feat: collect and report 1-min load average
This commit is contained in:
@@ -9,6 +9,7 @@ import (
|
||||
"bugs.watch/agent/internal/collectors"
|
||||
"bugs.watch/agent/internal/collectors/cpu"
|
||||
"bugs.watch/agent/internal/collectors/disk"
|
||||
"bugs.watch/agent/internal/collectors/loadavg"
|
||||
"bugs.watch/agent/internal/collectors/memory"
|
||||
"bugs.watch/agent/internal/collectors/uptime"
|
||||
"bugs.watch/agent/internal/config"
|
||||
@@ -32,6 +33,7 @@ func Run(ctx context.Context, cfg config.Config) {
|
||||
cpu.New(),
|
||||
memory.New(),
|
||||
disk.New(),
|
||||
loadavg.New(),
|
||||
}
|
||||
|
||||
mgr := dbmanager.New(cfg.DBConfigPath)
|
||||
|
||||
@@ -16,7 +16,9 @@ type Window struct {
|
||||
|
||||
uptime uint64
|
||||
|
||||
// DB metrics — populated only when a DB collector is registered.
|
||||
load float64
|
||||
|
||||
// DB metrics. populated only when a DB collector is registered.
|
||||
hasDB bool
|
||||
dbDriver uint8
|
||||
dbConnPctSum float64
|
||||
@@ -34,6 +36,7 @@ type Snapshot struct {
|
||||
MemAvg float64
|
||||
DiskAvg float64
|
||||
Uptime uint64
|
||||
Load float64
|
||||
|
||||
// nil when DB monitoring is not enabled.
|
||||
DBDriver *uint8
|
||||
@@ -66,6 +69,10 @@ func (w *Window) SetUptime(value uint64) {
|
||||
w.uptime = value
|
||||
}
|
||||
|
||||
func (w *Window) SetLoad(value float64) {
|
||||
w.load = value
|
||||
}
|
||||
|
||||
func (w *Window) AddDBConnPct(v float64) {
|
||||
w.dbConnPctSum += v
|
||||
w.dbConnPctCount++
|
||||
@@ -118,6 +125,7 @@ func (w *Window) Snapshot() Snapshot {
|
||||
MemAvg: memAvg,
|
||||
DiskAvg: diskAvg,
|
||||
Uptime: w.uptime,
|
||||
Load: w.load,
|
||||
}
|
||||
|
||||
if w.hasDB {
|
||||
|
||||
41
internal/collectors/loadavg/loadavg.go
Normal file
41
internal/collectors/loadavg/loadavg.go
Normal file
@@ -0,0 +1,41 @@
|
||||
package loadavg
|
||||
|
||||
import (
|
||||
"os"
|
||||
"strconv"
|
||||
"strings"
|
||||
|
||||
"bugs.watch/agent/internal/aggregate"
|
||||
)
|
||||
|
||||
type Collector struct{}
|
||||
|
||||
func New() *Collector {
|
||||
return &Collector{}
|
||||
}
|
||||
|
||||
func (c *Collector) Name() string {
|
||||
return "loadavg"
|
||||
}
|
||||
|
||||
// Collect reads /proc/loadavg and stores the 1-minute load average.
|
||||
// Failure is silently ignored by design.
|
||||
func (c *Collector) Collect(window *aggregate.Window) {
|
||||
data, err := os.ReadFile("/proc/loadavg")
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
// Format: "<load1> <load5> <load15> <running/total> <last_pid>"
|
||||
fields := strings.Fields(string(data))
|
||||
if len(fields) < 1 {
|
||||
return
|
||||
}
|
||||
|
||||
load1, err := strconv.ParseFloat(fields[0], 64)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
|
||||
window.SetLoad(load1)
|
||||
}
|
||||
@@ -20,7 +20,7 @@ import (
|
||||
sqlitecol "bugs.watch/agent/internal/collectors/sqlite"
|
||||
)
|
||||
|
||||
// Driver type identifiers — used as the db_driver field in the health payload
|
||||
// Driver type identifiers. used as the db_driver field in the health payload
|
||||
// so the dashboard can render the correct icon without parsing strings.
|
||||
const (
|
||||
DriverNone uint8 = 0
|
||||
|
||||
@@ -38,6 +38,7 @@ func (s *HTTPSender) Send(ctx context.Context, snap aggregate.Snapshot) error {
|
||||
MemAvg: snap.MemAvg,
|
||||
DiskAvg: snap.DiskAvg,
|
||||
Uptime: snap.Uptime,
|
||||
Load: snap.Load,
|
||||
HostID: s.hostID,
|
||||
Environment: s.env,
|
||||
DBDriver: snap.DBDriver,
|
||||
|
||||
@@ -22,11 +22,12 @@ func (s *LoggingSender) Send(ctx context.Context, snap aggregate.Snapshot) error
|
||||
return ctx.Err()
|
||||
default:
|
||||
log.Printf(
|
||||
"SEND cpu=%.2f%% mem=%.2f%% disk=%.2f%% uptime=%ds",
|
||||
"SEND cpu=%.2f%% mem=%.2f%% disk=%.2f%% uptime=%ds load=%.2f",
|
||||
snap.CPUAvg,
|
||||
snap.MemAvg,
|
||||
snap.DiskAvg,
|
||||
snap.Uptime,
|
||||
snap.Load,
|
||||
)
|
||||
return nil
|
||||
}
|
||||
|
||||
@@ -10,11 +10,12 @@ type SnapshotPayload struct {
|
||||
MemAvg float64 `json:"mem_avg"`
|
||||
DiskAvg float64 `json:"disk_avg"`
|
||||
Uptime uint64 `json:"uptime"`
|
||||
Load float64 `json:"load"`
|
||||
|
||||
HostID string `json:"host_id,omitempty"`
|
||||
Environment string `json:"environment,omitempty"`
|
||||
|
||||
// Optional — only present when DB monitoring is enabled.
|
||||
// Optional. only present when DB monitoring is enabled.
|
||||
// DBDriver is a numeric identifier: 1=mysql/mariadb 2=postgres 3=redis 4=mongodb 5=sqlite
|
||||
DBDriver *uint8 `json:"db_driver,omitempty"`
|
||||
DBConnPct *float64 `json:"db_conn_pct,omitempty"`
|
||||
|
||||
Reference in New Issue
Block a user