cannot import name 'NANOARROW_TIME_UNIT_SECOND' from 'oracledb.interchange.nanoarrow_bridge'

[Jaza]

Running in a docker container using the docker image: docker.io/library/python:3.11.4-alpine3.17

With cython-0.29.37

Installed oracledb 3.1.0

Getting this error:

tests/conftest.py:7: in <module>
    from mypkgfoo import get_db_cursor
/path/to/mypkgfoo/__init__.py:1: in <module>
    from . import pools
/path/to/mypkgfoo/pools.py:1: in <module>
    from mypkgfoo.constants import DbPoolAndCreatedTime, DbPoolKey
/path/to/mypkgfoo/constants.py:4: in <module>
    from oracledb import AsyncConnection, AsyncConnectionPool, AsyncCursor
/path/to/oracledb/__init__.py:319: in <module>
    from .interchange.dataframe import (
/path/to/oracledb/interchange/dataframe.py:34: in <module>
    from .column import OracleColumn
/path/to/oracledb/interchange/column.py:43: in <module>
    from .nanoarrow_bridge import (
E   ImportError: cannot import name 'NANOARROW_TIME_UNIT_SECOND' from 'oracledb.interchange.nanoarrow_bridge' 
    (/path/to/oracledb/interchange/nanoarrow_bridge.cpython-311-x86_64-linux-musl.so)

Not getting this error in other environments (that aren't running alpine).

First gave me this error 3 days ago (10 May 2025).

Was previously running with same docker image python:3.11.4-alpine3.17 , same cython-0.29.37 , same oracledb 3.1.0 , and it worked fine, I didn't get this error, 5 days ago (8 May 2025).

Anyone please able to help? Currently blocking me from running all my CI builds, really need to resolve.

[cjbj]

This got borked by the just-released Cython 3.1. Using the older Cython should work.

@anthony-tuininga fixed the problem in our internal repo at the end of last week, and no doubt will push out the patch to GitHub's main branch this week.

[anthony-tuininga]

@Jaza, a patch has been pushed. I created a new issue #493 to track it further. Builds are in progress if you want to try those. We will discuss when the next release takes place.

[Jaza]

Thanks @anthony-tuininga and @cjbj for looking into it so promptly, that has unblocked my internal CI pipelines!

[Jaza]

@cjbj thanks for your prompt reply!

I'm using Cython 0.29.37 (I pinned it to cython<3 ages ago, I can't remember why anymore, something broke for me back when Cython 3.0 was released). So how could the release of Cython 3.1 have broken python-oracledb for me?

Like I said, I confirmed via build logs that my python core version, cython version, and python-oracledb version were unchanged, between when everything was working fine for me, and when this nanoarrow error started happening.

Anyway, if you guys have a fix in the pipeline, looking forward to seeing it in main and in a release for python-oracledb sometime soon.

[Jaza]

Just tried bumping Cython to 3.1.0 . Same error.

[Jaza]

Tried Cython 3.0.12 . Same error.

[Jaza]

I can reproduce the error in other environments if I force python-oracledb to install from source (in my case, using poetry, the command is POETRY_INSTALLER_NO_BINARY=oracledb poetry install)

[Jaza]

tl;dr: no matter what environment I'm in (and no matter the cython version), I get the nanoarrow error if I install python-oracledb from source. No nanoarrow error if I install python-oracledb from an official pre-built binary.

Don't know what magic you guys are doing in github actions, that I'm not doing locally, to build a binary that doesn't result in that error. Wish I could use an official pre-built binary in my CI environment - but none of the pre-built binaries appears to be compatible with that python:3.11.4-alpine3.17 environment of mine.

[Jaza]

Ahhh, figured it out... it doesn't use whatever cython version I have in my project's virtualenv when building python-oracledb from source... it creates its own separate virtualenv, and installs the cython version specified in python-oracledb's pyproject.toml .

I changed python-oracledb's pyproject.toml (on my local machine) to specify cython<3.1.0 , and built it, and installed the resulting wheel. Works, no nanoarrow error.

Will see what I can do to get it installing like that in my CI environment. That will at least unblock me, until the python-oracledb guys release an official fix.

[cjbj]

@Jaza On another topic, how are you finding python-oracledb on Alpine? I assume you are using Thin mode?

[Jaza]

@cjbj python-oracledb on Alpine is pretty similar to how it is on Ubuntu, except that it seems to need to build from source, the "manylinux" artifacts don't seem to get accepted by pip / poetry (the artifact that it builds on Alpine is called "musl"). Yes, using Thin mode (although I previously used cx_oracle i.e. Thick mode also in Alpine - installing the Oracle Instant Client was a pain, but otherwise it was the same deal getting it working).

[cjbj]

@Jaza Thanks for the feedback. You must be keen on Alpine if you had bothered to install Instant Client !