Hive调优实践

Hive调优

1 文件格式的选择

ORC格式确实要比textFile要更适合于hive,查询速度会提高20-40%左右

例子1:

youtube1的文件格式是TextFIle,youtube3的文件格式是orc

hive> select videoId,uploader,age,views from youtube1 order by views limit 10;

Query ID = hadoop_20170710085454_6768a540-a0b3-4d98-92a0-f97d4eff8b42
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0070, Tracking URL = http://master:8088/proxy/application_1499153664137_0070/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0070
Hadoop job information for Stage-1: number of mappers: 6; number of reducers: 1
2017-07-10 08:55:18,434 Stage-1 map = 0%,  reduce = 0%
2017-07-10 08:55:56,924 Stage-1 map = 1%,  reduce = 0%, Cumulative CPU 11.99 sec
...
2017-07-10 08:56:33,719 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 62.7 sec
MapReduce Total cumulative CPU time: 1 minutes 2 seconds 700 msec
Ended Job = job_1499153664137_0070
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 6  Reduce: 1   Cumulative CPU: 62.7 sec   HDFS Read: 1086704175 HDFS Write: 323 SUCCESS
Total MapReduce CPU Time Spent: 1 minutes 2 seconds 700 msec
OK
P4c-EViSRsw ERNESTINEbrowning   1240    0
woMdGKHIg3o Maxwell739  1240    0
_jVH58-X4C4 nelenajolly 1240    0
k1iTl0Kh4DQ rachellelala    1240    0
8LdPZ_n1S4c GohanxVidel21   1240    0
PSex2TAkQC8 Qingy3  1254    0
YC20zP9p_wI aimeenmegan 1254    0
3XduLiQMMTM marshallgovindan    1240    0
IjeXG6yXXZ4 SenateurDupont1973  1254    0
sw4XgF1zkXE bablooian   1240    0
Time taken: 119.739 seconds, Fetched: 10 row(s)

hive> select videoId,uploader,age,views from youtube3 order by views limit 10;

Query ID = hadoop_20170710085959_e6d66799-0a8a-4696-bf93-a0abc3f00de0
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0071, Tracking URL = http://master:8088/proxy/application_1499153664137_0071/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0071
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 1
2017-07-10 08:59:43,472 Stage-1 map = 0%,  reduce = 0%
2017-07-10 09:00:12,555 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 2.99 sec
...
2017-07-10 09:00:52,499 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 50.2 sec
MapReduce Total cumulative CPU time: 50 seconds 200 msec
Ended Job = job_1499153664137_0071
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 1   Cumulative CPU: 50.2 sec   HDFS Read: 95047649 HDFS Write: 323 SUCCESS
Total MapReduce CPU Time Spent: 50 seconds 200 msec
OK
P4c-EViSRsw ERNESTINEbrowning   1240    0
woMdGKHIg3o Maxwell739  1240    0
_jVH58-X4C4 nelenajolly 1240    0
k1iTl0Kh4DQ rachellelala    1240    0
8LdPZ_n1S4c GohanxVidel21   1240    0
PSex2TAkQC8 Qingy3  1254    0
YC20zP9p_wI aimeenmegan 1254    0
3XduLiQMMTM marshallgovindan    1240    0
IjeXG6yXXZ4 SenateurDupont1973  1254    0
sw4XgF1zkXE bablooian   1240    0
Time taken: 109.776 seconds, Fetched: 10 row(s)

例子2:

hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube1 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710090404_46a79a3d-8863-4390-8898-7f82c5a3b7ab
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 5
Starting Job = job_1499153664137_0072, Tracking URL = http://master:8088/proxy/application_1499153664137_0072/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0072
Hadoop job information for Stage-1: number of mappers: 6; number of reducers: 5
2017-07-10 09:05:04,067 Stage-1 map = 0%,  reduce = 0%
2017-07-10 09:05:40,498 Stage-1 map = 6%,  reduce = 0%, Cumulative CPU 5.9 sec
...
2017-07-10 09:06:09,681 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 39.56 sec
MapReduce Total cumulative CPU time: 39 seconds 560 msec
Ended Job = job_1499153664137_0072
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0073, Tracking URL = http://master:8088/proxy/application_1499153664137_0073/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0073
Hadoop job information for Stage-2: number of mappers: 2; number of reducers: 1
2017-07-10 09:06:45,360 Stage-2 map = 0%,  reduce = 0%
...
2017-07-10 09:07:42,083 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 3.62 sec
MapReduce Total cumulative CPU time: 3 seconds 620 msec
Ended Job = job_1499153664137_0073
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 6  Reduce: 5   Cumulative CPU: 39.56 sec   HDFS Read: 1086732303 HDFS Write: 1188 SUCCESS
Stage-Stage-2: Map: 2  Reduce: 1   Cumulative CPU: 3.62 sec   HDFS Read: 8218 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 43 seconds 180 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
Blogs   447581
People  447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Howto   124885
Style   124885
Pets    86444
Animals 86444
Travel  82068
Events  82068
Education   54133
Technology  50925
Science 50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 201.127 seconds, Fetched: 25 row(s)

hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view
explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710090909_bfddc50b-665d-4296-9475-0cee55058c85
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 3
Starting Job = job_1499153664137_0074, Tracking URL = http://master:8088/proxy/application_1499153664137_0074/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0074
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 09:09:46,075 Stage-1 map = 0%,  reduce = 0%
...
2017-07-10 09:10:39,596 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 20.87 sec
MapReduce Total cumulative CPU time: 20 seconds 870 msec
Ended Job = job_1499153664137_0074
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0075, Tracking URL = http://master:8088/proxy/application_1499153664137_0075/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0075
Hadoop job information for Stage-2: number of mappers: 3; number of reducers: 1
2017-07-10 09:11:16,172 Stage-2 map = 0%,  reduce = 0%
...
2017-07-10 09:12:01,694 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 4.43 sec
MapReduce Total cumulative CPU time: 4 seconds 430 msec
Ended Job = job_1499153664137_0075
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 20.87 sec   HDFS Read: 47331548 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 3  Reduce: 1   Cumulative CPU: 4.43 sec   HDFS Read: 9228 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 25 seconds 300 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Pets    86444
Animals 86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 177.463 seconds, Fetched: 25 row(s)

