Ah I see - didn't catch that. Use local-cluster as per Aaron's suggestion!

On Sat, Nov 16, 2013 at 10:57 AM, Alex Boisvert <[email protected]> wrote:
> Thanks everybody.  I think I'll give Aaron's local-cluster suggestion a
> shot.
>
>
> On Sat, Nov 16, 2013 at 9:50 AM, Aaron Davidson <[email protected]> wrote:
>>
>> I was under the impression that he was using the same JVM for Spark and
>> other stuff, and wanted to limit how much of it Spark could use. Patrick's
>> solution is of course the right way to go if that's not the case.
>>
>>
>> On Sat, Nov 16, 2013 at 9:40 AM, Patrick Wendell <[email protected]>
>> wrote:
>>>
>>> If you are using local mode, you can just pass -Xmx32g to the JVM that
>>> is launching spark and it will have that much memory.
>>>
>>> On Fri, Nov 15, 2013 at 6:30 PM, Aaron Davidson <[email protected]>
>>> wrote:
>>> > One possible workaround would be to use the local-cluster Spark mode.
>>> > This
>>> > is normally used only for testing, but it will actually spawn a
>>> > separate
>>> > process for the executor. The format is:
>>> > new SparkContext("local-cluster[1,4,32000]")
>>> > This will spawn 1 Executor that is allocated 4 cores and 32GB
>>> > (approximated
>>> > as 32k MB). Since this is a separate process with its own JVM, you'd
>>> > probably want to just change your original JVM's memory to 32 GB.
>>> >
>>> > Note that since local-cluster mode more closely simulates a cluster,
>>> > it's
>>> > possible that certain issues like dependency problems may arise that
>>> > don't
>>> > appear when using local mode.
>>> >
>>> >
>>> > On Fri, Nov 15, 2013 at 11:43 AM, Alex Boisvert
>>> > <[email protected]>
>>> > wrote:
>>> >>
>>> >> When starting a local-mode Spark instance, e.g., new
>>> >> SparkContext("local[4]"), what memory configuration options are
>>> >> available/considered to limit Spark's memory usage?
>>> >>
>>> >> For instance, if I have a JVM with 64GB and would like to
>>> >> reserve/limit
>>> >> Spark to using only 32GB of the heap.
>>> >>
>>> >> thanks!
>>> >>
>>> >
>>
>>
>

Reply via email to