write_deltalake fails on Databricks volume

[Bernolt]

Environment

Delta-rs version: 0.17.4

Binding: python (pyarrow engine)

Environment:

• Cloud provider: Azure
• OS:
• Other: Databricks runtime 13.3 LTS

---

Bug

What happened: From a python application running on a Databricks cluster, I want to write to an append-only delta table. The function is called as follows:

write_deltalake(
      data=arrow_table,
      table_or_uri="/Volume/catalog/schema/volume_path/table_path",
      mode="append",
      overwrite_schema=False)

However, I am getting the below error:

OSError: Generic LocalFileSystem error: Unable to copy file from /Volumes/catalog/schema/volume_path/table_path/_delta_log/_commit_e964ab56-f56c-403a-b06d-fe2b6bcabf9d.json.tmp to /Volumes/catalog/schema/volume_path/table_path/_delta_log/00000000000000000000.json: Function not implemented (os error 38)

What you expected to happen: As Databricks supports copy/rename/delete operations, I would expect it to work. As far as I know Databricks use a Local File System API, which emulates a filesystem on top of a cloud storage.

How to reproduce it: I made the below notebook to reproduce the error. It needs to be run from a Databricks Runtime.

# Databricks notebook source
# MAGIC %sh
# MAGIC touch /Volumes/catalog/schema/volume/table_path/to_rename.tmp

# COMMAND ----------

# MAGIC %sh
# MAGIC mv /Volumes/catalog/schema/volume/table_path/to_rename.tmp /Volumes/catalog/schema/volume/table_path/renamed.todelete

# COMMAND ----------

# MAGIC %sh 
# MAGIC rm /Volumes/catalog/schema/volume/table_path/renamed.todelete

# COMMAND ----------

from deltalake import write_deltalake
import pyarrow as pa

# COMMAND ----------

arrow_table = pa.table([
    pa.array([2, 4, 5, 100]),
    pa.array(["Flamingo", "Horse", "Brittle stars", "Centipede"])
    ], names=['n_legs', 'animals'])

# COMMAND ----------

write_deltalake(table_or_uri = "/Volumes/catalog/schema/volume/table_path/reproduce_deltars_error_table_01",
                data = arrow_table,
                mode = "append",
                overwrite_schema=False)

More details:

[Bernolt]

It might not be a bug from delta-rs perspective, however, it would be helpful to have some insights on the underlying file system operation performed.

[ion-elgreco]

Afaik, databricks volumes are fuse mounted, so this is not an bug. If you want to write to mounted storage that doesn't support CopyIfNotExists, you can pass this to the writer:

storage_options = {"allow_unsafe_rename": "true"}

[Bernolt]

Thanks, solved my issue.