impel-cli 0.20.17 → 0.20.18

package/RELEASE_NOTES.md

@@ -1,5 +1,13 @@
 # Release notes
 
+## 0.20.18 — Recoverable Codex profiling cohorts
+
+- Widens the bounded retry window for transient current-tenant reads while
+  still rejecting an exact tenant mismatch before a Codex attempt starts.
+- Stops scheduling new work after an attempt-infrastructure failure, waits for
+  active workers to settle, and retains an explicitly incomplete private
+  aggregate so valid samples are not silently lost.
+
 ## 0.20.17 — Managed desktop task views
 
 - Enables the reviewed MCP Apps renderer only in exact managed Codex Desktop

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "impel-cli",
-  "version": "0.20.17",
+  "version": "0.20.18",
   "description": "Prepare isolated Claude and Codex workspaces for every accessible Impel tenant",
   "type": "module",
   "bin": {

package/scripts/profile-native-codex.mjs

@@ -14,6 +14,7 @@ const PROFILE_SCHEMA = "impel.native-codex-profile.v1";
 const DEFAULT_TIMEOUT_MS = 5 * 60 * 1000;
 const DEFAULT_GRACE_MS = 5_000;
 const MAX_CAPTURE_BYTES = 256 * 1024 * 1024;
+const TENANT_PREFLIGHT_RETRY_DELAYS_MS = [250, 750, 1_500];
 
 function privateDirectory(directory) {
   if (fs.existsSync(directory) && fs.lstatSync(directory).isSymbolicLink()) {
@@ -211,16 +212,17 @@ function runCaptured(command, args, { environment = process.env, timeoutMs = 60_
 }
 
 async function currentTenant(impelBinary, environment) {
-  for (let attempt = 0; attempt < 2; attempt += 1) {
+  for (let attempt = 0; attempt <= TENANT_PREFLIGHT_RETRY_DELAYS_MS.length; attempt += 1) {
     const invocation = impelInvocation(impelBinary, ["tenant", "current"], environment);
     const result = await runCaptured(invocation.command, invocation.args, { environment });
     const lines = result.stdout.split(/\r?\n/u).map((line) => line.trim()).filter(Boolean);
     if (result.code === 0 && lines.length === 1 && /^[A-Za-z0-9_.:-]{1,160}$/u.test(lines[0])) {
       return lines[0];
     }
-    if (attempt === 0) await delay(250);
+    const retryDelay = TENANT_PREFLIGHT_RETRY_DELAYS_MS[attempt];
+    if (retryDelay !== undefined) await delay(retryDelay);
   }
-  throw new Error("could not establish one exact current Impel tenant after two bounded attempts");
+  throw new Error("could not establish one exact current Impel tenant after bounded attempts");
 }
 
 export async function assertExpectedTenant({ expectedTenant, impelBinary = "impel", environment = process.env }) {
@@ -495,19 +497,38 @@ async function profileAttempt(options, sessionDir, index) {
   return record;
 }
 
-async function runWorkers(options, sessionDir) {
+function profilerFailureClass(error) {
+  if (/could not establish one exact current Impel tenant/u.test(error?.message || "")) {
+    return "tenant-preflight-unavailable";
+  }
+  if (/selected tenant .* does not match expected tenant/u.test(error?.message || "")) {
+    return "tenant-mismatch";
+  }
+  return "attempt-infrastructure";
+}
+
+export async function runWorkers(options, sessionDir, profileAttemptImpl = profileAttempt) {
   const results = new Array(options.attempts);
   let next = 0;
+  let abort = null;
   const worker = async () => {
     for (;;) {
+      if (abort) return;
       const index = next;
       next += 1;
       if (index >= options.attempts) return;
-      results[index] = await profileAttempt(options, sessionDir, index);
+      try {
+        results[index] = await profileAttemptImpl(options, sessionDir, index);
+      } catch (error) {
+        abort ||= {
+          failureClass: profilerFailureClass(error),
+          attemptId: `${options.cohort}-${String(index + 1 + options.attemptOffset).padStart(3, "0")}`,
+        };
+      }
     }
   };
   await Promise.all(Array.from({ length: options.concurrency }, () => worker()));
-  return results;
+  return { attempts: results.filter(Boolean), abort };
 }
 
 async function restoreTenant(originalTenant, options) {
@@ -553,7 +574,7 @@ async function main(argv) {
   privateDirectory(sessionDir);
   let restored = false;
   try {
-    const attempts = await runWorkers(options, sessionDir);
+    const result = await runWorkers(options, sessionDir);
     const aggregate = {
       schema: PROFILE_SCHEMA,
       sessionId,
@@ -564,12 +585,18 @@ async function main(argv) {
       cohort: options.cohort,
       cliVersion: options.hostBuild,
       promptSha256: options.promptSha256,
-      attempts,
+      complete: result.abort === null && result.attempts.length === options.attempts,
+      expectedAttempts: options.attempts,
+      abort: result.abort,
+      attempts: result.attempts,
     };
     const aggregatePath = path.join(sessionDir, "aggregate.json");
     privateWrite(aggregatePath, `${JSON.stringify(aggregate, null, 2)}\n`);
     process.stderr.write(`Private Codex profile written under ${sessionDir}\n`);
     process.stdout.write(`${JSON.stringify(aggregate)}\n`);
+    if (!aggregate.complete) {
+      throw new Error(`cohort aborted (${aggregate.abort?.failureClass || "incomplete"}); partial aggregate retained`);
+    }
   } finally {
     restored = await restoreTenant(originalTenant, options);
     if (!restored) process.stderr.write("profile-native-codex: could not verify restoration of the original tenant\n");