Will data & metadata stay in sync?¶

Here, we walk through different errors that can occur while saving artifacts & metadata records, and show that the LaminDB instance does not get corrupted by dangling metadata or artifacts. You could say transactions across data & metadata are ACID.

Setup¶

Install the lamindb Python package:

pip install 'lamindb[jupyter,aws]'

from laminci.db import setup_local_test_postgres

pgurl = setup_local_test_postgres()

💡 Created Postgres test instance: 'postgresql://postgres:[email protected]:5432/pgtest'
It runs in docker container 'pgtest'

!lamin init --db {pgurl} --storage ./test-acid

❗ using the sql database name for the instance name

💡 connected lamindb: testuser1/pgtest

import pytest
import lamindb as ln
from upath import UPath

💡 connected lamindb: testuser1/pgtest

Save error due to failed upload¶

Let’s try to save an artifact to a storage location without permission.

artifact = ln.Artifact.from_anndata(
    ln.core.datasets.anndata_mouse_sc_lymph_node(),
    description="Mouse Lymph Node scRNA-seq",
)

❗ no run & transform get linked, consider calling ln.track()

Because the public API only allows you to set a default storage for which you have permission, we need to hack it:

ln.setup.settings.storage._root = UPath("s3://nf-core-awsmegatests")

ln.settings.storage

StorageSettings(root='/home/runner/work/lamindb/lamindb/docs/faq/test-acid', uid='IhAi1ViQyVAH')

This raises a RuntimeError:

with pytest.raises(RuntimeError) as error:
    artifact.save()
print(error.exconly())

❗ could not upload artifact: Artifact(uid='rLkWcGE9ZP7qwfcAvRGd', description='Mouse Lymph Node scRNA-seq', suffix='.h5ad', type='dataset', accessor='AnnData', size=17177479, hash='JYwvsCMVfGYZLiZPSdHZNw', hash_type='md5', visibility=1, key_is_virtual=True, created_by_id=1, storage_id=1, updated_at='2024-07-26 14:36:03 UTC')

RuntimeError: Access Denied

Let’s now check that no metadata records were added to the database:

assert len(ln.Artifact.filter().all()) == 0

Save error during bulk creation¶

filepath = ln.core.datasets.file_jpg_paradisi05()
artifact = ln.Artifact(filepath, description="My image")
artifacts = [artifact, "this is not a record"]

❗ no run & transform get linked, consider calling ln.track()

This raises an exception:

with pytest.raises(Exception) as error:
    ln.save(artifacts)
print(error.exconly())

AttributeError: 'str' object has no attribute '_state'

Nothing got saved:

artifacts = ln.Artifact.filter().all()
assert len(artifacts) == 0

If a list of data objects is passed to ln.save() and the upload of one of these data objects fails, the successful uploads are maintained and a RuntimeError is raised, listing the successfully uploaded data objects up until that point.