AgentSkillsCN

cost-optimized-log-trace-sampling

在保持可观测性的前提下，通过日志与追踪采样策略优化成本。主动启用以下场景：(1) 配置追踪采样；(2) 降低日志量；(3) 头部采样与尾部采样；(4) 基于优先级的采样；(5) 成本优化。触发指令包括：“采样”“追踪采样”“日志采样”“成本优化”“头部采样”“尾部采样”“采样率”。

SKILL.md
--- frontmatter
name: cost-optimized-log-trace-sampling
version: "1.0"
description: >
  Sampling strategies for logs and traces to optimize costs while maintaining observability.
  PROACTIVELY activate for: (1) Configuring trace sampling, (2) Log volume reduction,
  (3) Head-based vs tail-based sampling, (4) Priority-based sampling, (5) Cost optimization.
  Triggers: "sampling", "trace sampling", "log sampling", "cost optimization", "head-based", "tail-based", "sample rate"
core-integration:
  techniques:
    primary: ["systematic_analysis"]
    secondary: ["structured_evaluation"]
  contracts:
    input: "none"
    output: "none"
  patterns: "none"
  rubrics: "none"

Cost-Optimized Log and Trace Sampling

This skill provides expertise in sampling strategies for managing observability costs.

Overview

Observability data can be expensive at scale. Smart sampling strategies maintain visibility while controlling costs.

Sampling Strategies

Head-Based Sampling

Decision made at trace start:

  • •Probabilistic: Sample X% of traces randomly
  • •Rate-limiting: Sample up to N traces per second
yaml
# OpenTelemetry config
sampler:
  type: traceidratio
  ratio: 0.1  # Sample 10%

Tail-Based Sampling

Decision made after trace completes:

  • •Error-based: Always sample traces with errors
  • •Latency-based: Sample slow traces (> p99 latency)
  • •Attribute-based: Sample specific user IDs or endpoints

Hybrid Approaches

  1. •Sample 100% of errors
  2. •Sample 100% of slow requests
  3. •Sample 10% of successful, fast requests

Log Sampling Strategies

  1. •Dynamic log levels: Debug in dev, warn in prod
  2. •Sample verbose logs: Log 1% of debug statements
  3. •Aggregate before sending: Count similar events locally

Cost Optimization Tips

  1. •Set appropriate retention periods
  2. •Use log aggregation to reduce volume
  3. •Filter noise before export
  4. •Use tiered storage (hot/warm/cold)

[Content to be expanded based on plugin_spec_agentient-observability.md specifications]