llm-cli-gateway 2.14.0-rc.1 → 2.14.1

package/dist/postgres-job-store-worker.js

@@ -1,6 +1,7 @@
 import { writeFileSync } from "node:fs";
 import { parentPort, workerData } from "node:worker_threads";
 let pool = null;
+const PG_NOW_MS = "(EXTRACT(EPOCH FROM now()) * 1000)::bigint";
 function signal(shared, status) {
     const view = new Int32Array(shared);
     Atomics.store(view, 0, status);
@@ -70,13 +71,26 @@ async function init() {
       owner_principal TEXT,
       transport TEXT NOT NULL DEFAULT 'process',
       http_status INTEGER,
-      payload_json TEXT
+      payload_json TEXT,
+      owner_instance TEXT,
+      lease_deadline BIGINT
     );
     CREATE INDEX IF NOT EXISTS idx_jobs_request_key ON jobs(request_key);
     CREATE INDEX IF NOT EXISTS idx_jobs_status ON jobs(status);
     CREATE INDEX IF NOT EXISTS idx_jobs_expires_at ON jobs(expires_at);
     CREATE INDEX IF NOT EXISTS idx_jobs_request_key_finished ON jobs(request_key, finished_at);
 
+    CREATE TABLE IF NOT EXISTS gateway_instances (
+      instance_id TEXT PRIMARY KEY,
+      role TEXT,
+      hostname TEXT,
+      pid INTEGER,
+      started_at BIGINT NOT NULL,
+      last_heartbeat BIGINT NOT NULL
+    );
+    CREATE INDEX IF NOT EXISTS idx_gateway_instances_heartbeat
+      ON gateway_instances(last_heartbeat);
+
     CREATE TABLE IF NOT EXISTS validation_runs (
       validation_id TEXT PRIMARY KEY,
       owner_principal TEXT NOT NULL,
@@ -116,6 +130,9 @@ async function init() {
     await getPool().query("ALTER TABLE jobs ADD COLUMN IF NOT EXISTS transport TEXT NOT NULL DEFAULT 'process'");
     await getPool().query("ALTER TABLE jobs ADD COLUMN IF NOT EXISTS http_status INTEGER");
     await getPool().query("ALTER TABLE jobs ADD COLUMN IF NOT EXISTS payload_json TEXT");
+    await getPool().query("ALTER TABLE jobs ADD COLUMN IF NOT EXISTS owner_instance TEXT");
+    await getPool().query("ALTER TABLE jobs ADD COLUMN IF NOT EXISTS lease_deadline BIGINT");
+    await getPool().query("CREATE INDEX IF NOT EXISTS idx_jobs_owner_status ON jobs(owner_instance, status)");
 }
 async function op(method, args) {
     switch (method) {
@@ -127,9 +144,9 @@ async function op(method, args) {
             await getPool().query(`INSERT INTO jobs (id, correlation_id, request_key, cli, args_json, output_format,
                            status, exit_code, stdout, stderr, output_truncated, error,
                            started_at, finished_at, pid, expires_at, owner_principal,
-                           transport, http_status, payload_json)
-         VALUES ($1, $2, $3, $4, $5, $6, 'running', NULL, '', '', FALSE, NULL,
-                 $7, NULL, $8, $9, $10, $11, NULL, $12)`, [
+                           transport, http_status, payload_json, owner_instance, lease_deadline)
+         VALUES ($1, $2, $3, $4, $5, $6, 'queued', NULL, '', '', FALSE, NULL,
+                 $7, NULL, $8, $9, $10, $11, NULL, $12, $13, ${PG_NOW_MS} + $14)`, [
                 input.id,
                 input.correlationId,
                 input.requestKey,
@@ -142,9 +159,106 @@ async function op(method, args) {
                 input.ownerPrincipal ?? null,
                 input.transport ?? "process",
                 input.payloadJson ?? null,
+                input.ownerInstance ?? null,
+                workerData.leaseTtlMs,
             ]);
             return null;
         }
+        case "markRunning": {
+            const [id, opts] = args;
+            const result = await getPool().query(`UPDATE jobs
+         SET status = 'running', pid = $2, lease_deadline = ${PG_NOW_MS} + $3::bigint
+         WHERE id = $1 AND status = 'queued'`, [id, opts?.pid ?? null, workerData.leaseTtlMs]);
+            return (result.rowCount ?? 0) > 0;
+        }
+        case "registerInstance": {
+            const meta = args[0];
+            await getPool().query(`INSERT INTO gateway_instances (instance_id, role, hostname, pid, started_at, last_heartbeat)
+         VALUES ($1, $2, $3, $4, ${PG_NOW_MS}, ${PG_NOW_MS})
+         ON CONFLICT (instance_id) DO UPDATE SET
+           role = EXCLUDED.role, hostname = EXCLUDED.hostname, pid = EXCLUDED.pid,
+           last_heartbeat = EXCLUDED.last_heartbeat`, [meta.instanceId, meta.role ?? null, meta.hostname ?? null, meta.pid ?? null]);
+            return null;
+        }
+        case "heartbeat": {
+            const instanceId = args[0];
+            await withClient(async (client) => {
+                await client.query("BEGIN");
+                try {
+                    await client.query(`UPDATE gateway_instances SET last_heartbeat = ${PG_NOW_MS} WHERE instance_id = $1`, [instanceId]);
+                    await client.query(`UPDATE jobs SET lease_deadline = ${PG_NOW_MS} + $2
