oxirs.toml

# OxiRS Configuration Example

[server]
host = "0.0.0.0"
port = 3030
admin_ui = true
cors = true

[datasets.default]
name = "default"
location = "./data/default"
read_only = false
shacl_shapes = []
services = []

[datasets.example]
name = "example"
location = "./data/example"
read_only = true
shacl_shapes = []
services = []

[security]
auth_required = false

[security.users.admin]
password_hash = "$2b$12$..."
roles = ["admin", "write", "read"]

[security.users.reader]
password_hash = "$2b$12$..."
roles = ["read"]

[features]
# Optional features
text_search = false
vector_search = false
rdf_star = false
clustering = false
streaming = false

[logging]
level = "info"
format = "json"
output = "stdout"

[performance]
cache_size = 104857600  # 100MB
query_timeout = 30000   # 30 seconds
max_concurrent_queries = 100

# Advanced Query Optimizer Configuration
[optimizer]
enabled = true
cache_size = 10000
parallel_optimization = true

# Machine Learning-based Cost Prediction
# Replaces heuristic-only estimation with learned models for up to 1.75x speedup
[optimizer.ml]
enabled = true

# Model type: "linear", "ridge", "gradient_boosting"
# ridge: L2 regularization (recommended for most cases)
# linear: Simple linear regression (faster, less accurate)
model_type = "ridge"

# Confidence threshold (0.0-1.0) for using ML predictions
# If model confidence < threshold, falls back to heuristic estimation
confidence_threshold = 0.7

# Retraining interval in hours
# Model will retrain periodically using accumulated execution feedback
training_interval_hours = 24

# Maximum training examples to keep in memory
# Older examples are evicted using sliding window strategy
max_training_examples = 10000

# Minimum examples required before initial training
# Lower values train faster but may be less accurate
min_examples_for_training = 100

# Feature normalization (z-score)
# Improves model accuracy for queries with varying scales
feature_normalization = true

# Automatic retraining when interval is reached
auto_retraining = true

# Paths for model persistence (relative to data directory)
model_persistence_path = "./data/ml_models/query_cost_predictor.json"
training_data_path = "./data/ml_training/training_data.json"

# Model Quality Thresholds
[optimizer.ml.quality]
# Minimum R² (coefficient of determination) score
# R² > 0.8 indicates excellent predictive accuracy
min_r_squared = 0.8

# Maximum Mean Absolute Error as percentage
# MAE < 20% means predictions are within 20% of actual costs on average
max_mae_percent = 20.0

# Maximum Root Mean Square Error as percentage
# RMSE < 30% indicates good overall prediction quality
max_rmse_percent = 30.0

# Enable automatic rollback if new model performs worse
# Protects against quality degradation during retraining
enable_rollback = true

# Distributed Cache Configuration (Phase 5.2)
# Redis-based L1+L2 cache with coherence protocol for multi-node clusters
[cache.distributed]
enabled = false  # Cluster only (enable for multi-node deployments)

# L1 Cache (Local In-Memory)
l1_max_size = 1000          # Maximum entries in local cache
l1_ttl_seconds = 300        # Time-to-live: 5 minutes

# L2 Cache (Redis)
l2_redis_url = "redis://localhost:6379"  # Redis connection URL
l2_ttl_seconds = 3600       # Time-to-live: 1 hour

# Compression for large values (>1KB)
compression = true          # Enables gzip compression

# Pub/Sub channel for cache invalidation
invalidation_channel = "oxirs:cache:invalidate"

# Cache Coherence Protocol
[cache.coherence]
# Protocol type: "pubsub" (eventual), "write-through" (strong), "write-behind" (eventual)
protocol = "pubsub"

# Consistency level: "eventual", "strong", "bounded-staleness"
consistency_level = "eventual"

# Maximum staleness in seconds (for bounded-staleness)
max_staleness_seconds = 60