1、日期 (hive没有date类型)
1)、获取时间
current_date 2019-05-07
current_timestamp/now() 2019-05-07 15:20:49.247
2)、从日期时间中提取字段
year(''),month(''),day/dayofmonth,hour(''),minute(''),second(''),dayofweek(),dayofyear(),weekofyear()
trunc('2009-02-12', 'MM')截取某部分的日期,其他部分默认为01第二个参数 ["year", "yyyy", "yy", "mon", "month", "mm"]
3)、日期时间转换
unix_timestamp返回当前时间的unix时间戳:
unix_timestamp()当前时间,unix_timestamp('2016-04-08', 'yyyy-MM-dd')。
from_unixtime(0, 'yyyy-MM-dd HH:mm:ss');时间戳对应时间,联合unix_timestamp使用可以把字符串转换成不同格式的日期。
to_date/date()返回时间字符串的日期部分。
4)、日期计算
months_between('1997-02-28 10:30:00', '1996-10-30')两个日期之间的月数
datediff('2009-07-31', '2009-07-30');两个日期之间的天数
add_months('2016-08-31', 1)日期加n月
last_day(date)当月的最后一天
date_add('2016-07-30', 1)
date_sub('2016-07-30', 1)
2、数值函数
round(3.1415926) 返回 BIGINT;round(double a, int d)返回DOUBLE
floor(double a)ceil(double a)
rand()取随机数函数
abs(double a) 绝对值
3、字符串函数
length()字符串长度
concat(‘abc’,'def’,'gh’)字符串链接
concat_ws(',','abc','def','gh')带分隔符字符串连接函数
upper('abSEd')
lower('abSEd')
trim(' abc ')
regexp_replace('foobar', 'oo|ar', '')正则表达式替换
get_json_object(string json_string,'$.owner')获取json中的某个值
presto. json_extract_scalar
from_json('{"k": "fruit", "v": "apple"}','k STRING, v STRING', map("",""))
json_tuple('{"name":"jack","server":"www.qq.com"}','server','name')
cast(1 as bigint)类型转换
substr(string A, int start, int len)
substr(string A, int start)
split(regexp_replace(msg,'\\\\x22','"'),' -')[1]
4、取最近一次所有数据
select id,loan_time as last_loan_time from
(
select a.id,loan_time,
row_number() over(partition by a.id order by loan_time desc) as xu
from
cs.repayment a
)
where xu=1
--默认为从起点到当前行
sum(sales) OVER(PARTITION BY user_type ORDER BY sales asc) AS sales_1,
--从起点到当前行,结果与sales_1不同。
sum(sales) OVER(PARTITION BY user_type ORDER BY sales asc ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS sales_2,
--当前行+往前3行
sum(sales) OVER(PARTITION BY user_type ORDER BY sales asc ROWS BETWEEN 3 PRECEDING AND CURRENT ROW) AS sales_3,
--当前行+往前3行+往后1行
sum(sales) OVER(PARTITION BY user_type ORDER BY sales asc ROWS BETWEEN 3 PRECEDING AND 1 FOLLOWING) AS sales_4,
--当前行+往后所有行
sum(sales) OVER(PARTITION BY user_type ORDER BY sales asc ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS sales_5,
--分组内所有行
SUM(sales) OVER(PARTITION BY user_type) AS sales_6
lag(device_id) over (order by sales) as default_before_one_line,
lead(device_id,2) over (order by sales) as after_two_line,
lag(device_id,2,'abc') over (order by sales) as before_two_line
5、默认值coalesce(name,"")
解决join空值中的坑。
6、多行拼接concat_ws
concat_ws('_',collect_set(promotion_id)collect_set的作用是对promotion_id去重
7、解析list为多行
json_extract
select union_id
,get_json_object(tag1,'$.url') audit_content
from (
select union_id
,substr(get_json_object(audit_content,'$.images'),2,length(get_json_object(audit_content,'$.images'))-2) audit_content
from table
WHERE pt = "${bizdate}" and get_json_object(audit_content,'$.images')<>'[]'
) ta lateral view explode(split(audit_content, ',')) r1 as tag1 ;
explore(c_map) b AS map_key, map_value;
presto
select id,new_name from table
cross join unnest(split(name,',')) as tmp(new_name )
alter table tablename add columns(columnname string) ;
删除外部表,先将外部表改为内部表
alter table test_external set TBLPROPERTIES('EXTERNAL'='false');
alter table test_external change column etl_time etl_time string COMMENT '数据时间';
alter table app.example_orc partition (dt="20180505") concatenate;合并小文件
