MVP’s Favorite Content: Microsoft Teams and DevOps

by Contributed | Aug 9, 2024 | Technology

This article is contributed. See the original author and article here.

In this blog series dedicated to Microsoft’s technical articles, we’ll highlight our MVPs’ favorite article along with their personal insights.

Onyinye Madubuko, M365 MVP, Ireland

Clear Teams cache – Microsoft Teams | Microsoft Learn

“This was helpful in solving new Teams application for users experiencing issues.”

*Relevant Blog: Teams Window keeps flickering and not launching (techiejournals.com)

Laurent Carlier, M365 MVP, France

Overview of meetings, webinars, and town halls – Microsoft Teams | Microsoft Learn

“Teams meetings have evolved significantly over the past few years, with the end of live Team events, the introduction of Town Halls, and the strengthening of Teams Premium features. It’s not always easy to understand what is and isn’t included in Teams Premium licences, or to explain the benefits of purchasing this new plan. This documentation and its comparison tables make my job a lot easier today.”

Edward Kuo, Microsoft Azure MVP, Taiwan

Introduction to Azure DevOps – Training | Microsoft Learn

“I am a DevOps expert and an Azure specialist, primarily responsible for guiding enterprises in using Azure DevOps and establishing DevOps teams.”

*Relevant Blog: DevOps – EK.Technology Learn (edwardkuo.dev)

Kazushi Kamegawa, Developer Technologies MVP, Japan

Managed DevOps Pools – The Origin Story – Engineering@Microsoft

“Using Azure Pipelines for CI/CD in a closed network environment requires the use of self-hosted agents, and managing these images was a very labor-intensive task. Even with automation, updates took 5-6 hours and had to be done once or twice a month. It was probably a challenge for everyone.

In this context, the announcement of the Managed DevOps Pools on this blog was very welcome news. It’s not just me; it’s likely the solution everyone was hoping for, and I am very much looking forward to it.”

(In Japanese: Azure Pipelinesを使って閉域環境でのCI/CDはセルフホストエージェントを使わなければならない上に、イメージの管理は非常に大変な作業でした。更新作業には自動化していても5-6時間かかる上に、月に1-2度は行わなくてはなりません。おそらく皆さん大変だったでしょう。

そんな中、Managed DevOps Poolのアナウンスが本ブログで行われました。私だけではなく、おそらく皆さんが望んだソリューションであり、大変期待しています。)

*Relevant event: Azure DevOpsオンライン Vol.11 ～ Managed DevOps Pool解説 – connpass

Monitoring GPU Metrics in AKS with Azure Managed Prometheus, DCGM Exporter and Managed Grafana

by Contributed | Aug 8, 2024 | Technology

This article is contributed. See the original author and article here.

Azure Monitor managed service for Prometheus provides a production-grade solution for monitoring without the hassle of installation and maintenance. By leveraging these managed services, we can focus on extracting insights from your metrics and logs rather than managing the underlying infrastructure.

The integration of essential GPU metrics—such as Framebuffer Memory Usage, GPU Utilization, Tensor Core Utilization, and SM Clock Frequencies—into Azure Managed Prometheus and Grafana enhances the visualization of actionable insights. This integration facilitates a comprehensive understanding of GPU consumption patterns, enabling more informed decisions regarding optimization and resource allocation.

Azure Managed Prometheus recently announced general availability of Operator and CRD support, which will enable customers to customize metrics collection and add scraping of metrics from workloads and applications using Service and Pod Monitors, similar to the OSS Prometheus Operator.

This blog will demonstrate how we leveraged the CRD/Operator support in Azure Managed Prometheus and used the Nvidia DCGM Exporter and Grafana to enable GPU monitoring.

GPU monitoring

As the use of GPUs has skyrocketed for deploying large language models (LLMs) for both inference and fine-tuning, monitoring these resources becomes critical to ensure optimal performance and utilization. Prometheus, an open-source monitoring and alerting toolkit, coupled with Grafana, a powerful dashboarding and visualization tool, provides an excellent solution for collecting, visualizing, and acting on these metrics.

Essential metrics such as Framebuffer Memory Usage, GPU Utilization, Tensor Core Utilization, and SM Clock Frequencies serve as fundamental indicators of GPU consumption, offering invaluable insights into the performance and efficiency of graphics processing units, and thereby enabling us to reduce our COGs and improve operations.

Using Nvidia’s DGCM Exporter with Azure Managed Prometheus

The DGCM Exporter is a tool developed by Nvidia to collect and export GPU metrics. It runs as a pod on Kubernetes clusters and gathers various metrics from Nvidia GPUs, such as utilization, memory usage, temperature, and power consumption. These metrics are crucial for monitoring and managing the performance of GPUs.

