## Summary Fusion’s core runtime now treats PostgreSQL as the authoritative metadata store without leaving current CLI, dashboard, desktop, or engine composition roots uncompilable between stack layers. This is the 99-file foundation for the larger cutover: subsequent PRs migrate the remaining consumers, plugins, and operator surfaces. ## Design decisions - Runtime store construction fails closed when an asynchronous PostgreSQL layer is unavailable; SQLite remains readable only at explicit migration and identity-recovery boundaries. - Project ownership is enforced across active, archived, workflow, mission, analytics, and plugin-schema data. - The small set of cross-package files in this layer are compatibility-critical call sites required for a green intermediate commit, not the complete consumer migration. - Schema migration 0008 remains assigned to session-advisor state from current `main`; mission lineage idempotency advances to 0009 so neither invariant can be skipped. ## Validation - All affected package typechecks pass: Core, Engine, Dashboard, CLI, and Desktop. - `pnpm test:gate` passes: 478 tests across the engine gate, PostgreSQL core gate, and CLI workflow shape. - The PR changes exactly 99 files. ## Stack This is the base PR. Engine/dashboard, CLI/desktop/ops, plugins, and docs/release follow as stacked PRs, each below 100 changed files. Related: #2105 <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * PostgreSQL is now the standard runtime backend, with embedded PostgreSQL enabled by default. * Added project-scoped storage for tasks, archives, chat sessions, missions, knowledge pages, and operational data. * Improved archived-task search, filtering, pagination, and restoration. * Added safer plugin schema initialization with validation and project isolation. * Added PostgreSQL-backed workflow, mission, validator, and dashboard capabilities. * **Bug Fixes** * Improved startup timeout cancellation and resource cleanup. * Prevented cross-project data access and phantom reservation cleanup errors. * Ensured archived tasks remain read-only and asynchronous writes complete reliably. * Retired SQLite opt-out settings with clear startup errors. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
180 lines
7.0 KiB
TypeScript
180 lines
7.0 KiB
TypeScript
/**
|
|
* createCliAgentRuntime — the per-project bootstrap that wires the CLI Agent
|
|
* Executor subsystem together (U-final integration).
|
|
*
|
|
* Every component (PTY session manager, telemetry hub, adapter registry, resume
|
|
* coordinator) is built with injection seams and tested in isolation; this
|
|
* factory is the single place that actually instantiates the live bundle and
|
|
* stitches the seams:
|
|
*
|
|
* - Builds a {@link CliSessionStore} over the project's existing PostgreSQL
|
|
* data layer (never opens a second connection).
|
|
* - Registers all bundled adapters into a fresh {@link CliAdapterRegistry} (a
|
|
* per-runtime registry, NOT the process-wide `defaultCliAdapterRegistry`, so
|
|
* multi-project boots never collide on duplicate-registration).
|
|
* - Constructs the {@link CliSessionManager} (PTY lifecycle) and
|
|
* {@link TelemetryHub} (per-session token registry, rebuilt from live store
|
|
* records on construction).
|
|
* - Constructs the {@link CliResumeCoordinator}, wiring `reattachTelemetry` to
|
|
* re-mint a hook token + rewrite the session's hook scripts on relaunch.
|
|
*
|
|
* It returns the {@link CliAgentRuntime} bundle the {@link TaskExecutor}
|
|
* consumes, plus the two narrow predicates the self-healing / stuck-task seams
|
|
* read, plus a `dispose` that tears the manager down cleanly (scoped SIGKILL of
|
|
* the runtime's own PTYs only — never the dashboard / port 4040).
|
|
*/
|
|
|
|
import { CliSessionStore } from "@fusion/core";
|
|
import type { AsyncDataLayer } from "@fusion/core";
|
|
import { CliAdapterRegistry } from "./adapter.js";
|
|
import { BUNDLED_CLI_ADAPTERS } from "./adapters/index.js";
|
|
import { CliSessionManager, type CliSessionManagerOptions } from "./session-manager.js";
|
|
import { TelemetryHub, type TelemetryHubOptions } from "./telemetry-hub.js";
|
|
import { CliResumeCoordinator } from "./resume-coordinator.js";
|
|
import { writeSessionHookScripts } from "./hook-scripts.js";
|
|
import type { CliAgentRuntime } from "../executor.js";
|
|
|
|
/** Options for {@link createCliAgentRuntime}. */
|
|
export interface CreateCliAgentRuntimeOptions {
|
|
/** The project's `.fusion` dir (scratch root for hook scripts). */
|
|
fusionDir: string;
|
|
/** The project's already-open PostgreSQL data layer (reused, never re-opened). */
|
|
asyncLayer: AsyncDataLayer;
|
|
/** Project this runtime drives (`cli_sessions.projectId`). */
|
|
projectId: string;
|
|
/**
|
|
* Absolute URL of the dashboard hook ingestion endpoint the generated hook
|
|
* scripts POST to (e.g. `http://127.0.0.1:4040/api/cli-agent/hooks`).
|
|
*/
|
|
hookEndpointUrl: string;
|
|
/** Optional override for the hook scratch-dir root (tests). */
|
|
hookDirRoot?: string;
|
|
/** Optional notification dispatch forwarded to the TelemetryHub. */
|
|
onNotification?: TelemetryHubOptions["onNotification"];
|
|
/**
|
|
* Test seams forwarded to the {@link CliSessionManager} (e.g. a mocked node-pty
|
|
* loader so runtime construction never touches a real PTY).
