| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package supervisord |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | "strings" |
| 8 | |
| 9 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 10 | ) |
| 11 | |
| 12 | func (c *Collector) collect() (map[string]int64, error) { |
| 13 | info, err := c.client.getAllProcessInfo() |
| 14 | if err != nil { |
| 15 | return nil, err |
| 16 | } |
| 17 | |
| 18 | ms := make(map[string]int64) |
| 19 | c.collectAllProcessInfo(ms, info) |
| 20 | |
| 21 | return ms, nil |
| 22 | } |
| 23 | |
| 24 | func (c *Collector) collectAllProcessInfo(ms map[string]int64, info []processStatus) { |
| 25 | c.resetCache() |
| 26 | ms["running_processes"] = 0 |
| 27 | ms["non_running_processes"] = 0 |
| 28 | for _, p := range info { |
| 29 | if _, ok := c.cache[p.group]; !ok { |
| 30 | c.cache[p.group] = make(map[string]bool) |
| 31 | c.addProcessGroupCharts(p) |
| 32 | } |
| 33 | if _, ok := c.cache[p.group][p.name]; !ok { |
| 34 | c.addProcessToCharts(p) |
| 35 | } |
| 36 | c.cache[p.group][p.name] = true |
| 37 | |
| 38 | ms["group_"+p.group+"_running_processes"] += 0 |
| 39 | ms["group_"+p.group+"_non_running_processes"] += 0 |
| 40 | if isProcRunning(p) { |
| 41 | ms["running_processes"] += 1 |
| 42 | ms["group_"+p.group+"_running_processes"] += 1 |
| 43 | } else { |
| 44 | ms["non_running_processes"] += 1 |
| 45 | ms["group_"+p.group+"_non_running_processes"] += 1 |
| 46 | } |
| 47 | id := procID(p) |
| 48 | ms[id+"_state_code"] = int64(p.state) |
| 49 | ms[id+"_exit_status"] = int64(p.exitStatus) |
| 50 | ms[id+"_uptime"] = calcProcessUptime(p) |
| 51 | ms[id+"_downtime"] = calcProcessDowntime(p) |
| 52 | } |
| 53 | c.cleanupCache() |
| 54 | } |
| 55 | |
| 56 | func (c *Collector) resetCache() { |
| 57 | for _, procs := range c.cache { |
| 58 | for name := range procs { |
| 59 | procs[name] = false |
| 60 | } |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | func (c *Collector) cleanupCache() { |
| 65 | for group, procs := range c.cache { |
| 66 | for name, ok := range procs { |
| 67 | if !ok { |
| 68 | c.removeProcessFromCharts(group, name) |
| 69 | delete(c.cache[group], name) |
| 70 | } |
| 71 | } |
| 72 | if len(c.cache[group]) == 0 { |
| 73 | c.removeProcessGroupCharts(group) |
| 74 | delete(c.cache, group) |
| 75 | } |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | func calcProcessUptime(p processStatus) int64 { |
| 80 | if !isProcRunning(p) { |
| 81 | return 0 |
| 82 | } |
| 83 | return int64(p.now - p.start) |
| 84 | } |
| 85 | |
| 86 | func calcProcessDowntime(p processStatus) int64 { |
| 87 | if isProcRunning(p) || p.stop == 0 { |
| 88 | return 0 |
| 89 | } |
| 90 | return int64(p.now - p.stop) |
| 91 | } |
| 92 | |
| 93 | func (c *Collector) addProcessGroupCharts(p processStatus) { |
| 94 | charts := newProcGroupCharts(p.group) |
| 95 | if err := c.Charts().Add(*charts...); err != nil { |
| 96 | c.Warning(err) |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | func (c *Collector) addProcessToCharts(p processStatus) { |
| 101 | id := procID(p) |
| 102 | for _, chart := range *c.Charts() { |
| 103 | var dimID string |
| 104 | switch chart.ID { |
| 105 | case fmt.Sprintf(groupProcessesStateCodeChartTmpl.ID, p.group): |
| 106 | dimID = id + "_state_code" |
| 107 | case fmt.Sprintf(groupProcessesExitStatusChartTmpl.ID, p.group): |
| 108 | dimID = id + "_exit_status" |
| 109 | case fmt.Sprintf(groupProcessesUptimeChartTmpl.ID, p.group): |
| 110 | dimID = id + "_uptime" |
| 111 | case fmt.Sprintf(groupProcessesDowntimeChartTmpl.ID, p.group): |
| 112 | dimID = id + "_downtime" |
| 113 | default: |
| 114 | continue |
| 115 | } |
| 116 | dim := &collectorapi.Dim{ID: dimID, Name: p.name} |
| 117 | if err := chart.AddDim(dim); err != nil { |
| 118 | c.Warning(err) |
| 119 | return |
| 120 | } |
| 121 | chart.MarkNotCreated() |
| 122 | } |
| 123 | } |
| 124 | |
| 125 | func (c *Collector) removeProcessGroupCharts(group string) { |
| 126 | prefix := "group_" + group |
| 127 | for _, c := range *c.Charts() { |
| 128 | if strings.HasPrefix(c.ID, prefix) { |
| 129 | c.MarkRemove() |
| 130 | c.MarkNotCreated() |
| 131 | } |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | func (c *Collector) removeProcessFromCharts(group, name string) { |
| 136 | id := procID(processStatus{name: name, group: group}) |
| 137 | for _, chart := range *c.Charts() { |
| 138 | var dimID string |
| 139 | switch chart.ID { |
| 140 | case fmt.Sprintf(groupProcessesStateCodeChartTmpl.ID, group): |
| 141 | dimID = id + "_state_code" |
| 142 | case fmt.Sprintf(groupProcessesExitStatusChartTmpl.ID, group): |
| 143 | dimID = id + "_exit_status" |
| 144 | case fmt.Sprintf(groupProcessesUptimeChartTmpl.ID, group): |
| 145 | dimID = id + "_uptime" |
| 146 | case fmt.Sprintf(groupProcessesDowntimeChartTmpl.ID, group): |
| 147 | dimID = id + "_downtime" |
| 148 | default: |
| 149 | continue |
| 150 | } |
| 151 | if err := chart.MarkDimRemove(dimID, true); err != nil { |
| 152 | c.Warning(err) |
| 153 | return |
| 154 | } |
| 155 | chart.MarkNotCreated() |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | func procID(p processStatus) string { |
| 160 | return fmt.Sprintf("group_%s_process_%s", p.group, p.name) |
| 161 | } |
| 162 | |
| 163 | func isProcRunning(p processStatus) bool { |
| 164 | // http://supervisord.org/subprocess.html#process-states |
| 165 | // STOPPED (0) |
| 166 | // STARTING (10) |
| 167 | // RUNNING (20) |
| 168 | // BACKOFF (30) |
| 169 | // STOPPING (40) |
| 170 | // EXITED (100) |
| 171 | // FATAL (200) |
| 172 | // UNKNOWN (1000) |
| 173 | return p.state == 20 |
| 174 | } |