AI AnalysisΒΆ
Robusta can integrate with Holmes GPT to analyze health issues on your cluster, and to run AI based root cause analysis for alerts.
Holmes can be installed with Robusta by adding enableHolmesGPT: true
to the Robusta generated_values.yaml
file.
When available, AI based investigations can be launched using the Ask Holmes
button in Slack. The results will be sent back as a new message.
![](../_images/robusta-holmes-investigation1.png)
On the Robusta UI, clicking the Find Root Cause
button will start an investigation and display its results.
![](../_images/ai-root-causeanalysis1.png)
ConfigurationΒΆ
Warning
Only GPT-4o is officially supported. We highly recommend using GPT-4o to get the most accurate results!
In order to include Holmes GPT
with you Robusta installation, add the following to your generated_values.yaml
In the examples below, we're assuming you created a Kubernetes secret
named holmes-secrets
to store sensitive variables.
To use Open AI (this is the default llm):
enableHolmesGPT: true
holmes:
additionalEnvVars:
- name: MODEL
value: gpt-4o
- name: OPENAI_API_KEY
valueFrom:
secretKeyRef:
name: holmes-secrets
key: openAiKey
To use Azure Open AI:
enableHolmesGPT: true
holmes:
additionalEnvVars:
- name: MODEL
value: azure/my-azure-deployment # the name of your azure deployment
- name: AZURE_API_VERSION
value: 2024-02-15-preview
- name: AZURE_API_BASE
value: https://my-org.openai.azure.com/ # base url of you azure deployment
- name: AZURE_API_KEY
valueFrom:
secretKeyRef:
name: holmes-secrets
key: azureOpenAiKey
To use Azure AI, follow the setup instructions below and then edit Robusta's Helm values. Do NOT skip the setup instructions, as they include a mandatory change rate-limits. Without this change, Holmes wont work.
Mandatory Setup for Azure AI
The following steps cover how to obtain the correct AZURE_API_VERSION value and how to increase the token limit to prevent rate limiting.
Go to your Azure portal and choose Azure OpenAI
Click your AI service
Click Go to Azure Open AI Studio
Choose Deployments
Select your Deployment
Click Open in Playground
Go to View Code
Choose Python and scroll to find the API VERSION. Copy this! You will need it for Robusta's Helm values.
Go back to Deployments, and click Edit Deployment
MANDATORY: Increase the token limit. Change this value to at least 450K tokens for Holmes to work properly. We recommend choosing the highest value available. (Holmes queries Azure AI infrequently but in bursts. Therefore the overall cost of using Holmes with Azure AI is very low, but you must increase the quota to avoid getting rate-limited on a single burst of requests.)
To use AWS Bedrock:
enableHolmesGPT: true
holmes:
enablePostProcessing: true
additionalEnvVars:
- name: MODEL
value: bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0 # your bedrock model
- name: AWS_REGION_NAME
value: us-east-1
- name: AWS_ACCESS_KEY_ID
valueFrom:
secretKeyRef:
name: holmes-secrets
key: awsAccessKeyId
- name: AWS_SECRET_ACCESS_KEY
valueFrom:
secretKeyRef:
name: holmes-secrets
key: awsSecretAccessKey