Remove o11y (#249)

* refactor(all-in-one): optimize image size by ~1.1G

- Replace envoy binary with optimized version from GitHub releases
  (734M -> 163M, saves 571M)
- Remove observability components by default:
  - prometheus (110M)
  - promtail (89M)
  - loki (64M)
  - grafana (305M)
- Add ENVOY_VERSION arg for flexible version control

Total image size reduced from ~2.4G to ~1.3G

Change-Id: I87cfa6299c9b9d9f5a7fd548cb6a997084093a73
Co-developed-by: Claude <noreply@anthropic.com>

* refactor(all-in-one): remove observability services from supervisord

Remove prometheus, promtail, loki, and grafana service configurations
since these components are no longer included in the image.

Change-Id: I7318cd025491da54d76b92a04373124d2017b7df
Co-developed-by: Claude <noreply@anthropic.com>

* refactor(all-in-one): set O11Y default to off in get-ai-gateway.sh

Change O11Y default value from 'on' to 'off' since observability
components are no longer included in the image by default.

Change-Id: I69203aaec31f37258701dedf689603759b728c2a
Co-developed-by: Claude <noreply@anthropic.com>

* refactor(all-in-one): remove unused observability startup scripts

Remove start-promtail.sh, start-grafana.sh, start-prometheus.sh,
and start-loki.sh since these components are no longer included.

Change-Id: Ic09a1dd54247acd939f89472eda19be3a154f510
Co-developed-by: Claude <noreply@anthropic.com>

Author: johnlanni

all-in-one/Dockerfile

@@ -3,20 +3,13 @@ ARG BASE_VERSION=2022-10-27T19-02-22
 ARG CORE_VERSION=2.1.11
 ARG CONSOLE_VERSION=2.1.11
 ARG APISERVER_VERSION=0.0.29
-ARG PROMETHEUS_VERSION=v2.40.7
-ARG PROMTAIL_VERSION=2.9.4
-ARG LOKI_VERSION=2.9.4
-ARG GRAFANA_VERSION=9.3.6
+ARG ENVOY_VERSION=2.1.11
 
 FROM ${HUB}/api-server:${APISERVER_VERSION} AS apiserver
 FROM ${HUB}/higress:${CORE_VERSION} AS controller
 FROM ${HUB}/pilot:${CORE_VERSION} AS pilot
 FROM ${HUB}/gateway:${CORE_VERSION} AS gateway
 FROM ${HUB}/console:${CONSOLE_VERSION} AS console
-FROM ${HUB}/prometheus:${PROMETHEUS_VERSION} AS prometheus
-FROM ${HUB}/promtail:${PROMTAIL_VERSION} AS promtail
-FROM ${HUB}/loki:${LOKI_VERSION} AS loki
-FROM ${HUB}/grafana:${GRAFANA_VERSION} AS grafana
 FROM ${HUB}/eclipse-temurin:21-jre AS jdk
 
 FROM ${HUB}/base:${BASE_VERSION}
@@ -35,7 +28,6 @@ COPY --from=pilot /usr/local/bin/higress-pilot-start.sh /usr/local/bin/higress-p
 COPY --from=gateway /var/lib/istio/envoy/*.json /var/lib/istio/envoy/
 COPY --from=gateway /var/lib/istio/envoy/*.so /var/lib/istio/envoy/
 COPY --from=gateway /usr/local/bin/pilot-agent /usr/local/bin/pilot-agent
-COPY --from=gateway /usr/local/bin/envoy /usr/local/bin/envoy
 COPY --from=gateway /usr/local/bin/higress-proxy-*.sh /usr/local/bin/
 RUN chmod a+x /usr/local/bin/higress-proxy-container-init.sh; \
   sed -i 's/1337/0/g' /usr/local/bin/higress-proxy-container-init.sh; \
@@ -50,27 +42,26 @@ ENV JAVA_HOME=/opt/java/openjdk
 COPY --from=jdk $JAVA_HOME $JAVA_HOME
 ENV PATH="${JAVA_HOME}/bin:${PATH}"
 
