Atualizou functions suspend/resume

X-Lovable-Edit-ID: edt-90e6d40e-f2d7-42a3-bf33-8354f6b62c2a Co-authored-by: domfelipe <53182096+domfelipe@users.noreply.github.com>
2026-08-07 12:36:50 +00:00 · 2026-04-23 19:06:20 +00:00 · 2026-04-23 19:06:20 +00:00 · a72b4b6624
commit a72b4b6624
parent 50a0c9042e 074513e5ff
5 changed files with 103 additions and 68 deletions
--- a/README.md
+++ b/README.md
@ -24,6 +24,16 @@ Já configuradas via `.env` (gerado automaticamente pelo Lovable Cloud):
 - `PADDLE_API_KEY` / `PADDLE_WEBHOOK_SECRET`
 - `SUPABASE_SERVICE_ROLE_KEY`
 - `RAILWAY_API_TOKEN` — token da Railway Public API, usado por `provision-agent`/`suspend-agent`/`resume-agent` para criar e gerenciar containers Hermes.
+
+  > **Suspend/Resume via flag, não via stop.** Railway não permite parar containers sem redeploy. As Edge Functions `suspend-agent`/`resume-agent` usam a env var `HERMES_SUSPENDED` + `serviceInstanceRedeploy`:
+  > - `suspend` → upsert `HERMES_SUSPENDED=true` e redeploy → container entra em `sleep infinity`.
+  > - `resume` → upsert `HERMES_SUSPENDED=""` e redeploy → container sobe normal.
+  >
+  > Isso depende do start command verificar a flag. Novos serviços já são provisionados com o comando correto (`HERMES_START_COMMAND` em `_shared/railway.ts`):
+  > ```bash
+  > /bin/bash -c 'if [ "$HERMES_SUSPENDED" = "true" ]; then echo "Agent suspended" && sleep infinity; fi && if [ -n "$HERMES_SOUL_OVERRIDE" ]; then echo "$HERMES_SOUL_OVERRIDE" > /opt/data/SOUL.md; fi && /opt/hermes/docker/entrypoint.sh gateway run'
+  > ```
+  > **Para serviços Railway existentes (provisionados antes desta mudança):** atualize manualmente o Start Command via Railway UI/Agent para o comando acima — caso contrário `suspend-agent` apenas marcará `status='suspended'` no banco mas o container continuará rodando.
 - `OPENROUTER_API_KEY` — **obrigatório**. Injetado em cada container Hermes provisionado para que o agente possa chamar os modelos `openrouter/google/gemma-4-*-it`. Sem isso o `provision-agent` retorna 500.
 - `ADMIN_TELEGRAM_BOT_TOKEN` — token do bot de admin (ex: `@mika_test2_bot`) usado pelo `payments-webhook` para enviar notificações de novos clientes, falhas de pagamento e cancelamentos.
 - `ADMIN_TELEGRAM_CHAT_ID` — chat ID do admin que recebe as notificações (ex: `179720882`).
--- a/supabase/functions/_shared/railway.ts
+++ b/supabase/functions/_shared/railway.ts
@ -2,6 +2,18 @@
 // Docs: https://docs.railway.com/reference/public-api
 const RAILWAY_GRAPHQL = "https://backboard.railway.app/graphql/v2";

+/**
+ * Start command padrão dos containers Hermes.
+ * - Verifica HERMES_SUSPENDED no início: se true, dorme infinitamente (agente "pausado")
+ * - Aplica HERMES_SOUL_OVERRIDE em /opt/data/SOUL.md se presente
+ * - Inicia o gateway Hermes
+ *
+ * IMPORTANTE: este comando deve ser idêntico ao configurado nos serviços Railway
+ * existentes. Para serviços antigos, atualize manualmente via UI/Agent do Railway.
+ */
+export const HERMES_START_COMMAND =
+  `/bin/bash -c 'if [ "$HERMES_SUSPENDED" = "true" ]; then echo "Agent suspended" && sleep infinity; fi && if [ -n "$HERMES_SOUL_OVERRIDE" ]; then echo "$HERMES_SOUL_OVERRIDE" > /opt/data/SOUL.md; fi && /opt/hermes/docker/entrypoint.sh gateway run'`;
+
 export interface RailwayError {
  message: string;
  path?: string[];
@ -92,28 +104,14 @@ export async function configureRailwayService(opts: {
  }

  // Agora as variáveis. Railway recomenda variableUpsert por chave.
-  const variableUpsert = `
-    mutation VariableUpsert($input: VariableUpsertInput!) {
-      variableUpsert(input: $input)
-    }
-  `;
  for (const [name, value] of Object.entries(opts.variables)) {
-    const r = await railwayQuery(
-      variableUpsert,
-      {
-        input: {
-          projectId: undefined, // será inferido pelo serviceId+environmentId
-          environmentId: opts.environmentId,
-          serviceId: opts.serviceId,
-          name,
-          value,
-        },
-      },
-      opts.token,
-    );
-    if (r.errors?.length) {
-      throw new Error(`variableUpsert(${name}) failed: ${JSON.stringify(r.errors)}`);
-    }
+    await upsertRailwayVariable({
+      token: opts.token,
+      serviceId: opts.serviceId,
+      environmentId: opts.environmentId,
+      name,
+      value,
+    });
  }
 }

