spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From shahid ashraf <sha...@trialx.com>
Subject Re: issue with spark.driver.maxResultSize parameter in spark 1.3
Date Mon, 02 Nov 2015 05:46:45 GMT
Is your process getting killed...
if yes then try to see using dmesg.

On Mon, Nov 2, 2015 at 8:17 AM, karthik kadiyam <
karthik.kadiyam.32@gmail.com> wrote:

> Did any one had issue setting spark.driver.maxResultSize value ?
>
> On Friday, October 30, 2015, karthik kadiyam <karthik.kadiyam.32@gmail.com>
> wrote:
>
>> Hi Shahid,
>>
>> I played around with spark driver memory too. In the conf file it was set
>> to " --driver-memory 20G " first. When i changed the spark driver
>> maxResultSize from default to 2g ,i changed the driver memory to 30G and
>> tired too. It gave we same error says "bigger than  (1024.0 MB) " .
>> spark.driver.maxResultSize
>> One other thing i observed is , in one of the tasks the data its trying
>> to process is more than 100 MB and that exceutor and task keeps losing
>> connection and doing retry. I tried increase the Tasks by repartition from
>> 120 to 240 to 480 also. Still i can see in one of my tasks it still is
>> trying to process more than 100 mb. Other task hardly process 1 mb to 10 mb
>> , some around 20 mbs, some have 0 mbs .
>>
>> Any idea how can i try to even the data distribution acrosss multiple
>> node.
>>
>> On Fri, Oct 30, 2015 at 12:09 AM, shahid ashraf <shahid@trialx.com>
>> wrote:
>>
>>> Hi
>>> I guess you need to increase spark driver memory as well. But that
>>> should be set in conf files
>>> Let me know if that resolves
>>> On Oct 30, 2015 7:33 AM, "karthik kadiyam" <karthik.kadiyam.32@gmail.com>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> In spark streaming job i had the following setting
>>>>
>>>>             this.jsc.getConf().set("spark.driver.maxResultSize", “0”);
>>>> and i got the error in the job as below
>>>>
>>>> User class threw exception: Job aborted due to stage failure: Total
>>>> size of serialized results of 120 tasks (1082.2 MB) is bigger than
>>>> spark.driver.maxResultSize (1024.0 MB)
>>>>
>>>> Basically i realized that as default value is 1 GB. I changed
>>>> the configuration as below.
>>>>
>>>> this.jsc.getConf().set("spark.driver.maxResultSize", “2g”);
>>>>
>>>> and when i ran the job it gave the error
>>>>
>>>> User class threw exception: Job aborted due to stage failure: Total
>>>> size of serialized results of 120 tasks (1082.2 MB) is bigger than
>>>> spark.driver.maxResultSize (1024.0 MB)
>>>>
>>>> So, basically the change i made is not been considered in the job. so
>>>> my question is
>>>>
>>>> - "spark.driver.maxResultSize", “2g” is this the right way to change
>>>> or any other way to do it.
>>>> - Is this a bug in spark 1.3 or something or any one had this issue
>>>> before?
>>>>
>>>>
>>


-- 
with Regards
Shahid Ashraf

Mime
View raw message