From e0bf4674bd84bdef6e8502aacf8f7525032256ef Mon Sep 17 00:00:00 2001
From: Gregor Zeitlinger <gregor.zeitlinger@grafana.com>
Date: Wed, 23 Apr 2025 12:00:38 +0200
Subject: [PATCH] otel health check extension is faster than trying to scrape
 prom metrics

---
 docker/Dockerfile          |  2 +-
 docker/otelcol-config.yaml |  6 ++++++
 docker/run-all.sh          | 30 +-----------------------------
 3 files changed, 8 insertions(+), 30 deletions(-)

diff --git a/docker/Dockerfile b/docker/Dockerfile
index 35cea57b..a7157923 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -40,7 +40,7 @@ RUN yum install -y unzip dos2unix jq
 # installs for the final image
 # see https://github.com/thomasdarimont/keycloak/blob/main/docs/guides/server/containers.adoc#installing-additional-rpm-packages
 RUN mkdir -p /mnt/rootfs
-RUN dnf install --installroot /mnt/rootfs curl-minimal jq procps --releasever 9 --setopt install_weak_deps=false --nodocs -y && \
+RUN dnf install --installroot /mnt/rootfs curl-minimal --releasever 9 --setopt install_weak_deps=false --nodocs -y && \
     dnf --installroot /mnt/rootfs clean all && \
     rpm --root /mnt/rootfs -e --nodeps setup
 
diff --git a/docker/otelcol-config.yaml b/docker/otelcol-config.yaml
index ba4fa425..beee00be 100644
--- a/docker/otelcol-config.yaml
+++ b/docker/otelcol-config.yaml
@@ -13,6 +13,11 @@ receivers:
           static_configs:
             - targets: ["localhost:8888"]
 
+extensions:
+  health_check:
+    endpoint: 0.0.0.0:13133
+    path: "/ready"
+
 processors:
   batch:
 
@@ -41,6 +46,7 @@ exporters:
     verbosity: detailed
 
 service:
+  extensions: [health_check]
   pipelines:
     traces:
       receivers: [otlp]
diff --git a/docker/run-all.sh b/docker/run-all.sh
index 2f589b75..b8a415de 100755
--- a/docker/run-all.sh
+++ b/docker/run-all.sh
@@ -43,15 +43,13 @@ services["loki"]="http://localhost:3100/ready"
 services["prometheus"]="http://localhost:9090/api/v1/status/runtimeinfo"
 services["tempo"]="http://localhost:3200/ready"
 services["pyroscope"]="http://localhost:4040/ready"
+services["otelcol"]="http://localhost:13133/ready"
 
 # Initialize service_ready status to false for all services
 for service in "${!services[@]}"; do
 	service_ready[$service]=false
 done
 
-# Also check OpenTelemetry collector separately (since it uses a different check method)
-service_ready["otelcol"]=false
-
 # Function to check if a service is ready
 check_service_ready() {
 	local service=$1
@@ -80,29 +78,6 @@ check_service_ready() {
 	return 1
 }
 
-# Function to check if OpenTelemetry collector is ready
-check_otelcol_ready() {
-	# Skip if already marked as ready
-	if [[ ${service_ready["otelcol"]} == true ]]; then
-		return 0
-	fi
-
-	# Check if collector is ready via Prometheus metric
-	if curl -sg 'http://localhost:9090/api/v1/query?query=otelcol_process_uptime_total{}' 2>/dev/null | jq -r .data.result[0].value[1] 2>/dev/null | grep '[0-9]' >/dev/null; then
-		# Calculate and display startup time
-		end_time=$(date +%s)
-		# shellcheck disable=SC2154
-		otelcol_start_time=${start_time_otelcol}
-		elapsed=$((end_time - otelcol_start_time))
-		elapsed_times["otelcol"]=$elapsed
-		service_ready["otelcol"]=true
-		echo "OpenTelemetry collector is up and running. Startup time: ${elapsed} seconds"
-		return 0
-	fi
-
-	return 1
-}
-
 # Wait for all services to be ready
 all_ready=false
 while [[ $all_ready == false ]]; do
@@ -111,9 +86,6 @@ while [[ $all_ready == false ]]; do
 		check_service_ready "$service" "${services[$service]}"
 	done
 
-	# Check OpenTelemetry collector
-	check_otelcol_ready
-
 	# Check if all services are ready
 	all_ready=true
 	for service in "${!service_ready[@]}"; do