[ https://issues.apache.org/jira/browse/PIG-3866?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14190592#comment-14190592 ]
Daniel Dai commented on PIG-3866: --------------------------------- [~r4_ruchir], if we turn classloader into thread local, why we still keep static PigContext.classloader? Will that cause race condition? FileLocalizer is a totally different change, do you want this part or this is accidentally included? > Create ThreadLocal classloader per PigContext > --------------------------------------------- > > Key: PIG-3866 > URL: https://issues.apache.org/jira/browse/PIG-3866 > Project: Pig > Issue Type: Wish > Components: impl > Affects Versions: 0.12.0 > Reporter: Ruchir > Fix For: 0.14.0 > > Attachments: diff > > > For shared Hadoop environment, where there are large number of customers who > submit PIG jobs for execution, it becomes very cumbersome to handle all > support issue (ex providing access to CLI machine to every user, limiting > user for submission etc). > There is need to provide REST based interface for submitting PIG job/view PIG > job status/view PIG request output, so that there is single entry point to > cluster. With this, users are abstracted of Hadoop configuration and need to > get access on machine. Also given it is hosted service, it is possible to > provide rate limiting, QoS etc to user. > We have designed such PIG self-service which provide REST APIs for all > operations and this webapp runs under web container. In order to run multiple > requests in same JVM, we need to create classloader per request thread. -- This message was sent by Atlassian JIRA (v6.3.4#6332)