From dfbdbf2a67def12ecb45187dab30f91dac13e73f Mon Sep 17 00:00:00 2001
From: Vincent Koc <vincentkoc@ieee.org>
Date: Wed, 12 Aug 2026 16:16:22 +0800
Subject: [PATCH] fix(provider-runtime): retry transient socket failures
 (#122535)

* test: cover UND_ERR_SOCKET retry recovery

Punchcard-Session: golden-cedar-brook-4q

* fix(provider-runtime): retry UND_ERR_SOCKET failures

Punchcard-Session: golden-cedar-brook-4q
---
 src/agents/run-wait.test.ts                  | 1 +
 src/infra/retryable-network-errors.ts        | 2 +-
 src/provider-runtime/operation-retry.test.ts | 1 +
 3 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/agents/run-wait.test.ts b/src/agents/run-wait.test.ts
index 344063256018..3385701f690f 100644
--- a/src/agents/run-wait.test.ts
+++ b/src/agents/run-wait.test.ts
@@ -838,6 +838,7 @@ describe("isRecoverableAgentWaitError", () => {
     "EHOSTUNREACH",
     "ENETUNREACH",
     "EAI_AGAIN",
+    "UND_ERR_SOCKET",
   ])("recovers from %s connection failures", (code) => {
     expect(isRecoverableAgentWaitError(`connect ${code} 127.0.0.1:443`)).toBe(true);
   });
diff --git a/src/infra/retryable-network-errors.ts b/src/infra/retryable-network-errors.ts
index f048fab9bb9f..e619adbe22d2 100644
--- a/src/infra/retryable-network-errors.ts
+++ b/src/infra/retryable-network-errors.ts
@@ -55,7 +55,7 @@ const TRANSIENT_NETWORK_MESSAGE_SNIPPETS = [
 ];
 
 const RETRYABLE_CONNECTION_ERROR_CODE_RE =
-  /\b(?:ECONNRESET|ECONNREFUSED|ETIMEDOUT|EPIPE|EHOSTUNREACH|ENETUNREACH|EAI_AGAIN)\b/i;
+  /\b(?:ECONNRESET|ECONNREFUSED|ETIMEDOUT|EPIPE|EHOSTUNREACH|ENETUNREACH|EAI_AGAIN|UND_ERR_SOCKET)\b/i;
 
 function isWrappedFetchFailedMessage(message: string): boolean {
   if (message === "fetch failed") {
diff --git a/src/provider-runtime/operation-retry.test.ts b/src/provider-runtime/operation-retry.test.ts
index 3e12125f14cf..42ab8e9602ab 100644
--- a/src/provider-runtime/operation-retry.test.ts
+++ b/src/provider-runtime/operation-retry.test.ts
@@ -51,6 +51,7 @@ describe("executeProviderOperationWithRetry", () => {
     "EHOSTUNREACH",
     "ENETUNREACH",
     "EAI_AGAIN",
+    "UND_ERR_SOCKET",
     "ENOTFOUND",
   ])("retries %s network failures from structured errors", async (code) => {
     const cause = Object.assign(new Error("connect failed"), { code });