---
apiVersion: v1
kind: ConfigMap
metadata:
name: tdg-config
namespace: tdg-system
labels:
app.kubernetes.io/name: tdg
app.kubernetes.io/component: configuration
data:
tdg-config.yaml: |
# TDG System Configuration
server:
host: 0.0.0.0
port: 8080
workers: 4
max_connections: 1000
request_timeout_seconds: 30
dashboard:
enabled: "true"
port: 8081
update_interval_seconds: 10
max_websocket_connections: 100
storage:
backend: sled
path: /data/tdg
cache_size_mb: 256
compression: "true"
hot_cache_ttl_seconds: 300
warm_tier_threshold_mb: 512
cold_tier_threshold_mb: 2048
metrics:
enabled: "true"
retention_hours: 168 # 1 week
aggregation_interval_seconds: 60
export_formats:
- prometheus
- json
alerts:
enabled: "true"
evaluation_interval_seconds: 10
max_active_alerts: 100
cooldown_period_seconds: 300
profiling:
enabled: "false" # Enable only when debugging
sample_interval_ms: 100
max_profiles_retained: 100
mcp:
enabled: "true"
max_concurrent_tools: 10
timeout_seconds: 60
logging:
level: info
format: json
output: stdout
alert-rules.yaml: |
# Default Alert Rules
rules:
- id: high_cpu
name: High CPU Usage
metric: cpu_usage_percent
condition: greater_than
threshold: 85.0
duration_seconds: 60
severity: warning
- id: critical_cpu
name: Critical CPU Usage
metric: cpu_usage_percent
condition: greater_than
threshold: 95.0
duration_seconds: 30
severity: critical
- id: high_memory
name: High Memory Usage
metric: memory_usage_mb
condition: greater_than
threshold: 7168.0 # 7GB
duration_seconds: 120
severity: warning
- id: critical_memory
name: Critical Memory Usage
metric: memory_usage_mb
condition: greater_than
threshold: 8192.0 # 8GB
duration_seconds: 60
severity: critical
- id: slow_analysis
name: Slow Analysis Performance
metric: avg_analysis_time_ms
condition: greater_than
threshold: 5000.0
duration_seconds: 300
severity: warning
- id: low_cache_hit
name: Low Cache Hit Ratio
metric: cache_hit_ratio
condition: less_than
threshold: 0.6
duration_seconds: 600
severity: info
- id: storage_full
name: Storage Near Capacity
metric: storage_usage_percent
condition: greater_than
threshold: 85.0
duration_seconds: 300
severity: warning