| 1 | package proxy |
| 2 | |
| 3 | import ( |
| 4 | "fmt" |
| 5 | "strings" |
| 6 | "sync" |
| 7 | "time" |
| 8 | ) |
| 9 | |
| 10 | // CallRecord holds metrics for a single proxy call. |
| 11 | type CallRecord struct { |
| 12 | Provider string |
| 13 | UpstreamDomain string |
| 14 | Endpoint string |
| 15 | Model string |
| 16 | StatusCode int |
| 17 | Stream bool |
| 18 | Duration time.Duration |
| 19 | InputTokens int |
| 20 | OutputTokens int |
| 21 | CacheCreation int |
| 22 | CacheRead int |
| 23 | Error bool |
| 24 | QuotaProbe bool |
| 25 | Timestamp time.Time |
| 26 | } |
| 27 | |
| 28 | // Metrics collects proxy call records for the session. |
| 29 | type Metrics struct { |
| 30 | mu sync.Mutex |
| 31 | records []CallRecord |
| 32 | } |
| 33 | |
| 34 | // DefaultMetrics is the package-level metrics instance. |
| 35 | var DefaultMetrics = &Metrics{} |
| 36 | |
| 37 | // Record appends a call record. |
| 38 | func (m *Metrics) Record(r CallRecord) { |
| 39 | if r.Timestamp.IsZero() { |
| 40 | r.Timestamp = time.Now() |
| 41 | } |
| 42 | m.mu.Lock() |
| 43 | m.records = append(m.records, r) |
| 44 | m.mu.Unlock() |
| 45 | } |
| 46 | |
| 47 | // Summary returns a formatted metrics summary. Returns empty string if no calls recorded. |
| 48 | func (m *Metrics) Summary() string { |
| 49 | m.mu.Lock() |
| 50 | records := make([]CallRecord, len(m.records)) |
| 51 | copy(records, m.records) |
| 52 | m.mu.Unlock() |
| 53 | |
| 54 | if len(records) == 0 { |
| 55 | return "" |
| 56 | } |
| 57 | |
| 58 | var ( |
| 59 | total int |
| 60 | errors int |
| 61 | quotaProbes int |
| 62 | totalIn int |
| 63 | totalOut int |
| 64 | totalCacheCre int |
| 65 | totalCacheRead int |
| 66 | totalDur time.Duration |
| 67 | |
| 68 | byModel = make(map[string]*modelStats) |
| 69 | byStatus = make(map[int]int) |
| 70 | byDomain = make(map[string]int) |
| 71 | byEndpoint = make(map[string]int) |
| 72 | ) |
| 73 | |
| 74 | for _, r := range records { |
| 75 | total++ |
| 76 | totalIn += r.InputTokens |
| 77 | totalOut += r.OutputTokens |
| 78 | totalCacheCre += r.CacheCreation |
| 79 | totalCacheRead += r.CacheRead |
| 80 | totalDur += r.Duration |
| 81 | |
| 82 | if r.Error { |
| 83 | errors++ |
| 84 | } |
| 85 | if r.QuotaProbe { |
| 86 | quotaProbes++ |
| 87 | } |
| 88 | |
| 89 | byStatus[r.StatusCode]++ |
| 90 | byDomain[r.UpstreamDomain]++ |
| 91 | byEndpoint[r.Endpoint]++ |
| 92 | |
| 93 | ms, ok := byModel[r.Model] |
| 94 | if !ok { |
| 95 | ms = &modelStats{} |
| 96 | byModel[r.Model] = ms |
| 97 | } |
| 98 | ms.count++ |
| 99 | ms.inputTokens += r.InputTokens |
| 100 | ms.outputTokens += r.OutputTokens |
| 101 | ms.cacheCreation += r.CacheCreation |
| 102 | ms.cacheRead += r.CacheRead |
| 103 | ms.totalDur += r.Duration |
| 104 | } |
| 105 | |
| 106 | var b strings.Builder |
| 107 | b.WriteString("\n") |
| 108 | b.WriteString(strings.Repeat("─", 60) + "\n") |
| 109 | b.WriteString(" Proxy Metrics Summary\n") |
| 110 | b.WriteString(strings.Repeat("─", 60) + "\n") |
| 111 | |
| 112 | fmt.Fprintf(&b, " Total requests: %d\n", total) |
| 113 | fmt.Fprintf(&b, " Errors: %d\n", errors) |
| 114 | fmt.Fprintf(&b, " Quota probes: %d\n", quotaProbes) |
| 115 | fmt.Fprintf(&b, " Total duration: %s\n", totalDur.Round(time.Millisecond)) |
| 116 | fmt.Fprintf(&b, " Total tokens: %d in / %d out\n", totalIn, totalOut) |
| 117 | if totalCacheCre > 0 || totalCacheRead > 0 { |
| 118 | fmt.Fprintf(&b, " Cache tokens: %d created / %d read\n", totalCacheCre, totalCacheRead) |
| 119 | } |
| 120 | |
| 121 | if len(byModel) > 0 { |
| 122 | b.WriteString("\n By Model:\n") |
| 123 | for model, ms := range byModel { |
| 124 | avg := time.Duration(0) |
| 125 | if ms.count > 0 { |
| 126 | avg = ms.totalDur / time.Duration(ms.count) |
| 127 | } |
| 128 | fmt.Fprintf(&b, " %-30s %3d req, %6d in / %6d out, avg %s\n", |
| 129 | model, ms.count, ms.inputTokens, ms.outputTokens, avg.Round(time.Millisecond)) |
| 130 | if ms.cacheCreation > 0 || ms.cacheRead > 0 { |
| 131 | fmt.Fprintf(&b, " %-30s cache: %d created / %d read\n", |
| 132 | "", ms.cacheCreation, ms.cacheRead) |
| 133 | } |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | if len(byStatus) > 0 { |
| 138 | b.WriteString("\n By Status Code:\n") |
| 139 | for code, count := range byStatus { |
| 140 | label := fmt.Sprintf("%d", code) |
| 141 | if code == 0 { |
| 142 | label = "N/A (error/intercepted)" |
| 143 | } |
| 144 | fmt.Fprintf(&b, " %-25s %d\n", label, count) |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | if len(byDomain) > 0 { |
| 149 | b.WriteString("\n By Upstream Domain:\n") |
| 150 | for domain, count := range byDomain { |
| 151 | if domain == "" { |
| 152 | domain = "(local)" |
| 153 | } |
| 154 | fmt.Fprintf(&b, " %-35s %d\n", domain, count) |
| 155 | } |
| 156 | } |
| 157 | |
| 158 | if len(byEndpoint) > 0 { |
| 159 | b.WriteString("\n By Endpoint:\n") |
| 160 | for ep, count := range byEndpoint { |
| 161 | fmt.Fprintf(&b, " %-35s %d\n", ep, count) |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | b.WriteString(strings.Repeat("─", 60) + "\n") |
| 166 | return b.String() |
| 167 | } |
| 168 | |
| 169 | type modelStats struct { |
| 170 | count int |
| 171 | inputTokens int |
| 172 | outputTokens int |
| 173 | cacheCreation int |
| 174 | cacheRead int |
| 175 | totalDur time.Duration |
| 176 | } |
| 177 | |