kafkat VS fluent-plugin-kafka

Few tips before I close this blog post, always have enough memory in Kafka server as it is very memory intensive and tends to shutdown gracefully every time it hits heap size limit without indicating anything in the logs. For monitoring your cluster use Kafka Manager, it does the job very well and also have KafkaT on server which saves you from running cumbersome builtin Kafka commands.

Before we moved to using the data from Kafka, we kept it running as our secondary datasource imported in to Big Query to compare the data count. The data count seemed to be better coming from Kafka indicating that we were losing some data via our S3 based pipeline. As we started adding more and more datasets to Kafka we eventually hit a road bump due to memory issue causing server to fall out of the quorum, which was initially hard to identify with errors in the Tron server. The issue 97 was opened over to discuss on Github in Fluentd Kafka Plugin repository. That got drilled down to the plugin's limitation to counter it bound to the use of the gem ruby-kafka, a fix PR294 was contributed to resolve the issue from happening in future. The fix is now part of the fluent-plugin-kafka as well which will benefit many others. Happy to say that the patch works well in production :).