Does anyone successfully used this on AWS?

[oscarnevarezleal]

Does anyone successfully used this on AWS?

[mjgpy3]

@oscarnevarezleal Yes, I have run it on an EC2 instance. Do you have any questions?

[flosell]

@oscarnevarezleal Are you maybe trying to run this on a memory-constrained instance?
I just found your issue trying to find out why kafka was crashing immediately after start when I deployed the container in a t2.nano instance but was working fine on my laptop.

Turns out, the container by default starts a JVM with -Xmx1G -Xms1G so the t2.nano instance just wasn't big enough. Moving to a bigger instance (t2.small) solved the issue for me. You can also set KAFKA_HEAP_OPTS to something like "-Xmx250m -Xms250m" if you are comfortable running with less memory.

[oscarnevarezleal]

@flosell That was exactly my problem, thnks!!

[chorsnell]

@mjgpy3 / @oscarnevarezleal / @flosell

How do you handle the ADVERTISED_HOST? I am using mine on EC2 but via ECS, so I can't be sure which instance it will be under.

In my local dev environment, I have to specify the IP like

  kafka:
    image: spotify/kafka
    hostname: kafka
    environment:
    - ADVERTISED_HOST=192.168.xxx.xxx
...

[oscarnevarezleal]

@poppahorse ADVERTISED_HOST should be setted with EC2´s public IP ( not a subnet IP )

Please note that AWS could ( might, most likely ) change your EC2 public IP without warning you so.

If you launch your container with a fixed ADVERTISED_HOST env, this would stop working from one moment to another. What I do in this cases is to assign a Elastic IP to this particular instance altought I don´t know how to apply ( scale ) same rule with multiple instances.

This is an example of docker-compose.yml using spotify/kafka & confluentinc/cp-kafka-rest togheter.

`---
version: '2'
services:
kafka:
image: spotify/kafka
environment:
ADVERTISED_HOST: YOUR_INSTANCE_PUBLIC_IP
ADVERTISED_PORT: 9092
labels:
- io.confluent.docker.testing=true
ports:
- "9092:9092"
- "2181:2181"

kafka-rest:
image: confluentinc/cp-kafka-rest:latest
environment:
KAFKA_REST_ZOOKEEPER_CONNECT: kafka:2181
KAFKA_REST_HOST_NAME: default-config
labels:
- io.confluent.docker.testing=true
links:
- kafka
ports:
- "8082:8082"
`

[aaparella]

Having a similar problem to @poppahorse, I think. Using this container in ECS and every time a consumer / producer tries to consume / produce to a topic there's a "LEADER_NOT_AVAILABLE" error message.

I have a producer and the kafka container in the same task, with the producer linked to the kafka instance, and ports 2181 and 9092 exposed on the kafka container. I have the ADVERTISED_HOST environment variable set to the IP of the container instance (I know that's not ideal this is just for proof of concept purposes).

Any advice?