From df6cf5ec639aa414e3cd013de3edb31f59fbbe04 Mon Sep 17 00:00:00 2001
From: Serge Petrenko <sergepetrenko@tarantool.org>
Date: Tue, 17 Jan 2023 14:18:52 +0300
Subject: [PATCH] raft: fix an assertion failure in raft_stop_candidate
There is a false assertion in raft_stop_candidate(): it assumes that the
node must always have a running timer whenever it sees the leader.
This is not true when the node is busy writing the new term on disk.
Cover the mentioned case in the assertion.
Closes #8169
NO_DOC=bugfix
Co-authored-by: Sergey Ostanevich <sergos@tarantool.org>
---
.../gh-8169-raft-assert-on-wal-write.md | 4 ++
src/lib/raft/raft.c | 6 ++-
test/unit/raft.c | 45 ++++++++++++++++++-
3 files changed, 52 insertions(+), 3 deletions(-)
create mode 100644 changelogs/unreleased/gh-8169-raft-assert-on-wal-write.md
diff --git a/changelogs/unreleased/gh-8169-raft-assert-on-wal-write.md b/changelogs/unreleased/gh-8169-raft-assert-on-wal-write.md
new file mode 100644
index 0000000000..e7f8752804
--- /dev/null
+++ b/changelogs/unreleased/gh-8169-raft-assert-on-wal-write.md
@@ -0,0 +1,4 @@
+## bugfix/raft
+
+* Fixed an assertion failure in case an election candidate is reconfigured to a
+ voter during an ongoning WAL write (gh-8169).
diff --git a/src/lib/raft/raft.c b/src/lib/raft/raft.c
index 87ef5b7871..4dccc7aea2 100644
--- a/src/lib/raft/raft.c
+++ b/src/lib/raft/raft.c
@@ -1163,9 +1163,11 @@ raft_stop_candidate(struct raft *raft)
} else {
/*
* Leader is seen and node is waiting for its death. Do not stop
- * the timer.
+ * the timer. If there is a write in progress the timer is
+ * stopped now, but will be re-started once the write completes.
*/
- assert(raft_ev_timer_is_active(&raft->timer));
+ assert(raft_ev_timer_is_active(&raft->timer) ||
+ raft->is_write_in_progress);
}
raft->state = RAFT_STATE_FOLLOWER;
raft_schedule_broadcast(raft);
diff --git a/test/unit/raft.c b/test/unit/raft.c
index b63183e888..923eac192c 100644
--- a/test/unit/raft.c
+++ b/test/unit/raft.c
@@ -2347,10 +2347,52 @@ raft_test_resign(void)
raft_finish_test();
}
+static void
+raft_test_candidate_disable_during_wal_write(void)
+{
+ raft_start_test(2);
+ /*
+ * There was a false-positive assertion failure in a special case: the
+ * node has just received a is_leader notification and is currently
+ * writing it on disk. At the same time it is configured as voter
+ * (gh-8169).
+ */
+ struct raft_node node;
+ raft_node_create(&node);
+ raft_node_cfg_is_candidate(&node, true);
+ raft_node_block(&node);
+ raft_node_send_leader(&node, 2, 2);
+ ok(raft_node_check_full_state(
+ &node,
+ RAFT_STATE_FOLLOWER /* State. */,
+ 2 /* Leader. */,
+ 1 /* Term. */,
+ 0 /* Vote. */,
+ 2 /* Volatile term. */,
+ 0 /* Volatile vote. */,
+ "{}" /* Vclock. */
+ ), "Leader is seen, but wal write is in progress");
+ raft_node_cfg_is_candidate(&node, false);
+ raft_node_unblock(&node);
+ ok(raft_node_check_full_state(
+ &node,
+ RAFT_STATE_FOLLOWER /* State. */,
+ 2 /* Leader. */,
+ 2 /* Term. */,
+ 0 /* Vote. */,
+ 2 /* Volatile term. */,
+ 0 /* Volatile vote. */,
+ "{0: 1}" /* Vclock. */
+ ), "State is persisted");
+
+ raft_node_destroy(&node);
+ raft_finish_test();
+}
+
static int
main_f(va_list ap)
{
- raft_start_test(19);
+ raft_start_test(20);
(void) ap;
fakeev_init();
@@ -2374,6 +2416,7 @@ main_f(va_list ap)
raft_test_split_vote();
raft_test_pre_vote();
raft_test_resign();
+ raft_test_candidate_disable_during_wal_write();
fakeev_free();
--
GitLab