|
| 1 | +// Copyright 2022 The Prometheus Authors |
| 2 | +// Licensed under the Apache License, Version 2.0 (the "License"); |
| 3 | +// you may not use this file except in compliance with the License. |
| 4 | +// You may obtain a copy of the License at |
| 5 | +// |
| 6 | +// http://www.apache.org/licenses/LICENSE-2.0 |
| 7 | +// |
| 8 | +// Unless required by applicable law or agreed to in writing, software |
| 9 | +// distributed under the License is distributed on an "AS IS" BASIS, |
| 10 | +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 11 | +// See the License for the specific language governing permissions and |
| 12 | +// limitations under the License. |
| 13 | + |
| 14 | +package collector |
| 15 | + |
| 16 | +import ( |
| 17 | + "context" |
| 18 | + "errors" |
| 19 | + "fmt" |
| 20 | + "sync" |
| 21 | + "time" |
| 22 | + |
| 23 | + "github.com/go-kit/log" |
| 24 | + "github.com/go-kit/log/level" |
| 25 | + "github.com/prometheus/client_golang/prometheus" |
| 26 | + "gopkg.in/alecthomas/kingpin.v2" |
| 27 | +) |
| 28 | + |
| 29 | +var ( |
| 30 | + factories = make(map[string]func(logger log.Logger) (Collector, error)) |
| 31 | + initiatedCollectorsMtx = sync.Mutex{} |
| 32 | + initiatedCollectors = make(map[string]Collector) |
| 33 | + collectorState = make(map[string]*bool) |
| 34 | + forcedCollectors = map[string]bool{} // collectors which have been explicitly enabled or disabled |
| 35 | +) |
| 36 | + |
| 37 | +const ( |
| 38 | + // Namespace for all metrics. |
| 39 | + namespace = "pg" |
| 40 | + |
| 41 | + defaultEnabled = true |
| 42 | + defaultDisabled = false |
| 43 | +) |
| 44 | + |
| 45 | +var ( |
| 46 | + scrapeDurationDesc = prometheus.NewDesc( |
| 47 | + prometheus.BuildFQName(namespace, "scrape", "collector_duration_seconds"), |
| 48 | + "postgres_exporter: Duration of a collector scrape.", |
| 49 | + []string{"collector"}, |
| 50 | + nil, |
| 51 | + ) |
| 52 | + scrapeSuccessDesc = prometheus.NewDesc( |
| 53 | + prometheus.BuildFQName(namespace, "scrape", "collector_success"), |
| 54 | + "postgres_exporter: Whether a collector succeeded.", |
| 55 | + []string{"collector"}, |
| 56 | + nil, |
| 57 | + ) |
| 58 | +) |
| 59 | + |
| 60 | +type Collector interface { |
| 61 | + Update(ctx context.Context, server *server, ch chan<- prometheus.Metric) error |
| 62 | +} |
| 63 | + |
| 64 | +func registerCollector(name string, isDefaultEnabled bool, createFunc func(logger log.Logger) (Collector, error)) { |
| 65 | + var helpDefaultState string |
| 66 | + if isDefaultEnabled { |
| 67 | + helpDefaultState = "enabled" |
| 68 | + } else { |
| 69 | + helpDefaultState = "disabled" |
| 70 | + } |
| 71 | + |
| 72 | + // Create flag for this collector |
| 73 | + flagName := fmt.Sprintf("collector.%s", name) |
| 74 | + flagHelp := fmt.Sprintf("Enable the %s collector (default: %s).", name, helpDefaultState) |
| 75 | + defaultValue := fmt.Sprintf("%v", isDefaultEnabled) |
| 76 | + |
| 77 | + flag := kingpin.Flag(flagName, flagHelp).Default(defaultValue).Action(collectorFlagAction(name)).Bool() |
| 78 | + collectorState[name] = flag |
| 79 | + |
| 80 | + // Register the create function for this collector |
| 81 | + factories[name] = createFunc |
| 82 | +} |
| 83 | + |
| 84 | +// PostgresCollector implements the prometheus.Collector interface. |
| 85 | +type PostgresCollector struct { |
| 86 | + Collectors map[string]Collector |
| 87 | + logger log.Logger |
| 88 | + |
| 89 | + servers map[string]*server |
| 90 | +} |
| 91 | + |
| 92 | +// NewPostgresCollector creates a new PostgresCollector. |
| 93 | +func NewPostgresCollector(logger log.Logger, dsns []string, filters ...string) (*PostgresCollector, error) { |
| 94 | + f := make(map[string]bool) |
| 95 | + for _, filter := range filters { |
| 96 | + enabled, exist := collectorState[filter] |
| 97 | + if !exist { |
| 98 | + return nil, fmt.Errorf("missing collector: %s", filter) |
| 99 | + } |
| 100 | + if !*enabled { |
| 101 | + return nil, fmt.Errorf("disabled collector: %s", filter) |
| 102 | + } |
| 103 | + f[filter] = true |
| 104 | + } |
| 105 | + collectors := make(map[string]Collector) |
| 106 | + initiatedCollectorsMtx.Lock() |
| 107 | + defer initiatedCollectorsMtx.Unlock() |
