@vm0/runner 3.3.0 → 3.3.2

package/index.js

@@ -5,7 +5,6 @@ import { program } from "commander";
 
 // src/commands/start.ts
 import { Command } from "commander";
-import { writeFileSync as writeFileSync2 } from "fs";
 import { dirname, join as join2 } from "path";
 
 // src/lib/config.ts
@@ -5915,7 +5914,8 @@ var runsMainContract = c3.router({
       201: createRunResponseSchema,
       400: apiErrorSchema,
       401: apiErrorSchema,
-      404: apiErrorSchema
+      404: apiErrorSchema,
+      429: apiErrorSchema
     },
     summary: "Create and execute agent run"
   }
@@ -7477,8 +7477,10 @@ var publicApiErrorTypeSchema = z20.enum([
   // Auth failure (401)
   "not_found_error",
   // Resource missing (404)
-  "conflict_error"
+  "conflict_error",
   // Resource conflict (409)
+  "rate_limit_error"
+  // Rate limit exceeded (429)
 ]);
 var publicApiErrorSchema = z20.object({
   error: z20.object({
@@ -7726,9 +7728,7 @@ var createRunRequestSchema = z23.object({
   // volume_name -> version
 });
 var runListQuerySchema = listQuerySchema.extend({
-  agentId: z23.string().optional(),
-  status: publicRunStatusSchema.optional(),
-  since: timestampSchema.optional()
+  status: publicRunStatusSchema.optional()
 });
 var publicRunsListContract = c17.router({
   list: {
@@ -7742,7 +7742,7 @@ var publicRunsListContract = c17.router({
       500: publicApiErrorSchema
     },
     summary: "List runs",
-    description: "List runs with optional filtering by agent, status, and time"
+    description: "List runs with optional filtering by status"
   },
   create: {
     method: "POST",
@@ -7755,6 +7755,7 @@ var publicRunsListContract = c17.router({
       400: publicApiErrorSchema,
       401: publicApiErrorSchema,
       404: publicApiErrorSchema,
+      429: publicApiErrorSchema,
       500: publicApiErrorSchema
     },
     summary: "Create run",
@@ -9628,9 +9629,9 @@ async function executeJob(context, config, options = {}) {
   }
 }
 
