hive-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Yong Zhang <>
Subject Hive 0.12 MAPJOIN hangs sometimes
Date Fri, 11 Mar 2016 14:43:21 GMT
Hi, Hive users:
Currently our Hadoop vendor comes with Hive 0.12. I know it is a kind of old version, but
upgrade still has some long path to go.
Right now, we are facing an issue in the Hive 0.12.
We have one ETL kind of steps implemented in Hive, and due to the data volume in this step,
we know that MAPJOIN is the right way to go, as one side of data is very small, but the other
size is much larger.
So below is the query example:
set hive.exec.compress.output=true;set parquet.compression=snappy;set mapred.reduce.tasks=1;set -Xmx4096m;set mapred.task.timeout=7200000;set;set
insert overwrite table a(dt='${hiveconf:run_date}', source='ip')select  /*+ MAPJOIN(trial_event)
The above query can be finished daily around 10 minutes, which we are very happy about it.
But sometimes, the query will be hang hours in the ETL, until we manually kill it.
I add the debug info in the Hive, and found the following message:
2016-03-11 09:11:52 Starting to launch local task to process map join;  maximum memory = 536870912SLF4J:
Failed to load class "org.slf4j.impl.StaticLoggerBinder".SLF4J: Defaulting to no-operation
(NOP) logger implementationSLF4J: See for
further details.16/03/11 09:11:55 DEBUG ipc.Client: IPC Client (-1284813870) connection to
namenode/ from etl: closed16/03/11 09:11:55 DEBUG ipc.Client: IPC Client
(-1284813870) connection to namenode/ from etl: stopped, remaining connections
Then there is no more log after that for hours.
If we don't use MAPJOIN, we won't face this issue, but the query will take 2.5 hours.
When this happens, I can see the NameNode works fine, I can run all kinds of "HDFS" operation
without any issue, while this query is hanging. What does this "IPC Client remaining connections
0" mean? If we cannot upgrade our Hive version as now, any workaround do we have?
View raw message