revision-id: 25cbeaf535e07cb54d5267471ae61a88a3cb6ae4 (fb-prod201903-267-g25cbeaf535e) parent(s): d4d71e07cadb223228474d66ad3277de8650a937 author: Sergei Petrunia committer: Sergei Petrunia timestamp: 2019-12-01 22:29:37 +0300 message: Fix rocksdb.rocksdb_timeout_rollback test with range locking In Range-Locking mode, the statement update t1 set a = a + 1 where a = 2; doesn't acquire a snapshot (because it reads committed data). This means, con2 will acquire its read shapshot only here: connection con2; select * from t1; when con1 has already committed. Because of that, it will see rows with a=11 and a=12, causing a test failure like this: @@ -58,6 +58,8 @@ 6 9 10 +11 +12 commit; select * from t1; a