[
https://issues.apache.org/jira/browse/HBASE-6111?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Robertson updated HBASE-6111:
---------------------------------
Description:
I have started seeing this issue in our environment. HBASE-1672 was closed as
non reproducible, so I cloned it here.
I have a 367M record table, compressed with snappy, and running a vanilla MR
SCAN with no filters spawns 441 Mappers. The cluster has 216 slots for
mappers, and the first wave all report 100% data-local mappers. As the second
wave of mappers come up they don't get run locally to the RS and data locality
drops.
This kills our environment, as it saturates the network at 120M which is very
clear on ganglia.
I am really happy to help diagnose this, but need some guidance on what to do.
I don't know enough yet about how task assignment works in MR to determine why
the machines are picking up random tasks for their second effort and not one
for the local RS.
was:
The number of data local map tasks while scanning a table is only about 10% of
the total map tasks...
My table had 280 regions and 13M records... The number of map tasks in the scan
job were equal to the number of regions (280). Only 25 of them were data local
tasks.
Environment: DN, TT and RS running on the same nodes, all using CDH3.
Ganglia monitoring everything. (was: DN, TT and RS running on the same nodes.)
Affects Version/s: (was: 0.19.3)
(was: 0.20.0)
0.20.2
0.92.0
Fix Version/s: (was: 0.20.0)
> CLONE - Map tasks not local to RS
> ---------------------------------
>
> Key: HBASE-6111
> URL: https://issues.apache.org/jira/browse/HBASE-6111
> Project: HBase
> Issue Type: Bug
> Components: mapred, master, regionserver
> Affects Versions: 0.20.2, 0.92.0
> Environment: DN, TT and RS running on the same nodes, all using CDH3.
> Ganglia monitoring everything.
> Reporter: Tim Robertson
>
> I have started seeing this issue in our environment. HBASE-1672 was closed
> as non reproducible, so I cloned it here.
> I have a 367M record table, compressed with snappy, and running a vanilla MR
> SCAN with no filters spawns 441 Mappers. The cluster has 216 slots for
> mappers, and the first wave all report 100% data-local mappers. As the
> second wave of mappers come up they don't get run locally to the RS and data
> locality drops.
> This kills our environment, as it saturates the network at 120M which is very
> clear on ganglia.
> I am really happy to help diagnose this, but need some guidance on what to
> do. I don't know enough yet about how task assignment works in MR to
> determine why the machines are picking up random tasks for their second
> effort and not one for the local RS.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
