fix(runner): persist codex home in session pvc (#242)

src/backend/codex-stdio.ts

@@ -1,7 +1,7 @@
 import { spawn, type ChildProcessWithoutNullStreams } from "node:child_process";
 import { createHash } from "node:crypto";
 import { accessSync, constants as fsConstants, readdirSync, readFileSync } from "node:fs";
-import { chmod, copyFile, mkdir } from "node:fs/promises";
+import { chmod, copyFile, mkdir, writeFile } from "node:fs/promises";
 import path from "node:path";
 import * as readline from "node:readline";
 import type { BackendEvent, BackendProfile, BackendTurnResult, CommandRecord, FailureKind, InitialPromptAssembly, JsonRecord, JsonValue, RunRecord, TerminalStatus } from "../common/types.js";
@@ -852,6 +852,7 @@ async function prepareProjectedCodexHome(codexHome: string, projectedHome: strin
       await copyFile(path.join(projectedHome, fileName), path.join(codexHome, fileName));
       await chmod(path.join(codexHome, fileName), 0o600);
     }
+    await normalizeCopiedCodexConfig(codexHome);
     return null;
   } catch (error) {
     const payload = {
@@ -875,6 +876,29 @@ async function prepareProjectedCodexHome(codexHome: string, projectedHome: strin
   }
 }
 
+async function normalizeCopiedCodexConfig(codexHome: string): Promise<void> {
+  const configPath = path.join(codexHome, "config.toml");
+  const modelCatalogPath = path.join(codexHome, "model-catalog.json");
+  if (!fileReadable(modelCatalogPath).readable) return;
+  let configToml = "";
+  try {
+    configToml = readFileSync(configPath, "utf8");
+  } catch {
+    return;
+  }
+  const normalized = configToml.replace(
+    /(^\s*model_catalog_json\s*=\s*")([^"]+)("\s*$)/mu,
+    (_match, prefix: string, _oldValue: string, suffix: string) => `${prefix}${tomlBasicStringValue(modelCatalogPath)}${suffix}`,
+  );
+  if (normalized === configToml) return;
+  await writeFile(configPath, normalized, { mode: 0o600 });
+  await chmod(configPath, 0o600);
+}
+
+function tomlBasicStringValue(value: string): string {
+  return value.replace(/\\/gu, "\\\\").replace(/"/gu, "\\\"");
+}
+
 function codexHomeReadiness(codexHome: string, profile: BackendProfile): BackendTurnResult | null {
   const auth = fileReadable(`${codexHome}/auth.json`);
   const config = fileReadable(`${codexHome}/config.toml`);

src/runner/k8s-job.ts

@@ -250,7 +250,7 @@ export function renderRunnerJobManifest(options: RunnerJobRenderOptions): { mani
 
 function runnerEnv(options: RunnerJobRenderOptions, context: { namespace: string; jobName: string; runnerJobId: string; runnerId: string; attemptId: string; sourceCommit: string; secretRefs: CredentialProjection[]; toolCredentials: ToolCredentialProjection[]; sessionPvc: RunnerSessionPvcOptions | undefined; runnerIdleTimeoutMs: number; missingTerminalAfterToolTimeoutMs: number; backendRetryMaxAttempts: number; backendRetryInitialBackoffMs: number; backendRetryMaxBackoffMs: number }): JsonRecord[] {
   const selectedSecret = context.secretRefs.find((item) => item.profile === options.run.backendProfile);
-  const codexHome = selectedSecret?.runtimeMountPath ?? defaultRuntimeHome(options.run.backendProfile);
+  const codexHome = runnerCodexHome(options.run.backendProfile, selectedSecret, context.sessionPvc);
   const bootRepoUrl = optionalString(options.bootRepoUrl) ?? defaultBootRepoUrl;
   return dedupeEnvVars([
     { name: "AGENTRUN_MGR_URL", value: options.managerUrl },
@@ -285,6 +285,7 @@ function runnerEnv(options: RunnerJobRenderOptions, context: { namespace: string
     { name: "AGENTRUN_RUNNER_POLL_INTERVAL_MS", value: "250" },
     { name: "HOME", value: "/home/agentrun" },
     { name: "CODEX_HOME", value: codexHome },
+    { name: "AGENTRUN_CODEX_HOME_STORAGE_KIND", value: context.sessionPvc ? "session-pvc" : "runner-home" },
     ...runnerOtelEnvVars(process.env),
     ...(selectedSecret ? [{ name: "AGENTRUN_CODEX_SECRET_HOME", value: selectedSecret.projectionMountPath }] : []),
     ...(context.sessionPvc ? [
@@ -580,6 +581,11 @@ function normalizeMountPath(value: string | undefined, profile: string): string
   return value;
 }
 
+function runnerCodexHome(profile: string, selectedSecret: CredentialProjection | undefined, sessionPvc: RunnerSessionPvcOptions | undefined): string {
+  if (!sessionPvc) return selectedSecret?.runtimeMountPath ?? defaultRuntimeHome(profile);
+  return `${sessionPvc.mountPath}/codex-home/${sanitizeVolumeName(profile)}`;
+}
+
 function defaultRuntimeHome(profile: string): string {
   return `/home/agentrun/.codex-${sanitizeVolumeName(profile)}`;
 }