lei.chen/myxce
1
0
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat(libxce-failover): reconnect recovery — ctx ERR->RUN reset + stale wbuf drain (fixes post-failover EINVAL) (task 3.2)

Browse source
This commit is contained in:
Rongsong Shen 2026-06-17 03:36:24 +00:00
parent 84c5e32c2c
commit 12284bc3f5
5 changed files with 170 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

35
src/lib/client_failover.c
View file

@ -159,6 +159,36 @@ int xce_failover_rotate(api_ctx_t *ac)
return rc < 0 ? rc : 0;
}
/* ════════════════════════════════════════════════════════════════════════
* Reconnect recovery — context-state reset (CI2) + stale-reply drain (CI1)
* ════════════════════════════════════════════════════════════════════════ */
int xcecli_ctx_recover(api_ctx_t* ac)
{
if (ac == NULL)
return -EINVAL;
/* CI2 — state reset: the transport-error path (handle_disconnect) left
* ctx->state = XCE_CTX_ERR, which the per-op XCE_CTX_RUN guard in xceapi.c
* rejects as -EINVAL — the reported bug. After a successful reconnect,
* restore RUN so the next xce_* attempt passes. ONLY flip ERR→RUN: a
* deliberately-stopped (DOWN) or not-yet-started (INIT/BUSY) ctx is left
* untouched. */
if (ac->ctx.state == XCE_CTX_ERR)
ac->ctx.state = XCE_CTX_RUN;
/* CI1 — stale-reply drain: handle_disconnect may have injected a stale
* transport status into the single-slot wbuf. Drain any pending entry with
* a NON-BLOCKING get so the next attempt's cbuf_get cannot consume a
* prior-attempt reply. cbuf_tryget returns -EAGAIN when the channel is
* already empty — the normal, no-op case. The reconnect closed the old fd
* first, so no further replies arrive from the dead connection. */
int stale = 0;
cbuf_tryget(&ac->ctx.wbuf, &stale);
return 0;
}
/* ════════════════════════════════════════════════════════════════════════
* XcecliRetryLoop — decision core (pure) + side-effecting wrapper
* ════════════════════════════════════════════════════════════════════════ */
@ -289,6 +319,11 @@ bool xcecli_retry_next(api_ctx_t* ac, int* rc, bool is_write,
if (rrc < 0) {
WARN("xcecli failover: rotate failed (rc=%d); retrying anyway", rrc);
} else {
/* Reconnect succeeded: repair the ctx before the next xce_* attempt
* (CI2 ERR→RUN state reset, CI1 stale-reply drain) so the per-op guard
* passes and no prior-attempt reply is consumed (1.2, 4.4). */
xcecli_ctx_recover(ac);
/* Log the rotation target (7.1). xce_failover_rotate consumes the
* redirect hint when present (so the post-rotate target is not always
* the sticky master); report the sticky master cursor as a best-effort