services/keep-balance/metrics.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package main
   6
   7 import (
   8         "fmt"
   9         "net/http"
  10         "sync"
  11
  12         "github.com/prometheus/client_golang/prometheus"
  13         "github.com/prometheus/client_golang/prometheus/promhttp"
  14 )
  15
  16 type observer interface{ Observe(float64) }
  17 type setter interface{ Set(float64) }
  18
  19 type metrics struct {
  20         reg         *prometheus.Registry
  21         statsGauges map[string]setter
  22         observers   map[string]observer
  23         setupOnce   sync.Once
  24         mtx         sync.Mutex
  25 }
  26
  27 func newMetrics() *metrics {
  28         return &metrics{
  29                 reg:         prometheus.NewRegistry(),
  30                 statsGauges: map[string]setter{},
  31                 observers:   map[string]observer{},
  32         }
  33 }
  34
  35 func (m *metrics) DurationObserver(name, help string) observer {
  36         m.mtx.Lock()
  37         defer m.mtx.Unlock()
  38         if obs, ok := m.observers[name]; ok {
  39                 return obs
  40         }
  41         summary := prometheus.NewSummary(prometheus.SummaryOpts{
  42                 Namespace: "arvados",
  43                 Name:      name,
  44                 Subsystem: "keepbalance",
  45                 Help:      help,
  46         })
  47         m.reg.MustRegister(summary)
  48         m.observers[name] = summary
  49         return summary
  50 }
  51
  52 // UpdateStats updates prometheus metrics using the given
  53 // balancerStats. It creates and registers the needed gauges on its
  54 // first invocation.
  55 func (m *metrics) UpdateStats(s balancerStats) {
  56         type gauge struct {
  57                 Value interface{}
  58                 Help  string
  59         }
  60         s2g := map[string]gauge{
  61                 "total":             {s.current, "current backend storage usage"},
  62                 "garbage":           {s.garbage, "garbage (unreferenced, old)"},
  63                 "transient":         {s.unref, "transient (unreferenced, new)"},
  64                 "overreplicated":    {s.overrep, "overreplicated"},
  65                 "underreplicated":   {s.underrep, "underreplicated"},
  66                 "lost":              {s.lost, "lost"},
  67                 "dedup_byte_ratio":  {s.dedupByteRatio(), "deduplication ratio, bytes referenced / bytes stored"},
  68                 "dedup_block_ratio": {s.dedupBlockRatio(), "deduplication ratio, blocks referenced / blocks stored"},
  69         }
  70         m.setupOnce.Do(func() {
  71                 // Register gauge(s) for each balancerStats field.
  72                 addGauge := func(name, help string) {
  73                         g := prometheus.NewGauge(prometheus.GaugeOpts{
  74                                 Namespace: "arvados",
  75                                 Name:      name,
  76                                 Subsystem: "keep",
  77                                 Help:      help,
  78                         })
  79                         m.reg.MustRegister(g)
  80                         m.statsGauges[name] = g
  81                 }
  82                 for name, gauge := range s2g {
  83                         switch gauge.Value.(type) {
  84                         case blocksNBytes:
  85                                 for _, sub := range []string{"blocks", "bytes", "replicas"} {
  86                                         addGauge(name+"_"+sub, sub+" of "+gauge.Help)
  87                                 }
  88                         case int, int64, float64:
  89                                 addGauge(name, gauge.Help)
  90                         default:
  91                                 panic(fmt.Sprintf("bad gauge type %T", gauge.Value))
  92                         }
  93                 }
  94         })
  95         // Set gauges to values from s.
  96         for name, gauge := range s2g {
  97                 switch val := gauge.Value.(type) {
  98                 case blocksNBytes:
  99                         m.statsGauges[name+"_blocks"].Set(float64(val.blocks))
 100                         m.statsGauges[name+"_bytes"].Set(float64(val.bytes))
 101                         m.statsGauges[name+"_replicas"].Set(float64(val.replicas))
 102                 case int:
 103                         m.statsGauges[name].Set(float64(val))
 104                 case int64:
 105                         m.statsGauges[name].Set(float64(val))
 106                 case float64:
 107                         m.statsGauges[name].Set(float64(val))
 108                 default:
 109                         panic(fmt.Sprintf("bad gauge type %T", gauge.Value))
 110                 }
 111         }
 112 }
 113
 114 func (m *metrics) Handler(log promhttp.Logger) http.Handler {
 115         return promhttp.HandlerFor(m.reg, promhttp.HandlerOpts{
 116                 ErrorLog: log,
 117         })
 118 }