You can integrate this exporter with Azure Managed Prometheus. The section below in blog describes the steps and changes needed to deploy the DCGM Exporter successfully.

Prerequisites

Before we jump straight to the installation, ensure your AKS cluster meets the following requirements:

1. GPU Node Pool: Add a node pool with the required VM SKU that includes GPU support.


2. GPU Driver: Ensure the NVIDIA Kubernetes device plugin driver is running as a DaemonSet on your GPU nodes.


3. Enable Azure Managed Prometheus and Azure Managed Grafana on your AKS cluster.

Refactoring Nvidia DCGM Exporter for AKS: Code Changes and Deployment Guide

Updating API Versions and Configurations for Seamless Integration

As per the official documentation, the best way to get started with DGCM Exporter is to install it using Helm. When installing over AKS with Managed Prometheus, you might encounter the below error:

Error: Installation Failed: Unable to build Kubernetes objects from release manifest: resource mapping not found for name: "dcgm-exporter-xxxxx" namespace: "default" from "": no matches for kind "ServiceMonitor" in version "monitoring.coreos.com/v1". Ensure CRDs are installed first.

To resolve this, follow these steps to make necessary changes in the DCGM code:

1. Clone the Project: Go to the GitHub repository of the DCGM Exporter and clone the project or download it to your local machine.


2. Navigate to the Template Folder: The code used to deploy the DCGM Exporter is located in the template folder within the deployment folder.


3. Modify the service-monitor.yaml File: Find the file service-monitor.yaml. The apiVersion key in this file needs to be updated from monitoring.coreos.com/v1 to azmonitoring.coreos.com/v1. This change allows the DCGM Exporter to use the Azure managed Prometheus CRD.

apiVersion: azmonitoring.coreos.com/v1

4. Handle Node Selectors and Tolerations: GPU node pools often have tolerations and node selector tags. Modify the values.yaml file in the deployment folder to handle these configurations:

nodeSelector:
  accelerator: nvidia

tolerations:
- key: "sku"
  operator: "Equal"
  value: "gpu"
  effect: "NoSchedule"

Helm: Packaging, Pushing, and Installation on Azure Container Registry

We followed the MS Learn documentation for pushing and installing the package through Helm on Azure Container Registry. For a comprehensive understanding, you can refer to the documentation. Here are the quick steps for installation:

After making all the necessary changes in the deployment folder on the source code, be on that directory to package the code. Log in to your registry to proceed further.

1. Package the Helm chart and login to your container registry:

helm package .
helm registry login  --username $USER_NAME --password $PASSWORD

2. Push the Helm Chart to the Registry:

helm push dcgm-exporter-3.4.2.tgz oci:///helm

3. Verify that the package has been pushed to the registry on Azure portal.

4. Install the chart and verify the installation:

helm install dcgm-nvidia oci:///helm/dcgm-exporter -n gpu-resources
#Check the installation on your AKS cluster by running:
helm list -n gpu-resources
#Verify the DGCM Exporter:
Kubectl get po -n gpu-resources
Kubectl get ds -n gpu-resources

You can now check that the DGCM Exporter is running on the GPU nodes as a DaemonSet.

Exporting GPU Metrics and Configuring Azure Managed Grafana Dashboard

Once the DGCM Exporter DaemonSet is running across all GPU node pools, you need to export the GPU metrics generated by this workload to Azure Managed Prometheus. This is accomplished by deploying a PodMonitor resource. Follow these steps:

1. Deploy the PodMonitor: Apply the following YAML configuration to deploy the PodMonitor:

apiVersion: azmonitoring.coreos.com/v1
kind: PodMonitor
metadata:
  name: nvidia-dcgm-exporter
  labels:
    app.kubernetes.io/name: nvidia-dcgm-exporter
spec:
  selector:
    matchLabels:
      app.kubernetes.io/name: nvidia-dcgm-exporter
  podMetricsEndpoints:
  - port: metrics
    interval: 30s
  podTargetLabels:

2. Check if the PodMonitor is deployed and running by executing:

kubectl get podmonitor -n 

3. Verify Metrics export: Ensure that the metrics are being exported to Azure Managed Prometheus on the portal by navigating to the “Metrics” page on your Azure Monitor Workspace.

Create the DGCM Dashboard on Azure Managed Grafana

The GitHub repository for the DGCM Exporter includes a JSON file for the Grafana dashboard. Follow the MS Learn documentation to import this JSON into your Managed Grafana instance.