我这里的比较是从CPU的耗时进行比较,至于Job的初始化等耗时由于存在不确定性,会影响比较的准确性所以暂时忽略

语句youtube1耗时youtube3耗时
select videoId,uploader,age,views from youtube1/youtube3 order by views limit 10;Total MapReduce CPU Time Spent: 1 minutes 2 seconds 700 msec(63s)Total MapReduce CPU Time Spent: 50 seconds 200 msec(51s)
select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube1/youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;Total MapReduce CPU Time Spent: 43 seconds 180 msec(44s)Total MapReduce CPU Time Spent: 25 seconds 300 msec(26s)


由上面表格我们可以看出以orc作为文件格式的表格的查询速度会比以textfile为文件格式的表格要提升20-40%左右

2 order by 和distribute by&sort by

如果在数据量比较大的情况下,也就是执行一个job时间比较长,那么在使用order by 进行topn之前先用distribute by 和 sort by 先局部进行topN,然后最终结果在使用order by效果会更好.如果你的job执行起来也就是几分钟的话,那就不建议使用,因为先使用distribute by和sort by 先 topN 再对结果order by 要启动三个 job来完成,MR的启动时间有点长,这是没必要的开销,比直接使用order by花费更多时间

2.1 使用 order by

hive> select * from youtube3 order by views limit 10;

Query ID = hadoop_20170710103535_4de243fb-f26b-42ab-8cc9-aef568b4a5cb
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0094, Tracking URL = http://master:8088/proxy/application_1499153664137_0094/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0094
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 1
2017-07-10 10:35:55,171 Stage-1 map = 0%,  reduce = 0%
2017-07-10 10:36:33,748 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 24.59 sec
...
2017-07-10 10:37:45,584 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 114.3 sec
MapReduce Total cumulative CPU time: 1 minutes 54 seconds 300 msec
Ended Job = job_1499153664137_0094
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 1   Cumulative CPU: 114.3 sec   HDFS Read: 574632526 HDFS Write: 2016 SUCCESS
Total MapReduce CPU Time Spent: 1 minutes 54 seconds 300 msec
OK
P4c-EViSRsw ERNESTINEbrowning   1240    ["Entertainment"]   66  0   0.0 0   0   ["b0b7yT9e7hM","UEq74FEphvc","PjQ2y_af08Y","myj6exBy5IQ","vKMljsGZluc","l4FzM1qImdg","mXw1UE2mOuk","gH_K5iqSdgA","InHt3_nkT7s","h3gMHd0XKnE","fXRt4ua3UsQ","T6-5K58vhl4","P1iwvBnOoHg","QAg12t6UvcQ","jy7zqy_Dzic","emUoDRc98ig","g66xEcC9kWY","AbYjCb1jHP8","RF0aRkdtiAM","BURRJkwFkqc"]
woMdGKHIg3o Maxwell739  1240    ["Pets","Animals"]  103 0   0.0 0   0   ["hSSqMOL6ThE","c2k9YCkcoaU","9idmedIj5pk","AUokp8o5aZc","Vhc9bGFKhds","LNQI20dunF0","BOlo7q7MSZA","9cGYnRGqgyM","_oO1WPlJUxc","XMqP8E6kfdY","qnR6BnVByVA","GkGd4RyWMYU","VHt8IZSDU1o","Z83w2lIJZVM","jmnugMwoP1Q","llfOxhnlrJU","JXUmvz4cIRA","Q8nHWUG7aAA","dqQOVI6hDCI","m80Qz8nwmHI"]
_jVH58-X4C4 nelenajolly 1240    ["Entertainment"]   205 0   0.0 0   0   ["VkNejMz7FeI","3AfW6V5EAWY","y8i8U9Ni6Dc","_I3Aeh_HHEg","sZBoBpxnoY8","fheFnYcUxLs","MzFWXWkmvz8","OQDRuqWsnp8","qGW43kf2QEU","u79DsR40LOA","X4TtK-3hryw","ANL8q16edOE","OMLKjvcs5EY","qohfnkV8J3Y","jLEFgyU6TdI","cEmyMSvPvVg","-1B8IkZkGOs","STUi3-cV4RQ","-IU8xCjTNW8","-OfnDq4HLPw"]
k1iTl0Kh4DQ rachellelala    1240    ["People","Blogs"]  48  0   0.0 0   0   []
8LdPZ_n1S4c GohanxVidel21   1240    ["Entertainment"]   271 0   0.0 0   0   ["Dt-0QZX5MII","K34GSodojaM","g6hYT815-9I","r5aYMYBEyPo","Bp_JAed6uTc","qe3PF-KyjCI","nWvwSYIUMpE","LxKS1beypxo","5ZmqPCjpmik","m_wYIbkkizc","9xDjgav1pkk","w5WJKrLR78M","nxj_3D3lIQ0","VSEV1IXg2hQ","gwEJPELAJiM","qDrBEST48_I","LymqYIU3E4U","2bhoPxDzU3w","nr_-AyOyfqk","Cg4qrZG1uVc"]
PSex2TAkQC8 Qingy3  1254    ["Entertainment"]   300 0   0.0 0   0   ["FOs2-r_Fikg","I2TrAyBGGcY","iHws_tdiK64","Yfr_wFmWHTI","FK-Z56YqTsU","Rmfy5-dPl6E","nsfOPbS2bSk","tz59v4WgXhA","ggPoCRixQvQ","hvtWJxJn_GI","qKD6VMiCSlI","O9pFAKOON7o","MZta_rn6o48","y2JV5Hs4WYg","kzP6HA-8MmA","60AR_R_UklQ","tNDVwd00Mow","N0lmZqXKJUw","dYxKJNYpjSA","EWBW5QuAedY"]
YC20zP9p_wI aimeenmegan 1254    ["Film","Animation"]    138 0   0.0 0   0   ["AuuCpfU1aNA","pfeU2WJPlnc","tdxYPHX7mhs","Vig624KxPPQ","mFWhAjQdjyI","7rbpfmqKfKI","Of8uv_etw1w","iuzYVo8L7KU","M_g220ZmoRo","oJ2YCvmg24M","PmxQP2h6Qvg","1u4HgPlGqrg","mGdoxToj5IA","hAnjWHP7KAc","rK66jzIvynM","RgK4BaBnIBU","B9Yp1lD4Ohg","vFRHyRmZ8c0","ZFoswQiXhwU","ptKLZlhz2SI"]
3XduLiQMMTM marshallgovindan    1240    ["Education"]   545 0   0.0 0   0   []
IjeXG6yXXZ4 SenateurDupont1973  1254    ["Howto","Style"]   467 0   0.0 0   0   []
sw4XgF1zkXE bablooian   1240    ["People","Blogs"]  47  0   0.0 0   0   []
Time taken: 153.26 seconds, Fetched: 10 row(s)

