spark-mcp/internal/analyzer/rules.go

package analyzer

import "fmt"

// Analyze runs all heuristic rules and returns the triggered findings.
func Analyze(in Input, t Thresholds) []Finding {
	var out []Finding
	out = append(out, checkDataSkew(in, t)...)
	out = append(out, checkGCPressure(in, t)...)
	out = append(out, checkBottleneck(in, t)...)
	return out
}

func checkDataSkew(in Input, t Thresholds) []Finding {
	var out []Finding
	for stage, m := range in.StageMetrics {
		if m.MinPartitionBytes <= 0 || m.MaxPartitionBytes <= 0 {
			continue
		}
		ratio := float64(m.MaxPartitionBytes) / float64(m.MinPartitionBytes)
		if ratio > t.DataSkewRatio {
			sev := SeverityWarning
			if ratio > t.DataSkewRatio*2 {
				sev = SeverityCritical
			}
			out = append(out, Finding{
				Rule:     "data_skew",
				Severity: sev,
				Stage:    stage,
				Evidence: map[string]any{
					"max_partition_bytes": m.MaxPartitionBytes,
					"min_partition_bytes": m.MinPartitionBytes,
					"ratio":               ratio,
					"threshold":           t.DataSkewRatio,
				},
				Message: fmt.Sprintf("stage %q 数据倾斜: max/min = %.1fx (阈值 %.1fx)", stage, ratio, t.DataSkewRatio),
			})
		}
	}
	return out
}

func checkGCPressure(in Input, t Thresholds) []Finding {
	var out []Finding
	var totalGC, totalCPU int64
	var worstExec string
	var worstRatio float64
	for id, m := range in.ExecutorMetrics {
		if m.CPUTimeMS <= 0 {
			continue
		}
		r := float64(m.GCTimeMS) / float64(m.CPUTimeMS)
		totalGC += m.GCTimeMS
		totalCPU += m.CPUTimeMS
		if r > worstRatio {
			worstRatio = r
			worstExec = id
		}
	}
	if totalCPU > 0 {
		aggRatio := float64(totalGC) / float64(totalCPU)
		if aggRatio > t.GCPressureRatio {
			sev := SeverityWarning
			if aggRatio > t.GCPressureRatio*2 {
				sev = SeverityCritical
			}
			out = append(out, Finding{
				Rule:     "gc_pressure",
				Severity: sev,
				Evidence: map[string]any{
					"aggregate_gc_ratio": aggRatio,
					"worst_executor":     worstExec,
					"worst_ratio":        worstRatio,
					"threshold":          t.GCPressureRatio,
				},
				Message: fmt.Sprintf("GC 压力过高: 聚合 GC/CPU 比率 = %.1f%% (阈值 %.1f%%, 最差 executor %s = %.1f%%)",
					aggRatio*100, t.GCPressureRatio*100, worstExec, worstRatio*100),
			})
		}
	}
	return out
}

func checkBottleneck(in Input, t Thresholds) []Finding {
	var out []Finding
	thresholdBytes := int64(t.BottleneckShuffleGB * (1 << 30))
	for stage, m := range in.StageMetrics {
		total := m.ShuffleReadBytes + m.ShuffleWriteBytes
		if total < 0 {
			continue // overflow guard
		}
		if total > thresholdBytes {
			sev := SeverityInfo
			if total > thresholdBytes*2 {
				sev = SeverityWarning
			}
			if total > thresholdBytes*5 {
				sev = SeverityCritical
			}
			out = append(out, Finding{
				Rule:     "bottleneck",
				Severity: sev,
				Stage:    stage,
				Evidence: map[string]any{
					"shuffle_read_bytes":  m.ShuffleReadBytes,
					"shuffle_write_bytes": m.ShuffleWriteBytes,
					"shuffle_total_gb":    float64(total) / (1 << 30),
					"threshold_gb":        t.BottleneckShuffleGB,
				},
				Message: fmt.Sprintf("stage %q 是潜在瓶颈: shuffle 总 %.1f GB (阈值 %.1f GB)",
					stage, float64(total)/(1<<30), t.BottleneckShuffleGB),
			})
		}
	}
	return out
}