@ -138,48 +136,67 @@ export async function deployRailwayService(opts: {
 }

 /**
- * Suspende ou retoma um serviço Railway usando `sleepApplication`.
- * Railway não aceita `numReplicas: 0` via serviceInstanceUpdate — o caminho
- * suportado para pausar é `sleepApplication: true` (e `false` para acordar).
- * `replicas <= 0` => sleep, `>= 1` => wake.
+ * Upsert de uma única variável de ambiente no serviço Railway.
+ * Para "remover" o efeito de uma variável boolean, passe value="" (string vazia).
 */
-export async function setRailwayReplicas(opts: {
+export async function upsertRailwayVariable(opts: {
  token: string;
  serviceId: string;
  environmentId: string;
-  replicas: number;
+  projectId?: string;
+  name: string;
+  value: string;
 }): Promise<void> {
-  const sleep = opts.replicas <= 0;
  const mutation = `
-    mutation ServiceInstanceUpdate($serviceId: String!, $environmentId: String!, $input: ServiceInstanceUpdateInput!) {
-      serviceInstanceUpdate(serviceId: $serviceId, environmentId: $environmentId, input: $input)
+    mutation VariableUpsert($input: VariableUpsertInput!) {
+      variableUpsert(input: $input)
    }
  `;
-  const res = await railwayQuery(
-    mutation,
-    {
-      serviceId: opts.serviceId,
-      environmentId: opts.environmentId,
-      input: { sleepApplication: sleep },
-    },
-    opts.token,
-  );
-  if (res.errors?.length) {
-    throw new Error(`setRailwayReplicas (sleepApplication=${sleep}) failed: ${JSON.stringify(res.errors)}`);
-  }
+  const input: Record<string, unknown> = {
+    environmentId: opts.environmentId,
+    serviceId: opts.serviceId,
+    name: opts.name,
+    value: opts.value,
+  };
+  if (opts.projectId) input.projectId = opts.projectId;

-  // Força redeploy para aplicar imediatamente o novo estado de sleep
-  try {
-    await deployRailwayService({
-      token: opts.token,
-      serviceId: opts.serviceId,
-      environmentId: opts.environmentId,
-    });
-  } catch (e) {
-    console.warn("setRailwayReplicas: redeploy after sleep change failed (non-fatal):", e);
+  const res = await railwayQuery(mutation, { input }, opts.token);
+  if (res.errors?.length) {
+    throw new Error(`variableUpsert(${opts.name}) failed: ${JSON.stringify(res.errors)}`);
  }
 }

+/**
+ * Suspende ou retoma um serviço Hermes via flag HERMES_SUSPENDED + redeploy.
+ * Railway não suporta stop sem redeploy — a abordagem oficial é controlar
+ * via env var consumida pelo start command (ver HERMES_START_COMMAND).
+ *
+ * suspend=true  → seta HERMES_SUSPENDED=true e redeploy (container fica em sleep infinity)
+ * suspend=false → seta HERMES_SUSPENDED="" e redeploy (container sobe normalmente)
+ */
+export async function setHermesSuspended(opts: {
+  token: string;
+  serviceId: string;
+  environmentId: string;
+  projectId?: string;
+  suspend: boolean;
+}): Promise<void> {
+  await upsertRailwayVariable({
+    token: opts.token,
+    serviceId: opts.serviceId,
+    environmentId: opts.environmentId,
+    projectId: opts.projectId,
+    name: "HERMES_SUSPENDED",
+    value: opts.suspend ? "true" : "",
+  });
+
+  await deployRailwayService({
+    token: opts.token,
+    serviceId: opts.serviceId,
+    environmentId: opts.environmentId,
+  });
+}
+
 /** Busca o environmentId do primeiro deployment de um serviço. Útil quando vps_pool_id está null. */
 export async function getServiceEnvironmentId(opts: {
  token: string;
--- a/supabase/functions/provision-agent/index.ts
+++ b/supabase/functions/provision-agent/index.ts
@ -10,6 +10,7 @@ import {
  configureRailwayService,
  deployRailwayService,
  deleteTelegramWebhook,
+  HERMES_START_COMMAND,
 } from "../_shared/railway.ts";

 interface RequestBody {
@ -191,7 +192,7 @@ Deno.serve(async (req) => {
      serviceId: railwayServiceId,
      environmentId: pool.railway_environment_id,
      image: "nousresearch/hermes-agent:latest",
-      startCommand: "/opt/hermes/docker/entrypoint.sh gateway run",
+      startCommand: HERMES_START_COMMAND,
      variables: envVars,
    });

