[PIG-2872] StoreFuncInterface.setStoreLocation get's a copy of a Configuration object

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 0.11
• Component/s: impl

Environment

Pig trunk, Hadoop 0.20.205 with Kerberos, ElasticSearch trunk, Wonderdog trunk

• Assignee: Unassigned
• Reporter: Evert Lammerts
• Created: 13/Aug/12 10:08
• Updated: 10/Apr/14 13:52

Description

When an implementation of StoreFuncInterface.setStoreLocation is called from JobControlCompiler.getJob, it is passed a copy of the Configuration that will be used for the Job that will be submitted:

JobControlCompiler.java

sFunc.setStoreLocation(st.getSFile().getFileName(), new org.apache.hadoop.mapreduce.Job(nwJob.getConfiguration()));

When a new org.apache.hadoop.mapreduce.Job is created it creates a copy of the Configuration object, as far as I know. Thus anything added to the Configuration object in the implementation of setStoreLocation will not be included in the Configuration of nwJob in JobControlCompiler.getJob.

I notice this goes wrong in Wonderdog, which needs to include the Elasticsearch configuration file in the DistributedCache. It is added to mapred.cache.files through setStoreLocation, but this setting doesn't make it back into the Job returned by JobControlCompiler.getJob, and is therefore never localized.

This might be intentional semantics within Pig, but I'm not familiar enough with StoreFuncs to know whether it is.

Issue Links

• is related to: PIG-2890 Revert PIG-2578
• relates to: PIG-2870 pigServer.openIterator fails for jobs with no input splits