[ https://issues.apache.org/jira/browse/SPARK-2142?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sandy Ryza resolved SPARK-2142. ------------------------------- Resolution: Not a Problem I ran some tests that indicated that only stop-the-world GC time gets included in the GC Time metric. > Give better indicator of how GC cuts into task time > --------------------------------------------------- > > Key: SPARK-2142 > URL: https://issues.apache.org/jira/browse/SPARK-2142 > Project: Spark > Issue Type: Improvement > Components: Spark Core > Reporter: Sandy Ryza > > The currently reported GC time is a somewhat misleading metric. > If I see a task that has taken 10 minutes with a GC time of 9 minutes, the > straightforward interpretation is that GC is making my job 90% slower. This > is true if all that time is spent in a full GC. But if a GC thread is only > running in the background, it might not be impacting my task at all. > What a user really wants to know is whether GC is the cause of their task > being slow. Counting the number of full GCs or measuring the amount of time > spent in stop-the-world GCs could be a more useful metric. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org