2.2 使用distribute by和sort by

hive> select a.* from (select * from youtube3 distribute by views sort by views limit 10) a order by views limit 10;

Query ID = hadoop_20170710104040_52c1a6e7-0dc7-4359-82d0-e566c536947a
Total jobs = 3
Launching Job 1 out of 3
Number of reduce tasks not specified. Defaulting to jobconf value of: 2
Starting Job = job_1499153664137_0095, Tracking URL = http://master:8088/proxy/application_1499153664137_0095/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0095
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 2
2017-07-10 10:41:35,382 Stage-1 map = 0%,  reduce = 0%
...
2017-07-10 10:44:04,783 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 116.58 sec
MapReduce Total cumulative CPU time: 1 minutes 56 seconds 580 msec
Ended Job = job_1499153664137_0095
Launching Job 2 out of 3
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0096, Tracking URL = http://master:8088/proxy/application_1499153664137_0096/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0096
Hadoop job information for Stage-2: number of mappers: 1; number of reducers: 1
2017-07-10 10:44:39,421 Stage-2 map = 0%,  reduce = 0%
2017-07-10 10:45:08,358 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 1.12 sec
2017-07-10 10:45:28,180 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 2.68 sec
MapReduce Total cumulative CPU time: 2 seconds 680 msec
Ended Job = job_1499153664137_0096
Launching Job 3 out of 3
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0097, Tracking URL = http://master:8088/proxy/application_1499153664137_0097/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0097
Hadoop job information for Stage-3: number of mappers: 1; number of reducers: 1
2017-07-10 10:46:05,072 Stage-3 map = 0%,  reduce = 0%
2017-07-10 10:46:33,177 Stage-3 map = 100%,  reduce = 0%, Cumulative CPU 1.07 sec
2017-07-10 10:47:00,963 Stage-3 map = 100%,  reduce = 100%, Cumulative CPU 2.4 sec
MapReduce Total cumulative CPU time: 2 seconds 400 msec
Ended Job = job_1499153664137_0097
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 2   Cumulative CPU: 116.58 sec   HDFS Read: 574635698 HDFS Write: 5128 SUCCESS
Stage-Stage-2: Map: 1  Reduce: 1   Cumulative CPU: 2.68 sec   HDFS Read: 11354 HDFS Write: 2314 SUCCESS
Stage-Stage-3: Map: 1  Reduce: 1   Cumulative CPU: 2.4 sec   HDFS Read: 9143 HDFS Write: 1965 SUCCESS
Total MapReduce CPU Time Spent: 2 minutes 1 seconds 660 msec
OK
dTPF85D4tEs Antoinnette71   1240    ["Entertainment"]   73  0   0.0 0   0   []
5e_flEf6uoU alexandreetmartin   1254    ["Comedy"]  137 0   0.0 0   0   ["e8F9txsHzQI","nibCFP4OB74","tH-7yll1t24","T5K17x93aMM","KxSCMsBipAM","JYEsGC71bmo","PAgQcF7eOt0","-jpkLH4RQ7I","UsY-Iu8QsqA","ru48hMHNZ08","D4aHmwmEBkw","vVYTYnT8zXE","wBcaCI0UMAE","qs1Queu8tGE","zsLRrBHIlZc","8lmWTQZtOsY","ztunItmPYzM","O8KUVYl8_wI","yNYTg9aZ1FY","MPF3JwAgNTY"]
I3bMdoqH2g4 lapino50    1254    ["Comedy"]  131 0   0.0 0   0   []
AvdoZL0U0Ko kristianbreen   1240    ["Comedy"]  121 0   0.0 0   0   []
9SAe80p-U48 BobRucklepuckle 1240    ["Gaming"]  170 0   0.0 0   0   ["z04IyYFXfP0","jcaDekCI408","bV3XZe0Az34","O7aYaJ8PtCs","I07LmzF46a8","zEl2frWgqdU","zBA-h4uM3cY","0t_mN96jRb8","bahgug-PFPs","VGDOZiVRUtg","OaXE6pMEsZc","A2Is6NIEuso","hNtqpVP5st4","TxZ2lzPbuDA","UFSc3rrAYLQ","58DEKFeWh4I","YDEQIbcDO1M","o9dyNqactV8","Usb-fZm56GM","XvV1yvSPJ_s"]
P2cQ-W9J99A 636141  1254    ["Comedy"]  77  0   0.0 0   0   ["74sW9OHvkCg","onKIV_-f6ew","XtHYdkH1flg","-SM5P5lz86A","zpz3HFBQR2o","U9lOI7n8sB4","w99Q5PLtjjE","5N6Cchi9Gtw","p9Bn2NSfzHw","DR2IaxcAAUU","BiGZd4BPlqw","0GMzzp_nJbw","e2KVuzAcsv0","jEz_k247tws","OEt50_BdIpQ","MV95TjaXkX8","c8lyUi8QKWs","P1OXAQHv09E","gtp8fgYhp50","L5-0k1HHh_k"]
JobgU5-dFy4 singindancinfool    1254    ["Entertainment"]   184 0   0.0 0   0   ["7MFG_yaFlvk","c--loq6HsDQ","CyrZR3BSH5o","Wni8rmr7rzs","XvFfRpQS4GU","ENSiblZIMV0","0enG7lzuaqM","c8y-zzaI4g8","Czfk0W888V0","dc1ls-qQrYg","3ATYGqE6Hww","U9DPjKzumYw","esEiXSOjs2c","WRCpHFSYXdA","32YyZMxFrBA","Abg7uBdwGJM","BqELlVTMv8o","2tbj33kpBig","LronHxxGti8","NtSl2jdPezI"]
YC20zP9p_wI aimeenmegan 1254    ["Film","Animation"]    138 0   0.0 0   0   ["AuuCpfU1aNA","pfeU2WJPlnc","tdxYPHX7mhs","Vig624KxPPQ","mFWhAjQdjyI","7rbpfmqKfKI","Of8uv_etw1w","iuzYVo8L7KU","M_g220ZmoRo","oJ2YCvmg24M","PmxQP2h6Qvg","1u4HgPlGqrg","mGdoxToj5IA","hAnjWHP7KAc","rK66jzIvynM","RgK4BaBnIBU","B9Yp1lD4Ohg","vFRHyRmZ8c0","ZFoswQiXhwU","ptKLZlhz2SI"]
vBGEkIG8Zdk hachem44    1254    ["Entertainment"]   122 0   0.0 0   0   []
P4eiE33eqOU BsmStudio   1254    ["Comedy"]  287 0   0.0 0   0   ["1_ao3c5jicU","RRMAZj-msOc","X04hwJyaO_4","4oee8Zn3hHI","60W-uUYnj3w","FtoJoBp03r4","_IMSMYLM0zs","22Sz120MAT8","F-B35V6Sszk","1AExOXwL0Ag","owRgISGt_iw","Wz0ZfHbPC-c","b7uLUY-FdKM","yevHaMC10IE","jGvBLJfEIcs","Ma4Ozief4k8","ZwnLErETvKU","oE4oeyRMcKI","IQnEUxtk-Pw"]
Time taken: 368.489 seconds, Fetched: 10 row(s)
语句直接使用order by先使用distribute by和sort by再使用order by
select * from youtube3 order by views limit 10;Total MapReduce CPU Time Spent: 1 minutes 54 seconds 300 msec(115s)
Time taken: 153.26 seconds, Fetched: 10 row(s)
select a.* from (select * from youtube3 distribute by views sort by views limit 10) a order by views limit 10;Total MapReduce CPU Time Spent: 2 minutes 1 seconds 660 msec(122s)
Time taken: 368.489 seconds, Fetched: 10 row(s)


