feat: use Cloud API access scopes instead of Service Account by defau…

…lt auth

Signed-off-by: Valentin Pichard <7628998+w3st3ry@users.noreply.github.com>

README.md

@@ -7,7 +7,7 @@ Tiny Go binary that aims to export Nvidia GPU metrics to GCP monitoring, based o
 ## Requirements ⚓
 
 * Your machine must be a GCE (Google Compute Engine) instance.
-* Your Service Account must have the `Monitoring Metric Writer` permission.
+* The `Cloud API access scopes` of the instance or `Service Account` must have the `Monitoring Metric Writer` permission.
 * You need the `nvidia-smi` binary installed on your GCE instance.
 
 
@@ -17,7 +17,7 @@ Protip: You can use a [machine learning image](https://cloud.google.com/ai-platf
 
 If you're root, you can install the latest binary version using the following script:
 ```bash
-$ bash -e < <(curl -sSL https://raw.githubusercontent.com/instadeepai/gcp-gpu-metrics/master/install-latest.sh)
+$ bash < <(curl -sSL https://raw.githubusercontent.com/instadeepai/gcp-gpu-metrics/master/install-latest.sh)
 ```
 
 Or, you can download a [release/binary from this page](https://github.com/instadeepai/gcp-gpu-metrics/releases) and install it manually.
@@ -32,7 +32,7 @@ $ gcp-gpu-metrics
 
 Available flags:
 
-* `--service-account-path string` | GCP service account path. (default "./service-account.json")
+* `--service-account-path string` | GCP service account path. (default "")
 * `--metrics-interval uint` | Fetch metrics interval in seconds. (default 10)
 * `--enable-nvidiasmi-pm` | Enable persistence mod for nvidia-smi. (default false)
 * `--version` | Display current version/release and commit hash.

hack/gcp-gpu-metrics.service

@@ -8,7 +8,7 @@ StartLimitInterval=3
 Type=simple
 Restart=on-failure
 RestartSec=10s
-ExecStart=/usr/local/bin/gcp-gpu-metrics --enable-nvidiasmi-pm --service-account-path /root/service-account.json
+ExecStart=/usr/local/bin/gcp-gpu-metrics --enable-nvidiasmi-pm
 
 [Install]
 WantedBy=multi-user.target

install-latest.sh

@@ -1,4 +1,6 @@
-#!/bin/bash -e
+#!/bin/bash
+
+set -e
 
 pushd /tmp/
 

main.go

@@ -11,8 +11,8 @@ import (
 var (
 	// flags related
 
-	flagDisplayVersion       bool
-	flagServiceAccountPath   string = "./service-account.json"
+	flagDisplayVersion       bool   = false
+	flagServiceAccountPath   string = ""
 	flagFetchMetricsInterval uint64 = 10
 	flagEnableNvidiasmipm    bool   = false
 
@@ -54,7 +54,7 @@ func evaluateEnvVars() {
 func main() {
 	evaluateEnvVars()
 
-	flag.BoolVar(&flagDisplayVersion, "version", false, "Display current version/release and commit hash.")
+	flag.BoolVar(&flagDisplayVersion, "version", flagDisplayVersion, "Display current version/release and commit hash.")
 	flag.StringVar(&flagServiceAccountPath, "service-account-path", flagServiceAccountPath, "GCP service account path.")
 	flag.Uint64Var(&flagFetchMetricsInterval, "metrics-interval", flagFetchMetricsInterval, "Fetch metrics interval in seconds.")
 	flag.BoolVar(&flagEnableNvidiasmipm, "enable-nvidiasmi-pm", flagEnableNvidiasmipm, "Enable persistant mod for nvidia-smi.")

metrics.go

@@ -31,7 +31,15 @@ type service struct {
 
 func newService(slog *syslog.Writer) (*service, error) {
 	ctx := context.Background()
-	client, err := monitoring.NewMetricClient(ctx, option.WithCredentialsFile(flagServiceAccountPath))
+
+	var client *monitoring.MetricClient
+	var err error
+
+	if flagServiceAccountPath == "" {
+		client, err = monitoring.NewMetricClient(ctx)
+	} else {
+		client, err = monitoring.NewMetricClient(ctx, option.WithCredentialsFile(flagServiceAccountPath))
+	}
 	if err != nil {
 		return nil, err
 	}