uswitch / bifrost Goto Github PK

Hi all -

I don't know much about Java / Clojure so it's a bit hard for me to give a great bug report. But basically, I update the default config to my S3 / Zookeeper settings and start Bifrost with Lein, per the README.

It starts and runs, makes Zookeeper connections, reads the metadata from my Kafka nodes, writes some data in /tmp/ as expected, and then my container's supervisor comes around and kills the process for too much memory usage (I have 1.8Gb free for use by Bifrost).

Is there a way to limit the memory usage? Or do I just need more memory? If this amount of usage is unexpected, any tips on debugging?

I tried changing uploaders-n to 1 hoping that would help, but no luck.

Thanks!
-Charlie

Use prismatic schema to ensure the system config is correct before starting.

Bifrost currently doesn't handle rebalancing events- it assumes a single consumer process.

When starting Bifrost if there's a ZooKeeper exception thrown (e.g. #<ZkTimeoutException org.I0Itec.zkclient.exception.ZkTimeoutException: Unable to connect to zookeeper server within timeout: 6000>) it hangs the creation of processes for consuming the partition's messages. Add some proper re-try with decay timeout handling.

Hi - I have some functionality questions about Bifrost:

1. Can Bifrost output data onto S3 in a format that is consumable by Redshift?
2. How often can files be written out? Every hour? Every minute?
3. Can Bifrost extract data by a data field in the Kafka data, rather than the Kafka created_at timestamp?

Thanks!

The baldr file-format is minimal and simple. We have since realised that tar-files offers a similar sequential file format that is slightly more difficult to implement readers from scratch for, but which have operating system support on many platforms, as well as in the Java standard library. It would be nice to have the option to store messages in tar files.