Merge branch 'fix/monitoring-restart-policies' into 'main'

fix: keep monitoring services running after reboot

See merge request postgres-ai/postgresai!246

Author: Sarumyan

cli/test/config-consistency.test.ts

@@ -24,6 +24,10 @@ type GrafanaDatasourceConfig = {
   datasources?: GrafanaDatasource[];
 };
 
+type DockerComposeConfig = {
+  services?: Record<string, { restart?: unknown }>;
+};
+
 // These UIDs are referenced by provisioned Grafana dashboards.
 // Changing them without updating dashboard JSON silently breaks panels.
 const expectedGrafanaDatasourceUids = new Map([
@@ -259,6 +263,32 @@ describe("Config consistency", () => {
     expect(envExample).toContain("openssl rand -base64 18");
   });
 
+  test("long-running Docker Compose services survive host reboot", () => {
+    const composeConfig = Bun.YAML.parse(
+      readFileSync(composePath, "utf8")
+    ) as DockerComposeConfig;
+    const services = composeConfig.services ?? {};
+
+    // All services should survive host reboot unless they are one-shot setup jobs.
+    const oneShotServices = new Set(["config-init", "sources-generator"]);
+    const serviceEntries = Object.entries(services);
+
+    expect(serviceEntries.length).toBeGreaterThan(0);
+
+    for (const serviceName of oneShotServices) {
+      expect(services[serviceName]).toBeDefined();
+      expect(services[serviceName]?.restart).toBeUndefined();
+    }
+
+    for (const [serviceName, serviceConfig] of serviceEntries) {
+      if (oneShotServices.has(serviceName)) {
+        continue;
+      }
+
+      expect(serviceConfig.restart).toBe("unless-stopped");
+    }
+  });
+
   test("pgwatch presets only reference configured metrics", () => {
     const pgwatchConfig = loadPgwatchConfig();
     const metricNames = new Set(Object.keys(pgwatchConfig.metrics));

docker-compose.yml

@@ -66,6 +66,7 @@ services:
     depends_on:
       config-init:
         condition: service_completed_successfully
+    restart: unless-stopped
 
   # Hot standby used by xmin-horizon integration coverage when explicitly started.
   target-standby:
@@ -99,6 +100,7 @@ services:
       - target_standby_data:/var/lib/postgresql/data
     depends_on:
       - target-db
+    restart: unless-stopped
 
   # Postgres Sink - Storage for metrics in PostgreSQL format
   # Note: pg_hba.conf is configured to allow passwordless connections (trust)
@@ -297,6 +299,7 @@ services:
       pgwatch-prometheus:
         condition: service_started
     command: ["bash", "/postgres_ai_configs/scripts/postgres-reports.sh"]
+    restart: unless-stopped
 
   # Self-monitoring components (may have reduced functionality on macOS)