[Tarantool-patches] [PATCH v2] replication: fix join vclock obtainment in relay_initial_join
Ilya Kosarev
i.kosarev at tarantool.org
Mon Oct 28 15:25:06 MSK 2019
join_vclock test could fail on huge load due to vclock advance
comparing to an actual WAL. Now we are clearly obtaining vclock on the
flushed state using wal_begin_checkpoint. It is also better to get max
index and index count in single request in join_vclock test. With fixes
mentioned above it is not fragile anymore.
Closes #4160
---
https://github.com/tarantool/tarantool/tree/i.kosarev/gh-4160-fix-join-vclock
https://github.com/tarantool/tarantool/issues/4160
src/box/relay.cc | 9 +++++----
test/replication/join_vclock.result | 5 +----
test/replication/join_vclock.test.lua | 3 +--
test/replication/suite.ini | 1 -
4 files changed, 7 insertions(+), 11 deletions(-)
diff --git a/src/box/relay.cc b/src/box/relay.cc
index e849fcf4f..d16cd1a56 100644
--- a/src/box/relay.cc
+++ b/src/box/relay.cc
@@ -306,13 +306,14 @@ relay_initial_join(int fd, uint64_t sync, struct vclock *vclock)
});
/*
- * Sync WAL to make sure that all changes visible from
- * the frozen read view are successfully committed.
+ * Make sure that current database state is flushed to
+ * the WAL and obtain corresponding vclock.
*/
- if (wal_sync() != 0)
+ struct wal_checkpoint checkpoint;
+ if (wal_begin_checkpoint(&checkpoint) != 0)
diag_raise();
- vclock_copy(vclock, &replicaset.vclock);
+ vclock_copy(vclock, &checkpoint.vclock);
/* Respond to the JOIN request with the current vclock. */
struct xrow_header row;
diff --git a/test/replication/join_vclock.result b/test/replication/join_vclock.result
index a9781073d..d6d9af783 100644
--- a/test/replication/join_vclock.result
+++ b/test/replication/join_vclock.result
@@ -67,10 +67,7 @@ test_run:cmd("switch replica1")
---
- true
...
-cnt = box.space.test.index[0]:count()
----
-...
-box.space.test.index.primary:max()[1] == cnt - 1
+box.space.test.index.primary:max()[1] == box.space.test.index.primary:count() - 1
---
- true
...
diff --git a/test/replication/join_vclock.test.lua b/test/replication/join_vclock.test.lua
index 0b60dffc2..a813ba31f 100644
--- a/test/replication/join_vclock.test.lua
+++ b/test/replication/join_vclock.test.lua
@@ -26,8 +26,7 @@ ch:get()
errinj.set("ERRINJ_RELAY_FINAL_SLEEP", false)
test_run:cmd("switch replica1")
-cnt = box.space.test.index[0]:count()
-box.space.test.index.primary:max()[1] == cnt - 1
+box.space.test.index.primary:max()[1] == box.space.test.index.primary:count() - 1
test_run:cmd("switch default")
replica_set.drop_all(test_run)
diff --git a/test/replication/suite.ini b/test/replication/suite.ini
index 384dac677..ed1de3140 100644
--- a/test/replication/suite.ini
+++ b/test/replication/suite.ini
@@ -12,7 +12,6 @@ long_run = prune.test.lua
is_parallel = True
pretest_clean = True
fragile = errinj.test.lua ; gh-3870
- join_vclock.test.lua ; gh-4160
long_row_timeout.test.lua ; gh-4351
skip_conflict_row.test.lua ; gh-4457
sync.test.lua ; gh-3835 gh-3877
--
2.17.1
More information about the Tarantool-patches
mailing list