Documentation
¶
Overview ¶
Package diagnose implements the rule-based half of FR-MCP-05's diagnose_failure / vegaload diagnose: a set of fixed heuristics over a report.Result that flag the failure patterns a human reviewing a load test would look for first, plus a pure-arithmetic threshold suggester. It knows nothing about the CLI, MCP, or any LLM — internal/llm's narrative connector is a separate, optional layer that takes these Findings as input (see cmd/vegaload/diagnose.go for how the two combine).
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Bucket ¶
type Bucket struct {
OffsetSeconds float64 `json:"offset_seconds"`
ErrorRate float64 `json:"error_rate"`
Requests int64 `json:"requests"`
Failed int64 `json:"failed"`
}
Bucket identifies the single worst time-series point by error rate, letting a caller point at *when* a run went bad rather than just that it did.
type Findings ¶
type Findings struct {
ZeroIterations bool `json:"zero_iterations"`
FailurePercent float64 `json:"failure_percent"`
HighFailureRate bool `json:"high_failure_rate"` // > 50%
LatencyLongTail bool `json:"latency_long_tail"` // p99 > 5x p50
WorstBucket *Bucket `json:"worst_bucket,omitempty"`
Transient bool `json:"transient,omitempty"` // failures concentrated in one bucket vs. spread out
Notes []string `json:"notes"`
}
Findings is the rule-based diagnosis of a single report.Result: a set of independently-triggered observations, each already phrased as a sentence a report or MCP tool response can show verbatim.
type Thresholds ¶
type Thresholds struct {
LatencyP95 string `json:"latency_p95"`
ErrorRate float64 `json:"error_rate"`
}
Thresholds is the output of SuggestThresholds: pass/fail criteria a future run of the same scenario could be gated on, derived from this run's own observed performance.
func SuggestThresholds ¶
func SuggestThresholds(res *report.Result) Thresholds
SuggestThresholds derives conservative pass/fail thresholds from a clean baseline run: p95 latency headroom'd by 20% and rounded to the nearest 10ms (so the suggestion reads as a round number, not a spurious-precision artifact of one run), and an error-rate ceiling of 1.5x what was observed, floored at 1% so a zero-failure baseline still yields a usable, nonzero threshold.