feat: add loki health command (backport release-3.6.x) (#20590)

Co-authored-by: Jackson Coelho <fcjack@gmail.com>

cmd/loki/health.go

@@ -0,0 +1,72 @@
+package main
+
+import (
+	"fmt"
+	"net/http"
+	"os"
+	"regexp"
+	"time"
+)
+
+const (
+	healthFlag       = "health"
+	defaultHealthURL = "http://localhost:3100/ready"
+	healthTimeout    = 5 * time.Second
+)
+
+// CheckHealth checks if args contain the -health flag
+func CheckHealth(args []string) bool {
+	pattern := regexp.MustCompile(`^-+` + healthFlag + `$`)
+	for _, a := range args {
+		if pattern.MatchString(a) {
+			return true
+		}
+	}
+	return false
+}
+
+// RunHealthCheck performs a health check against the /ready endpoint
+// Returns exit code 0 if healthy, 1 if unhealthy
+func RunHealthCheck(args []string) int {
+
+	url := getHealthURL(args)
+
+	client := &http.Client{
+		Timeout: healthTimeout,
+	}
+
+	resp, err := client.Get(url)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "Health check failed: %v\n", err)
+		return 1
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode == http.StatusOK {
+		fmt.Println("Loki is healthy")
+		return 0
+	}
+
+	fmt.Fprintf(os.Stderr, "Loki is unhealthy: status code %d\n", resp.StatusCode)
+	return 1
+}
+
+// getHealthURL extracts the URL from args or returns default
+// Looks for -health.url=<url> or -health.url <url>
+func getHealthURL(args []string) string {
+	urlPattern := regexp.MustCompile(`^-+health\.url[=:]?(.*)$`)
+	healthArgPattern := regexp.MustCompile(`^-`)
+
+	for i, a := range args {
+		if matches := urlPattern.FindStringSubmatch(a); matches != nil {
+			if matches[1] != "" {
+				return matches[1]
+			}
+			// Check next argument for the URL value
+			if i+1 < len(args) && !healthArgPattern.MatchString(args[i+1]) {
+				return args[i+1]
+			}
+		}
+	}
+	return defaultHealthURL
+}

cmd/loki/main.go

@@ -33,6 +33,12 @@ func main() {
 
 	var config loki.ConfigWrapper
 
+	// Health check command - runs before any config parsing
+	// Usage: loki -health [-health.url=http://localhost:3100/ready]
+	if CheckHealth(os.Args[1:]) {
+		os.Exit(RunHealthCheck(os.Args[1:]))
+	}
+
 	if loki.PrintVersion(os.Args[1:]) {
 		fmt.Println(version.Print("loki"))
 		os.Exit(0)

production/docker/docker-compose.yaml

@@ -14,10 +14,6 @@ services:
   init:
     image: &lokiImage grafana/loki:latest
     user: root
-    entrypoint:
-      - "chown"
-      - "10001:10001"
-      - "/loki"
     volumes:
       - ./loki:/loki
     networks:
@@ -57,6 +53,7 @@ services:
   # for testing purposes only, disable in production
   log-generator:
     image: mingrammer/flog
+    platform: linux/amd64
     command:
       - --loop
       - --format=json
@@ -132,6 +129,13 @@ services:
     deploy:
       mode: replicated
       replicas: 3
+    # Native healthcheck using loki -health command (no shell required)
+    healthcheck:
+      test: ["CMD", "/usr/bin/loki", "-health"]
+      start_period: 30s
+      interval: 30s
+      timeout: 10s
+      retries: 3
 
   loki-write:
     image: *lokiImage
@@ -154,6 +158,13 @@ services:
     deploy:
       mode: replicated
       replicas: 3
+    # Native healthcheck using loki -health command (no shell required)
+    healthcheck:
+      test: ["CMD", "/usr/bin/loki", "-health"]
+      start_period: 30s
+      interval: 30s
+      timeout: 10s
+      retries: 3
 
   loki-backend:
     image: *lokiImage
@@ -177,6 +188,13 @@ services:
     deploy:
       mode: replicated
       replicas: 3
+    # # Native healthcheck using loki -health command (no shell required)
+    healthcheck:
+      test: ["CMD", "/usr/bin/loki", "-health"]
+      start_period: 30s
+      interval: 30s
+      timeout: 10s
+      retries: 3
 
   # alertmanager to enable receiving alerts
   alertmanager: