Hi Abhishek
CDH comes with old "MR1" (set of jars) in addition to the newer "MR2" setup. Remember job tracker, tasktrackers? that's MR1. Resource manager, Node Manager, that's MR2. For a CDH tarball install, the MR1 jars are in share/hadoop/mapreduce1, and the MR2 jars are in share/hadoop/mapreduce2. There's even a share/hadoop/mapreduce sym link, and by default it points to mapreduce2. If I remember correctly, the file input formats (both mapred and mapreduce) packed in the MR2 jars support recursive dirs, while those packed in the MR1 jars do not. So if your cluster runs Job tracke/TaskTrackers instead of yarn (Resource manager/Node managers), then that's the problem. Even if your cluster runs YARN, this can still be the problem, *if*: * your hadoop/hive clients are setup such that MR1 jars end up in the classpath, and * hive runs your queries locally (e.g. a fetch task instead of a remote MR job (see HIVE-2925)). You can see what hive's classpath ends up to be by running "set system:java.class.path;" in your hive cli to see the classpath. Grep for "mapreduce1", or "-mr1-". hth Gabriel Balan On 5/20/2016 8:45 AM, Abhishek Dubey wrote:
Thanks Al and Mich. I know about hive partitioning and bucketing. I have created a table with files in subdirectories and setting these properties: set mapred.input.dir.recursive=true; set hive.mapred.supports.subdirectories=true; makes queries work nicely without partitioning on local vm. Below are few posts on the same topic : http://stackoverflow.com/questions/26767713/can-hive-recursively-descend-into-subdirectories-without-partitions-or-editing-h http://stackoverflow.com/questions/20756561/how-to-pick-up-all-data-into-hive-from-subdirectories https://joshuafennessy.com/2015/06/30/configure-apache-hive-to-recursively-search-directories-for-files/ Solution given in the above posts is to set the two properties mentioned above which I am already setting. But these properties are not working on CDH 5.3.3 cluster, I’m yet to try on other CDH versions. *Thanks & Regards,* *Abhishek Dubey*** *From:*Mich Talebzadeh [mailto:mich.talebza...@gmail.com] *Sent:* Friday, May 20, 2016 1:22 AM *To:* user <user@hive.apache.org> *Subject:* Re: Unable to pick data from subdirectories into hive table in CDH 5.3.3 agreed but it still needs to know where the hive top node directory starts from, which is normally under ../../ warehouse Dr Mich Talebzadeh LinkedIn ///https://www.linkedin.com/profile/view?id=AAEAAAAWh2gBxianrbJd6zP6AcPCCdOABUrV8Pw/ http://talebzadehmich.wordpress.com <http://talebzadehmich.wordpress.com/> On 19 May 2016 at 20:32, Al Pivonka <alpivo...@gmail.com <mailto:alpivo...@gmail.com>> wrote: Read about hive partitions and bucketing. Since your location has multiple directories, Hive needs to know how to traverse them.. Hope this helps. On May 19, 2016 5:51 AM, "Abhishek Dubey" <abhishek.du...@xoriant.com <mailto:abhishek.du...@xoriant.com>> wrote: Hi, In hdfs I have a directory structure like this. /user/hdfs/Data/Data1/File1 /user/hdfs/Data/Data2/File2 And I am creating an external table like: CREATE external TABLE db.tablename ( amt1 STRING, amt2 STRING, amt3 STRING ) row format delimited fields terminated by ',' location '/user/hdfs/Data/'; Also, I have set two properties: set mapred.input.dir.recursive=true; set hive.mapred.supports.subdirectories=true; This setup is working perfectly fine on my local single node vm, Having all vanilla apache installations and setup, But. on cloudera 5.3.3 cluster of 4 nodes, above mentioned properties for recursive lookup of sub directories for an external hive table is not working. In the Cloudera manager i have added the properties in Hive-Site.xml, deployed configuration and restarted Hive service but still not working. <property> <name> mapred.input.dir.recursive</name> <value>true</value> </property> <property> <name>hive.mapred.supports.subdirectories</name> <value>true</value> </property> When querying select * on CDH What i get is this, Zero rows. hive> select * from tablename; OK Time taken: 0.322 seconds hive> Whereas on local vm it is giving desired output. Is there anything else on CDH that we need to take care to pick data from subdirectories into hive table? Thanks in advance. *Abhishek Dubey*
-- The statements and opinions expressed here are my own and do not necessarily represent those of Oracle Corporation.