Metrics¶

Metrics are the core data units used in the Guance system for continuously tracking system status. They consist of three parts: numerical values, timestamps, and dimensional tags. They record quantifiable system characteristics (such as resource utilization and business throughput) in time series format, enabling multi-dimensional analysis through tags (e.g., host, service, region). This provides precise data support for real-time monitoring, performance optimization, and trend prediction.

Data Architecture¶

The metric data processing in Guance is divided into three layers:

1. Collection Layer: Handled by DataKit (a lightweight agent deployed in the user's environment, analogous to Prometheus Exporters, directly interfacing with data sources, undertaking core responsibilities of collection, preprocessing, and secure transmission). It is responsible for capturing raw metrics from sources such as hosts, applications, and middleware.

2. Transmission Layer: DataKit encrypts the data and sends it to the Guance data center via HTTP/HTTPS.

3. Storage and Analysis Layer: Guance cleanses, stores the data, and provides visualization and analysis capabilities.

Metric Data Composition¶

A complete metric data unit contains three core elements:

For example:

cpu,host=server01,core=0 usage_user=12.3,usage_system=5.7 1690524000000000000

• Measurement: cpu

• Tags: host=server01, core=0 (marking the source server and CPU core)

• Fields: usage_user=12.3 (user space CPU usage), usage_system=5.7 (system space usage)

• Timestamp: 1690524000000000000 (2023-07-28 12:00:00 UTC)

Core Concepts¶

Measurement¶

A Measurement is a collection of similar metrics, containing multiple metrics and tags. For example, the cpu Measurement includes CPU-related metrics such as usage_user, usage_system.

Metric¶

A Metric reflects the operational state of a system (e.g., CPU usage, page load time). It consists of a metric name (identifier) and a metric value (specific numerical value).

Tag¶

Attributes of the data point collection object are identified through tags, presented as key-value pairs, consisting of a tag key and a tag value. A data point can carry multiple tags, used for data filtering and grouping.

Time Series¶

A Time Series is uniquely determined by the metric name + tag combination. Within the same time series, metric values (fields) are arranged in chronological order by timestamp to form a sequence.

The number of time series depends on the number of combinations of metric names and tags. Each additional distinct value for a tag multiplies the number of combinations. Excessive tag values can lead to time series inflation, directly impacting storage costs and query performance. It is recommended to reasonably control the use of high-cardinality tags (e.g., user ID, request path).

Field¶

A Field is a specific numerical metric, such as usage=58.3. Metric values change over time, forming time series data.

Data Visualization¶

Metric data is visualized in the following forms:

• Metric Analysis: Perform metric queries and analysis based on various query methods such as DQL and PromQL.
• Metric Management: Display measurements, metrics, tags, and time series statistics in list form.
• Visual Charts: Use metric data in Dashboards to create time series charts, bar charts, pie charts, etc.

Getting Started¶

Related Features¶