How to clear broadcast variable from driver memory?

[Kevin Jung <it...@samsung.com>]

Hi,
I tried Broadcast.unpersist() on Spark 1.0.1 but MemoryStore(driver memory)
still allocated it.

//LOGS
//Block broadcast_0 stored as values to memory (estimated size 380.1 MB,
free 5.7 GB)
The free size of memory was same after calling unpersist.
Can I clear this?



--
View this message in context: http://apache-spark-user-list.1001560.n3.nabble.com/How-to-clear-broadcast-variable-from-driver-memory-tp13353.html
Sent from the Apache Spark User List mailing list archive at Nabble.com.

---------------------------------------------------------------------
To unsubscribe, e-mail: user-unsubscribe@spark.apache.org
For additional commands, e-mail: user-help@spark.apache.org

Re: How to clear broadcast variable from driver memory?

[Andrew Or <an...@databricks.com>]

Hi Kevin, there is currently no way to do this... Broadcast.unpersist()
only unpersists it from the executors, but not from the driver. However,
this is not that bad because the Spark automatically cleans up broadcasts
that are no longer used, even on the driver. So as long as there is no
memory pressure for Spark to clean things up, the broadcast will stick
around.

Thanks for pointing this out. We should probably expose a mechanism for
doing this.

Andrew


2014-09-03 4:56 GMT-07:00 Kevin Jung <it...@samsung.com>:

> Hi,
> I tried Broadcast.unpersist() on Spark 1.0.1 but MemoryStore(driver memory)
> still allocated it.
>
> //LOGS
> //Block broadcast_0 stored as values to memory (estimated size 380.1 MB,
> free 5.7 GB)
> The free size of memory was same after calling unpersist.
> Can I clear this?
>
>
>
> --
> View this message in context:
> http://apache-spark-user-list.1001560.n3.nabble.com/How-to-clear-broadcast-variable-from-driver-memory-tp13353.html
> Sent from the Apache Spark User List mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: user-unsubscribe@spark.apache.org
> For additional commands, e-mail: user-help@spark.apache.org
>
>