@vm0/runner 3.12.0 → 3.12.1

package/index.js

@@ -416,6 +416,37 @@ async function withFileLock(path9, fn, options) {
   }
 }
 
+// src/lib/utils/process.ts
+import { execSync } from "child_process";
+function isProcessRunning(pid) {
+  try {
+    process.kill(pid, 0);
+    return true;
+  } catch (err) {
+    if (err instanceof Error && "code" in err && err.code === "EPERM") {
+      return true;
+    }
+    return false;
+  }
+}
+function killProcessTree(pid) {
+  try {
+    const childPidsStr = execSync(`pgrep -P ${pid} 2>/dev/null || true`, {
+      encoding: "utf-8"
+    }).trim();
+    if (childPidsStr) {
+      const childPids = childPidsStr.split("\n").map((p) => parseInt(p, 10));
+      for (const childPid of childPids) {
+        if (!isNaN(childPid)) {
+          killProcessTree(childPid);
+        }
+      }
+    }
+    process.kill(pid, "SIGKILL");
+  } catch {
+  }
+}
+
 // src/lib/utils/exec.ts
 import { exec } from "child_process";
 import { promisify } from "util";
@@ -549,14 +580,6 @@ function makeNsName(runnerIdx, nsIdx) {
 function makeVethName(runnerIdx, nsIdx) {
   return `${VETH_PREFIX}${runnerIdx}-${nsIdx}`;
 }
-function isPidAlive(pid) {
-  try {
-    process.kill(pid, 0);
-    return true;
-  } catch {
-    return false;
-  }
-}
 async function deleteIptablesRulesByComment(comment) {
   const deleteFromTable = async (table) => {
     try {
@@ -625,7 +648,7 @@ var NetnsPool = class _NetnsPool {
       const data = read();
       const orphaned = [];
       for (const [runnerIdx, runner] of Object.entries(data.runners)) {
-        if (!isPidAlive(runner.pid)) {
+        if (!isProcessRunning(runner.pid)) {
           orphaned.push({
             runnerIdx,
             namespaces: Object.entries(runner.namespaces).map(
@@ -662,7 +685,7 @@ var NetnsPool = class _NetnsPool {
       const data = read();
       for (const { runnerIdx } of orphanedData) {
         const runner = data.runners[runnerIdx];
-        if (runner && !isPidAlive(runner.pid)) {
+        if (runner && !isProcessRunning(runner.pid)) {
           delete data.runners[runnerIdx];
         }
       }
@@ -1671,8 +1694,8 @@ var FirecrackerVM = class {
    * since we want to clean up as much as possible even if some parts fail.
    */
   async cleanup() {
-    if (this.process && !this.process.killed) {
-      this.process.kill("SIGKILL");
+    if (this.process && !this.process.killed && this.process.pid) {
+      killProcessTree(this.process.pid);
       this.process = null;
     }
     if (this.netns) {
@@ -7884,6 +7907,7 @@ var modelProviderTypeSchema = z19.enum([
   "minimax-api-key",
   "deepseek-api-key",
   "zai-api-key",
+  "azure-foundry",
   "aws-bedrock"
 ]);
 var modelProviderFrameworkSchema = z19.enum(["claude-code", "codex"]);
@@ -10472,12 +10496,12 @@ function createStatusUpdater(statusFilePath, state) {
 }
 
