fix(controller): cold start elects in ~1.7 s, not ~10 s; ctrlstatus appliedLsn on the primary (bug-1879, bug-1878)

bug-1879: on a fresh HA cluster the config-holding ctrl-1 started its
term-1 election before its peers ran; connect failures count as "no"
votes (bug-1423), so it lost at once and became a follower with no
leader. ctrl-2/ctrl-3 hold no config and are voters only (by design),
so nothing retried until ctrl-1's 10 s standby liveness timeout: every
cold start had no primary for ~10 s.
- ha_note_leaderless() marks a config-holding node that becomes a
  follower with no leader known; the liveness poll then retries the
  election (PreVote) after HA_LEADERLESS_RETRY_MS = 1 s, doubling up to
  heartbeat_timeout_ms. Primary contact or becoming leader clears it.
- ha_heartbeat_is_fresh() answers "no" while leaderless: last_activity
  then only holds the watchdog's arm time, and the failover gates took
  it for a live primary.
Live cold start: primary 1.7 s after the first node started (was ~10 s).

bug-1878: ctrlstatus_fill() reported ha->applied_lsn for every role, a
standby-only counter a primary never advances, so every primary showed
"appliedLsn: 0". A primary now reports persist->last_applied_lsn. The
controller-status design doc's "already correct in every role" note is
corrected.

Tests: 5 new in test_ha_failover_prevote.c, 1 new in test_ctrlstatus.c
(RED: expected 4900, was 0); controller suite 1256/0. HA soaks:
sequential 6 rounds PASS, chaotic 6 rounds PASS (zero divergence, zero
job loss).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Rongsong Shen 2026-09-28 07:35:45 +00:00
parent 16b422e303
commit aa7fc65a36
Signed by untrusted user: rshen
GPG key ID: 80533536B0EB651C
6 changed files with 260 additions and 6 deletions

View file

@ -90,6 +90,33 @@ static void cs_teardown_cluster(cluster_t* c)
/* ── tests ──────────────────────────────────────────────────── */
/* bug-1878: a primary never advances ha->applied_lsn (a standby-only
* counter: how far replicated WAL has been received), so ctrlstatus showed
* "committedLsn: 4900 / appliedLsn: 0" on every primary. A primary reports
* its apply cursor, persist->last_applied_lsn (the last entry applied). */
static void test_ctrlstatus_primary_reports_its_apply_cursor(void)
{
cluster_t c;
myconf_t mc;
cs_setup_cluster(&c, &mc);
atomic_store_explicit(&c.ha.role, (int32_t)CTRL_PRIMARY,
memory_order_release);
atomic_store_explicit(&c.ha.applied_lsn, 0, memory_order_release);
atomic_store_explicit(&c.persist.committed_lsn, 4900, memory_order_release);
atomic_store_explicit(&c.persist.last_applied_lsn, 4900,
memory_order_release);
xce_ctrlstatus_t st;
ctrlstatus_fill(&c, &st);
TEST_ASSERT_EQUAL_UINT64(4900, st.committedLsn);
TEST_ASSERT_EQUAL_UINT64(4900, st.appliedLsn);
TEST_ASSERT_TRUE_MESSAGE(st.appliedLsn <= st.committedLsn,
"a primary applies only committed entries");
ctrlstatus_free_fields(&st);
}
/* (a) PRIMARY: responderRole==PRIMARY, primaryName==self, peers/version
* populated; committed/applied/term reflect the atomics. (Req 2.1, 3.1, 3.2,
* 4.1, 4.2, 5.1, 6.1, 6.2, 6.4) */
@ -102,7 +129,9 @@ static void test_ctrlstatus_primary(void)
atomic_store_explicit(&c.ha.role, (int32_t)CTRL_PRIMARY,
memory_order_release);
atomic_store_explicit(&c.ha.term, 7, memory_order_release);
atomic_store_explicit(&c.ha.applied_lsn, 42, memory_order_release);
/* bug-1878: a primary's applied position is persist.last_applied_lsn;
* ha.applied_lsn (standby-only) is left at the 0 a live primary has */
atomic_store_explicit(&c.persist.last_applied_lsn, 41, memory_order_release);
atomic_store_explicit(&c.ha.committed_config_tag, 0xAABBCCDDEE001122ULL,
memory_order_release);
atomic_store_explicit(&c.persist.committed_lsn, 41, memory_order_release);
@ -120,7 +149,7 @@ static void test_ctrlstatus_primary(void)
TEST_ASSERT_NOT_NULL(st.primaryName);
TEST_ASSERT_EQUAL_STRING("ctrlA", st.primaryName);
TEST_ASSERT_EQUAL_UINT64(7, st.term);
TEST_ASSERT_EQUAL_UINT64(42, st.appliedLsn);
TEST_ASSERT_EQUAL_UINT64(41, st.appliedLsn);
TEST_ASSERT_EQUAL_UINT64(41, st.committedLsn);
TEST_ASSERT_EQUAL_UINT64(0xAABBCCDDEE001122ULL, st.configVersion);
TEST_ASSERT_EQUAL_UINT64(0xAABBCCDDEE001122ULL, st.committedConfigVersion);
@ -1098,6 +1127,7 @@ void test_ctrlstatus(void)
UnitySetTestFile(__FILE__);
RUN_TEST(test_ctrlstatus_format_time_ms);
RUN_TEST(test_ctrlstatus_primary);
RUN_TEST(test_ctrlstatus_primary_reports_its_apply_cursor);
RUN_TEST(test_ctrlstatus_standby_known_primary);
RUN_TEST(test_ctrlstatus_unknown_primary);
RUN_TEST(test_ctrlstatus_standby_self_vote_is_unknown);