-// src/commands/start.ts
-var activeRuns = /* @__PURE__ */ new Set();
-function writeStatusFile(statusFilePath, mode, startedAt) {
+// src/lib/runner/status.ts
+import { writeFileSync as writeFileSync2 } from "fs";
+function writeStatusFile(statusFilePath, mode, activeRuns, startedAt) {
   const status = {
     mode,
     active_runs: activeRuns.size,
@@ -9646,212 +9647,268 @@ function writeStatusFile(statusFilePath, mode, startedAt) {
     );
   }
 }
-async function executeJob2(context, config) {
-  console.log(`  Executing job ${context.runId}...`);
-  console.log(`  Prompt: ${context.prompt.substring(0, 100)}...`);
-  console.log(`  Compose version: ${context.agentComposeVersionId}`);
-  try {
-    const result = await executeJob(context, config);
-    console.log(
-      `  Job ${context.runId} execution completed with exit code ${result.exitCode}`
+function createStatusUpdater(statusFilePath, state) {
+  return () => {
+    writeStatusFile(
+      statusFilePath,
+      state.mode,
+      state.activeRuns,
+      state.startedAt
     );
-    if (result.exitCode !== 0 && result.error) {
-      console.log(`  Job ${context.runId} failed: ${result.error}`);
+  };
+}
+
+// src/lib/runner/setup.ts
+async function setupEnvironment(options) {
+  const { config } = options;
+  const datasetSuffix = process.env.AXIOM_DATASET_SUFFIX;
+  if (!datasetSuffix) {
+    throw new Error(
+      "AXIOM_DATASET_SUFFIX is required. Set to 'dev' or 'prod'."
+    );
+  }
+  initMetrics({
+    serviceName: "vm0-runner",
+    runnerLabel: config.name,
+    axiomToken: process.env.AXIOM_TOKEN,
+    environment: datasetSuffix
+  });
+  const networkCheck = checkNetworkPrerequisites();
+  if (!networkCheck.ok) {
+    console.error("Network prerequisites not met:");
+    for (const error of networkCheck.errors) {
+      console.error(`  - ${error}`);
     }
+    process.exit(1);
+  }
+  console.log("Setting up network bridge...");
+  await setupBridge();
+  console.log("Flushing bridge ARP cache...");
+  await flushBridgeArpCache();
+  console.log("Cleaning up orphaned proxy rules...");
+  await cleanupOrphanedProxyRules(config.name);
+  console.log("Cleaning up orphaned IP allocations...");
+  await cleanupOrphanedAllocations();
+  console.log("Initializing network proxy...");
+  initVMRegistry();
+  const proxyManager = initProxyManager({
+    apiUrl: config.server.url,
+    port: config.proxy.port,
+    caDir: config.proxy.ca_dir
+  });
+  let proxyEnabled = false;
+  try {
+    await proxyManager.start();
+    proxyEnabled = true;
+    console.log("Network proxy initialized successfully");
   } catch (err) {
-    const error = err instanceof Error ? err.message : "Unknown execution error";
-    console.error(`  Job ${context.runId} execution failed: ${error}`);
-    const result = await completeJob(config.server.url, context, 1, error);
-    console.log(`  Job ${context.runId} reported as ${result.status}`);
+    console.warn(
+      `Network proxy not available: ${err instanceof Error ? err.message : "Unknown error"}`
+    );
+    console.warn(
+      "Jobs with experimentalFirewall enabled will run without network interception"
+    );
   }
+  return { proxyEnabled };
 }
-var startCommand = new Command("start").description("Start the runner").option("--config <path>", "Config file path", "./runner.yaml").action(
-  // eslint-disable-next-line complexity -- TODO: refactor complex function
-  async (options) => {
-    try {
-      const config = loadConfig(options.config);
-      validateFirecrackerPaths(config.firecracker);
-      console.log("Config valid");
-      const datasetSuffix = process.env.AXIOM_DATASET_SUFFIX;
-      if (!datasetSuffix) {
-        throw new Error(
-          "AXIOM_DATASET_SUFFIX is required. Set to 'dev' or 'prod'."
-        );
+async function cleanupEnvironment(resources) {
+  if (resources.proxyEnabled) {
+    console.log("Stopping network proxy...");
+    await getProxyManager().stop();
+  }
+  console.log("Flushing metrics...");
+  await flushMetrics();
+  await shutdownMetrics();
+}
+
+// src/lib/runner/signals.ts
+function setupSignalHandlers(state, handlers) {
+  process.on("SIGINT", () => {
+    console.log("\nShutting down...");
+    handlers.onShutdown();
+    state.mode = "stopped";
+    handlers.updateStatus();
+  });
+  process.on("SIGTERM", () => {
+    console.log("\nShutting down...");
+    handlers.onShutdown();
+    state.mode = "stopped";
+    handlers.updateStatus();
+  });
+  process.on("SIGUSR1", () => {
+    if (state.mode === "running") {
+      console.log("\n[Maintenance] Entering drain mode...");
+      console.log(
+        `[Maintenance] Active jobs: ${state.activeRuns.size} (will wait for completion)`
+      );
+      state.mode = "draining";
+      handlers.updateStatus();
+    }
+  });
+}
+
+// src/lib/runner/runner.ts
+var Runner = class {
+  config;
+  statusFilePath;
+  state;
+  resources = null;
+  running = true;
+  updateStatus;
+  constructor(config, statusFilePath) {
+    this.config = config;
+    this.statusFilePath = statusFilePath;
+    this.state = {
+      mode: "running",
+      activeRuns: /* @__PURE__ */ new Set(),
+      jobPromises: /* @__PURE__ */ new Set(),
+      startedAt: /* @__PURE__ */ new Date()
+    };
+    this.updateStatus = createStatusUpdater(statusFilePath, this.state);
+  }
+  async start() {
+    this.resources = await setupEnvironment({ config: this.config });
+    setupSignalHandlers(this.state, {
+      onShutdown: () => {
+        this.running = false;
