topic Re: Issue in importing librosa library while using databricks runtime engine 11.2 in Data Engineering

Issue in importing librosa library while using databricks runtime engine 11.2

Chandru — Tue, 29 Nov 2022 14:08:09 GMT

I have installed the library via PyPI on the cluster. When we import the package on notebook, getting the following error

import librosa

OSError: cannot load library 'libsndfile.so': libsndfile.so: cannot open shared object file: No such file or directory

-werners- — Tue, 29 Nov 2022 16:20:36 GMT

it seems the linux OS is missing the libsndfile package.

This is apparently a dependency of librosa.

This can be done by using an init script:

Chandru — Tue, 29 Nov 2022 16:25:40 GMT

Thank you werners. Just figured that out and had an init script to sort out the issue. Below steps helped me to solve the issue.

dbutils.fs.mkdirs("dbfs:/cluster-init/scripts/")

dbutils.fs.put("/cluster-init/scripts/libsndfile-install.sh","""

#!/bin/bash

apt-get --yes install libsndfile1

""", True)

and added init script for the cluster as

dbfs:/cluster-init/scripts/libsndfile-install.sh

Flo — Fri, 19 Jul 2024 14:02:27 GMT

If anybody ends up here after 2024: the init file must now be placed in the workspace for the cluster to accept it.

So in Workspace, use Create/File to create the init script.

Then add it to the cluster config in
Compute - Your cluster - Advanced Config - init scripts