+             WHERE owner_instance = $1 AND status IN ('queued', 'running')`, [instanceId, workerData.leaseTtlMs]);
+                    await client.query("COMMIT");
+                }
+                catch (error) {
+                    await client.query("ROLLBACK");
+                    throw error;
+                }
+            });
+            return null;
+        }
+        case "deregisterInstance": {
+            await getPool().query("DELETE FROM gateway_instances WHERE instance_id = $1", [args[0]]);
+            return null;
+        }
+        case "selectStaleProcessCandidates": {
+            const result = await getPool().query(`SELECT j.id AS id, j.pid AS pid, j.transport AS transport,
+                j.owner_instance AS owner_instance, gi.hostname AS hostname
+         FROM jobs j
+         LEFT JOIN gateway_instances gi ON gi.instance_id = j.owner_instance
+         WHERE j.status IN ('queued', 'running')
+           AND j.transport = 'process'
+           AND j.pid IS NOT NULL
+           AND (j.lease_deadline IS NULL OR j.lease_deadline < ${PG_NOW_MS})`);
+            return result.rows.map(r => ({
+                id: r.id,
+                pid: r.pid,
+                transport: r.transport ?? "process",
+                ownerInstance: r.owner_instance ?? null,
+                hostname: r.hostname ?? null,
+            }));
+        }
+        case "recoverStaleJobs": {
+            const [leaseTtlMs, httpJobGraceMs, liveConfirmedIds] = args;
+            const excludeIds = liveConfirmedIds ?? [];
+            const httpGraceCutoffMs = "(" + PG_NOW_MS + " - $1::bigint)";
+            return await withClient(async (client) => {
+                await client.query("BEGIN");
+                try {
+                    if (excludeIds.length > 0) {
+                        await client.query(`UPDATE jobs SET lease_deadline = ${PG_NOW_MS} + $1::bigint, pid = NULL
+               WHERE status IN ('queued', 'running') AND id = ANY($2::text[])`, [leaseTtlMs, excludeIds]);
+                    }
+                    const orphan = await client.query(`UPDATE jobs
+             SET status = 'orphaned',
+                 error = COALESCE(error, 'owning gateway instance is no longer alive'),
+                 finished_at = COALESCE(finished_at, $2),
+                 expires_at = $3,
+                 lease_deadline = NULL
+             WHERE status IN ('queued', 'running')
+               AND (lease_deadline IS NULL OR lease_deadline < ${PG_NOW_MS})
+               AND (transport <> 'http'
+                    OR (EXTRACT(EPOCH FROM started_at::timestamptz) * 1000)::bigint < ${httpGraceCutoffMs})
+               AND NOT (id = ANY($4::text[]))
+             RETURNING id, correlation_id, started_at, stdout, stderr, exit_code, transport, http_status`, [
+                        httpJobGraceMs,
+                        new Date().toISOString(),
+                        new Date(Date.now() + workerData.retentionMs).toISOString(),
+                        excludeIds,
+                    ]);
+                    await client.query("COMMIT");
+                    return { orphaned: orphan.rows };
+                }
+                catch (error) {
+                    await client.query("ROLLBACK");
+                    throw error;
+                }
+            });
+        }
+        case "gcInstances": {
+            const result = await getPool().query(`DELETE FROM gateway_instances WHERE last_heartbeat < ${PG_NOW_MS} - $1::bigint`, [args[0]]);
+            return result.rowCount ?? 0;
+        }
         case "recordOutput":
             await getPool().query("UPDATE jobs SET stdout = $2, stderr = $3, output_truncated = $4 WHERE id = $1", args);
             return null;
@@ -154,8 +268,8 @@ async function op(method, args) {
             await getPool().query(`UPDATE jobs
          SET status = $2, exit_code = $3, stdout = $4, stderr = $5,
              output_truncated = $6, error = $7, finished_at = $8,
-             expires_at = $9, http_status = $10
-         WHERE id = $1`, [
+             expires_at = $9, http_status = $10, lease_deadline = NULL
+         WHERE id = $1 AND status IN ('queued', 'running', 'orphaned')`, [
                 input.id,
                 input.status,
                 input.exitCode,
@@ -178,7 +292,10 @@ async function op(method, args) {
             const result = await getPool().query(`SELECT * FROM jobs
          WHERE request_key = $1
            AND started_at >= $2
-           AND status IN ('running', 'completed')
+           AND (
+             status IN ('running', 'completed')
+             OR (status = 'queued' AND lease_deadline IS NOT NULL AND lease_deadline >= ${PG_NOW_MS})
+           )
          ORDER BY started_at DESC
          LIMIT 1`, [args[0], cutoff]);
             return result.rows[0] ?? null;

