[
https://issues.apache.org/jira/browse/HIVE-19937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16528117#comment-16528117
]
Sahil Takiar commented on HIVE-19937:
-------------------------------------
[[email protected]] thanks for the input! I didn't notice
{{CopyOnFirstWriteProperties}} so that helps a lot.
For {{CopyOnFirstWriteProperties}} it looks like all the properties will get
copied into the "super" {{Properties}} object when there is a write, do you
think it would be possible to just copy the mutated properties to the super
class, rather than all of them? My concern is that its probably a common case
where this properties object gets mutated, in which case copying the entire
thing would probably defeat the purpose of interning.
I also noticed that in {{PartitionDesc#internProperties}} only the value is
being interned, but not the key, was that intentional? Code is below:
{code:java}
private static void internProperties(Properties properties) {
for (Enumeration<?> keys = properties.propertyNames();
keys.hasMoreElements();) {
String key = (String) keys.nextElement();
String oldValue = properties.getProperty(key);
if (oldValue != null) {
properties.setProperty(key, oldValue.intern());
}
}
}
{code}
I'm working on creating a test that can easily measure the impact of this
change.
> Intern JobConf objects in Spark tasks
> -------------------------------------
>
> Key: HIVE-19937
> URL: https://issues.apache.org/jira/browse/HIVE-19937
> Project: Hive
> Issue Type: Improvement
> Components: Spark
> Reporter: Sahil Takiar
> Assignee: Sahil Takiar
> Priority: Major
> Attachments: HIVE-19937.1.patch
>
>
> When fixing HIVE-16395, we decided that each new Spark task should clone the
> {{JobConf}} object to prevent any {{ConcurrentModificationException}} from
> being thrown. However, setting this variable comes at a cost of storing a
> duplicate {{JobConf}} object for each Spark task. These objects can take up a
> significant amount of memory, we should intern them so that Spark tasks
> running in the same JVM don't store duplicate copies.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
