You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kylin.apache.org by Jie Tao <ji...@gameforge.com> on 2016/06/20 12:57:39 UTC

flush map output at step Build Base Cuboid Data

I have been working with a problem for several days but no success. 
Building base Cuboid hangs in flushing the map output (ca 200MB) from 
the memory buffer to the disk:

INFO [main] org.apache.hadoop.mapred.MapTask: Starting flush of map output
INFO [main] org.apache.hadoop.mapred.MapTask: Spilling map output
INFO [main] org.apache.hadoop.mapred.MapTask: bufstart = 0; bufend = 
190884986; bufvoid = 536870912
INFO [main] org.apache.hadoop.mapred.MapTask: kvstart = 
134217724(536870896); kvend = 116938212(467752848); length = 
17279513/33554432

I saw: Spilling map output was started and one spill.out was created in 
the disk (only 80K), but this process was stuck. Based on the MapTask 
source code, I shall see a Log info "Finished spill", which I did not 
see. Has anybody experience with this issue? Maybe the following 
properties are wrong?

<property>
     <name>mapreduce.task.io.sort.mb</name>
     <value>512</value>
   </property>
   <property>
     <name>mapreduce.task.io.sort.factor</name>
     <value>100</value>
   </property>

(Kylin also sets mapreduce.task.io.sort.mb but was not taken. Someone 
wrote that mapreduce.task.io.sort.factor shall also be set correspondingly)

Another question: why Kylin only uses two map tasks in the base Cuboid 
building? Can we specify more map tasks?

Cheers,

Jie

Re: flush map output at step Build Base Cuboid Data

Posted by Li Yang <li...@apache.org>.
Don't have similar experience. If you have access to Hadoop node, maybe
jstack debug the hanging process?

On Mon, Jun 20, 2016 at 9:58 PM, Jie Tao <ji...@gameforge.com> wrote:

> another guess: the two mappers need communication? I saw that both mappers
> have a progress of 0.667 and then not further. When I build the same cube
> with smaller data size I only see one map task and the cube was built
> successfully.
>
> Cheers,
>
> Jie
>
>
> Am 20.06.2016 um 14:57 schrieb Jie Tao:
>
>> I have been working with a problem for several days but no success.
>> Building base Cuboid hangs in flushing the map output (ca 200MB) from the
>> memory buffer to the disk:
>>
>> INFO [main] org.apache.hadoop.mapred.MapTask: Starting flush of map output
>> INFO [main] org.apache.hadoop.mapred.MapTask: Spilling map output
>> INFO [main] org.apache.hadoop.mapred.MapTask: bufstart = 0; bufend =
>> 190884986; bufvoid = 536870912
>> INFO [main] org.apache.hadoop.mapred.MapTask: kvstart =
>> 134217724(536870896); kvend = 116938212(467752848); length =
>> 17279513/33554432
>>
>> I saw: Spilling map output was started and one spill.out was created in
>> the disk (only 80K), but this process was stuck. Based on the MapTask
>> source code, I shall see a Log info "Finished spill", which I did not see.
>> Has anybody experience with this issue? Maybe the following properties are
>> wrong?
>>
>> <property>
>>     <name>mapreduce.task.io.sort.mb</name>
>>     <value>512</value>
>>   </property>
>>   <property>
>>     <name>mapreduce.task.io.sort.factor</name>
>>     <value>100</value>
>>   </property>
>>
>> (Kylin also sets mapreduce.task.io.sort.mb but was not taken. Someone
>> wrote that mapreduce.task.io.sort.factor shall also be set correspondingly)
>>
>> Another question: why Kylin only uses two map tasks in the base Cuboid
>> building? Can we specify more map tasks?
>>
>> Cheers,
>>
>> Jie
>>
>
>

Re: flush map output at step Build Base Cuboid Data

Posted by Jie Tao <ji...@gameforge.com>.
another guess: the two mappers need communication? I saw that both 
mappers have a progress of 0.667 and then not further. When I build the 
same cube with smaller data size I only see one map task and the cube 
was built successfully.

Cheers,

Jie

Am 20.06.2016 um 14:57 schrieb Jie Tao:
> I have been working with a problem for several days but no success. 
> Building base Cuboid hangs in flushing the map output (ca 200MB) from 
> the memory buffer to the disk:
>
> INFO [main] org.apache.hadoop.mapred.MapTask: Starting flush of map 
> output
> INFO [main] org.apache.hadoop.mapred.MapTask: Spilling map output
> INFO [main] org.apache.hadoop.mapred.MapTask: bufstart = 0; bufend = 
> 190884986; bufvoid = 536870912
> INFO [main] org.apache.hadoop.mapred.MapTask: kvstart = 
> 134217724(536870896); kvend = 116938212(467752848); length = 
> 17279513/33554432
>
> I saw: Spilling map output was started and one spill.out was created 
> in the disk (only 80K), but this process was stuck. Based on the 
> MapTask source code, I shall see a Log info "Finished spill", which I 
> did not see. Has anybody experience with this issue? Maybe the 
> following properties are wrong?
>
> <property>
>     <name>mapreduce.task.io.sort.mb</name>
>     <value>512</value>
>   </property>
>   <property>
>     <name>mapreduce.task.io.sort.factor</name>
>     <value>100</value>
>   </property>
>
> (Kylin also sets mapreduce.task.io.sort.mb but was not taken. Someone 
> wrote that mapreduce.task.io.sort.factor shall also be set 
> correspondingly)
>
> Another question: why Kylin only uses two map tasks in the base Cuboid 
> building? Can we specify more map tasks?
>
> Cheers,
>
> Jie