Hi Ian,
As discussed in KAFKA-3775, proper memory management is better than throttling
and we've made some steps towards that in 0.10.1 and 0.10.2 (reduce the memory
RocksDb uses, provide a global memory limit for buffers within streams).
The scenario you mention is possible, and needs to be ad
> Yes, the partitions reflect those of the input topic. You could try to
create the topic manually before streams start, however, that might not be
an ideal operational way of doing things (it's best if streams continues to
do these things automatically). I'd suggest the scaling out approach first.
Hi Ian,
Yes, the partitions reflect those of the input topic. You could try to create
the topic manually before streams start, however, that might not be an ideal
operational way of doing things (it's best if streams continues to do these
things automatically). I'd suggest the scaling out appro
Hi Eno,
Thanks for the fast response.
> It looks like you have a lot of partitions for the count store.
I believe this isn't configurable? They were auto created by the stream.
I'm assuming its mirrored based of the amount off partitions our input
topic has.
> The locking part was supposed to h
Hi Ian,
It looks like you have a lot of partitions for the count store. Each RocksDb
database uses off heap memory (around 60-70MB in 0.10.2) which will add up if
you have these many stores in one instance. One solution would be to scale out
your streams application by using another Kafka Strea
Hi All,
I'm using Kafka Client 10.2 with Kafka Streams.
I'm performing a groupByKey on a stream and seeing large files appear
within my state directory. Is this expected?
90M 1_0/rocksdb/content-count-store
82M 1_1/rocksdb/content-count-store
102M 1_10/rocksdb/content-count-store
86M 1_11/rocksd
