Important Warning

Update July 2016: This Tech Preview documentation from March 2016 is outdated and deprecated. Please use the latest Confluent Platform documentation instead.

Important Configuration OptionsΒΆ

Camus provides configurations to reduce overhead in operations. Below is an explanation of how these configurations work, and why they are needed.

  1. kafka.max.pull.hrs: This configuration limits timestamp duration for each topic partition pull from Kafka. Since Camus output is partitioned by timestamp, large pulls from Kafka may create a large number of files. For example, if you use hourly partition, a pull of 3 days data in a Camus job will simultaneously write to 72 files and some versions of Hadoop don’t behave well in this case. Therefore, the ability to limit the timestamp duration is needed.

    The way how this works is that the timestamp for each event is compared to the timestamp of the first event. If the timestamp is greater than duration set, the pull is stopped and the next Camus job will pickup from where it left off.

  2. kafka.max.pull.minutes.per.task: This configuration limits the pull time for MapReduce tasks. If the MapReduce task time is exceeded, the pulls are halted and the pull of this topic partition in the next Camus job will continue from the offset where it left off.

  3. kafka.max.historical.days: This configuration limits the pull for historic data. Events older than the specified number of days will be ignored. This is useful in for situations where a valid offset is not available and a topic partition is pulled from the earliest offset.

  4. etl.fail.on.errors: This configuration controls the behavior of a Camus job in case of exceptions. The default behavior is that Camus catches exceptions when decoding data and write them to the error file associated with the Camus job. This is reasonable as we don’t want a single bad record to fail the Camus job. However, if you want to ensure data compatibility, you may want to set this configuration with value true to make sure that a Camus job fails in case of exceptions.