ALTER TABLE my_partition_test_table DROP IF EXISTS PARTITION (p_loctype='MHA');
GET_JSON_OBJECT(tag_cont,'$[*].tag') alter table my_partition_test_table if not exists add partition (p_hour='2017113003', p_city='573', p_loctype='MHA');
查询除了当前这条数据的其他数据并合并
replace(replace(concat_ws(';',collect_set(concat(commentid,':',hot_score))over()),concat(';',commentid,':',hot_score),''),concat(commentid,':',hot_score,';'),'') as comment_list,
窗口中按指定条件排序:
1、将指定条件的数据查出来添加排序字段;
2、first_value(case when act_semantic='pv_out' then extra_info else null end) over(partition by device_id, sinfo['req_id'], sinfo['object_id'] order by case when act_semantic='pv_out' then log_time else 0 end desc) as extra_info,
split( regexp_replace(regexp_replace(GET_JSON_OBJECT(ext1, '$[*].accounts'), '\\[+\\[+\\{', ''), '\\}+\\]+\\]', ''), '\\}+\\]+\\,+\\[+\\{') )获取list里面所有accounts值
array_contains
array_distinct
cardinality presto查询数组/map大小
spark concat 连接两个数组
to_json map转json
str_to_map json 转map
MSCK REPAIR TABLE 外部表数据复原
解决hive外部表创建_success问题
"mapreduce.fileoutputcommitter.marksuccessfuljobs": false
array_intersect(array(1, 2), array(2, 3)) i,
array_union(array(1, 2), array(2, 3)) u,
array_except(array(1, 2), array(2, 3)) e;
数据治理
hadoop fs -du -s -h oss://bigdata/data_warehouse/xxxxxxxxx
hadoop fs -rm -r -skipTrash oss://bigdata/data_warehouse/xxxxxxxxx
hadoop fs -ls oss://bigdata/data_warehouse/xxxxxxxxx |wc -n 100
SparkSession
val conf = new SparkConf().setAppName("SparkWithHive_Sparkv21")
val sparkSession = SparkSession.builder().config(conf).enableHiveSupport().getOrCreate()
sql
spark.sql("") 返回DataFrame
DataFrame方法:
show(n)默认展示20条记录
collect()获取所有数据到数组返回Array对象,map(r=>())
collectAsList()获取所有数据到List
first, head, take, takeAsList:获取若干行记录
where($"")<=>filter($"")条件筛选
select("","")查询指定
distinct()返回一个不包含重复记录的DataFrame
withColumn("period",$"")新增一列,$""取列值,lit(1)给默认值。
withColumnRenamed("原名","新名")
$""的操作:
length($"")
$"".cast(String)
when($""=,)otherwise()
自定义方法
def findHentai(sex:String,dressing:String): String ={
if(sex =="boy" && dressing == "裙子") "变态" else "正常"
}
声明
var check_amt_udf = udf(findHentai _)
调用
check_amt_udf.apply($"",$"")
join
df1.join(df2):笛卡尔积
多个相同字段形式
df1.join(df2, Seq("id","name"))
不相同字段
df1.join(df2 , df1("id" ) === df2( "t1_id"))
指定join类型
df1.join(df2, Seq("id","name"),"left")
groupBy("").agg(
max("").alias(""),
collect_list("") as ""
)
类 object
方法def f():Unit={}
变量var
常量val
spark 循环
df.foreachPartition(partition => {partition.foreach(line => {line.get(line.fieldIndex(dimensionColumn))})})
var index=Map("t1111dacu_rgst"->"1111dacu_rgst")
index+=("t1212dacu_rgst"->"1212dacu_rgst")
index.foreach{case (k,v) => {}}
SparkSQL执行时参数优化 - 不会叫的喵· - 博客园
spark从hdfs里面获取数据分区数是不可设定的,如果后面有复杂的处理逻辑,需要使用宽依赖算子如distinct使stage拆分,再设置partition。或者使用分桶的方法。
spark优化:
spark.conf.set("spark.sql.shuffle.partitions",1000)
df.repartition(1000).persist()
Hive任务优化
1、 map一直等于0%,set hive.auto.convert.join = false转成reduce端的Common Join。
2、 reduce一直是99%,可能是reduce太少(默认是3亿),SET hive.exec.reducers.bytes.per.reducer=1000000000;单个reduce人处理的数据量或set mapred.reduce.tasks=80;。
3、 hive.groupby.skewindata group by数据倾斜。hive.map.aggr=true 这个配置项代表是否在map端进行聚合
4、 set hive.optimize.skewjoin = true; join数据倾斜set hive.skewjoin.key =4*(count-reducer)
5、 join or 用union all GROUP BY这样替代
6、 Error: Java heap space 设置内存set mapreduce.map.memory.mb=4096; set mapreduce.map.java.opts=-Xmx3600m;
PySpark之列操作 - 简书