 // src/lib/firecracker/network.ts
-import { execSync, exec as exec3 } from "child_process";
+import { execSync as execSync2, exec as exec3 } from "child_process";
 import { promisify as promisify3 } from "util";
 var execAsync3 = promisify3(exec3);
 function commandExists(cmd) {
   try {
-    execSync(`which ${cmd}`, { stdio: "ignore" });
+    execSync2(`which ${cmd}`, { stdio: "ignore" });
     return true;
   } catch {
     return false;
@@ -10492,7 +10516,7 @@ function checkNetworkPrerequisites() {
     }
   }
   try {
-    execSync("sudo -n true 2>/dev/null", { stdio: "ignore" });
+    execSync2("sudo -n true 2>/dev/null", { stdio: "ignore" });
   } catch {
     errors.push(
       "Root/sudo access required for network configuration. Please run with sudo or configure sudoers."
@@ -10518,17 +10542,6 @@ import path7 from "path";
 var logger11 = createLogger("RunnerLock");
 var DEFAULT_PID_FILE = runtimePaths.runnerPid;
 var currentPidFile = null;
-function isProcessRunning(pid) {
-  try {
-    process.kill(pid, 0);
-    return true;
-  } catch (err) {
-    if (err instanceof Error && "code" in err && err.code === "EPERM") {
-      return true;
-    }
-    return false;
-  }
-}
 function acquireRunnerLock(options = {}) {
   const pidFile = options.pidFile ?? DEFAULT_PID_FILE;
   const runDir = path7.dirname(pidFile);
@@ -10909,6 +10922,7 @@ var startCommand = new Command("start").description("Start the runner").option("
 // src/commands/doctor.ts
 import { Command as Command2 } from "commander";
 import { existsSync as existsSync5, readFileSync as readFileSync3, readdirSync as readdirSync2 } from "fs";
+import { execSync as execSync3 } from "child_process";
 
 // src/lib/firecracker/process.ts
 import { readdirSync, readFileSync as readFileSync2, existsSync as existsSync4 } from "fs";
@@ -10916,12 +10930,21 @@ import path8 from "path";
 function parseFirecrackerCmdline(cmdline) {
   const args = cmdline.split("\0");
   if (!args[0]?.includes("firecracker")) return null;
+  let filePath;
   const sockIdx = args.indexOf("--api-sock");
-  const socketPath = args[sockIdx + 1];
-  if (sockIdx === -1 || !socketPath) return null;
-  const match = socketPath.match(/vm0-([a-f0-9]+)\/firecracker\.sock$/);
+  if (sockIdx !== -1) {
+    filePath = args[sockIdx + 1];
+  }
+  if (!filePath) {
+    const configIdx = args.indexOf("--config-file");
+    if (configIdx !== -1) {
+      filePath = args[configIdx + 1];
+    }
+  }
+  if (!filePath) return null;
+  const match = filePath.match(/vm0-([a-f0-9]+)\//);
   if (!match?.[1]) return null;
-  return { vmId: createVmId(match[1]), socketPath };
+  return createVmId(match[1]);
 }
 function parseMitmproxyCmdline(cmdline) {
   if (!cmdline.includes("mitmproxy") && !cmdline.includes("mitmdump")) {
@@ -10949,9 +10972,9 @@ function findFirecrackerProcesses() {
     if (!existsSync4(cmdlinePath)) continue;
     try {
       const cmdline = readFileSync2(cmdlinePath, "utf-8");
-      const parsed = parseFirecrackerCmdline(cmdline);
-      if (parsed) {
-        processes.push({ pid, ...parsed });
+      const vmId = parseFirecrackerCmdline(cmdline);
+      if (vmId) {
+        processes.push({ pid, vmId });
       }
     } catch {
       continue;
@@ -10964,33 +10987,25 @@ function findProcessByVmId(vmId) {
   const vmIdStr = vmIdValue(vmId);
   return processes.find((p) => vmIdValue(p.vmId) === vmIdStr) || null;
 }
-function isProcessRunning2(pid) {
-  try {
-    process.kill(pid, 0);
-    return true;
-  } catch {
-    return false;
-  }
-}
 async function killProcess(pid, timeoutMs = 5e3) {
-  if (!isProcessRunning2(pid)) return true;
+  if (!isProcessRunning(pid)) return true;
   try {
     process.kill(pid, "SIGTERM");
   } catch {
-    return !isProcessRunning2(pid);
+    return !isProcessRunning(pid);
   }
   const startTime = Date.now();
   while (Date.now() - startTime < timeoutMs) {
-    if (!isProcessRunning2(pid)) return true;
+    if (!isProcessRunning(pid)) return true;
     await new Promise((resolve) => setTimeout(resolve, 100));
   }
-  if (isProcessRunning2(pid)) {
+  if (isProcessRunning(pid)) {
     try {
       process.kill(pid, "SIGKILL");
     } catch {
     }
   }
-  return !isProcessRunning2(pid);
+  return !isProcessRunning(pid);
 }
 function findMitmproxyProcess() {
   const procDir = "/proc";
@@ -11018,15 +11033,26 @@ function findMitmproxyProcess() {
   return null;
 }
 