View file

@ -352,6 +352,138 @@ static void test_config_less_node_does_not_start_prevote(void)
printf("test_config_less_node_does_not_start_prevote: PASS\n");
}
/* ── bug-1879: cold boot / lost election with no leader known ──────────── */
/*
* A config-holding node that loses its first election because its peers are
* not up yet (connect failures count as "no" votes, bug-1423) became a
* follower with nobody to follow, and only the 10 s liveness timeout ever
* retried: every cold start had no primary for ~10 s. Now such a node is
* marked leaderless and the liveness poll retries the election (via PreVote)
* after leaderless_retry_ms, doubling up to heartbeat_timeout_ms.
*/
static uint64_t now_ms(void)
{
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
return (uint64_t)ts.tv_sec * 1000u + (uint64_t)ts.tv_nsec / 1000000u;
}
static void test_leaderless_follower_retries_election_after_short_delay(void)
{
ha_state_t ha;
TEST_ASSERT_EQUAL_INT(0, ha_init(&ha, &g_cfg, 0));
ha_do_demote(&ha);
make_three_node(&ha);
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
ha.last_activity = ts.tv_sec; /* the 6 s timeout is far away */
atomic_store(&ha.leaderless, 1);
ha.leaderless_retry_ms = HA_LEADERLESS_RETRY_MS;
ha.leaderless_since_ms = now_ms() - HA_LEADERLESS_RETRY_MS - 50;
liveness_poll_ctx_t ctx;
memset(&ctx, 0, sizeof(ctx));
ctx.ha = &ha;
atomic_init(&ctx.cancelled, 0);
liveness_poll_fn_test_entry(&ctx);
TEST_ASSERT_EQUAL_INT_MESSAGE(1, ha.raft.pre_vote_in_progress,
"a leaderless config-holding node retries after leaderless_retry_ms");
TEST_ASSERT_EQUAL_UINT32_MESSAGE(2 * HA_LEADERLESS_RETRY_MS,
ha.leaderless_retry_ms, "the next retry waits twice as long");
ha_fini(&ha);
}
static void test_leaderless_retry_waits_for_its_delay(void)
{
ha_state_t ha;
TEST_ASSERT_EQUAL_INT(0, ha_init(&ha, &g_cfg, 0));
ha_do_demote(&ha);
make_three_node(&ha);
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
ha.last_activity = ts.tv_sec;
atomic_store(&ha.leaderless, 1);
ha.leaderless_retry_ms = HA_LEADERLESS_RETRY_MS;
ha.leaderless_since_ms = now_ms() - 100;
liveness_poll_ctx_t ctx;
memset(&ctx, 0, sizeof(ctx));
ctx.ha = &ha;
atomic_init(&ctx.cancelled, 0);
liveness_poll_fn_test_entry(&ctx);
TEST_ASSERT_EQUAL_INT(0, ha.raft.pre_vote_in_progress);
ha_fini(&ha);
}
static void test_leaderless_retry_backoff_is_capped(void)
{
ha_state_t ha;