package/dist/sqlite-driver.d.ts

@@ -9,7 +9,7 @@ export interface GatewayStatement {
 export interface GatewayDatabase {
     exec(sql: string): void;
     prepare(sql: string): GatewayStatement;
-    withTransaction<A extends unknown[]>(fn: (...args: A) => void): (...args: A) => void;
+    withTransaction<A extends unknown[], R = void>(fn: (...args: A) => R): (...args: A) => R;
     close(): void;
 }
 export declare function openDatabase(dbPath: string): GatewayDatabase;

package/dist/sqlite-driver.js

@@ -115,8 +115,9 @@ class GatewayDatabaseImpl {
             this.db.exec("BEGIN");
             this.inTransaction = true;
             try {
-                fn(...args);
+                const result = fn(...args);
                 this.db.exec("COMMIT");
+                return result;
             }
             catch (error) {
                 try {

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "llm-cli-gateway",
-  "version": "2.14.0-rc.1",
+  "version": "2.14.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "llm-cli-gateway",
-      "version": "2.14.0-rc.1",
+      "version": "2.14.1",
       "license": "MIT",
       "dependencies": {
         "@modelcontextprotocol/sdk": "^1.29.0",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llm-cli-gateway",
-  "version": "2.14.0-rc.1",
+  "version": "2.14.1",
   "mcpName": "io.github.verivus-oss/llm-cli-gateway",
   "description": "Local-first MCP gateway for cross-model review across native coding-agent CLIs and API-token LLMs with sessions, durable jobs, and validation receipts.",
   "license": "MIT",
@@ -58,6 +58,7 @@
     ".agents/skills/session-workflow/SKILL.md",
     ".agents/skills/secure-orchestration/SKILL.md",
     ".agents/skills/implement-review-fix/SKILL.md",
+    ".agents/skills/retrospective-walk/SKILL.md",
     ".agents/skills/public-demo-session/SKILL.md"
   ],
   "scripts": {