| 1 | package metrics |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "os" |
| 8 | "time" |
| 9 | |
| 10 | "github.com/carlosframework/platform/internal/release" |
| 11 | ) |
| 12 | |
| 13 | // Agg is a min/avg/max/count summary of one metric over some window. N |
| 14 | // counts only the samples that actually carried the metric (nil readings — |
| 15 | // a parser that couldn't read a value — are excluded, never treated as |
| 16 | // zero). |
| 17 | type Agg struct { |
| 18 | Min float64 `json:"min"` |
| 19 | Avg float64 `json:"avg"` |
| 20 | Max float64 `json:"max"` |
| 21 | N int `json:"n"` |
| 22 | } |
| 23 | |
| 24 | // HourAgg is one hour's rollup: whole-host metrics plus per-unit metrics. |
| 25 | // A metric that no sample in the hour carried is simply absent from the |
| 26 | // map — never a zero-valued Agg standing in for "no data". |
| 27 | type HourAgg struct { |
| 28 | Host map[string]Agg `json:"host"` // "cpu","mem","disk","load1" |
| 29 | Units map[string]map[string]Agg `json:"units"` // unit -> "cpu","mem_bytes" -> Agg |
| 30 | } |
| 31 | |
| 32 | // DayRollup is a box's rolled-up day, keyed by UTC hour ("00".."23") — |
| 33 | // the object stored under metrics/boxes/<box>/days/<date>.json. |
| 34 | type DayRollup struct { |
| 35 | V int `json:"v"` // 1 |
| 36 | Box string `json:"box"` |
| 37 | Date string `json:"date"` // UTC 2006-01-02 |
| 38 | Hours map[string]HourAgg `json:"hours"` |
| 39 | } |
| 40 | |
| 41 | // aggOf reduces vals to an Agg, reporting ok=false for an empty slice — the |
| 42 | // "this metric was never carried" case, which callers turn into "key |
| 43 | // absent from the map" rather than a zero-valued Agg. |
| 44 | func aggOf(vals []float64) (Agg, bool) { |
| 45 | if len(vals) == 0 { |
| 46 | return Agg{}, false |
| 47 | } |
| 48 | min, max, sum := vals[0], vals[0], 0.0 |
| 49 | for _, v := range vals { |
| 50 | if v < min { |
| 51 | min = v |
| 52 | } |
| 53 | if v > max { |
| 54 | max = v |
| 55 | } |
| 56 | sum += v |
| 57 | } |
| 58 | return Agg{Min: min, Avg: sum / float64(len(vals)), Max: max, N: len(vals)}, true |
| 59 | } |
| 60 | |
| 61 | // RollupHour aggregates a slice of raw samples (typically an hour's worth) |
| 62 | // into one HourAgg. Nil metric readings are excluded from N and, when a |
| 63 | // metric is nil in every sample, its key is absent from the returned map |
| 64 | // entirely (both at the host level and per-unit). Host metrics are |
| 65 | // "cpu","mem","disk","load1"; unit metrics are "cpu" (from |
| 66 | // UnitSample.CPUFrac) and "mem_bytes" (from MemBytes, cast to float64). |
| 67 | func RollupHour(samples []Sample) HourAgg { |
| 68 | var hostCPU, hostMem, hostDisk, hostLoad1 []float64 |
| 69 | for _, s := range samples { |
| 70 | if v := s.Host.CPUFrac; v != nil { |
| 71 | hostCPU = append(hostCPU, *v) |
| 72 | } |
| 73 | if v := s.Host.MemFrac; v != nil { |
| 74 | hostMem = append(hostMem, *v) |
| 75 | } |
| 76 | if v := s.Host.DiskFrac; v != nil { |
| 77 | hostDisk = append(hostDisk, *v) |
| 78 | } |
| 79 | if v := s.Host.Load1; v != nil { |
| 80 | hostLoad1 = append(hostLoad1, *v) |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | host := map[string]Agg{} |
| 85 | for key, vals := range map[string][]float64{ |
| 86 | "cpu": hostCPU, "mem": hostMem, "disk": hostDisk, "load1": hostLoad1, |
| 87 | } { |
| 88 | if agg, ok := aggOf(vals); ok { |
| 89 | host[key] = agg |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | unitCPU := map[string][]float64{} |
