konradmalik/dlenv repository overview

⁠DLEnv

Container for ML, AI, Deep Learning, reinforcement learning etc.

Can also be used to run spark with all these libraries available for example in pyspark by default (spark jobs won't get crashed due to discrepancies between spark driver/notebook environment and executors)

Includes spark 2.4.4 with hadoop 2.10.0 (custom hadoop for jupyter with scala 2.11 (almond) on a remote cluster).

Explanation:

​ In short - this kernel (at least version 0.6.0 for scala 2.11) requires http filesystem support in order to work.

​ This is provided by hadoop >-= 2.9, but standard spark ships with older version.

​ The effect is that local spark cluster works, but when connecting to remote - executors get destroyed with error that there is no scheme for http filesystem.

​ This docker image resolves this isssue and allows to use remote spark cluster with almond 0.6.0 and scala 2.11.

​ For sample usage see docker-compose.yml.

​ Remember to adjust cpu and memory on the worker.

For contents, included libraries etc. see the first couple of lines of the Dockerfile.

⁠Random notes

• scala kernel for jupyter is provided by almond. In prefdef it automatically imports local spark jars and downloads (if not present) almond libs, so only thing you need to do is to create spark session. Refer to almond's "usage-spark.md" document.

• to add additional dependencies to the notebook, use imports with ivy like this example:

import $ivy.`org.apache.hadoop::hadoop-client:2.10.0`

• all logs from 'org' domain are hidden by default to not to pollute notebook cells. If you want to enable them, use this example:

import org.apache.log4j.{Level, Logger}
Logger.getLogger("org").setLevel(Level.INFO)

• example snippet to create spark session in notebook (spark master can be remote!):

val spark = {
  NotebookSparkSession.builder()
    .master("spark://localhost:7077")
    .config("spark.executor.instances", "4")
    .config("spark.executor.memory", "2g")
    .getOrCreate()
}

⁠How to build:

Image is available on dockerhub (konradmalik/dlenv).

If you want to build locally, use "make":

$ make build

Then run either bash shell or jupyter server using provided shell scripts.

Run scripts are currently set up to autodelete after exit so all data that is not in the "data" folder will be lost!