| 108 | + for key, enabled := range collectorState { |
| 109 | + if !*enabled || (len(f) > 0 && !f[key]) { |
| 110 | + continue |
| 111 | + } |
| 112 | + if collector, ok := initiatedCollectors[key]; ok { |
| 113 | + collectors[key] = collector |
| 114 | + } else { |
| 115 | + collector, err := factories[key](log.With(logger, "collector", key)) |
| 116 | + if err != nil { |
| 117 | + return nil, err |
| 118 | + } |
| 119 | + collectors[key] = collector |
| 120 | + initiatedCollectors[key] = collector |
| 121 | + } |
| 122 | + } |
| 123 | + |
| 124 | + servers := make(map[string]*server) |
| 125 | + for _, dsn := range dsns { |
| 126 | + s, err := makeServer(dsn) |
| 127 | + if err != nil { |
| 128 | + return nil, err |
| 129 | + } |
| 130 | + servers[dsn] = s |
| 131 | + } |
| 132 | + |
| 133 | + return &PostgresCollector{ |
| 134 | + Collectors: collectors, |
| 135 | + logger: logger, |
| 136 | + servers: servers, |
| 137 | + }, nil |
| 138 | +} |
| 139 | + |
| 140 | +// Describe implements the prometheus.Collector interface. |
| 141 | +func (n PostgresCollector) Describe(ch chan<- *prometheus.Desc) { |
| 142 | + ch <- scrapeDurationDesc |
| 143 | + ch <- scrapeSuccessDesc |
| 144 | +} |
| 145 | + |
| 146 | +// Collect implements the prometheus.Collector interface. |
| 147 | +func (n PostgresCollector) Collect(ch chan<- prometheus.Metric) { |
| 148 | + ctx := context.TODO() |
| 149 | + wg := sync.WaitGroup{} |
| 150 | + wg.Add(len(n.servers)) |
| 151 | + for _, s := range n.servers { |
| 152 | + go func(s *server) { |
| 153 | + n.subCollect(ctx, s, ch) |
| 154 | + wg.Done() |
| 155 | + }(s) |
| 156 | + } |
| 157 | + wg.Wait() |
| 158 | +} |
| 159 | + |
| 160 | +func (n PostgresCollector) subCollect(ctx context.Context, server *server, ch chan<- prometheus.Metric) { |
| 161 | + wg := sync.WaitGroup{} |
| 162 | + wg.Add(len(n.Collectors)) |
| 163 | + for name, c := range n.Collectors { |
| 164 | + go func(name string, c Collector) { |
| 165 | + execute(ctx, name, c, server, ch, n.logger) |
| 166 | + wg.Done() |
| 167 | + }(name, c) |
| 168 | + } |
| 169 | + wg.Wait() |
| 170 | +} |
| 171 | + |
| 172 | +func execute(ctx context.Context, name string, c Collector, s *server, ch chan<- prometheus.Metric, logger log.Logger) { |
| 173 | + begin := time.Now() |
| 174 | + err := c.Update(ctx, s, ch) |
| 175 | + duration := time.Since(begin) |
| 176 | + var success float64 |
| 177 | + |
| 178 | + if err != nil { |
| 179 | + if IsNoDataError(err) { |
| 180 | + level.Debug(logger).Log("msg", "collector returned no data", "name", name, "duration_seconds", duration.Seconds(), "err", err) |
| 181 | + } else { |
| 182 | + level.Error(logger).Log("msg", "collector failed", "name", name, "duration_seconds", duration.Seconds(), "err", err) |
| 183 | + } |
| 184 | + success = 0 |
| 185 | + } else { |
| 186 | + level.Debug(logger).Log("msg", "collector succeeded", "name", name, "duration_seconds", duration.Seconds()) |
| 187 | + success = 1 |
| 188 | + } |
| 189 | + ch <- prometheus.MustNewConstMetric(scrapeDurationDesc, prometheus.GaugeValue, duration.Seconds(), name) |
| 190 | + ch <- prometheus.MustNewConstMetric(scrapeSuccessDesc, prometheus.GaugeValue, success, name) |
| 191 | +} |
| 192 | + |
| 193 | +// collectorFlagAction generates a new action function for the given collector |
| 194 | +// to track whether it has been explicitly enabled or disabled from the command line. |
| 195 | +// A new action function is needed for each collector flag because the ParseContext |
| 196 | +// does not contain information about which flag called the action. |
| 197 | +// See: https://github.com/alecthomas/kingpin/issues/294 |
| 198 | +func collectorFlagAction(collector string) func(ctx *kingpin.ParseContext) error { |
| 199 | + return func(ctx *kingpin.ParseContext) error { |
| 200 | + forcedCollectors[collector] = true |
| 201 | + return nil |
| 202 | + } |
| 203 | +} |
| 204 | + |
| 205 | +// ErrNoData indicates the collector found no data to collect, but had no other error. |
| 206 | +var ErrNoData = errors.New("collector returned no data") |
| 207 | + |
| 208 | +func IsNoDataError(err error) bool { |
| 209 | + return err == ErrNoData |
| 210 | +} |
0 commit comments