet browser tabs is not supported.
Workaround: Use only one browser tab when you import a configuration package.
It is not possible to downgrade the AI Inference Server via Edge Management to a previous version.
Workaround: Uninstall the existing application of your Edge device and install a previous version. The previously-imported/received pipelines and your mapping information will be lost, you must re-import and map again.
LogModule can write strings to the log. Any other type causes a fatal error in runtime and the pipeline is stopped, e.g.:
def run (data):
.... log.info([1,2,3])
After the application upgrade (via Edge Management), the "503 Service unavailable" message may be displayed. The updated version of the application does not work correctly. Workaround for the problem: Restart the application.
The application description (displayed in Edge Management) is not updated after an upgrade due to a problem with Edge Management.
The parameter data received from the data bus can be overloaded along with the variable data, if the Python script processes them too slowly. (Reason: The parameters and the variables are read by the same data connector). The internal queue can be fine-tuned if required. Please read the operating manual.
The maximum length of the pipeline description is 2000 characters (in pipeline_config.yml file).
Running the pipeline at the same time can cause the application to crash if the memory used by the internal Python runtime reaches the limit.
Running multiple pipelines can impact the performance (other calculations are required with the same CPU).
If the inputs come faster than the Python script can process them, a problem with insufficient work memory can occur. Workaround for the problem: Test your AI modules before you make them available in production.
In the Mozilla Firefox browser, the selected tab is not underlined on the Pipeline visualization, Settings and About & legal information pages.
NumPy: Only versions 1.x are supported. Versions 2.x are not yet supported.