由上面表格我们可以看出,使用distribute by和sort by 来优化order by 对于使用场景是有要求的,并不是拿来就用的,而是使用于数据量较大的场景,要不然单单是启动多两个job的时间就够你喝一壶了。数据量小直接使用order by即可。

3 group by 的数据倾斜处理

  • hive.map.aggr = true; 相当于combine
  • hive.groupby.skewindata = true;开启两个job,这个参数的修改跟在orderby 之前使用sort by和distribute by处理一样,数据量不大没必要开启

4 Join的优化

  • map join当小表的数据量比较小时,会自动启用mapjoin,性能能提高30%左右
    • hive.auto.convert.join = true;自动优化是否采用map join
    • hive.auto.convert.join.noconditionaltask = true; 根据小表的大小来决定是否开启map join
    • hive.auto.convert.join.noconditionaltask.size = 10000000;10M 当小表的值小于设定的值时启用map join,只有上一个参数为true时有用
  • bucketmap join
    • hive.optimize.bucketmapjoin=true; 当两表都采用桶分表,并且小表的桶数是大表桶数的倍数时启用,减少数据的遍历
  • 过滤数据后再join
  • 如果多表join的on条件一样,一起写会转化成一个job

4.1 开启MapJoin和关闭MapJoin的比较

4.1.1 不启动MapJoin

hive> set hive.auto.convert.join = false;
hive> set hive.auto.convert.join.noconditionaltask = false;
hive> set hive.optimize.bucketmapjoin=false;

查看表格结构:

hive> desc formatted user;

# col_name              data_type               comment             

uploader                string                                      
videos                  int                                         
friends                 int                                         

# Detailed Table Information         
Database:               default                  
Owner:                  hadoop                   
CreateTime:             Fri Jul 07 13:46:53 CST 2017     
LastAccessTime:         UNKNOWN                  
Protect Mode:           None                     
Retention:              0                        
Location:               hdfs://cluster1/user/hive/warehouse/user     
Table Type:             MANAGED_TABLE            
Table Parameters:        
    COLUMN_STATS_ACCURATE   true                
    numFiles                1                   
    numRows                 1192676             
    rawDataSize             121652944           
    totalSize               9783258             
    transient_lastDdlTime   1499406528          

# Storage Information        
SerDe Library:          org.apache.hadoop.hive.ql.io.orc.OrcSerde    
InputFormat:            org.apache.hadoop.hive.ql.io.orc.OrcInputFormat  
OutputFormat:           org.apache.hadoop.hive.ql.io.orc.OrcOutputFormat     
Compressed:             No                       
Num Buckets:            24                       
Bucket Columns:         [uploader]               
Sort Columns:           []                       
Storage Desc Params:         
    field.delim             \t                  
    serialization.format    \t                  
Time taken: 0.087 seconds, Fetched: 34 row(s)

hive> desc formatted youtube3;

# col_name              data_type               comment             

videoid                 string                                      
uploader                string                                      
age                     int                                         
category                array<string>                               
length                  int                                         
views                   int                                         
rate                    float                                       
ratings                 int                                         
comments                int                                         
relatedid               array<string>                               

# Detailed Table Information         
Database:               default                  
Owner:                  hadoop                   
CreateTime:             Fri Jul 07 13:28:56 CST 2017     
LastAccessTime:         UNKNOWN                  
Protect Mode:           None                     
Retention:              0                        
Location:               hdfs://cluster1/user/hive/warehouse/youtube3     
Table Type:             MANAGED_TABLE            
Table Parameters:        
    COLUMN_STATS_ACCURATE   true                
    numFiles                6                   
    numRows                 5134636             
    rawDataSize             7949154631          
    totalSize               574644356           
    transient_lastDdlTime   1499405660          

# Storage Information        
SerDe Library:          org.apache.hadoop.hive.ql.io.orc.OrcSerde    
InputFormat:            org.apache.hadoop.hive.ql.io.orc.OrcInputFormat  
OutputFormat:           org.apache.hadoop.hive.ql.io.orc.OrcOutputFormat     
Compressed:             No                       
Num Buckets:            8                        
Bucket Columns:         [uploader]               
Sort Columns:           []                       
Storage Desc Params:         
    colelction.delim        &                   
    field.delim             \t                  
    serialization.format    \t                  
Time taken: 0.086 seconds, Fetched: 42 row(s)

实例测试:

hive> select a.uploader,a.friends,b.videoId from user a join youtube3 b on a.uploader = b.uploader limit 10;