| 94 | unitMem := map[string][]float64{} |
| 95 | unitNames := map[string]bool{} |
| 96 | for _, s := range samples { |
| 97 | for name, u := range s.Units { |
| 98 | unitNames[name] = true |
| 99 | if u.CPUFrac != nil { |
| 100 | unitCPU[name] = append(unitCPU[name], *u.CPUFrac) |
| 101 | } |
| 102 | if u.MemBytes != nil { |
| 103 | unitMem[name] = append(unitMem[name], float64(*u.MemBytes)) |
| 104 | } |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | units := map[string]map[string]Agg{} |
| 109 | for name := range unitNames { |
| 110 | inner := map[string]Agg{} |
| 111 | if agg, ok := aggOf(unitCPU[name]); ok { |
| 112 | inner["cpu"] = agg |
| 113 | } |
| 114 | if agg, ok := aggOf(unitMem[name]); ok { |
| 115 | inner["mem_bytes"] = agg |
| 116 | } |
| 117 | if len(inner) > 0 { |
| 118 | units[name] = inner |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | return HourAgg{Host: host, Units: units} |
| 123 | } |
| 124 | |
| 125 | // UpsertRollup reads box's day object for hour's UTC date (a missing |
| 126 | // object is not an error — it starts a fresh DayRollup), sets the entry |
| 127 | // for hour's UTC hour bucket ("00".."23") to agg, and writes the day back |
| 128 | // with the same read-etag, Put, retry-once-on-ErrCASConflict pattern |
| 129 | // WriteRaw uses: a second conflict is not retried again and is returned to |
| 130 | // the caller. |
| 131 | func UpsertRollup(ctx context.Context, s release.Store, box string, hour time.Time, agg HourAgg) error { |
| 132 | date := hour.UTC().Format("2006-01-02") |
| 133 | hourKey := hour.UTC().Format("15") |
| 134 | key := DayKey(box, date) |
| 135 | |
| 136 | day, etag, err := currentDayRollup(ctx, s, key, box, date) |
| 137 | if err != nil { |
| 138 | return err |
| 139 | } |
| 140 | day.Hours[hourKey] = agg |
| 141 | body, err := json.MarshalIndent(day, "", " ") |
| 142 | if err != nil { |
| 143 | return err |
| 144 | } |
| 145 | if _, err := s.Put(ctx, key, body, etag); err != nil { |
| 146 | if !errors.Is(err, release.ErrCASConflict) { |
| 147 | return err |
| 148 | } |
| 149 | day, etag, err = currentDayRollup(ctx, s, key, box, date) |
| 150 | if err != nil { |
| 151 | return err |
| 152 | } |
| 153 | day.Hours[hourKey] = agg |
| 154 | body, err = json.MarshalIndent(day, "", " ") |
| 155 | if err != nil { |
| 156 | return err |
| 157 | } |
| 158 | if _, err := s.Put(ctx, key, body, etag); err != nil { |
| 159 | return err |
| 160 | } |
| 161 | } |
| 162 | return nil |
| 163 | } |
| 164 | |
| 165 | // currentDayRollup reads key's current DayRollup and etag for a |
| 166 | // replace-only Put, folding "doesn't exist yet" into a fresh DayRollup at |
| 167 | // ifMatch "" (create-only) — the same absent-means-empty convention |
| 168 | // currentRawEtag uses for raw chunks. |
| 169 | func currentDayRollup(ctx context.Context, s release.Store, key, box, date string) (DayRollup, string, error) { |
| 170 | body, etag, err := s.Get(ctx, key) |
| 171 | if err != nil { |
| 172 | if errors.Is(err, os.ErrNotExist) { |
| 173 | return DayRollup{V: 1, Box: box, Date: date, Hours: map[string]HourAgg{}}, "", nil |
| 174 | } |
| 175 | return DayRollup{}, "", err |
| 176 | } |
| 177 | var day DayRollup |
| 178 | if err := json.Unmarshal(body, &day); err != nil { |
| 179 | return DayRollup{}, "", err |
| 180 | } |
| 181 | if day.Hours == nil { |
| 182 | day.Hours = map[string]HourAgg{} |
| 183 | } |
| 184 | return day, etag, nil |
| 185 | } |
| 186 | |
amadan