Implement stack health diagnostics, timeout adjustments, and cleanup dependencies

2026-04-22 15:00:12 +02:00
parent f4952f16e6
commit fd830deda9
7 changed files with 54 additions and 23 deletions
@@ -3,16 +3,6 @@ FROM ${BASE_IMAGE}

 WORKDIR /opt/truckwash-edge-agent

-RUN set -eux; \
-    apt-get update; \
-    apt-get install -y --no-install-recommends \
-        curl \
-        ca-certificates \
-        libcurl4-openssl-dev \
-        libsqlite3-dev; \
-    docker-php-ext-install curl sqlite3; \
-    rm -rf /var/lib/apt/lists/*
-
 COPY agent.php /opt/truckwash-edge-agent/agent.php

 ENTRYPOINT ["php", "/opt/truckwash-edge-agent/agent.php"]
@@ -3,15 +3,6 @@ FROM ${BASE_IMAGE}

 WORKDIR /opt/truckwash-edge-agent

-RUN set -eux; \
-    apt-get update; \
-    apt-get install -y --no-install-recommends \
-        curl \
-        ca-certificates \
-        libcurl4-openssl-dev; \
-    docker-php-ext-install curl; \
-    rm -rf /var/lib/apt/lists/*
-
 COPY lan-worker.php /opt/truckwash-edge-agent/lan-worker.php

 CMD ["php", "-S", "0.0.0.0:8090", "/opt/truckwash-edge-agent/lan-worker.php"]
@@ -1,3 +1,5 @@
+version: "2.4"
+
 services:
  lan-worker:
    build:
@@ -8,6 +8,7 @@ COMPOSE_FILE="$INSTALL_DIR/docker-compose.gateway.yml"
 RUNTIME_DIR="$INSTALL_DIR/runtime"
 ROLLBACK_STATUS_PATH="$RUNTIME_DIR/rollback-status.json"
 STAGED_UPDATE_PATH="$RUNTIME_DIR/staged-update.json"
+STACK_HEALTHCHECK_TIMEOUT_SECONDS="${STACK_HEALTHCHECK_TIMEOUT_SECONDS:-120}"

 log() {
  printf '[gateway-launcher] %s\n' "$1"
@@ -28,6 +29,17 @@ compose_cmd() {
  return 1
 }

+print_compose_diagnostics() {
+  log "docker ps --format '{{.Names}} {{.Status}}'"
+  docker ps --format '{{.Names}} {{.Status}}' || true
+
+  log "compose ps"
+  compose_cmd -f "$COMPOSE_FILE" ps || true
+
+  log "compose logs --tail=80"
+  compose_cmd -f "$COMPOSE_FILE" logs --tail=80 || true
+}
+
 config_value() {
  local key="$1"
  local fallback="${2:-}"
@@ -137,14 +149,30 @@ healthcheck_stack() {
  curl -fsS http://127.0.0.1:8090/health >/dev/null
 }

+wait_for_stack_health() {
+  local timeout_seconds="${1:-$STACK_HEALTHCHECK_TIMEOUT_SECONDS}"
+  local elapsed=0
+
+  while [ "$elapsed" -lt "$timeout_seconds" ]; do
+    if healthcheck_stack; then
+      return 0
+    fi
+
+    sleep 2
+    elapsed=$((elapsed + 2))
+  done
+
+  return 1
+}
+
 reconcile_stack() {
  ensure_dirs
  log "Reconciling compose stack"
  apply_stack
-  sleep 5

-  if ! healthcheck_stack; then
+  if ! wait_for_stack_health "$STACK_HEALTHCHECK_TIMEOUT_SECONDS"; then
    log "Healthcheck failed after compose rollout; reverting to previous artifacts"
+    print_compose_diagnostics
    rollback_stack
    return 1
  fi
@@ -10,7 +10,7 @@ WorkingDirectory=/opt/truckwash-edge-agent
 ExecStart=/opt/truckwash-edge-agent/gateway-launcher.sh up
 ExecReload=/opt/truckwash-edge-agent/gateway-launcher.sh reconcile
 ExecStop=/opt/truckwash-edge-agent/gateway-launcher.sh down
-TimeoutStartSec=300
+TimeoutStartSec=900

 [Install]
 WantedBy=multi-user.target
@@ -4,8 +4,16 @@ it('builds the installer around the compose stack artifacts and management polli
    $managerSource = file_get_contents(app_path('classes/edge_gateway_manager.php'));
    $serviceSource = file_get_contents(app_path('resources/edge-gateway-agent/truckwash-edge-agent.service'));
    $stackServiceSource = file_get_contents(app_path('resources/edge-gateway-agent/truckwash-edge-gateway-stack.service'));
+    $launcherSource = file_get_contents(app_path('resources/edge-gateway-agent/gateway-launcher.sh'));
+    $composeSource = file_get_contents(app_path('resources/edge-gateway-agent/docker-compose.gateway.yml'));
+    $edgeDockerfileSource = file_get_contents(app_path('resources/edge-gateway-agent/Dockerfile.edge-agent'));
+    $workerDockerfileSource = file_get_contents(app_path('resources/edge-gateway-agent/Dockerfile.lan-worker'));

    expect($managerSource)->not->toBeFalse();
+    expect($launcherSource)->not->toBeFalse();
+    expect($composeSource)->not->toBeFalse();
+    expect($edgeDockerfileSource)->not->toBeFalse();
+    expect($workerDockerfileSource)->not->toBeFalse();
    expect($managerSource)->toContain("'operationPollTimeoutSeconds' => self::COMMAND_POLL_TIMEOUT_SECONDS");
    expect($managerSource)->toContain('fetch_http "Verify install token" "__VERIFY_URL__"');
    expect($managerSource)->toContain('fetch_http "Download PHP edge agent" "__AGENT_URL__" "$INSTALL_DIR/agent.php"');
@@ -31,6 +39,18 @@ it('builds the installer around the compose stack artifacts and management polli
    expect($managerSource)->not->toContain('"brokerUrl"');
    expect($serviceSource)->toContain('ExecStart=/usr/bin/php /opt/truckwash-edge-agent/agent.php --config /opt/truckwash-edge-agent/config.json');
    expect($stackServiceSource)->toContain('ExecStart=/opt/truckwash-edge-agent/gateway-launcher.sh up');
+    expect($stackServiceSource)->toContain('TimeoutStartSec=900');
+    expect($launcherSource)->toContain('STACK_HEALTHCHECK_TIMEOUT_SECONDS="${STACK_HEALTHCHECK_TIMEOUT_SECONDS:-120}"');
+    expect($launcherSource)->toContain('wait_for_stack_health');
+    expect($launcherSource)->toContain('compose_cmd -f "$COMPOSE_FILE" logs --tail=80 || true');
+    expect($composeSource)->toContain('version: "2.4"');
+    expect($composeSource)->toContain('condition: service_healthy');
+    expect($edgeDockerfileSource)->toContain('FROM ${BASE_IMAGE}');
+    expect($edgeDockerfileSource)->toContain('COPY agent.php /opt/truckwash-edge-agent/agent.php');
+    expect($edgeDockerfileSource)->not->toContain('docker-php-ext-install');
+    expect($workerDockerfileSource)->toContain('FROM ${BASE_IMAGE}');
+    expect($workerDockerfileSource)->toContain('COPY lan-worker.php /opt/truckwash-edge-agent/lan-worker.php');
+    expect($workerDockerfileSource)->not->toContain('docker-php-ext-install');
    expect($serviceSource)->not->toContain('node /opt/truckwash-edge-agent/agent.mjs');
 });