@sagentlab/navarch-runtime 0.1.77 → 0.1.78

package/dist/completion-retry.cjs

@@ -0,0 +1,48 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.COMPLETION_TRANSIENT_RETRY_DELAYS_MS = void 0;
+exports.isTransientCompletionError = isTransientCompletionError;
+exports.completeWithTransientRetry = completeWithTransientRetry;
+const api_cjs_1 = require("./api.cjs");
+/**
+ * Backoff before each re-send of a completion that failed in transit. About
+ * two minutes in total: long enough to ride out a Wi-Fi drop or an edge blip,
+ * short enough that the lease is still live when the report finally lands.
+ */
+exports.COMPLETION_TRANSIENT_RETRY_DELAYS_MS = [2_000, 5_000, 15_000, 30_000, 60_000];
+/** Statuses the edge or a gateway answers for an unreachable origin; the completion route never chose them. */
+const TRANSIENT_STATUSES = new Set([408, 429, 502, 503, 504, 520, 521, 522, 523, 524]);
+/**
+ * A failure that says nothing about the completion itself: the request never
+ * got an answer from the completion route. A plain 500 is the route's own
+ * answer and keeps the existing crash path.
+ */
+function isTransientCompletionError(err) {
+    return (err instanceof api_cjs_1.NavarchTransportError ||
+        (err instanceof api_cjs_1.NavarchApiError && TRANSIENT_STATUSES.has(err.status)));
+}
+/**
+ * Sends a lease completion, re-sending it while the control plane is
+ * unreachable. Without this, one dropped connection turned a finished session
+ * (pull request already open) into `Session crashed:`, which spent an
+ * infrastructure retry and discarded the report and evidence the review loop
+ * needed (carova#310). A re-send whose earlier copy did land gets the
+ * `already_released` 409, which callers already treat as success.
+ */
+async function completeWithTransientRetry(send, options = {}) {
+    const delays = options.delaysMs ?? exports.COMPLETION_TRANSIENT_RETRY_DELAYS_MS;
+    const sleep = options.sleep ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));
+    for (let attempt = 0;; attempt += 1) {
+        try {
+            await send();
+            return;
+        }
+        catch (err) {
+            const delayMs = delays[attempt];
+            if (delayMs === undefined || !isTransientCompletionError(err))
+                throw err;
+            options.onRetry?.(attempt + 1, delayMs, err);
+            await sleep(delayMs);
+        }
+    }
+}

package/dist/session.cjs

@@ -25,6 +25,7 @@ const delivery_cjs_1 = require("./delivery.cjs");
 const worktree_guard_cjs_1 = require("./worktree-guard.cjs");
 const adapter_watchdog_cjs_1 = require("./adapter-watchdog.cjs");
 const lease_heartbeat_cjs_1 = require("./lease-heartbeat.cjs");
+const completion_retry_cjs_1 = require("./completion-retry.cjs");
 const adapter_capacity_cjs_1 = require("./adapter-capacity.cjs");
 const agent_auth_cjs_1 = require("./agent-auth.cjs");
 const browser_evidence_cjs_1 = require("./browser-evidence.cjs");
@@ -730,7 +731,9 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
                 transcript_url: completion.transcript_url,
             };
             try {
-                await api.completeLease(leaseId, completion);
+                await (0, completion_retry_cjs_1.completeWithTransientRetry)(() => api.completeLease(leaseId, completion), {
+                    onRetry: (attempt, delayMs, err) => log.warn(`completion for ${leaseId} did not reach the control plane (${(0, redact_cjs_1.redactText)(String(err), knownSecrets)}); re-sending in ${delayMs / 1000}s (${attempt}).`),
+                });
                 if (deliveredPullRequest && completion.status === "completed" && gitWorktree.restoredHostedCheckpoint) {
                     // The pull request carries the hosted attempt's work now; a later
                     // hosted attempt must not resume the stale snapshot.
@@ -766,7 +769,7 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
                     log.error(`completion for ${leaseId} was rejected as ${rejection.code}, which is not remediable; failing the lease.`);
                 }
                 try {
-                    await api.completeLease(leaseId, {
+                    await (0, completion_retry_cjs_1.completeWithTransientRetry)(() => api.completeLease(leaseId, {
                         ...completion,
                         status: "failed",
                         report: remediable
@@ -774,7 +777,7 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
                             : `${redactedRejection}\n\n---\n\n${completion.report}`,
                         failure_summary: redactedRejection,
                         exit_status: "failed",
-                    });
+                    }));
                 }
                 catch (fallbackErr) {
                     // Same race as the primary completion above: if something else released

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@sagentlab/navarch-runtime",
-  "version": "0.1.77",
+  "version": "0.1.78",
   "description": "Navarch machine-side session manager: claims delivery tasks and runs them through local coding agents.",
   "type": "commonjs",
   "license": "MIT",