Can you check the parameters: mapred.min.split.size and dfs.block.size ? -----Original Message----- From: Tim Sell [mailto:[email protected]] Sent: Tuesday, February 23, 2010 11:26 AM To: [email protected] Subject: Re: Hive jobs only run with 1 map task
It happens on a table that is a single 30 gig tab separated file. It also happens on tables that are split over a hundreds files. On 23 February 2010 19:20, Namit Jain <[email protected]> wrote: > What is the size of the input data for the query ? > > Since you are using CombineHiveInputFormat, multiple files can be read by a > single mapper. > > > > -namit > > -----Original Message----- > From: Tim Sell [mailto:[email protected]] > Sent: Tuesday, February 23, 2010 11:14 AM > To: [email protected] > Subject: Re: Hive jobs only run with 1 map task > > If it helps looking at the job conf in the map reduce logs I noticed > mapred.input.format.class=org.apache.hadoop.hive.ql.io.CombineHiveInputFormat > > > On 23 February 2010 19:11, Tim Sell <[email protected]> wrote: >> Is hive.input.format set on the table? I'm not sure how to pull that >> out again. I know they are stored as text though. >> I should mention they do actually parse/process correctly. >> >> Here are all the set parameters >> >> hive> set; >> silent=off >> javax.jdo.option.ConnectionUserName=hive >> hive.exec.reducers.bytes.per.reducer=100000000 >> hive.mapred.local.mem=0 >> datanucleus.autoStartMechanismMode=checked >> hive.metastore.connect.retries=5 >> datanucleus.validateColumns=false >> hive.metastore.rawstore.impl=org.apache.hadoop.hive.metastore.ObjectStore >> datanucleus.autoCreateSchema=true >> javax.jdo.option.ConnectionPassword=hive >> datanucleus.validateConstraints=false >> datancucleus.transactionIsolation=read-committed >> datanucleus.validateTables=false >> hive.map.aggr.hash.min.reduction=0.5 >> datanucleus.storeManagerType=rdbms >> hive.exec.script.maxerrsize=100000 >> hive.merge.size.per.task=256000000 >> hive.test.mode.prefix=test_ >> hive.groupby.skewindata=false >> hive.default.fileformat=TextFile >> hive.script.auto.progress=false >> hive.groupby.mapaggr.checkinterval=100000 >> hive.hwi.listen.port=9999 >> datanuclues.cache.level2=true >> hive.hwi.war.file=${HIVE_HOME}/lib/hive-hwi.war >> hive.merge.mapfiles=true >> hive.exec.compress.output=false >> datanuclues.cache.level2.type=SOFT >> javax.jdo.option.ConnectionDriverName=com.mysql.jdbc.Driver >> hive.map.aggr=true >> hive.join.emit.interval=1000 >> hive.metastore.warehouse.dir=hdfs://master1.hadoop.last.fm:8020/user/hive/warehouse >> javax.jdo.PersistenceManagerFactoryClass=org.datanucleus.jdo.JDOPersistenceManagerFactory >> hive.mapred.mode=nonstrict >> hive.exec.scratchdir=/tmp/hive-${user.name} >> javax.jdo.option.NonTransactionalRead=true >> hive.metastore.local=true >> hive.test.mode.samplefreq=32 >> hive.test.mode=false >> javax.jdo.option.ConnectionURL=jdbc:mysql://10.101.1.35/hive?createDatabaseIfNotExist=true >> javax.jdo.option.DetachAllOnCommit=true >> hive.heartbeat.interval=1000 >> hive.map.aggr.hash.percentmemory=0.5 >> hive.exec.reducers.max=107 >> hive.hwi.listen.host=0.0.0.0 >> hive.exec.compress.intermediate=false >> hive.optimize.cp=true >> hive.optimize.ppd=true >> hive.session.id=tims_201002231907 >> hive.merge.mapredfiles=false >> >> ~Tim. >> >> On 23 February 2010 19:03, Namit Jain <[email protected]> wrote: >>> Can you check your input format ? >>> >>> Can you check the value of the parameter : >>> hive.input.format ? >>> >>> Can you send all the parameters ? >>> >>> >>> >>> Thanks, >>> -namit >>> >>> >>> >>> -----Original Message----- >>> From: Tim Sell [mailto:[email protected]] >>> Sent: Tuesday, February 23, 2010 11:00 AM >>> To: [email protected] >>> Subject: Hive jobs only run with 1 map task >>> >>> We just upgraded to hadoop 0.20 (from hadoop 0.18), impressively our >>> same hive package kept working against the new hadoop setup. >>> >>> Since the upgrade every hive starts with only 1 map task though. Even >>> after setting it with eg: set mapred.map.tasks=32; >>> We recompiled our hive setup against hadoop 0.20 and still get the same >>> issue. >>> >>> Any suggestions for something obvious we might have missed? >>> >>> ~Tim. >>> >> >