After importing the JSON, the dashboard displaying GPU metrics will be visible on Grafana.

Azure Infra Girls LATAM: Certificación AZ-900 (Azure Fundamentals)

by Contributed | Aug 7, 2024 | Technology

This article is contributed. See the original author and article here.

La computación en la nube está revolucionando diversas áreas de la tecnología, incluyendo programación, datos, inteligencia artificial (IA) y seguridad. Para ayudar a los profesionales a especializarse en esta área en constante evolución, Microsoft está lanzando la iniciativa Azure Infra Girls. Este programa ofrece una serie de cuatro clases en vivo, gratuitas y en español, que se llevarán a cabo del 3 al 24 de septiembre, a las 12:30pm (GMT-6, Ciudad de México).

 

REGÍSTRATE AQUÍ: aka.ms/AzureInfraGirlsLATAM

 

Durante estas sesiones, los participantes tendrán la oportunidad de profundizar sus conocimientos en computación en la nube y prepararse para la certificación AZ-900 (Azure Fundamentals), a través de una ruta de aprendizaje con cursos certificados en Microsoft Learn.

 

Todas nuestras sesiones comenzaran en base a la zona horaria de Ciudad de México. 

 

Sesión	Descripción
Conceptos de computación en la nube con Microsoft Azure  Sept 3, 12:30PM Mexico City (GMT-6)	En el primer episodio de Azure Infra Girls, aprenderás sobre los conceptos básicos de la programación en la nube. Comprenderás qué son las nubes públicas, privadas e híbridas, los beneficios y los tipos de servicios, como IaaS, PaaS, SaaS, Serverless, y cómo usar Azure para desarrollar tus aplicaciones.
Arquitectura y servicios de Azure  Sept 11, 12:30PM Mexico City (GMT-6)	En la segunda sesión, profundizaremos en los conceptos de arquitectura y servicios de Azure, con algunos ejercicios prácticos para crear un Escritorio Virtual (Azure Virtual Desktop) y alojar un recurso en Azure.
Administración y gobernanza de Azure  Sept 17, 12:30PM Mexico City (GMT-6)	En la tercera sesión, abordaremos los servicios de administración y gobernanza de Azure, analizando el control de costos, las funcionalidades y herramientas de gobernanza, el cumplimiento y la supervisión.
Simulacion del examen Azure Fundamentals AZ-900   Sept 24, 12:30PM Mexico City (GMT-6)	En esta sesión, realizaremos un simulacro del examen con preguntas relacionadas con los temas. Estarán relacionados, sobre conceptos de nube, arquitectura y servicios de Azure. Asimismo, con administración y gobernanza.

Para aquellos que desean profundizar en la computación en la nube y prepararse para la certificación AZ-900 Azure Fundamentals, tenemos una ruta de aprendizaje completa disponible en Microsoft Learn. Esta ruta aborda las principales áreas de la certificación, permitiéndote estudiar de forma gratuita y a tu propio ritmo. Además, al completar los cursos, podrás obtener certificados que pueden ser añadidos a tu perfil de LinkedIn, destacando tus nuevas habilidades y conocimientos.

Los módulos de Microsoft Learn para la certificación AZ-900 incluyen:

• Descripción de los conceptos de la nube (25-30 %)


• Descripción de la arquitectura y los servicios de Azure (35-40 %)


• Descripción de la administración y la gobernanza de Azure (30-35 %)

Práctica para el examen:

• Evaluación de sus conocimientos: estas evaluaciones le proporcionarán una infomacion general del estilo, la redacción y la dificultad de las preguntas que probablemente verá en el examen. A través de estas valoraciones, puede evaluar su preparación, determinar dónde necesita preparación adicional y llenar los vacíos de conocimiento para aumentar la probabilidad de aprobar el examen.


• Demostración de experiencia: aquí puede experimentar con el aspecto del examen antes de realizarlo. Podrá interactuar con diferentes tipos de preguntas en la misma interfaz de usuario que usará durante el examen.

Learn about AppJetty’s ISV Success for Business Applications solution in Microsoft AppSource

by Contributed | Aug 7, 2024 | Technology

This article is contributed. See the original author and article here.

Microsoft ISV Success for Business Applications offers platforms, resources, and support designed to help partners develop, publish, and market business apps. Learn more about this offer from AppJetty:

MappyField 365: MappyField 365 is a powerful geo-mapping plugin for Microsoft Dynamics 365 that boosts business productivity with advanced features like live tracking, geographic data visualization, proximity search, auto-scheduling, auto check-ins, territory management, and heat maps. Accelerate your business across organizations with location intelligence from AppJetty.