Skip to content

Latest commit

 

History

History
239 lines (149 loc) · 6.22 KB

README.md

File metadata and controls

239 lines (149 loc) · 6.22 KB
Raw

PiB outer-loop: Vision on Edge (VoE)

Introduction

  • VoE is an open-source tool that builds vision-based intelligent edge solutions using Machine Learning. Visit this repo for more information.
  • Deploying VoE is very similar process to deploying other applications in the outer-loop labs, but there are a few dependencies that must be configured first.

Note: The VoE app is being deprecated and will be replaced by a new version. It is used here for demonstrative purposes to show how to deploy a more complex app with PiB.

Lab Prerequisites

Fleet Configuration Prerequisites

  • There are a few dependencies and prerequisites that must be configured before deploying the VoE app.
  • Azure Resources
    • Azure IoT Hub
    • Azure Cognitive Services
  • Fleet VM Configuration
    • At least 8 cores
    • Kubernetes secrets

Create Azure Resources

Define resource name variables

export VOE_HUB_NAME=voe-hub-$MY_BRANCH

export VOE_RG=voe-rg-$MY_BRANCH

export VOE_AZ_COG_SVC_NAME=voe-acs-$MY_BRANCH

Login to Azure CLI

az login --use-device-code

# option if your Codespace is configured with SP credentials
flt az login

Create Azure IoT Hub

# add azure-iot extension
az extension add -n azure-iot

az iot hub create --resource-group $VOE_RG --name $VOE_HUB_NAME

Create Azure Cognitive Services

# you may have to create a cognitive services multi-service account in the azure portal to fulfill the requirement to agree to the responsible AI terms for the resource
az cognitiveservices account create --kind CognitiveServices --name $VOE_AZ_COG_SVC_NAME --resource-group $VOE_RG --sku S0 --location yourlocation

Update fleet creation script

  • Add the following lines to vm/setup/pre-flux.sh and replace the values in [] with the names of the resources created above.

    • This will run on the fleet vm/s during setup and will create the voe namespace and required k8s secret.
    • The VM uses Managed Identity to retrieve the connection string values from the IoT Hub.

  • Do NOT run this fence!

# add the iot extension
az extension add -n azure-iot

# azure iot secrets
echo "IOTHUB_CONNECTION_STRING=$(az iot hub connection-string show --hub-name [your-voe-hub-name] -o tsv)" > "$HOME/.ssh/iot.env"
echo "IOTEDGE_DEVICE_CONNECTION_STRING=$(az iot hub device-identity connection-string show --hub-name [your-voe-hub-name] --device-id "$(hostname)" -o tsv)" >> "$HOME/.ssh/iot.env"

# azure cognitive services secrets
echo "ENDPOINT=$(az cognitiveservices account show -n [your-voe-acs-name] -g [your-voe-rg] --query properties.endpoint -o tsv)" > "$HOME/.ssh/acs.env"
echo "TRAINING_KEY=$(az cognitiveservices account keys list -n [your-voe-acs-name] -g [your-voe-rg] --query key1 -o tsv)" >> "$HOME/.ssh/acs.env"

kubectl create ns voe

kubectl create secret generic azure-env --from-env-file "$HOME/.ssh/iot.env" --from-env-file="$HOME/.ssh/acs.env" -n voe
  • After updating the script, push your changes
git add .
git commit -m "updated pre-flux.sh"
git push

Create a fleet

  • The VoE application requires at least 8 cores, this must be specified at fleet creation with the --cores flag.
  • Before creating the fleet, a Managed Identity (MI) must be configured. See setup docs for instructions.
# before creating the cluster, make sure PIB_MI is set
# MI is required for the voe k8s secrets to be created properly
flt env PIB_MI

# set MY_CLUSTER
export MY_CLUSTER=central-tx-voe-$MY_BRANCH

# create your cluster
flt create -c $MY_CLUSTER --cores 8

Add iot devices (clusters in fleet) to IoT Hub

You must run this command for each cluster in the fleet

az iot hub device-identity create --ee -n $VOE_HUB_NAME -d $MY_CLUSTER

Create apps/voe directory

  • Add voe to the apps/ directory
cd $PIB_BASE/labs/advanced-labs

cp -aR voe ../../apps

# commit and push changes
git add ../../
git commit -m "Adding voe to apps/"
git push

Deploy the VoE app to your fleet

# start in the apps/voe directory
cd $PIB_BASE/apps/voe

# make sure repo is up to date
git pull

# add all clusters as a target
flt targets add all

# other options
# flt targets add region:central
# flt targets add zone:central-tx
# flt targets add district:central-tx-voe

# deploy voe via ci-cd and GitOps Automation
flt targets deploy

Verify VoE deployment

# should see voe added
git pull

# force flux to reconcile
flt sync

# check the cluster for voe
flt check app voe

# check the status of the pods running on the cluster
# it can take ~2 minutes for all of the voe services to be ready
flt exec kic pods

# curl healthz and readyz endpoints
flt curl /healthz
flt curl /readyz

Navigate to VoE in the browser

  • Get the FQDN of your cluster
  • Copy and paste the FQDN into your browser

    • You should get the VoE home page

      # display the FQDN
echo $MY_CLUSTER.$PIB_SSL

# if dns/ssl is not configured
# use the cluster IP
cat $PIB_BASE/clusters/$MY_CLUSTER.yaml | grep domain

VoE Application

  • As mentioned, the VoE application is a complex application and is made up of 6 services (CVCapture, Inference, Predict, RTSPSim, Upload, and Web).
  • Because of the complexity, there is additional ingress configuration required to ensure the different services can communicate as needed.
  • The Inference service is dependent on 4 of the other services, we use initContainers to enforce the ordering.
  • The Web, Upload, and RTSPSim services require a persistent volume claim (pvc).

Delete Your Cluster

  • Once you're finished with the workshop and experimenting, delete your cluster
# start in the root of your repo
cd $PIB_BASE
git pull
flt delete $MY_CLUSTER
rm ips
git commit -am "deleted cluster"
git push