The following sections detail updates to the AWS Panorama Appliance software, including changes to the operation system, AWS Panorama libraries, the application SDK, and the application container image.
Release date 2024-09-27
Type: Mandatory
Sagemaker Neo Model nodes are not supported on this image. Applications currently using Sagemaker Neo model nodes will need to be rebuilt without them. Jetpack 4.6 is not supported on 8.0.29, although you can still deploy Jetpack 4.6 Neo models on older images. It’s recommended that you rebuild your application without the Neo models for use with Jetpack 5.1.1, using OpenGPU to run models directly on the GPU instead. Alternatively, you may decline the update to 8.0.29 and keep using Neo model nodes.
*While the 8.0.29 image is marked as "Mandatory", you will not be forced to upgrade to it. All regular functions are available under the old image, including managing (deploying) applications and OTA from one pre-8.0.29 image to another pre-8.0.29 image. However, you cannot OTA from a post-8.0.29 image to a previous, older image.
Change base image from Ubuntu 18 to Ubuntu 20 Software versions: Python - 3.8.10 NumPy - 1.24.4
Image URI: public.ecr.aws/panorama/panorama-application:1.2.1-arm64v8
Release date 2023-12-28
Type: Optional
With this update, the appliance uses additional AWS services to manage software updates. If you restrict network communication outbound from the appliance, or connect it to a private VPC subnet, you must allow access to additional endpoints and ports before applying the update.
- Amazon ECR service and Docker registry endpoints
- AWS IoT Core credential provider endpoints
- AWS IoT Core data plane endpoints (additional ports)
For details on ports and endpoints used by the AWS Panorama Appliance, see Network setup.
If you connect your appliance to a private VPC subnet, create VPC endpoints for these services and, for IoT Core, add an additional Route53 record set for subdomains of the credentials endpoint. For more information, see Using VPC endpoints.
This repo provides a CloudFormation template that demonstrates how to configure VPC endpoints, hosted zones, and record sets in your private subnets: vpc-appliance.yml
Fixed an issue that causes connectivity issues in application containers when using custom DNS settings.
Release date: 2023-11-08
Type: Optional
This release is recalled due to an issue that causes connectivity issues in application containers when using custom DNS settings.
If your appliance is affected, you can revert to the previous version 6.2.1 by choosing Install software update in the device settings menu. For more information, see Update the appliance software.
Release date: 2023-10-17
Application SDK: 1.2.0
Python : 3.8.0
NumPy : 1.24.3
Image URI: public.ecr.aws/panorama/panorama-application:1.2.0-py3.8
This release updates Python to version 3.8 in the base image. For all available tags, see https://gallery.ecr.aws/panorama/panorama-application
Release date: 2023-09-06
Type: Optional
Fixed an issue that caused the appliance to appear offline after a network interruption.
Release date: 2023-07-06
Type: Required
Updated error handling to improve interoperability with media gateways.
Updates to improve system stability, performance, and security.
Release date: 2023-03-31
Type: Optional
Fixed a race condition that caused disk corruption during shutdown, and improved autorecovery from disk corruption.
Improved error handling for unsupported (non-H.264) RTSP stream encodings.
Improved camera stream connection retry behavior to improve interoperability with media gateways.
Release date: 2023-01-23
Type: Optional
Improved error handling for input port configuration issues.
Fixed an issue that caused applications to crash after rebooting the device.
Release date: 2022-11-16
Type: Optional
Improved format and usability of application logs.
Fixed an issue that caused application state to be incorrect after power loss.
Fixed an issue that caused application files to remain on the appliance after the application instance is removed.
Fixed an issue that caused storage and memory metrics to be shown in incorrect units in Amazon CloudWatch.
Release date: 2022-11-16
Application SDK: 1.2.0
Python : 3.7.5
NumPy : 1.18.2
Version 1.2 of the application SDK adds a timeout parameter to video_in.get(). If you specify a timeout, the SDK raises TimeoutException if no frames are received within the specified number of seconds. The default behavior, which blocks the thread until a frame is received, is unchanged.
The SDK now sets the AWS_REGION environment variable to the device's AWS Region. You can use the AWS SDK for Python in your application without specifying a Region in code or your container image, and the device's Region is used automatically.
Improved error handling for self.call() to check model name against manifest and raise KeyError if the model is not in the application.
Improved error handling for put() to raise ValueError if the input frame list is empty.
Fixed a race condition in video_in.get() that caused the SDK to return old frames with is_cached set instead of blocking until at least one new frame is received. is_cached should only be set when a batch of frames from multiple sources includes some old frames and some new.
Release date: 2022-10-13
Type: Optional
BREAKING This release upgrades NVIDIA JetPack software on the appliance. After the update, deployed models will stop working due to the changes. Redeploy your applications to compile models against the new software version. If you use GPU access in your application container, recompile your application with libraries supported by JetPack 4.6.2.
You can use the AWS Panorama console or API to reboot an appliance remotely. For more information, see Reboot an appliance and Manage appliances with the AWS Panorama API.