+      },
+      updateStatus: this.updateStatus
+    });
+    console.log(
+      `Starting runner '${this.config.name}' for group '${this.config.group}'...`
+    );
+    console.log(`Max concurrent jobs: ${this.config.sandbox.max_concurrent}`);
+    console.log(`Status file: ${this.statusFilePath}`);
+    console.log("Press Ctrl+C to stop");
+    console.log("");
+    this.updateStatus();
+    await this.runMainLoop();
+    if (this.state.jobPromises.size > 0) {
+      console.log(
+        `Waiting for ${this.state.jobPromises.size} active job(s) to complete...`
+      );
+      await Promise.all(this.state.jobPromises);
+    }
+    await cleanupEnvironment(this.resources);
+    this.state.mode = "stopped";
+    this.updateStatus();
+    console.log("Runner stopped");
+    process.exit(0);
+  }
+  async runMainLoop() {
+    while (this.running) {
+      if (this.state.mode === "draining") {
+        if (this.state.activeRuns.size === 0) {
+          console.log("[Maintenance] All jobs completed, exiting drain mode");
+          this.running = false;
+          break;
+        }
+        if (this.state.jobPromises.size > 0) {
+          await Promise.race(this.state.jobPromises);
+          this.updateStatus();
+        }
+        continue;
       }
-      initMetrics({
-        serviceName: "vm0-runner",
-        runnerLabel: config.name,
-        axiomToken: process.env.AXIOM_TOKEN,
-        environment: datasetSuffix
-      });
-      const networkCheck = checkNetworkPrerequisites();
-      if (!networkCheck.ok) {
-        console.error("Network prerequisites not met:");
-        for (const error of networkCheck.errors) {
-          console.error(`  - ${error}`);
+      if (this.state.activeRuns.size >= this.config.sandbox.max_concurrent) {
+        if (this.state.jobPromises.size > 0) {
+          await Promise.race(this.state.jobPromises);
+          this.updateStatus();
         }
-        process.exit(1);
+        continue;
       }
-      console.log("Setting up network bridge...");
-      await setupBridge();
-      console.log("Flushing bridge ARP cache...");
-      await flushBridgeArpCache();
-      console.log("Cleaning up orphaned proxy rules...");
-      await cleanupOrphanedProxyRules(config.name);
-      console.log("Cleaning up orphaned IP allocations...");
-      await cleanupOrphanedAllocations();
-      console.log("Initializing network proxy...");
-      initVMRegistry();
-      const proxyManager = initProxyManager({
-        apiUrl: config.server.url,
-        port: config.proxy.port,
-        caDir: config.proxy.ca_dir
-      });
-      let proxyEnabled = false;
       try {
-        await proxyManager.start();
-        proxyEnabled = true;
-        console.log("Network proxy initialized successfully");
-      } catch (err) {
-        console.warn(
-          `Network proxy not available: ${err instanceof Error ? err.message : "Unknown error"}`
-        );
-        console.warn(
-          "Jobs with experimentalFirewall enabled will run without network interception"
+        const job = await withRunnerTiming(
+          "poll",
+          () => pollForJob(this.config.server, this.config.group)
         );
-      }
-      const statusFilePath = join2(dirname(options.config), "status.json");
-      const startedAt = /* @__PURE__ */ new Date();
-      const state = { mode: "running" };
-      const updateStatus = () => {
-        writeStatusFile(statusFilePath, state.mode, startedAt);
-      };
-      console.log(
-        `Starting runner '${config.name}' for group '${config.group}'...`
-      );
-      console.log(`Max concurrent jobs: ${config.sandbox.max_concurrent}`);
-      console.log(`Status file: ${statusFilePath}`);
-      console.log("Press Ctrl+C to stop");
-      console.log("");
-      updateStatus();
-      let running = true;
-      process.on("SIGINT", () => {
-        console.log("\nShutting down...");
-        running = false;
-        state.mode = "stopped";
-        updateStatus();
-      });
-      process.on("SIGTERM", () => {
-        console.log("\nShutting down...");
-        running = false;
-        state.mode = "stopped";
-        updateStatus();
-      });
-      process.on("SIGUSR1", () => {
-        if (state.mode === "running") {
-          console.log("\n[Maintenance] Entering drain mode...");
-          console.log(
-            `[Maintenance] Active jobs: ${activeRuns.size} (will wait for completion)`
+        if (!job) {
+          await new Promise(
+            (resolve) => setTimeout(resolve, this.config.sandbox.poll_interval_ms)
           );
-          state.mode = "draining";
-          updateStatus();
-        }
-      });
-      const jobPromises = /* @__PURE__ */ new Set();
-      while (running) {
-        if (state.mode === "draining") {
-          if (activeRuns.size === 0) {
-            console.log(
-              "[Maintenance] All jobs completed, exiting drain mode"
-            );
-            running = false;
-            break;
-          }
-          if (jobPromises.size > 0) {
-            await Promise.race(jobPromises);
-            updateStatus();
-          }
-          continue;
-        }
-        if (activeRuns.size >= config.sandbox.max_concurrent) {
-          if (jobPromises.size > 0) {
-            await Promise.race(jobPromises);
-            updateStatus();
-          }
           continue;
         }
-        try {
-          const job = await withRunnerTiming(
-            "poll",
-            () => pollForJob(config.server, config.group)
