Merge pull request #215 from aaditagrawal/codex/sync-20260808-l1-server-settle-a13c

sync: port upstream server settle, tunnel/update safety, and early web polish (pingdotgg#5482–pingdotgg#5486)

Author: cursor[bot]

apps/mobile/src/features/threads/PendingApprovalCard.tsx

@@ -14,8 +14,10 @@ export interface PendingApprovalCardProps {
 }
 
 export function PendingApprovalCard(props: PendingApprovalCardProps) {
+  // Opaque for the same reason as PendingUserInputCard: nothing blurs the feed
+  // behind this card, so a translucent surface bleeds messages through it.
   return (
-    <View className="gap-2.5 rounded-[20px] border border-neutral-200 bg-neutral-100/80 p-4 dark:border-white/6 dark:bg-neutral-900/80">
+    <View className="gap-2.5 rounded-[20px] border border-neutral-200 bg-neutral-100 p-4 dark:border-white/6 dark:bg-neutral-900">
       <Text className="font-t3-bold text-2xs uppercase tracking-[1.1px] text-sky-700 dark:text-sky-300">
         Approval needed
       </Text>

apps/mobile/src/features/threads/PendingUserInputCard.tsx

@@ -24,8 +24,11 @@ export interface PendingUserInputCardProps {
 }
 
 export function PendingUserInputCard(props: PendingUserInputCardProps) {
+  // The surface is opaque on purpose: the card floats over the thread feed
+  // with no blur behind it, so a translucent background renders the questions
+  // on top of whatever message happens to sit underneath.
   return (
-    <View className="gap-2.5 rounded-[20px] border border-neutral-200 bg-neutral-100/80 p-4 dark:border-white/6 dark:bg-neutral-900/80">
+    <View className="gap-2.5 rounded-[20px] border border-neutral-200 bg-neutral-100 p-4 dark:border-white/6 dark:bg-neutral-900">
       <Text className="font-t3-bold text-2xs uppercase tracking-[1.1px] text-sky-700 dark:text-sky-300">
         User input needed
       </Text>

apps/server/src/cloud/bootService.test.ts

@@ -31,7 +31,7 @@ it("keeps systemd pinned to the stable launcher rather than a versioned server",
   });
 
   expect(unit).toContain("ExecStart=/usr/bin/node /home/theo/.t3/runtime/service-launcher.mjs");
-  expect(unit).toContain("KillMode=control-group");
+  expect(unit).toContain("KillMode=mixed");
   expect(unit).not.toContain("versions/1.2.3");
 });
 

apps/server/src/cloud/bootService.ts

@@ -64,7 +64,9 @@ export function renderBootServiceUnit(plan: BootServicePlan): string {
     `Environment=T3CODE_HOME=${quoteSystemdValue(plan.baseDir)}`,
     `Environment=${BOOT_SERVICE_UNIT_ENV}=${BOOT_SERVICE_UNIT_FILE}`,
     `ExecStart=${quoteSystemdValue(plan.nodePath)} ${quoteSystemdValue(plan.launcherPath)}`,
-    "KillMode=control-group",
+    // Let the launcher mark an explicit stop before it signals the server.
+    // systemd still SIGKILLs the whole cgroup if graceful shutdown times out.
+    "KillMode=mixed",
     "Restart=always",
     "RestartSec=5",
     `StandardOutput=append:${escapeSystemdSpecifiers(plan.logPath)}`,

apps/server/src/cloud/http.test.ts

@@ -1,7 +1,10 @@
 import * as NodeServices from "@effect/platform-node/NodeServices";
 import { describe, expect, it } from "@effect/vitest";
 import * as Effect from "effect/Effect";
+import * as FileSystem from "effect/FileSystem";
+import * as Layer from "effect/Layer";
 import * as Option from "effect/Option";
+import * as Path from "effect/Path";
 import * as PlatformError from "effect/PlatformError";
 import * as Tracer from "effect/Tracer";
 import {
@@ -15,6 +18,14 @@ import { EnvironmentId } from "@t3tools/contracts";
 import { RelayClientTracer } from "@t3tools/shared/relayTracing";
 import * as EnvironmentAuth from "../auth/EnvironmentAuth.ts";
 import * as ServerSecretStore from "../auth/ServerSecretStore.ts";
+import * as ServerConfigModule from "../config.ts";
+import { writeServiceState } from "../serviceLauncher.ts";
+import {
+  SERVICE_LAUNCHER_PROTOCOL,
+  SERVICE_STATE_FILE,
+  SERVICE_STOP_MARKER_FILE,
+  type ServiceUpdateRecord,
+} from "./serviceProtocol.ts";
 import * as ServerEnvironment from "../environment/ServerEnvironment.ts";
 import { CLOUD_CLI_DESIRED_LINK_SECRET } from "./CliState.ts";
 import * as CliTokenManager from "./CliTokenManager.ts";
@@ -24,6 +35,7 @@ import {
   consumeCloudReplayGuards,
   isSupportedLinkProviderKind,
   linkProofScopes,
+  pendingServiceUpdateExists,
   reconcileDesiredCloudLink,
   releaseManagedTunnelOnShutdown,
 } from "./http.ts";
@@ -256,6 +268,23 @@ describe("releaseManagedTunnelOnShutdown", () => {
     readonly respond?: () => Response;
   }
 
