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

fix(libxce-failover): known-dead ctx (XCE_CTX_ERR) triggers reconnect instead of surfacing -EINVAL

Browse source 
Live 3-node E2E revealed the reported 'Invalid argument' bug was still
reachable: after the primary dies, handle_disconnect leaves ctx.state=
XCE_CTX_ERR, so a FRESH xce_* call returns -EINVAL from the xceapi guard
before reaching a primary. The classifier mapped -EINVAL to TERMINAL, so the
loop bailed without rotating/recovering. decide() now detects the known-dead
ctx and treats it as a prompt reconnect cause (RETRY_NOW), surfacing the
distinct cluster-not-ready code on budget exhaustion, never -EINVAL.
Regression test: test_decide_known_dead_ctx_retries_not_terminal.
This commit is contained in:
Rongsong Shen 2026-06-17 04:46:01 +00:00
parent c616f41364
commit fa65e794ec
2 changed files with 63 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

41
src/lib/testsuite/test_libxce_retry.c
View file

@ -1022,6 +1022,46 @@ static void test_decide_transport_exhaustion_keeps_last_rc(void)
printf("test_decide_transport_exhaustion_keeps_last_rc: PASS\n");
}
/*
* Regression for the live-E2E bug: after the connected primary dies,
* handle_disconnect leaves ctx.state == XCE_CTX_ERR, so a FRESH xce_* call
* returns -EINVAL from the per-op xceapi guard BEFORE reaching a primary.
* decide() must treat the known-dead ctx as a prompt reconnect cause
* (RETRY_NOW, no backoff) rather than classifying -EINVAL as TERMINAL, and on
* budget exhaustion surface the distinct cluster-not-ready code, never -EINVAL.
*/
static void test_decide_known_dead_ctx_retries_not_terminal(void)
{
xce_client_config_t conf;
api_ctx_t ac;
fo_make_client_masters(&conf);
fo_make_api_ctx(&ac, &conf);
xce_retry_config_t cfg = fo_test_retry_cfg();
xcecli_retry_t st = XCECLI_RETRY_INIT(&ac);
xce_fo_action_t action;
/* Known-dead ctx + the guard's -EINVAL must RETRY (reconnect), not surface. */
ac.ctx.state = XCE_CTX_ERR;
int rc = -EINVAL;
bool a = xcecli_retry_decide(&ac, &rc, false, &cfg, &st, &action);
TEST_ASSERT_TRUE(a);
TEST_ASSERT_EQUAL_INT(XCE_FO_RETRY_NOW, action); /* prompt rotate, no backoff */
TEST_ASSERT_EQUAL_INT(0, st.delay_ms);
/* ctx stays ERR (no real reconnect in this unit) → keep retrying until the
* budget is exhausted, then surface -E_XCE_NOT_READY (never -EINVAL). */
int guard = 0;
do {
rc = -EINVAL;
a = xcecli_retry_decide(&ac, &rc, false, &cfg, &st, &action);
} while (a && ++guard < 100);
TEST_ASSERT_FALSE(a);
TEST_ASSERT_EQUAL_INT(-E_XCE_NOT_READY, rc);
printf("test_decide_known_dead_ctx_retries_not_terminal: PASS\n");
}
/* Req 6.3: when MANY clients fail over simultaneously they must NOT retry in
* lockstep — randomized (jittered) backoff staggers their delays. Each client
* carries its OWN retry carrier (seeded from its own api_ctx pointer, exactly as
@ -1657,6 +1697,7 @@ void test_libxce_retry(void)
RUN_TEST(test_decide_terminal_surfaces_unchanged);
RUN_TEST(test_decide_dup_on_resent_write_is_success);
RUN_TEST(test_decide_transport_exhaustion_keeps_last_rc);
RUN_TEST(test_decide_known_dead_ctx_retries_not_terminal);
RUN_TEST(test_decide_jitter_staggers_many_clients);
RUN_TEST(test_recover_resets_err_state_to_run);
RUN_TEST(test_recover_leaves_run_state_untouched);