数据分布均匀是保证GP高效并行处理能力的基础。因此定义表时，如果选用HASH分布策略，保证数据分布均匀是获取高性能的关键所在。
选择的依据遵从三大原则：
第一个就是首先保证前面提到的所有节点数据存放是均匀的。
第二，如果经常进行大表连接，那么尽量把连接键定义成数据分布键（如果多个列作为数据分布键，他们应该都出现在连接中，否则还是会造成无效广播），这样尽量减少无效的interconnect。
第三，尽量保证where条件产生的结果集的存储也尽量是均匀的。
常用语句：
1. 查看表使用空间
SELECT *
FROM PUBLIC.dba_segments
WHERE owner LIKE 'owber_name'
AND table_name LIKE '%table_name%'
ORDER BY table_name;
2. GP中查看分区：
select partitionname,partitionboundary from pg_partitions where tablename='table_name';
select partitionname,partitionboundary from pg_catalog.pg_partitions where tablename='table_name';
3. 查看正在运行的sql
select * from pg_stat_activity;
4. 修改表的owner语句
Alter table table_name owner to owner_name;
5. 增加表分区
ALTER TABLE table_name ADD PARTITION P20091001 START (DATE '2009-10-01') INCLUSIVE END (DATE '2009-10-02') EXCLUSIVE WITH(appendonly=true,compresslevel=5);
6. 修改列类型
ALTER TABLE table_name ALTER COLUMN a TYPE varchar(2048);
7. 修改distributed 列
alter table table_name set distributed by(column_1);
8.explain和explain analyze是有区别的。explain是基于统计，给出执行计划，他不会真正执行SQL，因此没有执行统计。explain analyze会真正执行SQL因此会提供执行过程中真实资源消耗统计，但是对在线系统可能影响会较大
9.查看对象大小(表、索引、数据库等)
select pg_size_pretty(pg_relation_size(’$schema.$table’));
10.查看用户(非系统)表和索引
select * from pg_stat_user_tables;
seect * from pg_stat_user_indexes;
11.查看表分区
select b.nspname||’.'||a.relname as tablename, d.parname as partname
from pg_class a, pg_namespace b, pg_partition c, pg_partition_rule d
where a.relnamespace = b.oid
and b.nspname = ‘$schema’
and a.relname = ‘$table’
and a.oid = c.parrelid
and c.oid = d.paroid
order by parname;
12.查看Distributed key
select b.attname
from pg_class a, pg_attribute b, pg_type c, gp_distribution_policy d, pg_namespace e
where d.localoid = a.oid
and a.relnamespace = e.oid
and e.nspname = ‘$schema’
and a.relname=’$table’
and a.oid = b.attrelid
and b.atttypid = c.oid
and b.attnum > 0
and b.attnum = any(d.attrnums)
order by attnum;
13.查看当前存活的查询
select procpid as pid, sess_id as session, usename as user, current_query as query, waiting,
date_trunc(’second’, query_start) as start_time, client_addr as useraddr
from pg_stat_activity
where datname =’$PGDATABASE’
and current_query not like ‘%from pg_stat_activity%where datname =%’
order by start_time;
14.表上被用作外键的字段名
select f.conname, pg_get_constraintdef(f.oid), t2.relname
from pg_class t, pg_class t2, pg_constraint f
where f.confrelid = t.oid
and f.conrelid = t2.oid
and f.contype = ‘f’
and t.relname = ‘$table’;
