This commit is contained in:
2026-10-11 12:24:01 +02:00
commit 9d4c8b7a6f
10 changed files with 356 additions and 0 deletions

1
.gitignore vendored Normal file
View File

@@ -0,0 +1 @@
.gocache

42
config/config.go Normal file
View File

@@ -0,0 +1,42 @@
// config/config.go
package config
import (
"os"
"time"
"gopkg.in/yaml.v3"
)
type Config struct {
ScrapeInterval time.Duration `yaml:"scrape_interval"`
Readers []ReaderCfg `yaml:"readers"`
Exporters []ExporterCfg `yaml:"exporters"`
}
type ReaderCfg struct {
Type string `yaml:"type"`
Endpoint string `yaml:"endpoint,omitempty"`
Model string `yaml:"model,omitempty"`
}
type ExporterCfg struct {
Type string `yaml:"type"`
Addr string `yaml:"addr,omitempty"`
JSONFile string `yaml:"json_file,omitempty"`
}
func Load(path string) (*Config, error) {
data, err := os.ReadFile(path)
if err != nil {
return nil, err
}
var c Config
if err := yaml.Unmarshal(data, &c); err != nil {
return nil, err
}
if c.ScrapeInterval == 0 {
c.ScrapeInterval = 5 * time.Second
}
return &c, nil
}

13
config/config.yaml Normal file
View File

@@ -0,0 +1,13 @@
# config/config.yaml
scrape_interval: 5s
readers:
- type: nvidia_smi
exporters:
- type: dev
json_file: "/tmp/nvllama-metrics.json" # optional, omit for stdout only
# - type: prometheus
# addr: ":9464"
# - type: telegraf
# addr: "udp://localhost:8094"

25
core/metric.go Normal file
View File

@@ -0,0 +1,25 @@
package core
import (
"context"
"time"
)
type Metric struct {
Name string
Help string
Type string // "counter", "gauge", "histogram"
Value float64
Labels map[string]string
Timestamp time.Time
}
type Reader interface {
Name() string
Read(ctx context.Context) ([]Metric, error)
}
type Exporter interface {
Name() string
Write(ctx context.Context, metrics []Metric) error
}

55
exporters/dev.go Normal file
View File

@@ -0,0 +1,55 @@
package exporters
import (
"context"
"encoding/json"
"fmt"
"os"
"sort"
"time"
"git.govekar.net/LovroG05/nvllama-watchdog/core"
)
type DevExporter struct {
jsonFile string // optional: path to dump JSON, empty = stdout only
}
func NewDevExporter(jsonFile string) *DevExporter {
return &DevExporter{jsonFile: jsonFile}
}
func (e *DevExporter) Name() string { return "dev" }
func (e *DevExporter) Write(_ context.Context, ms []core.Metric) error {
// Sort for stable output
sort.Slice(ms, func(i, j int) bool {
if ms[i].Name != ms[j].Name {
return ms[i].Name < ms[j].Name
}
return fmt.Sprint(ms[i].Labels) < fmt.Sprint(ms[j].Labels)
})
// Human-readable to stdout
fmt.Printf("\n── %s ── (%d metrics)\n", time.Now().Format("15:04:05"), len(ms))
for _, m := range ms {
fmt.Printf(" %-40s %12.4f %v\n", m.Name, m.Value, m.Labels)
}
// Optional JSON dump (overwrite each cycle, for `jq` inspection)
if e.jsonFile != "" {
payload := struct {
Timestamp string `json:"timestamp"`
Count int `json:"count"`
Metrics []core.Metric `json:"metrics"`
}{
Timestamp: time.Now().UTC().Format(time.RFC3339),
Count: len(ms),
Metrics: ms,
}
data, _ := json.MarshalIndent(payload, "", " ")
_ = os.WriteFile(e.jsonFile, data, 0644)
}
return nil
}

75
exporters/prometheus.go Normal file
View File

@@ -0,0 +1,75 @@
// exporters/prometheus.go
package exporters
import (
"context"
"fmt"
"net/http"
"sort"
"strings"
"sync"
"git.govekar.net/LovroG05/nvllama-watchdog/core"
)
type PrometheusExporter struct {
mu sync.RWMutex
metrics []core.Metric
addr string
}
func NewPrometheusExporter(addr string) *PrometheusExporter {
return &PrometheusExporter{addr: addr}
}
func (e *PrometheusExporter) Name() string { return "prometheus" }
func (e *PrometheusExporter) Write(_ context.Context, ms []core.Metric) error {
e.mu.Lock()
e.metrics = ms
e.mu.Unlock()
return nil
}
// Handler serves the /metrics HTTP endpoint.
func (e *PrometheusExporter) Handler() http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
e.mu.RLock()
defer e.mu.RUnlock()
w.Header().Set("Content-Type", "text/plain; version=0.0.4")
// group by name for HELP/TYPE (simplified)
sort.Slice(e.metrics, func(i, j int) bool {
return e.metrics[i].Name < e.metrics[j].Name
})
var b strings.Builder
for _, m := range e.metrics {
b.WriteString(fmt.Sprintf("%s%s %g\n", m.Name, formatLabels(m.Labels), m.Value))
}
fmt.Fprint(w, b.String())
})
}
func formatLabels(l map[string]string) string {
if len(l) == 0 {
return " "
}
keys := make([]string, 0, len(l))
for k := range l {
keys = append(keys, k)
}
sort.Strings(keys)
parts := make([]string, len(keys))
for i, k := range keys {
parts[i] = fmt.Sprintf(`%s="%s"`, k, l[k])
}
return " {" + strings.Join(parts, ",") + "}"
}
func (e *PrometheusExporter) Start() error {
return http.ListenAndServe(e.addr, e.Handler())
}
func (e *PrometheusExporter) Addr() string {
return e.addr
}

