| 1 | // Package metrics is the pure half of host-process monitoring: every |
| 2 | // function here parses a string or a pair of prior readings and MUST fail |
| 3 | // toward nil, never a guessed zero. A zero CPU fraction and "we couldn't |
| 4 | // read /proc/stat" are different facts, and a caller several layers up |
| 5 | // (status pages, alerting) cannot tell them apart once a parser has already |
| 6 | // collapsed one into the other — so a garbled line, a missing field, or a |
| 7 | // counter that has gone backwards (reboot, cgroup recreated) all return nil |
| 8 | // rather than 0 or the previous value. Nothing in this package touches the |
| 9 | // filesystem or the OS: callers own reading /proc/stat, /proc/loadavg, and |
| 10 | // cgroup v2 files, and pass the resulting text or prior/current readings in. |
| 11 | package metrics |
| 12 | |
| 13 | import ( |
| 14 | "strconv" |
| 15 | "strings" |
| 16 | "time" |
| 17 | ) |
| 18 | |
| 19 | // CPUCounters is a monotonic reading of /proc/stat's aggregate cpu line. |
| 20 | type CPUCounters struct{ Busy, Total uint64 } |
| 21 | |
| 22 | // ParseProcStat parses the aggregate "cpu" line of /proc/stat (fields after |
| 23 | // "cpu" are user nice system idle iowait irq softirq steal, with optional |
| 24 | // guest/guest_nice columns already folded into user/nice by the kernel on |
| 25 | // modern systems). Total is the sum of the first 8 fields; Busy is Total |
| 26 | // minus idle and iowait — deliberately only the first 8, so a kernel that |
| 27 | // adds guest columns is never double-counted. Any missing "cpu" prefix, too |
| 28 | // few fields, or a non-numeric field -> nil. |
| 29 | func ParseProcStat(text string) *CPUCounters { |
| 30 | for _, line := range strings.Split(text, "\n") { |
| 31 | fields := strings.Fields(line) |
| 32 | if len(fields) < 9 || fields[0] != "cpu" { |
| 33 | continue |
| 34 | } |
| 35 | var vals [8]uint64 |
| 36 | ok := true |
| 37 | for i := 0; i < 8; i++ { |
| 38 | v, err := strconv.ParseUint(fields[i+1], 10, 64) |
| 39 | if err != nil { |
| 40 | ok = false |
| 41 | break |
| 42 | } |
| 43 | vals[i] = v |
| 44 | } |
| 45 | if !ok { |
| 46 | continue |
| 47 | } |
| 48 | var total uint64 |
| 49 | for _, v := range vals { |
| 50 | total += v |
| 51 | } |
| 52 | idle, iowait := vals[3], vals[4] |
| 53 | return &CPUCounters{Busy: total - idle - iowait, Total: total} |
| 54 | } |
| 55 | return nil |
| 56 | } |
| 57 | |
| 58 | // CPUFrac turns two /proc/stat readings into a fraction of CPU busy over the |
| 59 | // interval between them. Both readings must be non-nil, cur.Total must |
| 60 | // exceed prev.Total, and cur.Busy must not be behind prev.Busy — counters |
| 61 | // that went backwards mean the box rebooted or the reading is otherwise |
| 62 | // invalid, and reporting a fraction from that would be a guess dressed up as |
| 63 | // a measurement, so it's nil instead. The Busy check matters on its own: |
| 64 | // Total can still advance while Busy alone regresses, and since both are |
| 65 | // unsigned, cur.Busy-prev.Busy would silently wrap into a huge garbage |
| 66 | // value rather than go negative. |
| 67 | func CPUFrac(prev, cur *CPUCounters) *float64 { |
| 68 | if prev == nil || cur == nil || cur.Total <= prev.Total || cur.Busy < prev.Busy { |
| 69 | return nil |
| 70 | } |
| 71 | frac := float64(cur.Busy-prev.Busy) / float64(cur.Total-prev.Total) |
| 72 | return &frac |
| 73 | } |
| 74 | |
| 75 | // ParseLoadAvg parses /proc/loadavg's first three whitespace-separated |
| 76 | // fields (1/5/15-minute load averages). Any missing or non-numeric field |
| 77 | // fails the whole reading -> all three results nil, since a partial load |
| 78 | // average (e.g. l1 present but l5 garbled) is not a value worth publishing |