Query ID = hadoop_20170710092626_a2f303f3-8ee5-4747-a6c0-96ccb5d2f0d7
Total jobs = 1
Launching Job 1 out of 1
Number of reduce tasks not specified. Estimated from input data size: 3
Starting Job = job_1499153664137_0077, Tracking URL = http://master:8088/proxy/application_1499153664137_0077/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0077
Hadoop job information for Stage-1: number of mappers: 5; number of reducers: 3
2017-07-10 09:26:41,012 Stage-1 map = 0%,  reduce = 0%
...
2017-07-10 09:27:49,553 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 66.63 sec
MapReduce Total cumulative CPU time: 1 minutes 6 seconds 630 msec
Ended Job = job_1499153664137_0077
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 5  Reduce: 3   Cumulative CPU: 66.63 sec   HDFS Read: 86211567 HDFS Write: 698 SUCCESS
Total MapReduce CPU Time Spent: 1 minutes 6 seconds 630 msec
OK
a0000a  7   6NmdrPmWjSU
a00393977   0   _EEwnI7pCCk
a007plan    0   VlDabTbGLNc
a02030203   0   AaY3pbdfbUM
a02030203   0   P28i2Xu4WB8
a02030203   0   LFzKFOWOHBg
a042538 14  YHu_VlY_p_E
a042538 14  8ECefR4g0Tg
a04297  0   2rqgJ1FNflo
a04297  0   iT_yac8DcQk
Time taken: 109.44 seconds, Fetched: 10 row(s)

4.1.2 开启自动MapJoin优化

hive> set hive.auto.convert.join = true;
hive> set hive.auto.convert.join.noconditionaltask = true;

hive> select a.uploader,a.friends,b.videoId from user a join youtube3 b on a.uploader = b.uploader limit 10;

Query ID = hadoop_20170710093232_6b9bf92d-5d02-4ce6-bb35-a58625a8f03b
Total jobs = 1
17/07/10 09:32:35 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Execution log at: /tmp/hadoop/hadoop_20170710093232_6b9bf92d-5d02-4ce6-bb35-a58625a8f03b.log
2017-07-10 09:32:37 Starting to launch local task to process map join;  maximum memory = 518979584
2017-07-10 09:32:50 Processing rows:    200000  Hashtable size: 199999  Memory usage:   76725768    percentage: 0.148
...
2017-07-10 09:32:56 Processing rows:    1100000 Hashtable size: 1099999 Memory usage:   316646056   percentage: 0.61
2017-07-10 09:32:56 Dump the side-table for tag: 0 with group count: 1192676 into file: file:/tmp/hadoop/81308db8-add0-47bf-9a70-ae8e83f4f05c/hive_2017-07-10_09-32-30_881_1416539360570079360-1/-local-10003/HashTable-Stage-3/MapJoin-mapfile160--.hashtable
2017-07-10 09:32:59 Uploaded 1 File to: file:/tmp/hadoop/81308db8-add0-47bf-9a70-ae8e83f4f05c/hive_2017-07-10_09-32-30_881_1416539360570079360-1/-local-10003/HashTable-Stage-3/MapJoin-mapfile160--.hashtable (36232906 bytes)
2017-07-10 09:32:59 End of local task; Time Taken: 21.563 sec.
Execution completed successfully
MapredLocal task succeeded
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_1499153664137_0079, Tracking URL = http://master:8088/proxy/application_1499153664137_0079/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0079
Hadoop job information for Stage-3: number of mappers: 4; number of reducers: 0
2017-07-10 09:33:42,458 Stage-3 map = 0%,  reduce = 0%
2017-07-10 09:34:16,719 Stage-3 map = 50%,  reduce = 0%, Cumulative CPU 13.33 sec
2017-07-10 09:34:27,090 Stage-3 map = 100%,  reduce = 0%, Cumulative CPU 27.05 sec
MapReduce Total cumulative CPU time: 27 seconds 50 msec
Ended Job = job_1499153664137_0079
MapReduce Jobs Launched: 
Stage-Stage-3: Map: 4   Cumulative CPU: 27.05 sec   HDFS Read: 20408350 HDFS Write: 1094 SUCCESS
Total MapReduce CPU Time Spent: 27 seconds 50 msec
OK
maxmagpies  22  9UagGiEP_kU
lilsteps97  2   hKE3F0cLl_M
cokerish    6   mZpSi9Sfs2k
mikecrazy03 17  a-nkbgF3Wcw
xoPaiigexo  21  CmCl_hSXM80
r4nd0mUs3r  133 2Yke3hdSeJQ
popkorn615  39  sc88PbADsk8
popkorn615  39  bGNzyCCK2Uo
tuxkiller007    0   s1n2y3c2z9k
popkorn615  39  norcLxmpS7o
Time taken: 118.299 seconds, Fetched: 10 row(s)

4.1.3 开启BucketMapJoin优化

hive> set hive.optimize.bucketmapjoin=true;

hive> select a.uploader,a.friends,b.videoId from user a join youtube3 b on a.uploader = b.uploader limit 10;

Query ID = hadoop_20170710093535_dddfa9db-cae4-4a9d-bbe0-0dba73ee1cb5
Total jobs = 1
17/07/10 09:36:02 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Execution log at: /tmp/hadoop/hadoop_20170710093535_dddfa9db-cae4-4a9d-bbe0-0dba73ee1cb5.log
2017-07-10 09:36:05 Starting to launch local task to process map join;  maximum memory = 518979584
2017-07-10 09:36:18 Processing rows:    200000  Hashtable size: 199999  Memory usage:   76897640    percentage: 0.148
...
2017-07-10 09:36:23 Processing rows:    1100000 Hashtable size: 1099999 Memory usage:   315574944   percentage: 0.608
2017-07-10 09:36:23 Dump the side-table for tag: 0 with group count: 1192676 into file: file:/tmp/hadoop/81308db8-add0-47bf-9a70-ae8e83f4f05c/hive_2017-07-10_09-35-58_131_4163936080161520499-1/-local-10003/HashTable-Stage-3/MapJoin-mapfile170--.hashtable
2017-07-10 09:36:26 Uploaded 1 File to: file:/tmp/hadoop/81308db8-add0-47bf-9a70-ae8e83f4f05c/hive_2017-07-10_09-35-58_131_4163936080161520499-1/-local-10003/HashTable-Stage-3/MapJoin-mapfile170--.hashtable (36232906 bytes)
2017-07-10 09:36:26 End of local task; Time Taken: 21.374 sec.
Execution completed successfully
MapredLocal task succeeded
Launching Job 1 out of 1
Number of reduce tasks is set to 0 since there's no reduce operator
Starting Job = job_1499153664137_0080, Tracking URL = http://master:8088/proxy/application_1499153664137_0080/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0080
Hadoop job information for Stage-3: number of mappers: 4; number of reducers: 0
2017-07-10 09:37:09,550 Stage-3 map = 0%,  reduce = 0%
2017-07-10 09:37:43,845 Stage-3 map = 25%,  reduce = 0%, Cumulative CPU 12.14 sec
2017-07-10 09:37:44,884 Stage-3 map = 50%,  reduce = 0%, Cumulative CPU 12.58 sec
2017-07-10 09:37:54,187 Stage-3 map = 100%,  reduce = 0%, Cumulative CPU 25.41 sec
MapReduce Total cumulative CPU time: 25 seconds 410 msec
Ended Job = job_1499153664137_0080
MapReduce Jobs Launched: 
Stage-Stage-3: Map: 4   Cumulative CPU: 25.41 sec   HDFS Read: 20408350 HDFS Write: 1094 SUCCESS
Total MapReduce CPU Time Spent: 25 seconds 410 msec
OK
maxmagpies  22  9UagGiEP_kU
lilsteps97  2   hKE3F0cLl_M
cokerish    6   mZpSi9Sfs2k
mikecrazy03 17  a-nkbgF3Wcw
xoPaiigexo  21  CmCl_hSXM80
r4nd0mUs3r  133 2Yke3hdSeJQ
popkorn615  39  sc88PbADsk8
popkorn615  39  bGNzyCCK2Uo
tuxkiller007    0   s1n2y3c2z9k
popkorn615  39  norcLxmpS7o
Time taken: 117.119 seconds, Fetched: 10 row(s)