|
|
*/
|
|
managerOptions?: Pick<
|
|
CliSessionManagerOptions,
|
|
"loadPty" | "scrollbackBytes" | "concurrencyCeiling" | "highWatermark" | "injectionQuietWindowMs"
|
|
>;
|
|
}
|
|
|
|
/**
|
|
* The full bootstrapped CLI-agent runtime: the executor bundle, the predicates
|
|
* the self-healing + stuck-task seams read, the resume coordinator, and dispose.
|
|
*/
|
|
export interface BootstrappedCliAgentRuntime {
|
|
/** The bundle threaded into {@link TaskExecutorOptions.cliAgentRuntime}. */
|
|
bundle: CliAgentRuntime;
|
|
/** Engine-start orphan recovery sweep (call after engine start; errors logged). */
|
|
resumeCoordinator: CliResumeCoordinator;
|
|
/**
|
|
* Self-healing seam: whether a worktree path backs a resume-eligible session
|
|
* record (so idle-worktree sweeps treat it as in-use). Delegates to the resume
|
|
* coordinator's reservation set.
|
|
*/
|
|
isWorktreeResumeReserved: (worktreePath: string) => boolean;
|
|
/**
|
|
* Stuck-task seam: whether a task's live CLI session is `waitingOnInput`
|
|
* (expected idleness — suppress stuck flagging). Reads the live store.
|
|
*/
|
|
isCliSessionWaitingOnInput: (taskId: string) => boolean;
|
|
/** Tear down the PTY manager (scoped SIGKILL of this runtime's PTYs only). */
|
|
dispose: () => Promise<void>;
|
|
}
|
|
|
|
/**
|
|
* Construct the per-project CLI-agent runtime bundle. Pure construction — no IO
|
|
* beyond the store's reads against the supplied Database; spawning a PTY or
|
|
* running recovery is the caller's job (`resumeCoordinator.recoverOnStart()`).
|
|
*/
|
|
export async function createCliAgentRuntime(
|
|
options: CreateCliAgentRuntimeOptions,
|
|
): Promise<BootstrappedCliAgentRuntime> {
|
|
const { asyncLayer, projectId, hookEndpointUrl } = options;
|
|
|
|
// FNXC:CliAgentPostgres 2026-07-14-12:00:
|
|
// Hydrate the project-scoped cache before state machines or recovery inspect
|
|
// it; mutations remain ordered through the shared PostgreSQL data layer.
|
|
const store = await CliSessionStore.create(asyncLayer, projectId);
|
|
|
|
// 2. A per-runtime registry with every bundled adapter (not the process-wide
|
|
// singleton — avoids duplicate-registration across multi-project boots).
|
|
const registry = new CliAdapterRegistry();
|
|
for (const adapter of BUNDLED_CLI_ADAPTERS) {
|
|
registry.register(adapter);
|
|
}
|
|
|
|
// 3. PTY session manager.
|
|
const manager = new CliSessionManager({
|
|
registry,
|
|
store,
|
|
...options.managerOptions,
|
|
});
|
|
|
|
// 4. Telemetry hub — rebuilds its per-session token registry from live store
|
|
// records on construction.
|
|
const hub = new TelemetryHub({
|
|
store,
|
|
onNotification: options.onNotification,
|
|
});
|
|
|
|
// 5. Resume coordinator. On relaunch, re-mint a hook token and rewrite the
|
|
// session's hook scripts so the resumed CLI POSTs with a fresh, valid token.
|
|
const resumeCoordinator = new CliResumeCoordinator({
|
|
store,
|
|
manager,
|
|
registry,
|
|
reattachTelemetry: async (session) => {
|
|
const token = hub.issueToken(session.id);
|
|
await writeSessionHookScripts({
|
|
sessionId: session.id,
|
|
token,
|
|
endpointUrl: hookEndpointUrl,
|
|
dir: hookScriptDir(options, session.id),
|
|
});
|
|
},
|
|
});
|
|
|
|
const bundle: CliAgentRuntime = {
|
|
manager,
|
|
hub,
|
|
registry,
|
|
store,
|
|
projectId,
|
|
hookEndpointUrl,
|
|
hookDirRoot: options.hookDirRoot,
|
|
};
|
|
|
|
return {
|
|
bundle,
|
|
resumeCoordinator,
|
|
isWorktreeResumeReserved: (worktreePath: string) =>
|
|
resumeCoordinator.resumeReservedWorktrees().has(worktreePath),
|
|
isCliSessionWaitingOnInput: (taskId: string) => {
|
|
// A task's live session is "waiting on input" when any of its session
|
|
// records is in the waitingOnInput state. Defensive: a store error means
|
|
// "not waiting" (the stuck detector's own guard re-asserts this too).
|
|
try {
|
|
return store
|
|
.listByTask(taskId)
|
|
.some((s) => s.agentState === "waitingOnInput");
|
|
} catch {
|
|
return false;
|
|
}
|
|
},
|
|
dispose: async () => {
|
|
manager.dispose();
|
|
await store.flush();
|
|
},
|
|
};
|
|
}
|
|
|
|
/** Resolve the per-session hook scratch dir under the configured root. */
|
|
function hookScriptDir(options: CreateCliAgentRuntimeOptions, sessionId: string): string {
|
|
const root = options.hookDirRoot ?? `${options.fusionDir}/cli-agent/hooks`;
|
|
return `${root}/${sessionId}`;
|
|
}
|