Load Detection Rules

Load detection rules live in automated-rules/load-level/ and define how cluster load is classified as: - low - medium - high

This load level is then used by normal rules that define load_level_thresholds.

Example

rule_type = "load_detection"
display_name = "cluster_volume"

[[metrics]]
name = "containers"
source = "rox_sensor_num_containers_in_clusterentities_store"
weight = 1.0

[[metrics]]
name = "pods"
source = "rox_sensor_num_pods_in_store"
weight = 0.5

[[thresholds]]
level = "low"
max_value = 100

[[thresholds]]
level = "medium"
min_value = 100
max_value = 500

[[thresholds]]
level = "high"
min_value = 500

Worked Example (Step by Step)

Assume the current metrics are:

rox_sensor_num_containers_in_clusterentities_store 240
rox_sensor_num_pods_in_store 120

Using the weights from the rule:

• containers weight = 1.0
• pods weight = 0.5

Now compute:

1. weighted_sum = (240 * 1.0) + (120 * 0.5) = 240 + 60 = 300
2. total_weight = 1.0 + 0.5 = 1.5
3. normalized_value = weighted_sum / total_weight = 300 / 1.5 = 200

Compare normalized_value = 200 to thresholds:

• low: < 100
• medium: >= 100 and < 500
• high: >= 500

Result: medium load level.

How Score Calculation Works

The detector computes:

1. weighted_sum = Σ(metric_value * weight)
2. total_weight = Σ(weight)
3. normalized_value = weighted_sum / total_weight
4. Match normalized_value against threshold ranges

Practical Guidance

• Start with 1-3 metrics that represent cluster size/pressure.
• Use larger weights for more trusted indicators.
• Keep thresholds simple first, then tune from production observations.
• Re-test normal rules after load detection changes, because rule statuses may shift with new load levels.