---
id: "bug-608"
status: "fixed"
status_updated: "2026-09-24"
status_note: ""
component: "lib"
severity: "unknown"
timestamp: "2026-06-16T10:36:00Z"
last_seen: "2026-09-19T09:00:00Z"
occurrences: 1
files: ["src/lib/request.c"]
tags: ["libxce", "failover", "ha", "E_XCE_REDIRECT", "retry", "client", "protocol-mismatch"]
related: []
legacy: true
---

# bug-608: libxce client never fails over after primary killed; xce_joblist returns 'Invalid argument'…

## Description
libxce client never fails over after primary killed; xce_joblist returns 'Invalid argument' (EINVAL) indefinitely

## Environment
_Not recorded (migrated from buglog.json)._

## Reproduction
_Not recorded (migrated from buglog.json)._

## Root cause
Two gaps: (1) xce_request_with_retry() (request.c:1956) is implemented+unit-tested but NO production path calls it - every xce_*_req wrapper calls plain xce_request() (request.c:1385 etc). (2) The retry primitive keys on E_XCE_STANDBY + reply.redirect_host/port (request.c:2010), but the real controller emits a dedicated E_XCE_REDIRECT status with data_which=nodata (join.c:264-266) and never populates those trailer fields. Controller non-primary taxonomy is E_XCE_REDIRECT / E_XCE_NOMASTER / E_XCE_NOT_READY / E_XCE_STANDBY (errnox.h:22-25). So the client both bypasses the wrapper AND, even if wired, would not detect the controller's actual redirect signal.

## Fix
Spec libxce-failover: wire all production request paths through transparent failover; classify on actual controller statuses (REDIRECT=fast-rotate to find existing primary, address-if-present else rotate master list; NOMASTER/NOT_READY=backoff then cluster-not-ready; STANDBY=rotate); realign the existing retry unit tests to the emitted statuses. Implementation pending (requirements phase).

## Verification
_Not recorded (migrated from buglog.json)._

## References
_Not recorded (migrated from buglog.json)._