Hacker News Comments on
Distributed Systems in One Lesson by Tim Berglund
Devoxx Poland
·
Youtube
·
2
HN points
·
1
HN comments
- This course is unranked · view top recommended courses
Hacker News Stories and Comments
All the comments and stories posted to Hacker News that reference this video.https://www.youtube.com/watch?v=Y6Ev8GIlbxc&t=28m15sJust saw this and might explain why Hadoop is out of the spotlight. In summary, Spark and Kafka seem to be better? I'm not sure as I'm just starting to enter this field.
⬐ nathairtrasGetting this out of the way first, I've only started exploring non-Hadoop/non-HDFS Spark execution beyond some limited Amazon EMR work, but I'm interested in learning more about it. What follows is a combination of work experience and armchair research in the evenings. But I'm not claiming to be an expert.Have grown to really appreciate Spark in the Hadoop space. Started with plans to go with Impala, then went to Hive due to stability concerns, and finally to Spark due to speed / flexibility. You can write code against a data frame, or write Spark SQL, so you still have SQL.
HDFS has benefits over other storage approaches, if you are running Spark in the same cluster you get data proximity. But you can go with a different storage back-end. That costs in performance. "Performance of multiple query and enrichment jobs concurrently executed resulted in 90% longer execution times." https://redhatstorage.redhat.com/2018/06/25/why-spark-on-cep...
Unless you really have BIG data, you're invoking a lot of maintenance overhead to support a cluster when you may do just fine without.
Haven't had the freedom to explore other possibilities until recently, very interested in how Spark on k8s is working out. (Same comment could be made here as above and elsewhere - do you really need k8s? But I want to play with k8s and learn more about it, so... for that purpose I 'need' it.) https://spark.apache.org/docs/latest/running-on-kubernetes.h...
And there's always the cloud route. You can run an EMR job that uses files in s3. There is a cost, but you do not need to support a cluster in the same way. Or if you're feeling adventurous, use Lambda. https://www.qubole.com/blog/spark-on-aws-lambda/
And Spark isn't the only option. Have started learning about Dask, also looks very interesting for performing some of the same tasks. https://dask.org