Skip to content
This repository has been archived by the owner on Dec 2, 2021. It is now read-only.

Latest commit

 

History

History
91 lines (55 loc) · 4.94 KB

MonitorAzureResources.md

File metadata and controls

91 lines (55 loc) · 4.94 KB

Azure Resources and Azure Machine Learning Observability

You can collect logs and visualize Azure Resources for observability. In this documentation, we illustrate which option you can use to achieve project goal.

Observability Architecture

Even though each team may have different access permission to each Azure resource and observability requirements, you can use same technology to collect logs, then customize the visualization and alerts for each team.

Observability-choice

Customer may have their own tools for observabilty. In such case, you can consider log integration between Azure Monitor and their system.


Azure Machine Learning project group types

Typically, there are at least four types of groups in Machine Learning projects. Each team have different requirements for observability.

  • Software Engineer
  • Data Scientist
  • Quality Assuarance
  • Business Analyst

Software Engineer

SE typically works on entire sysetm and they will monitor application status using Azure dashboard, with alert.

Data Scientist

Data Scientists uses Azure Machine Learning to build and publish models. They should have full access to Azure Machine Learning workspace with detail logs and charts. In addition, they will monitor Model Training and Evaluation results using Azure Dashboards, with alert.

Quality Assurance

Quality Assurance team may run the model performance test built by Data Scientist team by using different set of test dataset. You may want to restrict their access to Azure Machine Learning service so that they cannot see the code to build the model, as they just need to run the model accuracy tests.

  • Not see the details on Azure DevOps and Azure ML
  • Run a performance experiment pipeline
  • See experiment results on Azure Dashboard
  • Receive notifications when experiment is started and ended

Business Analyst

Business analyst will monitor Model Performance & usage monitor by using Azure Dashboard using data sourced from Azure Monitor

Observability Technical Resource

Azure Monitor

Azure Application Insights

System Metrics

Azure Dashboard

Azure Alerts

Prometheus Exposition format

AzureML Training pipeline metric

AzureML Run context logging