search

caldempsey / docker-notebook-spark-s3

Template CI friendly local development environment featuring Spark Clusters + Blob Storage + a Notebook for prototyping data feature delivery.
3 stars 0 forks source link

[BUG] Clustered Spark fails to write _delta_log via a Notebook without granting the Notebook data access #3

Open caldempsey opened 4 months ago

caldempsey commented 4 months ago

Describe the problem

Reproduced in the notebook on https://github.com/caldempsey/docker-notebook-spark-s3/pull/6

At present we have set up a Jupyter Notebook w/ PySpark connected to a Spark cluster, where the Spark instance is intended to perform writes to a Delta table. I'm observing that the Spark instance fails to complete the writes if the Jupyter Notebook doesn't have access to the data location.

This behavior seems counterintuitive to me as I expect the Spark instance to handle data writes independently of the Jupyter Notebook's access to the data.

Steps to reproduce

Via the repo provided:

  1. Clone the repo
  2. Run the notebook. Observe we can write delta tables.
  3. Delete everything in the notebook's data folder.
  4. Remove infra-delta-lake/localhost/docker-compose.yml:63 ./../../notebook-data-lake/data:/data, which prevents the notebook from accessing /data at the same target shared with the Spark Master and Workers on their local filesystem.

Observed results

When the notebook has access to /data (but is a connected application not a member of the cluster), Delta Tables write successfully with _delta_log.

When the notebook does not have access to /data it complains that it can't write _delta_log, but parquet files still get written! 

Py4JJavaError: An error occurred while calling o56.save.
: org.apache.spark.sql.delta.DeltaIOException: [DELTA_CANNOT_CREATE_LOG_PATH] Cannot create file:/data/delta_table_of_dog_owners/_delta_log
    at org.apache.spark.sql.delta.DeltaErrorsBase.cannotCreateLogPathException(DeltaErrors.scala:1534)
    at org.apache.spark.sql.delta.DeltaErrorsBase.cannotCreateLogPathException$(DeltaErrors.scala:1533)
    at org.apache.spark.sql.delta.DeltaErrors$.cannotCreateLogPathException(DeltaErrors.scala:3203)
    at org.apache.spark.sql.delta.DeltaLog.createDirIfNotExists$1(DeltaLog.scala:443)```

Expected results

Expect the _delta_log to be written regardless of whether the Notebook has access to the target filesystem.

Further details

Since this error is surfacing from PySpark I'm wondering if either the Notebook instance is somehow electing itself master via PySpark or if there's a bug in delta lake where you can’t write delta tables without the application call-site having access to the location. Neither of these sound right but I can't think of a third way.

Feel free to have a gander or submit a PR 🙏 !

Environment information

caldempsey commented 4 months ago

Could yield answers: https://stackoverflow.com/questions/45997150/can-i-run-a-pyspark-jupyter-notebook-in-cluster-deploy-mode?rq=2

caldempsey commented 3 months ago

This needs to be fixed with a refinement to the overall architecture as above. It's not a bug, just that PySpark can only run in client/driver mode when connecting to a standalone cluster.

caldempsey commented 3 months ago

Databricks might also have a solution for this with their latest DataLake connectors. Kind of a game-changer in the space. Something to read on.