diff --git a/changelogs/unreleased/gh-8168-double-term-bump-on-promote.md b/changelogs/unreleased/gh-8168-double-term-bump-on-promote.md
new file mode 100644
index 0000000000000000000000000000000000000000..f8370313bc3c267649b20826fa820630bf47fd13
--- /dev/null
+++ b/changelogs/unreleased/gh-8168-double-term-bump-on-promote.md
@@ -0,0 +1,4 @@
+## bugfix/raft
+
+* Fixed nodes configured with `election_mode` = "manual" sometimes bumping the
+ election term excessively after their promotion (gh-8168).
diff --git a/src/lib/raft/raft.c b/src/lib/raft/raft.c
index 1e5ff2a9d47d741920e6c2b37cefc55519aea8a2..75ae79207f4dfab295a2e6e81d9c6297da5a9b7e 100644
--- a/src/lib/raft/raft.c
+++ b/src/lib/raft/raft.c
@@ -333,7 +333,12 @@ raft_sm_schedule_new_election(struct raft *raft);
static inline void
raft_sm_election_update(struct raft *raft)
{
- if (!raft->is_candidate)
+ /*
+ * The node might be promoted for current term, in which case
+ * is_candidate would be true. But it's not enough. If is_cfg_candidate
+ * is false, the node would give up as soon as new term starts.
+ */
+ if (!raft->is_cfg_candidate)
return;
/*
* Pre-vote protection. Every node must agree that the leader is gone.
@@ -615,8 +620,10 @@ raft_process_msg(struct raft *raft, const struct raft_msg *req, uint32_t source)
/*
* No need for pre-vote checks when the leader
* deliberately told us it's resigning.
+ * Note, the only case when automatic elections are
+ * allowed is when the node is configured candidate.
*/
- if (raft->is_candidate)
+ if (raft->is_cfg_candidate)
raft_sm_schedule_new_election(raft);
}
return 0;
@@ -951,7 +958,7 @@ static void
raft_sm_schedule_new_election(struct raft *raft)
{
say_info("RAFT: begin new election round");
- assert(raft->is_candidate);
+ assert(raft->is_cfg_candidate);
/* Everyone is a follower until its vote for self is persisted. */
raft_sm_schedule_new_term(raft, raft->volatile_term + 1);
raft_sm_schedule_new_vote(raft, raft->self, raft->vclock);
diff --git a/test/replication-luatest/gh_8168_extra_term_bump_test.lua b/test/replication-luatest/gh_8168_extra_term_bump_test.lua
new file mode 100644
index 0000000000000000000000000000000000000000..f607b01508dd8c2d419d5815efe9869e4eae8c9d
--- /dev/null
+++ b/test/replication-luatest/gh_8168_extra_term_bump_test.lua
@@ -0,0 +1,46 @@
+local t = require('luatest')
+local replica_set = require('luatest.replica_set')
+local server = require('luatest.server')
+
+local g = t.group('gh-8168')
+
+g.before_each(function(cg)
+ cg.replica_set = replica_set:new{}
+ local box_cfg = {
+ replication = {
+ server.build_listen_uri('server1'),
+ server.build_listen_uri('server2'),
+ },
+ election_mode = 'manual',
+ replication_timeout = 0.1,
+ }
+ cg.server1 = cg.replica_set:build_and_add_server{
+ alias = 'server1',
+ box_cfg = box_cfg,
+ }
+ box_cfg.election_timeout = 1e-9
+ box_cfg.read_only = true
+ cg.server2 = cg.replica_set:build_and_add_server{
+ alias = 'server2',
+ box_cfg = box_cfg,
+ }
+ cg.replica_set:start()
+end)
+
+g.after_each(function(cg)
+ cg.replica_set:drop()
+end)
+
+g.test_double_term_bump_on_promote = function(cg)
+ cg.server1:wait_for_election_leader()
+ local term = cg.server1:get_election_term()
+ local ok, _ = cg.server2:exec(function()
+ return pcall(box.ctl.promote)
+ end)
+ t.assert(ok, 'Promote succeeded')
+ t.assert_equals(cg.server2:get_election_term(), term + 1,
+ 'Promote bumped term once')
+ t.assert_equals(cg.server2:exec(function()
+ return box.info.election.state
+ end), 'leader', 'Server 2 is the leader')
+end
diff --git a/test/unit/raft.c b/test/unit/raft.c
index 923eac192cac5666c63fe685099b730afc850a89..c2934f262aba740b40dd2b5829ff8a60f3ac0765 100644
--- a/test/unit/raft.c
+++ b/test/unit/raft.c
@@ -1569,7 +1569,7 @@ raft_test_too_long_wal_write(void)
static void
raft_test_promote_restore(void)
{
- raft_start_test(12);
+ raft_start_test(21);
struct raft_node node;
raft_node_create(&node);
@@ -1617,6 +1617,41 @@ raft_test_promote_restore(void)
raft_node_restore(&node);
ok(!node.raft.is_candidate, "not a candidate");
+ /*
+ * The node doesn't schedule new elections in the next round after
+ * promotion.
+ */
+ raft_node_cfg_is_candidate(&node, false);
+ raft_node_cfg_is_enabled(&node, true);
+ raft_node_cfg_election_quorum(&node, 2);
+ raft_node_promote(&node);
+
+ is(node.raft.state, RAFT_STATE_CANDIDATE, "became candidate");
+ is(node.raft.term, 5, "new term");
+
+ /* Wait for the election timeout. */
+ double ts = raft_time();
+ raft_run_next_event();
+ ok(raft_time() - ts >= node.raft.election_timeout,
+ "election timeout passed");
+ is(node.raft.state, RAFT_STATE_CANDIDATE, "still a candidate");
+ is(node.raft.term, 5, "do not bump term");
+
+ is(raft_node_send_leader(
+ &node,
+ 5 /* Term. */,
+ 2 /* Source. */
+ ), 0, "another leader is accepted");
+
+ is(raft_node_send_follower(
+ &node,
+ 5 /* Term. */,
+ 2 /* Source. */
+ ), 0, "leader resign is accepted");
+
+ is(node.raft.state, RAFT_STATE_FOLLOWER, "stay follower");
+ is(node.raft.term, 5, "do not bump term");
+
raft_node_destroy(&node);
raft_finish_test();
}
diff --git a/test/unit/raft.result b/test/unit/raft.result
index d32947a469b51dd228129e564355ae8a65bb452d..2fdca67ef3e0f179852bd02c9be94695cb1f5068 100644
--- a/test/unit/raft.result
+++ b/test/unit/raft.result
@@ -265,7 +265,7 @@ ok 13 - subtests
ok 14 - subtests
*** raft_test_too_long_wal_write: done ***
*** raft_test_promote_restore ***
- 1..12
+ 1..21
ok 1 - became leader after promotion
ok 2 - restore drops a non-candidate leader to a follower
ok 3 - became leader after promotion
@@ -278,6 +278,15 @@ ok 14 - subtests
ok 10 - still old term
ok 11 - not a candidate
ok 12 - not a candidate
+ ok 13 - became candidate
+ ok 14 - new term
+ ok 15 - election timeout passed
+ ok 16 - still a candidate
+ ok 17 - do not bump term
+ ok 18 - another leader is accepted
+ ok 19 - leader resign is accepted
+ ok 20 - stay follower
+ ok 21 - do not bump term
ok 15 - subtests
*** raft_test_promote_restore: done ***
*** raft_test_bump_term_before_cfg ***