5
go.mod Normal file
View File

@@ -0,0 +1,5 @@
module git.govekar.net/LovroG05/nvllama-watchdog
go 1.27.1
require gopkg.in/yaml.v3 v3.0.1

3
go.sum Normal file
View File

@@ -0,0 +1,3 @@
gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

84
main.go Normal file
View File

@@ -0,0 +1,84 @@
package main
import (
"context"
"log"
"os/signal"
"syscall"
"time"
"git.govekar.net/LovroG05/nvllama-watchdog/config"
"git.govekar.net/LovroG05/nvllama-watchdog/core"
"git.govekar.net/LovroG05/nvllama-watchdog/exporters"
"git.govekar.net/LovroG05/nvllama-watchdog/readers"
)
func main() {
cfg, err := config.Load("config/config.yaml")
if err != nil {
log.Fatalf("config: %v", err)
}
// Build readers
var rs []core.Reader
for _, rc := range cfg.Readers {
switch rc.Type {
case "nvidia_smi":
rs = append(rs, readers.NewNvidiaSmiReader())
}
}
// Build exporters
var es []core.Exporter
var prom *exporters.PrometheusExporter
for _, ec := range cfg.Exporters {
switch ec.Type {
case "prometheus":
prom = exporters.NewPrometheusExporter(ec.Addr)
es = append(es, prom)
case "dev":
es = append(es, exporters.NewDevExporter(ec.JSONFile))
case "telegraf":
// es = append(es, exporters.NewTelegraf(ec.Addr))
}
}
// Start Prometheus HTTP listener in background
if prom != nil {
go func() {
log.Printf("prometheus exporter listening on %s", prom.Addr())
if err := prom.Start(); err != nil {
log.Fatalf("prometheus: %v", err)
}
}()
}
// Scrape loop
ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
defer stop()
ticker := time.NewTicker(cfg.ScrapeInterval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
var all []core.Metric
for _, r := range rs {
ms, err := r.Read(ctx)
if err != nil {
log.Printf("reader %s: %v", r.Name(), err)
continue
}
all = append(all, ms...)
}
for _, e := range es {
if err := e.Write(ctx, all); err != nil {
log.Printf("exporter %s: %v", e.Name(), err)
}
}
}
}
}

53
readers/nvidia_smi.go Normal file
View File

@@ -0,0 +1,53 @@
// readers/nvidia_smi.go
package readers
import (
"context"
"os/exec"
"strconv"
"strings"
"time"
"git.govekar.net/LovroG05/nvllama-watchdog/core"
)
type NvidiaSmiReader struct{}
func NewNvidiaSmiReader() *NvidiaSmiReader { return &NvidiaSmiReader{} }
func (r *NvidiaSmiReader) Name() string { return "nvidia_smi" }
func (r *NvidiaSmiReader) Read(ctx context.Context) ([]core.Metric, error) {
query := "index,name,utilization.gpu,memory.used,memory.total,temperature.gpu,power.draw"
out, err := exec.CommandContext(ctx, "nvidia-smi",
"--query-gpu="+query, "--format=csv,noheader,nounits").Output()
if err != nil {
return nil, err
}
var metrics []core.Metric
for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") {
parts := strings.Split(line, ",")
if len(parts) < 7 {
continue
}
idx := strings.TrimSpace(parts[0])
name := strings.TrimSpace(parts[1])
util, _ := strconv.ParseFloat(strings.TrimSpace(parts[2]), 64)
memUsed, _ := strconv.ParseFloat(strings.TrimSpace(parts[3]), 64)
memTotal, _ := strconv.ParseFloat(strings.TrimSpace(parts[4]), 64)
temp, _ := strconv.ParseFloat(strings.TrimSpace(parts[5]), 64)
power, _ := strconv.ParseFloat(strings.TrimSpace(parts[6]), 64)
base := map[string]string{"gpu": idx, "gpu_name": name}
metrics = append(metrics,
core.Metric{Name: "gpu_utilization_percent", Value: util, Labels: base, Timestamp: time.Now()},
core.Metric{Name: "gpu_memory_used_mib", Value: memUsed, Labels: base, Timestamp: time.Now()},
core.Metric{Name: "gpu_memory_total_mib", Value: memTotal, Labels: base, Timestamp: time.Now()},
core.Metric{Name: "gpu_temperature_celsius", Value: temp, Labels: base, Timestamp: time.Now()},
core.Metric{Name: "gpu_power_watts", Value: power, Labels: base, Timestamp: time.Now()},
)
}
return metrics, nil
}