4.1.4 总结

使用语句:select a.uploader,a.friends,b.videoId from user a join youtube3 b on a.uploader = b.uploader limit 10;

配置开启关闭
map join(将数据放到缓存中)End of local task; Time Taken: 21.563 sec.
Total MapReduce CPU Time Spent: 27 seconds 50 msec
CPU总耗时(22+28 = 50s)
Total MapReduce CPU Time Spent: 1 minutes 6 seconds 630 msec
CPU总耗时66s
bucket map joinEnd of local task; Time Taken: 21.374 sec.
Total MapReduce CPU Time Spent: 25 seconds 410 msec
CPU总耗时(22+26)48s
End of local task; Time Taken: 21.563 sec.
Total MapReduce CPU Time Spent: 27 seconds 50 msec
CPU总耗时(22+28)50s


由上面数据我们可以看出,map join在cpu耗时方面还是有所减少,而bucket map join相对于map join的提升就不明显,可能原因还是在于数据量问题,我们可以看出map join相对于普通的reduce join执行速度还是有提升的,而且数据量越大的时候又是会更明显。但是数据量不大的情况,有文件的读取上传等操作导致他的优势不明显,可能反而最终耗时更长。

5 参数配置:map数和reduce数的修改

5.1 map数的设置

5.1.1 小文件太多怎么办,会导致map数太多,不断创建和销毁任务会导致无谓的开销

  • set hive.input.format = org.apache.hadoop.hive.ql.io.CombineHiveInputFormat 用于合并小文件
  • set mapred.max.split.size = 128000000 限制文件的大小
  • set hive.merge.mapfiles = true; map任务后对文件进行合并
  • set hive.merge.mapredfiles = true; mapreduce任务后对文件进行合并

5.1.2 map数太少怎么办,设置map.tasks数

5.1.3 map任务和mapreduce任务后文件的合并与否的比较

1)不进行合并的设置
hive> set hive.merge.mapfiles=false;
hive> set hive.merge.mapredfiles = false;

hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710095252_9c97d548-c734-4df3-8d78-541c7e4ec66f
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 3
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0082, Tracking URL = http://master:8088/proxy/application_1499153664137_0082/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0082
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 09:53:26,240 Stage-1 map = 0%,  reduce = 0%
2017-07-10 09:53:54,547 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 2.05 sec
...
2017-07-10 09:54:22,238 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 21.08 sec
MapReduce Total cumulative CPU time: 21 seconds 80 msec
Ended Job = job_1499153664137_0082
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0083, Tracking URL = http://master:8088/proxy/application_1499153664137_0083/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0083
Hadoop job information for Stage-2: number of mappers: 1; number of reducers: 1
2017-07-10 09:54:59,413 Stage-2 map = 0%,  reduce = 0%
...
2017-07-10 09:55:56,728 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 2.43 sec
MapReduce Total cumulative CPU time: 2 seconds 430 msec
Ended Job = job_1499153664137_0083
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 21.08 sec   HDFS Read: 47331544 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 1  Reduce: 1   Cumulative CPU: 2.43 sec   HDFS Read: 5826 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 23 seconds 510 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Animals 86444
Pets    86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 196.576 seconds, Fetched: 25 row(s)
2)进行合并的设置
hive> set hive.merge.mapfiles=true;
hive> set hive.merge.mapredfiles=true;

hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710090909_bfddc50b-665d-4296-9475-0cee55058c85
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 3

Starting Job = job_1499153664137_0074, Tracking URL = http://master:8088/proxy/application_1499153664137_0074/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0074
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 09:09:46,075 Stage-1 map = 0%,  reduce = 0%
2017-07-10 09:10:15,149 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 1.99 sec
...
2017-07-10 09:10:39,596 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 20.87 sec
MapReduce Total cumulative CPU time: 20 seconds 870 msec
Ended Job = job_1499153664137_0074
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
Starting Job = job_1499153664137_0075, Tracking URL = http://master:8088/proxy/application_1499153664137_0075/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0075
Hadoop job information for Stage-2: number of mappers: 3; number of reducers: 1
2017-07-10 09:11:16,172 Stage-2 map = 0%,  reduce = 0%
...
2017-07-10 09:12:01,694 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 4.43 sec
MapReduce Total cumulative CPU time: 4 seconds 430 msec
Ended Job = job_1499153664137_0075
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 20.87 sec   HDFS Read: 47331548 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 3  Reduce: 1   Cumulative CPU: 4.43 sec   HDFS Read: 9228 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 25 seconds 300 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Pets    86444
Animals 86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 177.463 seconds, Fetched: 25 row(s)

5.2 reduce数的配置

reduce数太多或者太少都不好,太多的话会导致后续的job的map数过多,太少会导致浪费资源

reduce数由这三个参数来决定

  • set hive.exec.reducers.bytes.per.reducer=1024000000;当数据量超过该值时才会进行切分reduce,要不只有一个
  • set hive.exec.reducer.max=999;默认为999
  • set mapreduce.job.reduces = 3;直接设置,无视上述参数设定

