Resource Recommendations

The Resource Recommender feature in Agentkube IDE provides intelligent CPU and memory resource recommendations for your Kubernetes workloads. This built-in feature analyzes historical usage patterns to suggest optimal resource requests and limits, helping you right-size your containers and improve cluster efficiency. By analyzing usage data collected from Prometheus metrics, the Resource Recommender ensures your resource allocations match actual workload requirements, preventing both over-provisioning and resource contention.

​

How to Use

​

Prerequisites

To use the Resource Recommender feature, ensure:

• Prometheus is integrated with Agentkube IDE
• Prometheus is scraping cAdvisor metrics for container resource usage
• kube-state-metrics is deployed for Kubernetes object metadata
• Metric retention period covers at least 14 days for optimal recommendations
• Scrape interval is configured appropriately (typically 15-30 seconds)

To access resource recommendations in Agentkube IDE:

1. Navigate to the Workloads section
2. Select Pods from the workload types
3. Click on the Action button for the desired pod
4. Select Telemetry from the dropdown menu
5. Navigate to the Recommendation tab
6. View CPU and memory recommendations based on historical data

​

Recommendation Strategy

Agentkube implements an intelligent recommendation strategy that balances resource efficiency with workload stability.

​

CPU Recommendations

Request Calculation:

• Based on the 95th percentile of historical CPU usage
• Allows pods to burst above the request when needed
• No CPU limit is set by default to permit bursting

Required Metric: container_cpu_usage_seconds_total

​

Memory Recommendations

Request/Limit Calculation:

• Based on peak memory usage over the historical period
• Adds a 15% buffer to prevent OOM kills
• Sets both request and limit to the same value for predictability

Required Metric: container_memory_working_set_bytes

​

Data Requirements

The Resource Recommender requires 100+ historical datapoints to generate reliable recommendations. By default, it analyzes 14 days of historical data to capture regular usage patterns, peak load scenarios, and daily/weekly cycles.

​

Metrics Used

The Resource Recommender leverages the following Prometheus metrics for analysis:

​

Core Metrics

​

Supporting Metrics

​

OOM Kill Detection

When Out-Of-Memory (OOM) kill events are detected, the Resource Recommender automatically adjusts recommendations:

• Applies a 25% memory buffer (instead of 15%)
• Considers historical memory limits that were exceeded
• Helps prevent future OOM kills

Required Metrics:

• kube_pod_container_resource_limits
• kube_pod_container_status_last_terminated_reason