Add a healthcheck to the Docker container image

core/docker/Dockerfile

@@ -35,3 +35,5 @@ EXPOSE 8080
 USER trino:trino
 ENV LANG en_US.UTF-8
 CMD ["/usr/lib/trino/bin/run-trino"]
+HEALTHCHECK --interval=10s --timeout=5s --start-period=10s \
+  CMD /usr/lib/trino/bin/health-check

core/docker/bin/health-check

@@ -0,0 +1,32 @@
+#!/bin/bash
+
+set -euo pipefail
+
+function get_property() {
+    grep "^$1=" "$2" | cut -d'=' -f2
+}
+
+scheme=http
+port=8080
+
+config=/etc/trino/config.properties
+# prefer to use http even if https is enabled
+if [ "$(get_property 'http-server.http.enabled' "$config")" == "false" ]; then
+    scheme=https
+    port=$(get_property http-server.https.port "$config")
+else
+    port=$(get_property http-server.http.port "$config")
+fi
+
+endpoint="$scheme://localhost:$port/v1/info"
+
+# add --insecure to disable certificate verification in curl, in case a self-signed certificate is being used
+if ! info=$(curl --fail --silent --show-error --insecure "$endpoint"); then
+    echo >&2 "Server is not responding to requests"
+    exit 1
+fi
+
+if ! grep -q '"starting":\s*false' <<<"$info" >/dev/null; then
+    echo >&2 "Server is starting"
+    exit 1
+fi

core/docker/container-test.sh

@@ -1,5 +1,7 @@
+#!/usr/bin/env bash
+
 function cleanup {
-    if [[ ! -z ${CONTAINER_ID:-} ]]; then
+    if [[ -n ${CONTAINER_ID:-} ]]; then
         docker rm -f "${CONTAINER_ID}"
     fi
 }
@@ -14,12 +16,11 @@ function test_trino_starts {
     local CONTAINER_NAME=$1
     local PLATFORM=$2
     # We aren't passing --rm here to make sure container is available for inspection in case of failures
-    CONTAINER_ID=$(docker run -d --platform ${PLATFORM} "${CONTAINER_NAME}")
+    CONTAINER_ID=$(docker run -d --platform "${PLATFORM}" "${CONTAINER_NAME}")
 
     set +e
     I=0
-    until RESULT=$(docker exec "${CONTAINER_ID}" trino --execute "SELECT 'success'" 2>/dev/null)
-    do
+    until docker inspect "${CONTAINER_ID}" --format "{{json .State.Health.Status }}" | grep -q '"healthy"'; do
         if [[ $((I++)) -ge ${QUERY_RETRIES} ]]; then
             echo "🚨 Too many retries waiting for Trino to start"
             echo "Logs from ${CONTAINER_ID} follow..."
@@ -28,6 +29,9 @@ function test_trino_starts {
         fi
         sleep ${QUERY_PERIOD}
     done
+    if ! RESULT=$(docker exec "${CONTAINER_ID}" trino --execute "SELECT 'success'" 2>/dev/null); then
+        echo "🚨 Failed to execute a query after Trino container started"
+    fi
     set -e
 
     cleanup
@@ -41,17 +45,17 @@ function test_javahome {
     local CONTAINER_NAME=$1
     local PLATFORM=$2
     # Check if JAVA_HOME works
-    docker run --rm --platform ${PLATFORM} "${CONTAINER_NAME}" \
-        /bin/bash -c '$JAVA_HOME/bin/java -version' &> /dev/null
+    docker run --rm --platform "${PLATFORM}" "${CONTAINER_NAME}" \
+        /bin/bash -c '$JAVA_HOME/bin/java -version' &>/dev/null
 
-    [[ "$?" == "0" ]]
+    [[ $? == "0" ]]
 }
 
 function test_container {
     local CONTAINER_NAME=$1
     local PLATFORM=$2
     echo "🐢 Validating ${CONTAINER_NAME} on platform ${PLATFORM}..."
-    test_javahome ${CONTAINER_NAME} ${PLATFORM}
-    test_trino_starts ${CONTAINER_NAME} ${PLATFORM}
+    test_javahome "${CONTAINER_NAME}" "${PLATFORM}"
+    test_trino_starts "${CONTAINER_NAME}" "${PLATFORM}"
     echo "🎉 Validated ${CONTAINER_NAME} on platform ${PLATFORM}"
 }