[
https://issues.apache.org/jira/browse/TAJO-991?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14100329#comment-14100329
]
ASF GitHub Bot commented on TAJO-991:
-------------------------------------
Github user jinossy commented on a diff in the pull request:
https://github.com/apache/tajo/pull/107#discussion_r16338681
--- Diff: tajo-dist/src/main/conf/tajo-env.sh ---
@@ -68,4 +74,7 @@ export TAJO_WORKER_STANDBY_MODE=true
# It must be required to use HCatalogStore
# export HIVE_HOME=
-# export HIVE_JDBC_DRIVER_DIR=
\ No newline at end of file
+# export HIVE_JDBC_DRIVER_DIR=
+
+# Tajo PullServer mode. the default mode is dedicated mode(dedicated or
embedded)
+ export TAJO_PULLSERVER_MODE=dedicated
--- End diff --
How about changing ‘TAJO_PULLSERVER_STANDALONE = true’ instead of
dedicated/embeded ?
> Running PullServer on a dedicated JVM process which separates from worker.
> --------------------------------------------------------------------------
>
> Key: TAJO-991
> URL: https://issues.apache.org/jira/browse/TAJO-991
> Project: Tajo
> Issue Type: Sub-task
> Components: data shuffle
> Reporter: Hyoungjun Kim
> Assignee: Hyoungjun Kim
>
> Currently PullServer runs on TajoWorker. If a TajoWorker starts Garbage
> Collector, all fetching tasks which run on other worker are paused during
> Garbage Collection. Running PullServer on a dedicated JVM process which
> separates from worker is helpful avoiding stop the all cluster by GC.
> PullServer uses small memory and doesn't share any information and status
> with TajoWorker.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
