What happens if I save the same artifacts & records twice?¶

LaminDB’s operations are idempotent in the sense defined in this document.

This allows you to re-run a notebook or script without erroring or duplicating data. Similar behavior holds for human data entry.

Summary¶

Metadata records¶

If you try to create any metadata record (Record) and search_names is True (the default):

LaminDB will warn you if a record with similar name exists and display a table of similar existing records.
You can then decide whether you’d like to save a record to the database or rather query an existing one from the table.
If a name already has an exact match in a registry, LaminDB will return it instead of creating a new record. For versioned entities, also the version must be passed.

If you set search_names to False, you’ll directly populate the DB.

Data: artifacts & collections¶

If you try to create a Artifact object from the same content, depending on artifact_if_hash_exists,

you’ll get an existing object, if creation.artifact_if_hash_exists = "warn_return_existing" (the default)
you’ll get an error, if creation.artifact_if_hash_exists = "error"
you’ll get a warning and a new object, if creation.artifact_if_hash_exists = "warn_create_new"

Examples¶

# !pip install 'lamindb[jupyter]'
!lamin init --storage ./test-idempotency

→ connected lamindb: testuser1/test-idempotency

import lamindb as ln
import pytest

ln.track("ANW20Fr4eZgM0000")

→ connected lamindb: testuser1/test-idempotency

→ notebook imports: lamindb==0.76.15 pytest==8.3.3

→ created Transform('ANW20Fr4'), started new Run('g9XHksTd') at 2024-11-07 11:05:19 UTC

Metadata records¶

assert ln.settings.creation.search_names

Let us add a first record to the ULabel registry:

label = ln.ULabel(name="My project 1")
label.save()

ULabel(uid='UBxzEzKP', name='My project 1', created_by_id=1, run_id=1, created_at=2024-11-07 11:05:21 UTC)

If we create a new record, we’ll automatically get search results that give clues on whether we are prone to duplicating an entry:

label = ln.ULabel(name="My project 1a")

! record with similar name exists! did you mean to load it?

	uid	name	description	reference	reference_type	run_id	created_at	created_by_id
id
1	UBxzEzKP	My project 1	None	None	None	1	2024-11-07 11:05:21.668306+00:00	1

label.save()

ULabel(uid='0CWrtfep', name='My project 1a', created_by_id=1, run_id=1, created_at=2024-11-07 11:05:21 UTC)

In case we match an existing name directly, we’ll get the existing object:

label = ln.ULabel(name="My project 1")

→ returning existing ULabel record with same name: 'My project 1'

If we save it again, it will not create a new entry in the registry:

label.save()

ULabel(uid='UBxzEzKP', name='My project 1', created_by_id=1, run_id=1, created_at=2024-11-07 11:05:21 UTC)

Now, if we create a third record, we’ll get two alternatives:

label = ln.ULabel(name="My project 1b")

! records with similar names exist! did you mean to load one of them?

	uid	name	description	reference	reference_type	run_id	created_at	created_by_id
id
1	UBxzEzKP	My project 1	None	None	None	1	2024-11-07 11:05:21.668306+00:00	1
2	0CWrtfep	My project 1a	None	None	None	1	2024-11-07 11:05:21.722018+00:00	1

If we prefer to not perform a search, e.g. for performance reasons or too noisy logging, we can switch it off.

ln.settings.creation.search_names = False

label = ln.ULabel(name="My project 1c")

In this walkthrough, switch it back on:

ln.settings.creation.search_names = True

Data: artifacts and collections¶

Warn upon trying to re-ingest an existing artifact¶

assert ln.settings.creation.artifact_if_hash_exists == "warn_return_existing"

filepath = ln.core.datasets.file_fcs()

Create an Artifact:

artifact = ln.Artifact(filepath, description="My fcs artifact").save()

Create an Artifact from the same path:

artifact2 = ln.Artifact(filepath, description="My fcs artifact")

→ returning existing artifact with same hash: Artifact(uid='uIm1rkVoSWeBnu7n0000', is_latest=True, description='My fcs artifact', suffix='.fcs', size=6785467, hash='KCEXRahJ-Ui9Y6nksQ8z1A', _hash_type='md5', visibility=1, _key_is_virtual=True, storage_id=1, transform_id=1, run_id=1, created_by_id=1, created_at=2024-11-07 11:05:22 UTC)

It gives us the existing object:

assert artifact.id == artifact2.id
assert artifact.run == artifact2.run
assert len(artifact._previous_runs.all()) == 0

If you save it again, nothing will happen (the operation is idempotent):

artifact2.save()

Artifact(uid='uIm1rkVoSWeBnu7n0000', is_latest=True, description='My fcs artifact', suffix='.fcs', size=6785467, hash='KCEXRahJ-Ui9Y6nksQ8z1A', _hash_type='md5', visibility=1, _key_is_virtual=True, storage_id=1, transform_id=1, run_id=1, created_by_id=1, created_at=2024-11-07 11:05:22 UTC)

In the hidden cell below, you’ll see how this interplays with data lineage.

Error upon trying to re-ingest an existing artifact¶

ln.settings.creation.artifact_if_hash_exists = "error"

In this case, you’ll not be able to create an object from the same content:

with pytest.raises(FileExistsError):
    artifact3 = ln.Artifact(filepath, description="My new fcs artifact")

Warn and create a new artifact¶

Lastly, let us discuss the following setting:

ln.settings.creation.artifact_if_hash_exists = "warn_create_new"

In this case, you’ll create a new object:

artifact4 = ln.Artifact(filepath, description="My new fcs artifact").save()

! creating new Artifact object despite existing artifact with same hash: Artifact(uid='uIm1rkVoSWeBnu7n0000', is_latest=True, description='My fcs artifact', suffix='.fcs', size=6785467, hash='KCEXRahJ-Ui9Y6nksQ8z1A', _hash_type='md5', visibility=1, _key_is_virtual=True, storage_id=1, transform_id=1, run_id=1, created_by_id=1, created_at=2024-11-07 11:05:22 UTC)

! record with similar description exists! did you mean to load it?

	uid	version	is_latest	description	key	suffix	type	size	hash	n_objects	n_observations	_hash_type	_accessor	visibility	_key_is_virtual	storage_id	transform_id	run_id	created_at	created_by_id
id
1	uIm1rkVoSWeBnu7n0000	None	True	My fcs artifact	None	.fcs	None	6785467	KCEXRahJ-Ui9Y6nksQ8z1A	None	None	md5	None	1	True	1	1	1	2024-11-07 11:05:22.642840+00:00	1

You can verify that it’s a new entry by comparing the ids:

assert artifact4.id != artifact.id

ln.Artifact.filter(hash="KCEXRahJ-Ui9Y6nksQ8z1A").df()

	uid	version	is_latest	description	key	suffix	type	size	hash	n_objects	n_observations	_hash_type	_accessor	visibility	_key_is_virtual	storage_id	transform_id	run_id	created_at	created_by_id
id
1	uIm1rkVoSWeBnu7n0000	None	True	My fcs artifact	None	.fcs	None	6785467	KCEXRahJ-Ui9Y6nksQ8z1A	None	None	md5	None	1	True	1	1	1	2024-11-07 11:05:22.642840+00:00	1
2	rvq145AiBrolHKR60000	None	True	My new fcs artifact	None	.fcs	None	6785467	KCEXRahJ-Ui9Y6nksQ8z1A	None	None	md5	None	1	True	1	1	2	2024-11-07 11:05:24.390262+00:00	1

!rm -rf ./test-idempotency
!lamin delete --force test-idempotency