设置reduces

hive> set mapreduce.job.reduces;
mapreduce.job.reduces=-1

5.2.1 未设置reduce数时,也就是reduce是根据节点数据量大小进行自动切分

结果如下:

hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710095252_9c97d548-c734-4df3-8d78-541c7e4ec66f
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 3
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0082, Tracking URL = http://master:8088/proxy/application_1499153664137_0082/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0082
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 09:53:26,240 Stage-1 map = 0%,  reduce = 0%
2017-07-10 09:53:54,547 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 2.05 sec
...
2017-07-10 09:54:22,238 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 21.08 sec
MapReduce Total cumulative CPU time: 21 seconds 80 msec
Ended Job = job_1499153664137_0082
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0083, Tracking URL = http://master:8088/proxy/application_1499153664137_0083/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0083
Hadoop job information for Stage-2: number of mappers: 1; number of reducers: 1
2017-07-10 09:54:59,413 Stage-2 map = 0%,  reduce = 0%
2017-07-10 09:55:27,659 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 1.1 sec
2017-07-10 09:55:56,728 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 2.43 sec
MapReduce Total cumulative CPU time: 2 seconds 430 msec
Ended Job = job_1499153664137_0083
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 21.08 sec   HDFS Read: 47331544 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 1  Reduce: 1   Cumulative CPU: 2.43 sec   HDFS Read: 5826 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 23 seconds 510 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Animals 86444
Pets    86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 196.576 seconds, Fetched: 25 row(s)
hive> set hive.merge.mapfiles=true;
hive> set hive.merge.mapredfiles=true;
hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;
Query ID = hadoop_20170710095757_8fc6f5e5-6b3a-4618-b8b0-57ec16ea0225
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Estimated from input data size: 3
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0084, Tracking URL = http://master:8088/proxy/application_1499153664137_0084/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0084
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 09:58:31,781 Stage-1 map = 0%,  reduce = 0%
...
2017-07-10 09:59:24,606 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 20.81 sec
MapReduce Total cumulative CPU time: 20 seconds 810 msec
Ended Job = job_1499153664137_0084
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0085, Tracking URL = http://master:8088/proxy/application_1499153664137_0085/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0085
Hadoop job information for Stage-2: number of mappers: 1; number of reducers: 1
2017-07-10 10:00:00,596 Stage-2 map = 0%,  reduce = 0%
2017-07-10 10:00:29,530 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 1.08 sec
2017-07-10 10:00:58,684 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 2.67 sec
MapReduce Total cumulative CPU time: 2 seconds 670 msec
Ended Job = job_1499153664137_0085
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 20.81 sec   HDFS Read: 47331712 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 1  Reduce: 1   Cumulative CPU: 2.67 sec   HDFS Read: 5826 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 23 seconds 480 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Animals 86444
Pets    86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 187.501 seconds, Fetched: 25 row(s)

5.2.2 将reduce数设置为3

结果如下:

hive> set mapreduce.job.reduces=3;
hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710100303_395d684c-fdd7-4ac6-a1f1-d7aa9269dd2a
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Defaulting to jobconf value of: 3
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0086, Tracking URL = http://master:8088/proxy/application_1499153664137_0086/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0086
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 3
2017-07-10 10:04:03,317 Stage-1 map = 0%,  reduce = 0%
2017-07-10 10:04:32,522 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 2.0 sec
2017-07-10 10:04:36,916 Stage-1 map = 34%,  reduce = 0%, Cumulative CPU 6.11 sec
2017-07-10 10:04:41,375 Stage-1 map = 50%,  reduce = 0%, Cumulative CPU 7.62 sec
2017-07-10 10:04:48,683 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 16.05 sec
2017-07-10 10:04:50,754 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 17.43 sec
2017-07-10 10:04:53,926 Stage-1 map = 100%,  reduce = 67%, Cumulative CPU 19.1 sec
2017-07-10 10:04:57,047 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 20.75 sec
MapReduce Total cumulative CPU time: 20 seconds 750 msec
Ended Job = job_1499153664137_0086
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0087, Tracking URL = http://master:8088/proxy/application_1499153664137_0087/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0087
Hadoop job information for Stage-2: number of mappers: 2; number of reducers: 1
2017-07-10 10:05:32,722 Stage-2 map = 0%,  reduce = 0%
2017-07-10 10:05:59,576 Stage-2 map = 50%,  reduce = 0%, Cumulative CPU 0.93 sec
2017-07-10 10:06:00,604 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 2.03 sec
2017-07-10 10:06:19,183 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 3.61 sec
MapReduce Total cumulative CPU time: 3 seconds 610 msec
Ended Job = job_1499153664137_0087
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 3   Cumulative CPU: 20.75 sec   HDFS Read: 47331712 HDFS Write: 996 SUCCESS
Stage-Stage-2: Map: 2  Reduce: 1   Cumulative CPU: 3.61 sec   HDFS Read: 7527 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 24 seconds 360 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
People  447581
Blogs   447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Pets    86444
Animals 86444
Travel  82068
Events  82068
Education   54133
Technology  50925
Science 50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 177.949 seconds, Fetched: 25 row(s)

5.2.3 将reduce设为5的结果

hive> set mapreduce.job.reduces=5;
hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710100909_13293e63-288b-43d2-aa71-b50fd3992a9d
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Defaulting to jobconf value of: 5
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0088, Tracking URL = http://master:8088/proxy/application_1499153664137_0088/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0088
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 5
2017-07-10 10:10:23,022 Stage-1 map = 0%,  reduce = 0%
2017-07-10 10:10:54,170 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 3.81 sec
2017-07-10 10:10:56,290 Stage-1 map = 50%,  reduce = 0%, Cumulative CPU 9.27 sec
2017-07-10 10:10:58,461 Stage-1 map = 75%,  reduce = 0%, Cumulative CPU 11.33 sec
2017-07-10 10:11:05,880 Stage-1 map = 88%,  reduce = 0%, Cumulative CPU 14.86 sec
2017-07-10 10:11:07,987 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 15.2 sec
2017-07-10 10:11:14,235 Stage-1 map = 100%,  reduce = 20%, Cumulative CPU 16.87 sec
2017-07-10 10:11:18,373 Stage-1 map = 100%,  reduce = 40%, Cumulative CPU 18.56 sec
2017-07-10 10:11:19,412 Stage-1 map = 100%,  reduce = 60%, Cumulative CPU 20.24 sec
2017-07-10 10:11:21,495 Stage-1 map = 100%,  reduce = 80%, Cumulative CPU 21.66 sec
2017-07-10 10:11:22,525 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 23.08 sec
MapReduce Total cumulative CPU time: 23 seconds 80 msec
Ended Job = job_1499153664137_0088
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0089, Tracking URL = http://master:8088/proxy/application_1499153664137_0089/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0089
Hadoop job information for Stage-2: number of mappers: 2; number of reducers: 1
2017-07-10 10:12:00,374 Stage-2 map = 0%,  reduce = 0%
2017-07-10 10:12:27,396 Stage-2 map = 50%,  reduce = 0%, Cumulative CPU 0.91 sec
2017-07-10 10:12:28,425 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 2.03 sec
2017-07-10 10:12:46,169 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 3.61 sec
MapReduce Total cumulative CPU time: 3 seconds 610 msec
Ended Job = job_1499153664137_0089
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 5   Cumulative CPU: 23.08 sec   HDFS Read: 47339903 HDFS Write: 1188 SUCCESS
Stage-Stage-2: Map: 2  Reduce: 1   Cumulative CPU: 3.61 sec   HDFS Read: 8218 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 26 seconds 690 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
Blogs   447581
People  447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Style   124885
Howto   124885
Animals 86444
Pets    86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Nonprofits  16925
Activism    16925
Gaming  10182
Time taken: 184.918 seconds, Fetched: 25 row(s)

