There is no need to create a new producer instance for each write request. In doing so you lose the advantages of the buffering and batching that the producer offers. In your use case I would recommend having a single running producer and tuning the batch size and linger.ms settings if you find that the producer is using too much memory.
On Mon, Mar 13, 2017 at 5:05 AM, Amit K <amitk....@gmail.com> wrote: > Hi, > > I am using simple kafka producer (java based, version 0.9.0.0) in an > application where I receive lot of hits (about 50 per seconds, in much like > servlet way) on application that has kafka producer. Per request comes > different set of records. > > I am using only one instance of kafka producer to push the records to kafka > cluster. Is that good way to use kafka producer? As it is mentioned in > documentation, the kafkaproducer can be shared across multiple threads. > > Or should be there one kafka producer created to handle one request? > > Is there any best practice documents/guidelines to follow for using simple > java Kafka producer api? > > Thanks in advance for your responses. > > Thanks, > Amit > -- Robert Quinlivan Software Engineer, Signal
