detectk-collectors-clickhouse 0.1.0

ClickHouse collector and storage for DetectK

detectk-collectors-clickhouse

ClickHouse collector and storage for DetectK.

Installation

pip install detectk-collectors-clickhouse

Features

• ClickHouseCollector: Collect metrics from ClickHouse queries
• ClickHouseStorage: Store metric history in ClickHouse (dtk_datapoints and dtk_detections tables)
• Auto-registration in DetectK registries
• Connection pooling and error handling
• Partitioned tables for performance

Usage

As Collector

# config.yaml
name: "sessions_10min"

collector:
  type: "clickhouse"
  params:
    host: "localhost"
    database: "analytics"
    query: |
      SELECT
        count() as value,
        now() as timestamp
      FROM sessions
      WHERE timestamp > now() - INTERVAL 10 MINUTE

detector:
  type: "threshold"
  params:
    threshold: 1000

alerter:
  type: "mattermost"
  params:
    webhook_url: "${MATTERMOST_WEBHOOK}"

As Storage

# config.yaml
storage:
  enabled: true
  type: "clickhouse"
  params:
    host: "localhost"
    database: "default"
    save_detections: false  # Optional: save detection results

Multiple Detectors (A/B Testing)

# config.yaml - Compare multiple detection strategies
name: "sessions_10min"

collector:
  type: "clickhouse"
  params:
    host: "localhost"
    database: "analytics"
    query: "SELECT count() as value, now() as timestamp FROM sessions"

# Multiple detectors with auto-generated IDs
detectors:
  - type: "mad"
    params:
      window_size: "30 days"
      n_sigma: 3.0
    # ID auto-generated: e.g., "a1b2c3d4"

  - type: "mad"
    params:
      window_size: "30 days"
      n_sigma: 5.0
    # ID auto-generated: e.g., "b2c3d4e5" (different from above)

  - id: "zscore_7d"  # Manual ID override
    type: "zscore"
    params:
      window_size: "7 days"

alerter:
  type: "mattermost"
  params:
    webhook_url: "${MATTERMOST_WEBHOOK}"

storage:
  enabled: true
  type: "clickhouse"
  params:
    host: "localhost"
    database: "default"
    save_detections: true  # Save all detector results for comparison

How it works:

• Each detector gets a unique ID (auto-generated 8-char hash or manual)
• All detector results are saved to dtk_detections with their detector_id
• Alert sent if ANY detector finds anomaly (configurable in future)
• Query results: SELECT * FROM dtk_detections WHERE metric_name = 'sessions_10min' ORDER BY detected_at, detector_id

Configuration

Collector Parameters

• host: ClickHouse server host (default: localhost)
• port: ClickHouse server port (default: 9000)
• database: Database name (default: default)
• user: Username (optional)
• password: Password (optional)
• query: SQL query returning value and optionally timestamp columns
• timeout: Query timeout in seconds (default: 30)
• secure: Use SSL connection (default: false)

Storage Parameters

• host: ClickHouse server host (default: localhost)
• port: ClickHouse server port (default: 9000)
• database: Database name (default: default)
• user: Username (optional)
• password: Password (optional)
• timeout: Query timeout in seconds (default: 30)
• secure: Use SSL connection (default: false)
• save_detections: Save detection results to dtk_detections table (default: false)

Storage Schema

dtk_datapoints

Collected metric values (required for detection):

CREATE TABLE dtk_datapoints (
    id UInt64,
    metric_name String,
    collected_at DateTime64(3),
    value Float64,
    context String  -- JSON string
) ENGINE = MergeTree()
PARTITION BY toYYYYMM(collected_at)
ORDER BY (metric_name, collected_at);

dtk_detections

Detection results (optional, for audit/cooldown):

CREATE TABLE dtk_detections (
    id UInt64,
    metric_name String,
    detector_id String,  -- Unique detector identifier (for multi-detector support)
    detected_at DateTime64(3),
    value Float64,
    is_anomaly UInt8,
    anomaly_score Nullable(Float64),
    lower_bound Nullable(Float64),
    upper_bound Nullable(Float64),
    direction Nullable(String),
    percent_deviation Nullable(Float64),
    detector_type String,
    detector_params String,  -- JSON with full params for transparency
    alert_sent UInt8,
    alert_reason Nullable(String),
    alerter_type Nullable(String),
    context String  -- JSON
) ENGINE = MergeTree()
PARTITION BY toYYYYMM(detected_at)
ORDER BY (metric_name, detector_id, detected_at);

Multi-Detector Support: The detector_id field allows storing results from multiple detectors for the same metric. Each detector gets a unique ID (auto-generated or manual), enabling A/B testing and parameter tuning.

Tables are created automatically on first use.

License

MIT