--- a/supabase/functions/resume-agent/index.ts
+++ b/supabase/functions/resume-agent/index.ts
@ -1,11 +1,11 @@
 // resume-agent
-// Retoma o serviço Railway (numReplicas=1) e marca agent_instance.status='active'.
+// Retoma o serviço Hermes removendo HERMES_SUSPENDED (string vazia) e disparando redeploy.
 // Disparado automaticamente quando subscription volta para active/trialing,
 // ou manualmente pelo painel admin.

 import { createClient } from "https://esm.sh/@supabase/supabase-js@2.45.4";
 import { corsHeaders } from "../_shared/cors.ts";
-import { setRailwayReplicas, getServiceEnvironmentId } from "../_shared/railway.ts";
+import { setHermesSuspended, getServiceEnvironmentId } from "../_shared/railway.ts";

 const SUPABASE_URL = Deno.env.get("SUPABASE_URL")!;
 const SUPABASE_SERVICE_ROLE_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!;
@ -55,15 +55,17 @@ Deno.serve(async (req) => {
    return jsonResponse(200, { ok: true, already_active: true });
  }

-  // Resolve environmentId: prioriza vps_pool, fallback para query no Railway
+  // Resolve environmentId/projectId
  let environmentId: string | null = null;
+  let projectId: string | undefined;
  if (agent.vps_pool_id) {
    const { data: pool } = await supabase
      .from("vps_pool")
-      .select("railway_environment_id")
+      .select("railway_environment_id, railway_project_id")
      .eq("id", agent.vps_pool_id)
      .maybeSingle();
    environmentId = pool?.railway_environment_id ?? null;
+    projectId = pool?.railway_project_id ?? undefined;
  }
  if (!environmentId) {
    environmentId = await getServiceEnvironmentId({
@ -76,16 +78,17 @@ Deno.serve(async (req) => {
  }

  try {
-    await setRailwayReplicas({
+    await setHermesSuspended({
      token: RAILWAY_API_TOKEN,
      serviceId: agent.railway_service_id,
      environmentId,
-      replicas: 1,
+      projectId,
+      suspend: false,
    });
  } catch (e) {
    const msg = e instanceof Error ? e.message : String(e);
-    console.error("resume-agent: setRailwayReplicas failed:", msg);
-    return jsonResponse(500, { error: "railway scale-up failed", detail: msg });
+    console.error("resume-agent: setHermesSuspended failed:", msg);
+    return jsonResponse(500, { error: "railway resume failed", detail: msg });
  }

  await supabase
--- a/supabase/functions/suspend-agent/index.ts
+++ b/supabase/functions/suspend-agent/index.ts
@ -1,11 +1,12 @@
 // suspend-agent
-// Pausa o serviço Railway (numReplicas=0) e marca agent_instance.status='suspended'.
+// Pausa o serviço Hermes setando HERMES_SUSPENDED=true e disparando redeploy.
+// O start command verifica essa flag e entra em `sleep infinity` (agente pausado).
 // Disparado automaticamente quando subscription muda para canceled/past_due/unpaid/paused,
 // ou manualmente pelo painel admin.

 import { createClient } from "https://esm.sh/@supabase/supabase-js@2.45.4";
 import { corsHeaders } from "../_shared/cors.ts";
-import { setRailwayReplicas, getServiceEnvironmentId } from "../_shared/railway.ts";
+import { setHermesSuspended, getServiceEnvironmentId } from "../_shared/railway.ts";

 const SUPABASE_URL = Deno.env.get("SUPABASE_URL")!;
 const SUPABASE_SERVICE_ROLE_KEY = Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!;
@ -56,15 +57,17 @@ Deno.serve(async (req) => {
    return jsonResponse(200, { ok: true, no_container: true });
  }

-  // Resolve o environmentId: prioriza vps_pool, fallback para query no Railway
+  // Resolve o environmentId/projectId via vps_pool, fallback Railway query
  let environmentId: string | null = null;
+  let projectId: string | undefined;
  if (agent.vps_pool_id) {
    const { data: pool } = await supabase
      .from("vps_pool")
-      .select("railway_environment_id")
+      .select("railway_environment_id, railway_project_id")
      .eq("id", agent.vps_pool_id)
      .maybeSingle();
    environmentId = pool?.railway_environment_id ?? null;
+    projectId = pool?.railway_project_id ?? undefined;
  }
  if (!environmentId) {
    environmentId = await getServiceEnvironmentId({
@ -77,16 +80,17 @@ Deno.serve(async (req) => {
  }

  try {
-    await setRailwayReplicas({
+    await setHermesSuspended({
      token: RAILWAY_API_TOKEN,
      serviceId: agent.railway_service_id,
      environmentId,
-      replicas: 0,
+      projectId,
+      suspend: true,
    });
  } catch (e) {
    const msg = e instanceof Error ? e.message : String(e);
-    console.error("suspend-agent: setRailwayReplicas failed:", msg);
-    return jsonResponse(500, { error: "railway scale-down failed", detail: msg });
+    console.error("suspend-agent: setHermesSuspended failed:", msg);
+    return jsonResponse(500, { error: "railway suspend failed", detail: msg });
  }

  await supabase