Geni not usable wit latest Databricks

[behrica]

As mentioned here #332

I have an issue with geni on databricks. Apparently this call https://github.com/zero-one-group/geni/blob/e6eaeb937d7598648e1986928d35f5bea336441f/src/clojure/zero_one/geni/spark.clj#L37

which is execute even when using an existing spark session / context, fails on databricks:

IllegalArgumentException: Path must be absolute: target/checkpoint/3f38a4a8-51e9-47fc-a1d1-7c0f3e2f2520 at com.databricks.common.path.AbstractPath$.fromHadoopPath(AbstractPath.scala:114) at com.databricks.backend.daemon.data.client.DBFSV2.resolveAndGetFileSystem(DatabricksFileSystemV2.scala:148) at com.databricks.backend.daemon.data.client.DatabricksFileSystemV2.resolve(DatabricksFileSystemV2.scala:773) at com.databricks.backend.daemon.data.client.Databrick

If I understand the current code correctly, it cannot be avoided that this call is made, neither can the directory be changed.

My comments in #332 where based n the assumption that the code does wrongly "creates" an other session, but I think it's no true. It does get the session from databrcks, but tries to set the checkpoint directory on the existing session/context, and this fails now. Maybe it did work with older databricks/spark versions.

[behrica]

This issue makes it impossible to use geni with databricks.

[behrica]

I digged into it. While I was always wondering, if setting options here : https://github.com/zero-one-group/geni/blob/e6eaeb937d7598648e1986928d35f5bea336441f/src/clojure/zero_one/geni/defaults.clj#L6

is the "right thing" to do "in general", as there might always be Spark environments which do not support some of them.

Docu says:

Gets an existing SparkSession or, if there is no existing one, creates a new one based on the options set in this builder. In case an existing SparkSession is returned, the config options specified in this builder will be applied to the existing SparkSession.

so the options are always used even if an existing spark session is used, which to me is questionable, specially as it can not be disabled.

The setCheckpointDir is always called as well, even when using an existing spark session. And apparently my databricks cluster does not support to change the checkpoint dir to the unchangable default of "target/checkpoint/" (which is clearly a development setting and I clearly don't want to even call in my cluster) My suggestion would be to not set neither options nor checkpoint-dir by default.

[behrica]

on my fork I made a change which fixes this: https://github.com/behrica/geni/commit/16080dd0156da170b301984ce79ff6ca699fc084

[anthony-khong]

Hi Carsten @behrica, would you like to make a PR, and then I'm happy to merge it. Let me know if we need to change the CI workflows as well.