-# Install Prometheus
-COPY --from=prometheus /bin/prometheus /usr/local/bin/prometheus
-
-# Install Promtail
-COPY --from=promtail /usr/bin/promtail /usr/local/bin/promtail
-
-# Install Loki
-COPY --from=loki /usr/bin/loki /usr/local/bin/loki
-
-# Install Grafana
-COPY --from=grafana /usr/share/grafana /usr/share/grafana
-COPY --from=grafana /run.sh /usr/local/bin/grafana.sh
-
 # Install supervisord, logrotate, cron and initialize related folders
-RUN arch="$(dpkg --print-architecture)"; arch="${arch##*-}"; \
+# Also download optimized envoy binary
+ARG ENVOY_VERSION
+RUN arch="$(dpkg --print-architecture)"; \
+  case "$arch" in \
+    amd64) envoy_arch="amd64" ;; \
+    arm64) envoy_arch="arm64" ;; \
+    *) echo "Unsupported architecture: $arch" && exit 1 ;; \
+  esac; \
   apt-get update --allow-unauthenticated; \
   apt-get install --no-install-recommends -y --allow-unauthenticated \
     wget supervisor logrotate cron; \
   apt-get upgrade -y --allow-unauthenticated; \
   apt-get clean; \
   rm -rf /var/log/*log /var/lib/apt/lists/* /var/log/apt/* /var/lib/dpkg/*-old /var/cache/debconf/*-old; \
+  wget -q "https://github.com/higress-group/proxy/releases/download/v${ENVOY_VERSION}/envoy-${envoy_arch}.tar.gz" -O /tmp/envoy.tar.gz && \
+  tar -xzf /tmp/envoy.tar.gz -C / && \
+  rm /tmp/envoy.tar.gz && \
+  chmod +x /usr/local/bin/envoy; \
+  arch="${arch##*-}"; \
   wget https://github.com/mikefarah/yq/releases/latest/download/yq_linux_$arch -O /usr/local/bin/yq && chmod +x /usr/local/bin/yq; \
   mkdir -p /var/log/higress; \
   mkdir /data;
@@ -83,10 +74,6 @@ COPY ./gateway/podinfo /etc/istio/pod
 COPY ./scripts /usr/local/bin
 COPY ./apiserver/config /app/kubeconfig
 COPY ./config /opt/data/defaultConfig
-COPY ./prometheus /etc/prometheus
-COPY ./promtail /etc/promtail
-COPY ./loki /etc/loki
-COPY ./grafana /etc/grafana
 
 EXPOSE 8080 8443 8001
 

all-in-one/get-ai-gateway.sh

@@ -1582,7 +1582,7 @@ AUTO_ROUTING_DEFAULT_MODEL=${AUTO_ROUTING_DEFAULT_MODEL}"
 
   cat <<EOF >$DATA_FOLDER/$CONFIG_FILENAME
 MODE=full
-O11Y=on
+O11Y=off
 CONFIG_TEMPLATE=ai-gateway
 GATEWAY_HTTP_PORT=${GATEWAY_HTTP_PORT}
 GATEWAY_HTTPS_PORT=${GATEWAY_HTTPS_PORT}

all-in-one/scripts/start-grafana.sh

@@ -66,50 +66,3 @@ stdout_logfile_maxbytes=10MB
 redirect_stderr=true
 environment=
 
-[program:prometheus]
-directory=/
-command=bash /usr/local/bin/start-prometheus.sh
-priority=2000
-autostart=true
-autorestart=unexpected
-startsecs=1
-stdout_logfile=/var/log/higress/prometheus.log
-stdout_logfile_maxbytes=10MB
-redirect_stderr=true
-environment=
-
-[program:promtail]
-directory=/
-command=bash /usr/local/bin/start-promtail.sh
-priority=2100
-autostart=true
-autorestart=unexpected
-startsecs=1
-stdout_logfile=/var/log/higress/promtail.log
-stdout_logfile_maxbytes=10MB
-redirect_stderr=true
-environment=HOSTNAME="higress-gateway"
-
-[program:loki]
-directory=/
-command=bash /usr/local/bin/start-loki.sh
-priority=2200
-autostart=true
-autorestart=unexpected
-startsecs=1
-stdout_logfile=/var/log/higress/loki.log
-stdout_logfile_maxbytes=10MB
-redirect_stderr=true
-environment=
-
-[program:grafana]
-directory=/
-command=bash /usr/local/bin/start-grafana.sh
-priority=3000
-autostart=true
-autorestart=unexpected
-startsecs=1
-stdout_logfile=/var/log/higress/grafana.log
-stdout_logfile_maxbytes=10MB
-redirect_stderr=true
-environment=