-          );
-          if (!job) {
-            await new Promise(
-              (resolve) => setTimeout(resolve, config.sandbox.poll_interval_ms)
-            );
-            continue;
-          }
-          console.log(`Found job: ${job.runId}`);
-          try {
-            const context = await withRunnerTiming(
-              "claim",
-              () => claimJob(config.server, job.runId)
-            );
-            console.log(`Claimed job: ${context.runId}`);
-            activeRuns.add(context.runId);
-            updateStatus();
-            const jobPromise = executeJob2(context, config).catch((error) => {
-              console.error(
-                `Job ${context.runId} failed:`,
-                error instanceof Error ? error.message : "Unknown error"
-              );
-            }).finally(() => {
-              activeRuns.delete(context.runId);
-              jobPromises.delete(jobPromise);
-              updateStatus();
-            });
-            jobPromises.add(jobPromise);
-          } catch (error) {
-            console.log(
-              `Could not claim job ${job.runId}:`,
-              error instanceof Error ? error.message : "Unknown error"
-            );
-          }
-        } catch (error) {
-          console.error(
-            "Polling error:",
-            error instanceof Error ? error.message : "Unknown error"
-          );
-          await new Promise((resolve) => setTimeout(resolve, 2e3));
-        }
+        console.log(`Found job: ${job.runId}`);
+        await this.processJob(job.runId);
+      } catch (error) {
+        console.error(
+          "Polling error:",
+          error instanceof Error ? error.message : "Unknown error"
+        );
+        await new Promise((resolve) => setTimeout(resolve, 2e3));
       }
-      if (jobPromises.size > 0) {
-        console.log(
-          `Waiting for ${jobPromises.size} active job(s) to complete...`
+    }
+  }
+  async processJob(runId) {
+    try {
+      const context = await withRunnerTiming(
+        "claim",
+        () => claimJob(this.config.server, runId)
+      );
+      console.log(`Claimed job: ${context.runId}`);
+      this.state.activeRuns.add(context.runId);
+      this.updateStatus();
+      const jobPromise = this.executeJob(context).catch((error) => {
+        console.error(
+          `Job ${context.runId} failed:`,
+          error instanceof Error ? error.message : "Unknown error"
         );
-        await Promise.all(jobPromises);
-      }
-      if (proxyEnabled) {
-        console.log("Stopping network proxy...");
-        await getProxyManager().stop();
-      }
-      console.log("Flushing metrics...");
-      await flushMetrics();
-      await shutdownMetrics();
-      state.mode = "stopped";
-      updateStatus();
-      console.log("Runner stopped");
-      process.exit(0);
+      }).finally(() => {
+        this.state.activeRuns.delete(context.runId);
+        this.state.jobPromises.delete(jobPromise);
+        this.updateStatus();
+      });
+      this.state.jobPromises.add(jobPromise);
     } catch (error) {
-      if (error instanceof Error) {
-        console.error(`Error: ${error.message}`);
-      } else {
-        console.error("An unknown error occurred");
-      }
-      process.exit(1);
+      console.log(
+        `Could not claim job ${runId}:`,
+        error instanceof Error ? error.message : "Unknown error"
+      );
     }
   }
-);
+  async executeJob(context) {
+    console.log(`  Executing job ${context.runId}...`);
+    console.log(`  Prompt: ${context.prompt.substring(0, 100)}...`);
+    console.log(`  Compose version: ${context.agentComposeVersionId}`);
+    try {
+      const result = await executeJob(context, this.config);
+      console.log(
+        `  Job ${context.runId} execution completed with exit code ${result.exitCode}`
+      );
+      if (result.exitCode !== 0 && result.error) {
+        console.log(`  Job ${context.runId} failed: ${result.error}`);
+      }
+    } catch (err) {
+      const error = err instanceof Error ? err.message : "Unknown execution error";
+      console.error(`  Job ${context.runId} execution failed: ${error}`);
+      const result = await completeJob(
+        this.config.server.url,
+        context,
+        1,
+        error
+      );
+      console.log(`  Job ${context.runId} reported as ${result.status}`);
+    }
+  }
+};
+
+// src/commands/start.ts
+var startCommand = new Command("start").description("Start the runner").option("--config <path>", "Config file path", "./runner.yaml").action(async (options) => {
+  try {
+    const config = loadConfig(options.config);
+    validateFirecrackerPaths(config.firecracker);
+    console.log("Config valid");
+    const statusFilePath = join2(dirname(options.config), "status.json");
+    const runner = new Runner(config, statusFilePath);
+    await runner.start();
+  } catch (error) {
+    if (error instanceof Error) {
+      console.error(`Error: ${error.message}`);
+    } else {
+      console.error("An unknown error occurred");
+    }
+    process.exit(1);
+  }
+});
 
 // src/commands/doctor.ts
 import { Command as Command2 } from "commander";
@@ -10437,7 +10494,7 @@ var benchmarkCommand = new Command4("benchmark").description(
 });
 
 // src/index.ts
-var version = true ? "3.3.0" : "0.1.0";
+var version = true ? "3.3.2" : "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": "3.3.0",
+  "version": "3.3.2",
   "description": "Self-hosted runner for VM0 agents",
   "repository": {
     "type": "git",