TEST_ASSERT_EQUAL_INT(0, ha_init(&ha, &g_cfg, 0));
ha_do_demote(&ha);
make_three_node(&ha);
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
ha.last_activity = ts.tv_sec;
atomic_store(&ha.leaderless, 1);
ha.leaderless_retry_ms = g_cfg.heartbeat_timeout_ms;
ha.leaderless_since_ms = now_ms() - g_cfg.heartbeat_timeout_ms - 50;
liveness_poll_ctx_t ctx;
memset(&ctx, 0, sizeof(ctx));
ctx.ha = &ha;
atomic_init(&ctx.cancelled, 0);
liveness_poll_fn_test_entry(&ctx);
TEST_ASSERT_EQUAL_UINT32(g_cfg.heartbeat_timeout_ms, ha.leaderless_retry_ms);
ha_fini(&ha);
}
static void test_primary_contact_clears_leaderless(void)
{
ha_state_t ha;
TEST_ASSERT_EQUAL_INT(0, ha_init(&ha, &g_cfg, 0));
ha_do_demote(&ha);
atomic_store(&ha.leaderless, 1);
ha.leaderless_retry_ms = 4 * HA_LEADERLESS_RETRY_MS;
ha_on_heartbeat_received(&ha);
TEST_ASSERT_EQUAL_INT(0, atomic_load(&ha.leaderless));
TEST_ASSERT_EQUAL_UINT32(HA_LEADERLESS_RETRY_MS, ha.leaderless_retry_ms);
ha_fini(&ha);
}
static void test_only_a_config_holder_without_leader_is_leaderless(void)
{
ha_state_t ha;
cluster_t c;
myconf_t conf;
TEST_ASSERT_EQUAL_INT(0, ha_init(&ha, &g_cfg, 0));
memset(&c, 0, sizeof(c));
memset(&conf, 0, sizeof(conf));
c.myconf = &conf;
conf.config_version_tag = 0; /* voter only */
ha_note_leaderless(&ha, &c, -1);
TEST_ASSERT_EQUAL_INT(0, atomic_load(&ha.leaderless));
conf.config_version_tag = 0x1234; /* can lead, leader known */
ha_note_leaderless(&ha, &c, 1);
TEST_ASSERT_EQUAL_INT(0, atomic_load(&ha.leaderless));
ha_note_leaderless(&ha, &c, -1); /* can lead, no leader */
TEST_ASSERT_EQUAL_INT(1, atomic_load(&ha.leaderless));
TEST_ASSERT_EQUAL_UINT32(HA_LEADERLESS_RETRY_MS, ha.leaderless_retry_ms);
ha_fini(&ha);
}
/* ── Suite entry point ──────────────────────────────────────────────────── */
void test_ha_failover_prevote(void)
@ -367,6 +499,11 @@ void test_ha_failover_prevote(void)
RUN_TEST(test_heartbeat_without_round_is_plain_noop);
RUN_TEST(test_failover_prevote_win_reaches_real_election);
RUN_TEST(test_config_less_node_does_not_start_prevote);
RUN_TEST(test_leaderless_follower_retries_election_after_short_delay);
RUN_TEST(test_leaderless_retry_waits_for_its_delay);
RUN_TEST(test_leaderless_retry_backoff_is_capped);
RUN_TEST(test_primary_contact_clears_leaderless);
RUN_TEST(test_only_a_config_holder_without_leader_is_leaderless);
ha_prevote_rearm_liveness_hook = NULL;
}