+  // Writes the launcher's durable state file into this test's baseDir with
+  // the launcher's own writer; the release reads it to detect an in-flight
+  // update handoff.
+  const writeLauncherState = (update: ServiceUpdateRecord) =>
+    Effect.gen(function* () {
+      const path = yield* Path.Path;
+      const config = yield* ServerConfigModule.ServerConfig;
+      const statePath = path.join(config.baseDir, "runtime", SERVICE_STATE_FILE);
+      yield* Effect.promise(() =>
+        writeServiceState(statePath, {
+          protocol: SERVICE_LAUNCHER_PROTOCOL,
+          activeVersion: "0.0.30",
+          update,
+        }),
+      );
+    });
+
   const provideReleaseHarness =
     (harness: ReleaseHarness) =>
     <A, E, R>(effect: Effect.Effect<A, E, R>) =>
@@ -306,6 +335,14 @@ describe("releaseManagedTunnelOnShutdown", () => {
             }),
           ),
         ),
+        // The release consults the launcher state file under the configured
+        // baseDir, so every harness run gets a scoped temp baseDir.
+        Effect.provide(
+          ServerConfigModule.layerTest("/", { prefix: "t3-http-release-test-" }).pipe(
+            Layer.provideMerge(NodeServices.layer),
+          ),
+        ),
+        Effect.scoped,
       );
 
   // The persisted state of a CLI-managed link whose tunnel is releasable.
@@ -390,6 +427,84 @@ describe("releaseManagedTunnelOnShutdown", () => {
     }).pipe(provideReleaseHarness({ store, applyConfigCalls, requests }));
   });
 
+  it.effect("keeps the tunnel when shutdown hands off to a pending update", () => {
+    const { store, values } = makeMemorySecretStore(managedLinkSecrets);
+    const applyConfigCalls: Array<unknown> = [];
+    const requests: Array<HttpClientRequest.HttpClientRequest> = [];
+
+    return Effect.gen(function* () {
+      yield* writeLauncherState({
+        id: "update-1",
+        fromVersion: "0.0.30",
+        targetVersion: "0.0.31",
+        dbPath: "/tmp/state.sqlite",
+        status: "pending",
+      });
+
+      const released = yield* releaseManagedTunnelOnShutdown();
+
+      // The launcher restarts a server immediately, so the tunnel is not
+      // orphaned; keeping it avoids the hostname route re-propagation that
+      // dominates update downtime. The stored config must survive so the
+      // next boot respawns the connector against the same tunnel.
+      expect(released).toBe(false);
+      expect(applyConfigCalls).toEqual([]);
+      expect(requests).toEqual([]);
+      expect(values.has(CLOUD_ENDPOINT_RUNTIME_CONFIG)).toBe(true);
+    }).pipe(provideReleaseHarness({ store, applyConfigCalls, requests }));
+  });
+
+  it.effect("still releases a pending update when the launcher is stopping", () => {
+    // `t3 service uninstall` or `systemctl stop` during the pending window:
+    // the launcher writes its stop marker before signalling the child, so no
+    // replacement server is coming and the tunnel must not be kept.
+    const { store, values } = makeMemorySecretStore(managedLinkSecrets);
+    const applyConfigCalls: Array<unknown> = [];
+    const requests: Array<HttpClientRequest.HttpClientRequest> = [];
+
+    return Effect.gen(function* () {
+      yield* writeLauncherState({
+        id: "update-1",
+        fromVersion: "0.0.30",
+        targetVersion: "0.0.31",
+        dbPath: "/tmp/state.sqlite",
+        status: "pending",
+      });
+      const fs = yield* FileSystem.FileSystem;
+      const path = yield* Path.Path;
+      const config = yield* ServerConfigModule.ServerConfig;
+      yield* fs.writeFileString(path.join(config.baseDir, "runtime", SERVICE_STOP_MARKER_FILE), "");
+
+      expect(yield* pendingServiceUpdateExists).toBe(true);
+      const released = yield* releaseManagedTunnelOnShutdown();
+
+      expect(released).toBe(true);
+      expect(requests).toHaveLength(1);
+      expect(values.has(CLOUD_ENDPOINT_RUNTIME_CONFIG)).toBe(false);
+    }).pipe(provideReleaseHarness({ store, applyConfigCalls, requests }));
+  });
+
+  it.effect("still releases when the recorded update already settled", () => {
+    const { store, values } = makeMemorySecretStore(managedLinkSecrets);
+    const applyConfigCalls: Array<unknown> = [];
+    const requests: Array<HttpClientRequest.HttpClientRequest> = [];
+
+    return Effect.gen(function* () {
+      yield* writeLauncherState({
+        id: "update-1",
+        fromVersion: "0.0.30",
+        targetVersion: "0.0.31",
+        status: "committed",
+      });
+
+      const released = yield* releaseManagedTunnelOnShutdown();
+
+      expect(released).toBe(true);
+      expect(requests).toHaveLength(1);
+      expect(values.has(CLOUD_ENDPOINT_RUNTIME_CONFIG)).toBe(false);
+    }).pipe(provideReleaseHarness({ store, applyConfigCalls, requests }));
+  });
+
   it.effect("keeps a runtime config that a fast restart replaced mid-release", () => {
     const { store, values } = makeMemorySecretStore(managedLinkSecrets);
     const applyConfigCalls: Array<unknown> = [];

