create table t4
(
pk1 int not null,
pk2 int not null,
col1 int not null,
col2 int,
primary key PK(pk1,pk2),
unique key U(pk1, col1)
) engine=rocksdb;
insert into t4 values (1,1,1,1), (2,2,2,2),(3,3,3,3);
select * from t4 force index (U) where pk1 in (1,2,3);
Putting a breakpoint in ha_rocksdb::index_read_map, one can find that:
(gdb) p kd->name.c_str()
$35 = 0x7fffa006a188 "U"
(gdb)
(gdb) p kd->get_m_key_parts()
$36 = 4
(gdb) p table->key_info[active_index].actual_key_parts
$37 = 2
get_m_key_parts()==4, because the unique key has these key parts:
Summary:
Fixed it: Now, a Unique secondary index has a suffix of PK columns,
but it includes only columns that are not explicitly present in the
index definition (InnoDB does the same internally).
Test Plan: Run mtr
Reviewers: hermanlee4, maykov, jtolmer, yoshinorim
Differential Revision: https://reviews.facebook.net/D38619
Summary:
Fixed it: Now, a Unique secondary index has a suffix of PK columns,
but it includes only columns that are not explicitly present in the
index definition (InnoDB does the same internally).
Test Plan: Run mtr
Reviewers: hermanlee4, maykov, jtolmer, yoshinorim
Differential Revision: https://reviews.facebook.net/D38619
Consider an example:
Putting a breakpoint in ha_rocksdb::index_read_map, one can find that:
get_m_key_parts()==4, because the unique key has these key parts:
pk1,col1,extended(pk1,pk2)
pk1 should not be there.