+// src/lib/runner/types.ts
+import { z as z30 } from "zod";
+var RunnerModeSchema = z30.enum(["running", "draining", "stopping", "stopped"]);
+var RunnerStatusSchema = z30.object({
+  mode: RunnerModeSchema,
+  active_runs: z30.number(),
+  active_run_ids: z30.array(z30.string()),
+  started_at: z30.string(),
+  updated_at: z30.string()
+});
+
 // src/commands/doctor.ts
-function displayRunnerStatus(statusFilePath) {
+function displayRunnerStatus(statusFilePath, warnings) {
   if (!existsSync5(statusFilePath)) {
     console.log("Mode: unknown (no status.json)");
     return null;
   }
   try {
-    const status = JSON.parse(
-      readFileSync3(statusFilePath, "utf-8")
+    const status = RunnerStatusSchema.parse(
+      JSON.parse(readFileSync3(statusFilePath, "utf-8"))
     );
     console.log(`Mode: ${status.mode}`);
     if (status.started_at) {
@@ -11037,10 +11063,11 @@ function displayRunnerStatus(statusFilePath) {
     return status;
   } catch {
     console.log("Mode: unknown (status.json unreadable)");
+    warnings.push({ message: "status.json exists but cannot be parsed" });
     return null;
   }
 }
-async function checkApiConnectivity(config) {
+async function checkApiConnectivity(config, warnings) {
   console.log("API Connectivity:");
   try {
     await pollForJob(config.server, config.group);
@@ -11051,6 +11078,9 @@ async function checkApiConnectivity(config) {
     console.log(
       `    Error: ${error instanceof Error ? error.message : "Unknown error"}`
     );
+    warnings.push({
+      message: `Cannot connect to API: ${error instanceof Error ? error.message : "Unknown error"}`
+    });
   }
 }
 async function checkNetwork(config, warnings) {
@@ -11086,8 +11116,7 @@ function buildJobInfo(status, processes) {
       jobs.push({
         runId,
         vmId,
-        hasProcess: !!proc,
-        pid: proc?.pid
+        firecrackerPid: proc?.pid
       });
     }
   }
@@ -11101,13 +11130,61 @@ function displayRuns(jobs, maxConcurrent) {
   }
   console.log("  Run ID                                VM ID       Status");
   for (const job of jobs) {
-    const statusText = job.hasProcess ? `\u2713 Running (PID ${job.pid})` : "\u26A0\uFE0F No process";
+    const statusText = job.firecrackerPid ? `\u2713 Running (PID ${job.firecrackerPid})` : "\u26A0\uFE0F No process";
     console.log(`  ${job.runId}  ${job.vmId}    ${statusText}`);
   }
 }
-function detectOrphanResources(jobs, processes, workspaces, statusVmIds, warnings) {
+async function findOrphanNetworkNamespaces(warnings) {
+  let allNamespaces = [];
+  try {
+    const output = execSync3("ip netns list 2>/dev/null || true", {
+      encoding: "utf-8"
+    });
+    allNamespaces = output.split("\n").map((line) => line.split(" ")[0] ?? "").filter((ns) => ns.startsWith(NS_PREFIX));
+  } catch (err) {
+    warnings.push({
+      message: `Failed to list network namespaces: ${err instanceof Error ? err.message : "Unknown error"}`
+    });
+    return [];
+  }
+  if (allNamespaces.length === 0) {
+    return [];
+  }
+  const registryPath = runtimePaths.netnsRegistry;
+  if (!existsSync5(registryPath)) {
+    return allNamespaces;
+  }
+  try {
+    return await withFileLock(registryPath, async () => {
+      const registry = RegistrySchema.parse(
+        JSON.parse(readFileSync3(registryPath, "utf-8"))
+      );
+      const aliveNamespaces = /* @__PURE__ */ new Set();
+      for (const [runnerIdx, runner] of Object.entries(registry.runners)) {
+        if (isProcessRunning(runner.pid)) {
+          for (const nsIdx of Object.keys(runner.namespaces)) {
+            aliveNamespaces.add(`${NS_PREFIX}${runnerIdx}-${nsIdx}`);
+          }
+        }
+      }
+      const orphans = [];
+      for (const ns of allNamespaces) {
