在维护旧数据库的时候经常碰到非常的查询,多数都是两方面的原因。 1)没有加索引 2)查询语句导致索引用不上 3)过多的连接数据库
例子1:
在一个大型的计算中原来每天要花费半小时才能完成,对计算的过程进行仔细的分析,发现下面的语句花费了很长时间
select sum(order_qty - delivery_qty - reduce_confirm_qty - lost_qty ) qty from circle_ordering where sku = '" . $sku . "' AND submit_status = 5 AND order_type = 'AIR'
通过explain 这条语句,仔细分析数据库才知道并没有相关的索引作用在这条查询语句上,这样导致了这条sql是全表查询。于是对这三列(sku, submit_status, order_type)新建索引. 重新执行后,整个程序只用了10份钟就完成了。
例子2:
select a.ebay_id, b.ebay_id as ebay_subid, from_unixtime(a.ebay_paidtime) as ebay_paidtime,
a.ebay_account, a.ebay_countryname, c.store_name as warehouse, a.ebay_carrier,
b.sku, b.ebay_amount, a.ebay_currency, b.ebay_itemprice,
b.shipingfee, ((b.ebay_itemprice*b.ebay_amount)+b.shipingfee) as total_amount, ebay_postcode,
b.item_promotion_discount_amount, b.ship_promotion_discount_amount
from ebay_order a left join ebay_orderdetail b on(a.ebay_ordersn=b.ebay_ordersn)
left join ebay_store c on (a.ebay_warehouse = c.id)
where a.ebay_combine !=1 and (a.resend_org_ebay_id=0 or a.resend_org_ebay_id is null) and
b.ebay_amount >0 and a.ebay_warehouse !='' and a.ebay_user='manwei'
and
(
a.ebay_paidtime between UNIX_TIMESTAMP('".$astart."') and UNIX_TIMESTAMP('".$aend."')
or
(a.ebay_paidtime not between UNIX_TIMESTAMP('".$astart_p."') and UNIX_TIMESTAMP('".$aend_p."') and
a.shippedtime between UNIX_TIMESTAMP('".$astart_p."') and UNIX_TIMESTAMP('".$aend_p."')) ";
if($last_ebay_id!='') $data .= " or a.ebay_id >='".$last_ebay_id."'";
$data .= ") order by a.ebay_id, b.ebay_id ";
注意这个复杂的查询语句的条件
第一个条件 (a.ebay_paidtime between UNIX_TIMESTAMP('".$astart."') and UNIX_TIMESTAMP('".$aend."') 由于在ebay_paidtime字段有索引,如果只有这个条件,查询速度很快,查询一次不到一秒。但是因为后面还有两个条件使用了 or, 这样导致会导致了对ebay_order进行了全表查询,而这个表有3百多万条数据,所以查询非常慢。 (有这个说法 :验证在两个相同字段之间使用or不会导致全表扫描,只有出现不同字段自建使用or时会导致全表扫描。但我没有验证过。)
根据业务需求我们把三个用or 连接的查询条件拆出来,分别进行查询,最后用union语句连起来。这样查询的效率得到了大大的提高。修改后的查询如下
$data1 ="select " . $fields_list . "
from ebay_order a left join ebay_orderdetail b on(a.ebay_ordersn=b.ebay_ordersn)
left join ebay_store c on (a.ebay_warehouse = c.id)
where a.ebay_combine !=1 and (a.resend_org_ebay_id=0 or a.resend_org_ebay_id is null) and
b.ebay_amount >0 and a.ebay_warehouse !='' and a.ebay_user='manwei'
and a.ebay_paidtime between UNIX_TIMESTAMP('".$astart."') and UNIX_TIMESTAMP('".$aend."')";
$data2 = "select " . $fields_list . "
from ebay_order a left join ebay_orderdetail b on(a.ebay_ordersn=b.ebay_ordersn)
left join ebay_store c on (a.ebay_warehouse = c.id)
where a.ebay_combine !=1 and (a.resend_org_ebay_id=0 or a.resend_org_ebay_id is null) and
b.ebay_amount >0 and a.ebay_warehouse !='' and a.ebay_user='manwei'
and (
a.shippedtime between UNIX_TIMESTAMP('".$astart_p."') and UNIX_TIMESTAMP('".$aend_p."') and
a.ebay_paidtime not between UNIX_TIMESTAMP('".$astart."') and UNIX_TIMESTAMP('".$aend."')
)";
if($last_ebay_id!='') {
$data3 = "select " . $fields_list . "
from ebay_order a left join ebay_orderdetail b on(a.ebay_ordersn=b.ebay_ordersn)
left join ebay_store c on (a.ebay_warehouse = c.id)
where a.ebay_combine !=1 and (a.resend_org_ebay_id=0 or a.resend_org_ebay_id is null) and
b.ebay_amount >0 and a.ebay_warehouse !='' and a.ebay_user='manwei'
and a.ebay_id >='" .$last_ebay_id ."'";
}
$data = "(" . $data1 . ")";
if($data2 != "") $data = $data . " union (". $data2 . ")";
if($data3 != "") $data = $data . " union (". $data3 . ")";
小插曲,当我们分析data2的时候,无论如何给shippedtime加索引,只要查询shippedtime都是全表查询。仔细分析才知道原来在数据库设计的时候,这个shippedtime的字段是varchar, 程序把时间戳保存成这种类型,自然没有办法使用适合我们需要的索引,解决的方法是通过alter语句先把shippedtime改成int 类型,再增加一个索引到这个字段。这样这个查询慢的问题就彻底得到解决了。
例子3:
$data = $isfesdb->query($data);
$quan = $isfesdb->num_rows($data);
for($i=0;$i<$quan;$i++){
{
...
$vv = "select goods_name, goods_weight from ebay_goods where goods_sn='".$sku[$i]."' limit 1";
$vv = $isfesdb->execute($vv);
$vv = $isfesdb->getResultArray($vv);
if(count($vv)==0){
...
$sku[$i] = str_replace('-FBA-FR','',$sku[$i]);
...
}
...
}
从代码上看,这个只是很简单的查询,ebay_goods也有索引,应该很快就能查询到结果。但实际上整个流程跑下来很慢。仔细分析原因是因为$quan的数字太大,导致了for循环超过了10000次,这样导致了$vv这个查询进行了10000次。所以单独查一条没有性能问题,但是如果多次重复这样的查询就会引起性能问题。
解决的方法就是在for循环的前面先查询ebay_goods全表,把这个表记录到一个数组,然后在for循环里使用素组的数据。因为ebay_goods这个数组只有几千条记录,这个方法是可行的。 修改程序变成:
$vv = $isfesdb->query("select goods_sn, goods_name, goods_weight from ebay_goods");
$vv_quan = $isfesdb->num_rows($vv);
$vv_result = $isfesdb->getResultArray($vv);
for($i=0; $i<$vv_quan; $i++) {
$goods_array[$vv_result[$i]['goods_sn']] = array($vv_result[$i]['goods_name'], $vv_result[$i]['goods_weight']);
}
for($i=0;$i<$quan;$i++)
{
...
if(!array_key_exists($sku[$i], $goods_array)){
...
$sku[$i] = str_replace('-FBA-FR','',$sku[$i]);
...
}
...
}
我们采用数组的方法后,查询也比旧方法效率提高好几倍。这是因为现在我们的服务器配置的内存是足够大的,PHP的运行也是足够快的。瓶颈就在于php在等待mysql的查询结果。所以我们先用一次查询把数据库结果组成了数组。