| 79 | // on its own. |
| 80 | func ParseLoadAvg(text string) (l1, l5, l15 *float64) { |
| 81 | fields := strings.Fields(text) |
| 82 | if len(fields) < 3 { |
| 83 | return nil, nil, nil |
| 84 | } |
| 85 | v1, err1 := strconv.ParseFloat(fields[0], 64) |
| 86 | v5, err5 := strconv.ParseFloat(fields[1], 64) |
| 87 | v15, err15 := strconv.ParseFloat(fields[2], 64) |
| 88 | if err1 != nil || err5 != nil || err15 != nil { |
| 89 | return nil, nil, nil |
| 90 | } |
| 91 | return &v1, &v5, &v15 |
| 92 | } |
| 93 | |
| 94 | // ParseCgroupCPUUsec parses cpu.stat's usage_usec line. Missing key or a |
| 95 | // non-numeric value -> nil. |
| 96 | func ParseCgroupCPUUsec(text string) *int64 { |
| 97 | for _, line := range strings.Split(text, "\n") { |
| 98 | fields := strings.Fields(line) |
| 99 | if len(fields) != 2 || fields[0] != "usage_usec" { |
| 100 | continue |
| 101 | } |
| 102 | v, err := strconv.ParseInt(fields[1], 10, 64) |
| 103 | if err != nil { |
| 104 | return nil |
| 105 | } |
| 106 | return &v |
| 107 | } |
| 108 | return nil |
| 109 | } |
| 110 | |
| 111 | // ParseCgroupInt parses a single cgroup v2 scalar file such as |
| 112 | // memory.current or memory.peak. cgroup v2 uses the literal string "max" to |
| 113 | // mean "no limit set" — a real value, but not a number, so it maps to nil |
| 114 | // exactly like a parse failure would, since callers only want a byte count |
| 115 | // or nothing. |
| 116 | func ParseCgroupInt(text string) *int64 { |
| 117 | s := strings.TrimSpace(text) |
| 118 | if s == "" || s == "max" { |
| 119 | return nil |
| 120 | } |
| 121 | v, err := strconv.ParseInt(s, 10, 64) |
| 122 | if err != nil { |
| 123 | return nil |
| 124 | } |
| 125 | return &v |
| 126 | } |
| 127 | |
| 128 | // ParseCgroupIO parses io.stat, summing rbytes and wbytes across every |
| 129 | // per-device line (a cgroup can have I/O against several block devices at |
| 130 | // once, and the caller wants one whole-cgroup total). A line missing both |
| 131 | // keys is simply skipped rather than failing the whole parse — io.stat's |
| 132 | // per-device format is otherwise stable — but if no line anywhere yields |
| 133 | // either key, both results are nil rather than a false zero. |
| 134 | func ParseCgroupIO(text string) (rbytes, wbytes *int64) { |
| 135 | var rsum, wsum int64 |
| 136 | found := false |
| 137 | for _, line := range strings.Split(text, "\n") { |
| 138 | for _, field := range strings.Fields(line) { |
| 139 | key, val, ok := strings.Cut(field, "=") |
| 140 | if !ok { |
| 141 | continue |
| 142 | } |
| 143 | switch key { |
| 144 | case "rbytes": |
| 145 | if v, err := strconv.ParseInt(val, 10, 64); err == nil { |
| 146 | rsum += v |
| 147 | found = true |
| 148 | } |
| 149 | case "wbytes": |
| 150 | if v, err := strconv.ParseInt(val, 10, 64); err == nil { |
| 151 | wsum += v |
| 152 | found = true |
| 153 | } |
| 154 | } |
| 155 | } |
| 156 | } |
| 157 | if !found { |
| 158 | return nil, nil |
| 159 | } |
| 160 | return &rsum, &wsum |
| 161 | } |
| 162 | |
| 163 | // UnitCPUFrac turns two cgroup cpu.stat usage_usec readings and the elapsed |
| 164 | // wall time between them into a fraction of a CPU core busy. Both readings |
| 165 | // must be non-nil, cur must not be behind prev (cgroup recreated, counter |
| 166 | // reset), and elapsed must be strictly positive — any of those failing |
| 167 | // would turn a division into either a guess or a crash, so it's nil instead. |
| 168 | func UnitCPUFrac(prevUsec, curUsec *int64, elapsed time.Duration) *float64 { |
| 169 | if prevUsec == nil || curUsec == nil || *curUsec < *prevUsec || elapsed <= 0 { |
| 170 | return nil |
| 171 | } |
| 172 | frac := float64(*curUsec-*prevUsec) / 1e6 / elapsed.Seconds() |
| 173 | return &frac |
| 174 | } |
| 175 | |
amadan