You can pause and resume camera streams with the AWS Panorama API. When a camera stream is paused, your application doesn't receive images from that stream. You can connect multiple streams to an application and use pause and resume to switch between them. For more information, see Manage applications with the AWS Panorama API
Improved format and usability of deployment-related logs.
Fixed an issue that caused applications with multiple input streams to only display 1 output, instead of tiling the output.
Fixed an issue that caused application logs to have incorrect timestamps in CloudWatch logs.
Fixed an issue that allowed two code nodes to run side-by-side in one application.
Release date: 2022-08-24
Type: Optional
Fixed an issue that causes the device to go offline and require a factory reset. To learn how to perform a factory reset, see Power and reset buttons.
Fixed an issue that prevents you from removing faulty applications from the appliance.
You can use a USB drive to get an encrypted log image off of the device. The AWS Panorama service team can decrypt the logs on your behalf and assist in debugging. For more information, see Egressing logs from a device.
Release date: 2022-06-23
Type: Optional
Fixed an issue that caused cameras to disconnect with a dqBuffer(131) Failed due to Resource temporarily unavailable after retry error.
Fixed an issue that caused an apparmor="DENIED" error to appear in appliances' kern_log log stream.
Updates to improve system stability, performance, and security.
Release date: 2022-05-05
Type: Optional
The console_output log now rotates automatically and uses a maximum of 11 MB of storage on the appliance.
For more information, see Viewing AWS Panorama logs.
Version 1.1 of the application SDK adds is_cached and stream_id attributes to the media type.
Improved support for the threading library. The application SDK doesn't block other threads while waiting for inference with node.call.
Removed unused components to reduce size of application container images.
Release date: 2022-03-29
Application SDK: 1.1.0
Python : 3.7.5
NumPy : 1.18.2
Reference docs: resources/applicationsdk-reference.md
Release date: 2022-03-24
Type: Required
You can access the graphics processor (GPU) on the device to use GPU-accelerated libraries, or run machine learning models in your application code. To turn on GPU access, you add GPU access as a requirement to the package configuration after building your application code container.
For more information, see Using the GPU.
You can monitor or debug applications locally by running an HTTP server or other listener alongside your application code. To serve external traffic, you map ports on the device to ports on your application container.
A new sample application, debug-server, demonstrates how to use inbound ports to serve HTTP traffic. It uses multithreading to run application code, an HTTP server, and an HTTP client simultaneously. After running for a few minutes, the application sends an HTTP request to the device over the local network that signals it to restart the application code.
For more information, see Serving inbound traffic.
Added kern_log to appliance log streams to capture Linux kernel events. For more information, see Viewing AWS Panorama logs.
Release date: 2022-02-22
Type: Optional
Updated system software and libraries.
Removed unused packages to reduce image size and complexity.
Applied fix for CVE-2021-4034.
Release date: 2022-01-13
Type: Mandatory
You can now configure the AWS Panorama Appliance to use a specific NTP server for clock syncronization. Configure NTP settings during appliance setup with other networking settings.
Added the frameworkVersion setting for model descriptor files in #7. For TensorFlow and PyTorch models, you can specify the version of the framework that you used to build the model. This setting is passed on to SageMaker Neo (FrameworkVersion parameter) when AWS Panorama compiles the model. SageMaker Neo currently supports the following values:
TensorFlow: 1.5, 2.4
PyTorch: 1.4, 1.5, 1.6, 1.7, 1.8
The appliance now writes logs to the USB drive during provisioning. This feature is available when you reprovision a device after updating it to this version.
Fixed an issue that caused the appliance to stop trying to connect to inactive cameras after a few minutes.
Fixed an issue that sometimes caused active nodes to be reported as inactive.
Fixed an issue that sometimes caused applications to crash.
Fixed an issue that sometimes caused the appliance software to crash.
Release date: 2021-11-08
Type: Mandatory
Updated device logs to include the full log type (INFO, WARN, ERROR). You can filter logs in the Amazon CloudWatch Logs console by entering WARN or ERROR in the search bar of a log stream (case sensitive).
Updated device controller logs (occ_log) to match the timestamp in the log to the timestamp in CloudWatch logs.
For more information about logs, see Viewing AWS Panorama event logs in CloudWatch Logs
Added precisionMode setting to the model descriptor schema. You can set this field to FP16 to run the model with 16-bit floating-point precision, instead of 32-bit precision. Using lower precision significantly reduces inference time, but can affect accuracy.
When you update your model's descriptor file, reimport the model with the application CLI to apply the change. For an example descriptor file and update script, see the sample application.
Model descriptor: packages/123456789012-SQUEEZENET_PYTORCH_V1-1.0/descriptor.json
Update script: update-model-config.sh
Release date: 2021-10-20
Type: Mandatory
Adds support for application base image v1.0.
Version 1.0 of the application SDK replaces the application SDK from preview and changes the programming model for applications.
Application SDK: 1.0.0
Python : 3.7.5
NumPy : 1.18.2
Reference docs: resources/applicationsdk-reference.md