+        if (!aliveNamespaces.has(ns)) {
+          orphans.push(ns);
+        }
+      }
+      return orphans;
+    });
+  } catch (err) {
+    warnings.push({
+      message: `Failed to read netns registry: ${err instanceof Error ? err.message : "Unknown error"}`
+    });
+    return [];
+  }
+}
+async function detectOrphanResources(jobs, processes, workspaces, statusVmIds, warnings) {
   for (const job of jobs) {
-    if (!job.hasProcess) {
+    if (!job.firecrackerPid) {
       warnings.push({
         message: `Run ${job.vmId} in status.json but no Firecracker process running`
       });
@@ -11121,6 +11198,12 @@ function detectOrphanResources(jobs, processes, workspaces, statusVmIds, warning
       });
     }
   }
+  const orphanNetns = await findOrphanNetworkNamespaces(warnings);
+  for (const ns of orphanNetns) {
+    warnings.push({
+      message: `Orphan network namespace: ${ns} (runner process not running)`
+    });
+  }
   for (const ws of workspaces) {
     const vmId = runnerPaths.extractVmId(ws);
     if (!processVmIds.has(vmId) && !statusVmIds.has(vmId)) {
@@ -11157,9 +11240,9 @@ var doctorCommand = new Command2("doctor").description("Diagnose runner health,
     const workspacesDir = runnerPaths.workspacesDir(config.base_dir);
     const warnings = [];
     console.log(`Runner: ${config.name}`);
-    const status = displayRunnerStatus(statusFilePath);
+    const status = displayRunnerStatus(statusFilePath, warnings);
     console.log("");
-    await checkApiConnectivity(config);
+    await checkApiConnectivity(config, warnings);
     console.log("");
     await checkNetwork(config, warnings);
     console.log("");
@@ -11168,7 +11251,13 @@ var doctorCommand = new Command2("doctor").description("Diagnose runner health,
     const { jobs, statusVmIds } = buildJobInfo(status, processes);
     displayRuns(jobs, config.sandbox.max_concurrent);
     console.log("");
-    detectOrphanResources(jobs, processes, workspaces, statusVmIds, warnings);
+    await detectOrphanResources(
+      jobs,
+      processes,
+      workspaces,
+      statusVmIds,
+      warnings
+    );
     displayWarnings(warnings);
     process.exit(warnings.length > 0 ? 1 : 0);
   } catch (error) {
@@ -11251,8 +11340,8 @@ var killCommand = new Command3("kill").description("Force terminate a run and cl
       }
       if (runId && existsSync6(statusFilePath)) {
         try {
-          const status = JSON.parse(
-            readFileSync4(statusFilePath, "utf-8")
+          const status = RunnerStatusSchema.parse(
+            JSON.parse(readFileSync4(statusFilePath, "utf-8"))
           );
           const oldCount = status.active_runs;
           status.active_run_ids = status.active_run_ids.filter(
@@ -11309,8 +11398,8 @@ function resolveRunId(input, statusFilePath) {
   }
   if (existsSync6(statusFilePath)) {
     try {
-      const status = JSON.parse(
-        readFileSync4(statusFilePath, "utf-8")
+      const status = RunnerStatusSchema.parse(
+        JSON.parse(readFileSync4(statusFilePath, "utf-8"))
       );
       const match = status.active_run_ids.find(
         (id) => id.startsWith(input)
@@ -11658,7 +11747,7 @@ var snapshotCommand = new Command5("snapshot").description("Generate a Firecrack
 );
 
 // src/index.ts
-var version = true ? "3.12.0" : "0.1.0";
+var version = true ? "3.12.1" : "0.1.0";
 program.name("vm0-runner").version(version).description("Self-hosted runner for VM0 agents");
 program.addCommand(startCommand);
 program.addCommand(doctorCommand);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vm0/runner",
-  "version": "3.12.0",
+  "version": "3.12.1",
   "description": "Self-hosted runner for VM0 agents",
   "repository": {
     "type": "git",