apps/server/src/cloud/http.ts

@@ -46,7 +46,9 @@ import * as DateTime from "effect/DateTime";
 import * as Crypto from "effect/Crypto";
 import * as Duration from "effect/Duration";
 import * as Effect from "effect/Effect";
+import * as FileSystem from "effect/FileSystem";
 import * as Option from "effect/Option";
+import * as Path from "effect/Path";
 import * as Schema from "effect/Schema";
 import * as HttpEffect from "effect/unstable/http/HttpEffect";
 import { HttpServerRequest, HttpServerResponse } from "effect/unstable/http";
@@ -56,8 +58,14 @@ import * as HttpApiBuilder from "effect/unstable/httpapi/HttpApiBuilder";
 import * as EnvironmentAuth from "../auth/EnvironmentAuth.ts";
 import * as ServerSecretStore from "../auth/ServerSecretStore.ts";
 import { requireEnvironmentScope } from "../auth/http.ts";
+import * as ServerConfig from "../config.ts";
 import * as ServerEnvironment from "../environment/ServerEnvironment.ts";
 import * as ManagedEndpointRuntime from "./ManagedEndpointRuntime.ts";
+import {
+  SERVICE_STATE_FILE,
+  SERVICE_STOP_MARKER_FILE,
+  serviceStateHasPendingUpdate,
+} from "./serviceProtocol.ts";
 import {
   CLOUD_ENDPOINT_RUNTIME_CONFIG,
   CLOUD_LINKED_USER_ID,
@@ -627,6 +635,37 @@ export const reconcileDesiredCloudLink = Effect.fn("environment.cloud.reconcileD
   },
 );
 
+// The launcher owns this durable state, so read it directly both when a trial
+// decides whether it owns pre-activation cleanup and while a server tears down.
+export const pendingServiceUpdateExists = Effect.gen(function* () {
+  const config = yield* ServerConfig.ServerConfig;
+  const fs = yield* FileSystem.FileSystem;
+  const path = yield* Path.Path;
+  const runtimeDir = path.join(config.baseDir, "runtime");
+  const stateText = yield* fs
+    .readFileString(path.join(runtimeDir, SERVICE_STATE_FILE))
+    .pipe(Effect.option);
+  return Option.isSome(stateText) && serviceStateHasPendingUpdate(stateText.value);
+});
+
+// A pending update alone is not proof a replacement server is coming: an
+// explicit launcher stop (`t3 service uninstall`, `systemctl stop`) during
+// the pending window also tears this server down. The launcher marks that case
+// just before it signals the child, so pending + no marker is the handoff.
+const pendingUpdateHandoffExists = Effect.gen(function* () {
+  if (!(yield* pendingServiceUpdateExists)) {
+    return false;
+  }
+  const config = yield* ServerConfig.ServerConfig;
+  const fs = yield* FileSystem.FileSystem;
+  const path = yield* Path.Path;
+  const runtimeDir = path.join(config.baseDir, "runtime");
+  const stopping = yield* fs
+    .exists(path.join(runtimeDir, SERVICE_STOP_MARKER_FILE))
+    .pipe(Effect.orElseSucceed(() => false));
+  return !stopping;
+});
+
 // Cloudflare bills per provisioned tunnel, so an environment that goes offline
 // must not leave its tunnel behind. Releasing deletes only the tunnel — the
 // relay keeps the link and its hostname reservation, and the next startup's
@@ -649,6 +688,19 @@ export const releaseManagedTunnelOnShutdown = Effect.fn(
   if (!(yield* readCliDesiredCloudLink) || (yield* readCliDesiredLinkMode) !== "managed") {
     return false;
   }
+  // A shutdown that hands off to a pending remote update is not the
+  // environment going offline: the launcher immediately brings a server back
+  // (the new version, or the old one after a rollback). Deleting the tunnel
+  // here forces that server to provision a replacement UUID, and the public
+  // hostname's route to the new tunnel takes 1-2 minutes to propagate — the
+  // dominant cost of an update restart. Keep the tunnel instead: the next
+  // boot respawns the connector from the stored config and is reachable as
+  // soon as it connects, and the reconcile confirms the still-live tunnel
+  // without replacing it.
+  if (yield* pendingUpdateHandoffExists) {
+    yield* Effect.logInfo("Keeping the managed tunnel across the update restart");
+    return false;
+  }
   const token = yield* dependencies.cliTokenManager.getExisting;
   if (Option.isNone(token)) {
     return false;

