>> > We're
> >> > have a Spark streaming job, and we're reading a reasonable amount of
> >> > data
> >> > from Kafka (40 GB / minute or so). We would like to move to using the
> >> > Kafka
> >> > 0.10 client to avoid requiring our (0
gt; > Kafka
>> > 0.10 client to avoid requiring our (0.10.2.1) Kafka brokers from having
>> > to
>> > modify formats.
>> >
>> > We've run into https://issues.apache.org/jira/browse/SPARK-19185,
>> > 'ConcurrentModificationExceptions with CachedKafkaConsumers'. I've
&
to avoid requiring our (0.10.2.1) Kafka brokers from having
> to
> > modify formats.
> >
> > We've run into https://issues.apache.org/jira/browse/SPARK-19185,
> > 'ConcurrentModificationExceptions with CachedKafkaConsumers'. I've
> tried to
> > work around it as follows
asonable amount of data
> from Kafka (40 GB / minute or so). We would like to move to using the Kafka
> 0.10 client to avoid requiring our (0.10.2.1) Kafka brokers from having to
> modify formats.
>
> We've run into https://issues.apache.org/jira/browse/SPARK-19185,
>
brokers from
having to modify formats.
We've run into https://issues.apache.org/jira/browse/SPARK-19185,
'ConcurrentModificationExceptions with CachedKafkaConsumers'. I've tried
to work around it as follows:
1. Disabled consumer caching. This increased the total job time from ~1
minute per batch