5.2.4 将reduce数设置为2

hive> set mapreduce.job.reduces=2;
hive> select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

Query ID = hadoop_20170710101515_d3f7eb83-0edf-463b-90d6-306b6eed2079
Total jobs = 2
Launching Job 1 out of 2
Number of reduce tasks not specified. Defaulting to jobconf value of: 2
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0090, Tracking URL = http://master:8088/proxy/application_1499153664137_0090/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0090
Hadoop job information for Stage-1: number of mappers: 4; number of reducers: 2
2017-07-10 10:16:03,018 Stage-1 map = 0%,  reduce = 0%
2017-07-10 10:16:31,593 Stage-1 map = 25%,  reduce = 0%, Cumulative CPU 1.99 sec
2017-07-10 10:16:34,769 Stage-1 map = 50%,  reduce = 0%, Cumulative CPU 5.8 sec
2017-07-10 10:16:51,865 Stage-1 map = 63%,  reduce = 0%, Cumulative CPU 14.03 sec
2017-07-10 10:16:52,986 Stage-1 map = 75%,  reduce = 0%, Cumulative CPU 14.28 sec
2017-07-10 10:16:54,084 Stage-1 map = 84%,  reduce = 0%, Cumulative CPU 15.01 sec
2017-07-10 10:16:55,138 Stage-1 map = 84%,  reduce = 13%, Cumulative CPU 15.39 sec
2017-07-10 10:16:56,210 Stage-1 map = 100%,  reduce = 13%, Cumulative CPU 16.39 sec
2017-07-10 10:16:57,298 Stage-1 map = 100%,  reduce = 50%, Cumulative CPU 17.73 sec
2017-07-10 10:16:58,330 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 19.42 sec
MapReduce Total cumulative CPU time: 19 seconds 420 msec
Ended Job = job_1499153664137_0090
Launching Job 2 out of 2
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapreduce.job.reduces=<number>
Starting Job = job_1499153664137_0091, Tracking URL = http://master:8088/proxy/application_1499153664137_0091/
Kill Command = /home/hadoop/app/hadoop/bin/hadoop job  -kill job_1499153664137_0091
Hadoop job information for Stage-2: number of mappers: 1; number of reducers: 1
2017-07-10 10:17:35,020 Stage-2 map = 0%,  reduce = 0%
2017-07-10 10:18:01,896 Stage-2 map = 100%,  reduce = 0%, Cumulative CPU 0.9 sec
2017-07-10 10:18:31,798 Stage-2 map = 100%,  reduce = 100%, Cumulative CPU 2.45 sec
MapReduce Total cumulative CPU time: 2 seconds 450 msec
Ended Job = job_1499153664137_0091
MapReduce Jobs Launched: 
Stage-Stage-1: Map: 4  Reduce: 2   Cumulative CPU: 19.42 sec   HDFS Read: 47327614 HDFS Write: 900 SUCCESS
Stage-Stage-2: Map: 1  Reduce: 1   Cumulative CPU: 2.45 sec   HDFS Read: 5475 HDFS Write: 356 SUCCESS
Total MapReduce CPU Time Spent: 21 seconds 870 msec
OK
Entertainment   1304724
Music   1274825
Comedy  449652
Blogs   447581
People  447581
Film    442109
Animation   442109
Sports  390619
Politics    186753
News    186753
Autos   169883
Vehicles    169883
Howto   124885
Style   124885
Animals 86444
Pets    86444
Travel  82068
Events  82068
Education   54133
Science 50925
Technology  50925
UNA 42928
Activism    16925
Nonprofits  16925
Gaming  10182
Time taken: 190.953 seconds, Fetched: 25 row(s)

使用语句:select tagId, count(a.videoid) as sum from (select videoid,tagId from youtube3 lateral view explode(category) catetory as tagId) a group by a.tagId order by sum desc;

reduce数的选择只能通过不断地调整来达到一个最优的方案

reduce数CPU耗时详细情况
不设置时,自动切分为324sStage-Stage-1: Map: 4 Reduce: 3 Cumulative CPU: 21.08 sec HDFS Read: 47331544 HDFS Write: 996 SUCCESS Stage-Stage-2: Map: 1 Reduce: 1 Cumulative CPU: 2.43 sec HDFS Read: 5826 HDFS Write: 356 SUCCESS Total MapReduce CPU Time Spent: 23 seconds 510 msec
设置为526sMapReduce Jobs Launched: Stage-Stage-1: Map: 4 Reduce: 5 Cumulative CPU: 23.08 sec HDFS Read: 47339903 HDFS Write: 1188 SUCCESS Stage-Stage-2: Map: 2 Reduce: 1 Cumulative CPU: 3.61 sec HDFS Read: 8218 HDFS Write: 356 SUCCESS Total MapReduce CPU Time Spent: 26 seconds 690 msec
设置为222sMapReduce Jobs Launched: Stage-Stage-1: Map: 4 Reduce: 2 Cumulative CPU: 19.42 sec HDFS Read: 47327614 HDFS Write: 900 SUCCESS Stage-Stage-2: Map: 1 Reduce: 1 Cumulative CPU: 2.45 sec HDFS Read: 5475 HDFS Write: 356 SUCCESS Total MapReduce CPU Time Spent: 21 seconds 870 msec


所以在此条查询中reduce最佳是2

  • 3
    点赞
  • 15
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值