apps/server/src/cloud/serviceProtocol.ts

@@ -5,6 +5,10 @@ export const SERVICE_LAUNCHER_PROTOCOL = 2 as const;
 export const SERVICE_LAUNCHER_CONTEXT_ENV = "T3_SERVICE_LAUNCHER_CONTEXT";
 export const SERVICE_LAUNCHER_FILE = "service-launcher.mjs";
 export const SERVICE_STATE_FILE = "service-state.json";
+/** Written by the launcher just before an explicit stop kills its child, so
+    the child can tell "the service is going away" from "the launcher is about
+    to start my replacement" while a pending update is recorded. */
+export const SERVICE_STOP_MARKER_FILE = ".service-stopping";
 
 export interface PendingServiceUpdate {
   readonly id: string;

apps/server/src/orchestration/ActivityPayloadProjection.test.ts

@@ -44,6 +44,60 @@ describe("projectActivityPayload agent-field survival", () => {
     expect(data.somethingClientNeverReads).toBeUndefined();
   });
 
+  it("slims Codex-shaped mcp_tool_call items to rendered fields plus a result summary", () => {
+    const projected = projectActivityPayload(
+      activity({
+        itemType: "mcp_tool_call",
+        data: {
+          item: {
+            type: "mcpToolCall",
+            id: "item-1",
+            tool: "fetch_pr",
+            server: "github",
+            status: "completed",
+            arguments: { pr: 42 },
+            durationMs: 1200,
+            result: {
+              content: [{ type: "text", text: `PR body line one\n${"x".repeat(5000)}` }],
+              structuredContent: { huge: "y".repeat(5000) },
+            },
+            _meta: { internal: true },
+          },
+        },
+      }),
+    );
+    const data = (projected.payload as Record<string, unknown>).data as Record<string, unknown>;
+    const item = data.item as Record<string, unknown>;
+    expect(item.tool).toBe("fetch_pr");
+    expect(item.server).toBe("github");
+    expect(item.arguments).toEqual({ pr: 42 });
+    expect(item._meta).toBeUndefined();
+    expect(item.result).toEqual({ content: "PR body line one" });
+    expect(JSON.stringify(projected.payload).length).toBeLessThan(500);
+  });
+
+  it("slims Claude-shaped mcp_tool_call data (toolName/input/result block)", () => {
+    const projected = projectActivityPayload(
+      activity({
+        itemType: "mcp_tool_call",
+        data: {
+          toolName: "mcp__github__fetch_pr",
+          input: { pr: 42 },
+          result: {
+            type: "tool_result",
+            tool_use_id: "toolu_1",
+            content: [{ type: "text", text: `first line of output\n${"z".repeat(5000)}` }],
+          },
+        },
+      }),
+    );
+    const data = (projected.payload as Record<string, unknown>).data as Record<string, unknown>;
+    expect(data.toolName).toBe("mcp__github__fetch_pr");
+    expect(data.input).toEqual({ pr: 42 });
+    expect(data.result).toEqual({ content: "first line of output" });
+    expect(JSON.stringify(projected.payload).length).toBeLessThan(500);
+  });
+
   it("passes task lifecycle payloads (no data field) through untouched", () => {
     const source = activity({
       taskId: "task-9",