Description:
Hi all:
we find when partition table have a great many columns(like 300 cols),access second index,if needed data distribute in different partition, cpu high load in funcation build_template_field. like,
create table t(
id1 int,
id2 int
primary key(id1),
key(id2)
)engine=innodb
partition by range(id1)(
partition p0 values less than(100),
partition p1 values less than(200),
partition p2 values less than(300));
insert into tpar10col values(1,1);
insert into tpar10col values(101,1);
insert into tpar10col values(201,1);
insert into tpar10col values(2,2);
insert into tpar10col values(3,2);
insert into tpar10col values(4,2);
insert into tpar10col values(7,2);
insert into tpar10col values(8,2);
insert into tpar10col values(9,2);
insert into tpar10col values(10,2);
if we use statement "select * from t where id2=1",this statement will access p0,p1,p2。access every partition may call funcation ha_innobase::build_template。
ha_innobase::build_template loop every filed,if in write_set and read_set will call build_template_field funcation to find clust_rec_field_no(loop every field in clust index) and rec_field_no,because clust index include all filed ,so cpu high load here. perf top output like:
53.98% mysqld [.] build_template_field
10.40% [kernel] [k] _raw_spin_unlock_irqrestore
4.12% [kernel] [k] copy_user_generic_unrolled
1.95% mysqld [.] ha_innobase::build_template
1.48% mysqld [.] row_sel_store_mysql_field_func
can we save mysql_row_templ_t.clust_rec_field_no ,if next parition access,we copy it to new mysql_row_templ_t.clust_rec_field_no,not loop every field in clust index.in all parition this value maybe is same.
thanks.
How to repeat:
see the attachment
Suggested fix:
can we save mysql_row_templ_t.clust_rec_field_no ,if next parition access,we copy it to new mysql_row_templ_t.clust_rec_field_no,not loop every field in clust index.in all parition this value maybe is same.
Description: Hi all: we find when partition table have a great many columns(like 300 cols),access second index,if needed data distribute in different partition, cpu high load in funcation build_template_field. like, create table t( id1 int, id2 int primary key(id1), key(id2) )engine=innodb partition by range(id1)( partition p0 values less than(100), partition p1 values less than(200), partition p2 values less than(300)); insert into tpar10col values(1,1); insert into tpar10col values(101,1); insert into tpar10col values(201,1); insert into tpar10col values(2,2); insert into tpar10col values(3,2); insert into tpar10col values(4,2); insert into tpar10col values(7,2); insert into tpar10col values(8,2); insert into tpar10col values(9,2); insert into tpar10col values(10,2); if we use statement "select * from t where id2=1",this statement will access p0,p1,p2。access every partition may call funcation ha_innobase::build_template。 ha_innobase::build_template loop every filed,if in write_set and read_set will call build_template_field funcation to find clust_rec_field_no(loop every field in clust index) and rec_field_no,because clust index include all filed ,so cpu high load here. perf top output like: 53.98% mysqld [.] build_template_field 10.40% [kernel] [k] _raw_spin_unlock_irqrestore 4.12% [kernel] [k] copy_user_generic_unrolled 1.95% mysqld [.] ha_innobase::build_template 1.48% mysqld [.] row_sel_store_mysql_field_func can we save mysql_row_templ_t.clust_rec_field_no ,if next parition access,we copy it to new mysql_row_templ_t.clust_rec_field_no,not loop every field in clust index.in all parition this value maybe is same. thanks. How to repeat: see the attachment Suggested fix: can we save mysql_row_templ_t.clust_rec_field_no ,if next parition access,we copy it to new mysql_row_templ_t.clust_rec_field_no,not loop every field in clust index.in all parition this value maybe is same.