Bug: Load PreProcessor with split_by: None from config fails

[ugm2]

The issue

When trying to load the PreProcessor node with split_by: None from config fails.

Reproduce the issue

To reproduce the issue, execute this script with the latest Haystack version and python 3.8:

from haystack import Pipeline
from haystack.nodes.preprocessor import PreProcessor

processor = PreProcessor(
    split_by=None,
)

pipeline = Pipeline()
pipeline.add_node(processor, name="Preprocessor", inputs=["File"])

config = pipeline.get_config()
pipeline = Pipeline.load_from_config(config, pipeline_name="indexing")

The error

Where it fails:

haystack/haystack/pipelines/config.py

Lines 307 to 331 in 7290196

	Draft7Validator(schema).validate(instance=pipeline_config)
	break
	except ValidationError as validation:
	# If the validation comes from an unknown node, try to find it and retry:
	if list(validation.relative_schema_path) == ["properties", "components", "items", "anyOf"]:
	if validation.instance["type"] not in loaded_custom_nodes:
	logger.info(
	f"Missing definition for node of type {validation.instance['type']}. Looking into local classes..."
	)
	missing_component_class = BaseComponent.get_subclass(validation.instance["type"])
	schema = inject_definition_in_schema(node_class=missing_component_class, schema=schema)
	loaded_custom_nodes.append(validation.instance["type"])
	continue
	# A node with the given name was in the schema, but something else is wrong with it.
	# Probably it references unknown classes in its init parameters.
	raise PipelineSchemaError(
	f"Node of type {validation.instance['type']} found, but it failed validation. Possible causes:\n"
	" - The node is missing some mandatory parameter\n"
	" - Wrong indentation of some parameter in YAML\n"
	"See the stacktrace for more information."
	) from validation

The actual error:

On instance['components'][0]:
    {'name': 'Preprocessor',
     'params': {'split_by': None},
     'type': 'PreProcessor'}
Traceback (most recent call last):
  File "/home/paperspace/repositories/smart-search/env/lib/python3.8/site-packages/haystack/pipelines/config.py", line 338, in validate_schema
    Draft7Validator(schema).validate(instance=pipeline_config)
  File "/home/paperspace/repositories/smart-search/env/lib/python3.8/site-packages/jsonschema/validators.py", line 304, in validate
    raise error
jsonschema.exceptions.ValidationError: {'name': 'Preprocessor', 'type': 'PreProcessor', 'params': {'split_by': None}} is not valid under any of the given schemas

On instance['components'][0]:
    {'name': 'Preprocessor',
     'params': {'split_by': None},
     'type': 'PreProcessor'}

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "haystack_bug_test.py", line 12, in <module>
    pipeline = Pipeline.load_from_config(config, pipeline_name="indexing")
  File "/home/paperspace/repositories/smart-search/env/lib/python3.8/site-packages/haystack/pipelines/base.py", line 1913, in load_from_config
    validate_config(pipeline_config, strict_version_check=strict_version_check)
  File "/home/paperspace/repositories/smart-search/env/lib/python3.8/site-packages/haystack/pipelines/config.py", line 259, in validate_config
    validate_schema(
  File "/home/paperspace/repositories/smart-search/env/lib/python3.8/site-packages/haystack/pipelines/config.py", line 366, in validate_schema
    raise PipelineSchemaError(
haystack.errors.PipelineSchemaError: Node of type PreProcessor found, but it failed validation. Possible causes:
 - The node is missing some mandatory parameter
 - Wrong indentation of some parameter in YAML
See the stacktrace for more information.

[julian-risch]

@ugm2 Thank you also for reporting this bug, Unai! The problem seems to be the type of the split_by parameter indicated here: https://github.com/deepset-ai/haystack/blob/main/haystack/nodes/preprocessor/preprocessor.py#L54 and the schema file generated based on that with the following line:

haystack/haystack/json-schemas/haystack-pipeline-main.schema.json

Line 4286 in 7290196

"type": "string"

[julian-risch]

Hi @ugm2 here is a PR that should fix the issue: #3389 If you want you can try it out by checking out the branch of the PR even before we merge it into main.

[ugm2]

Hey @julian-risch ! I just checked out to the branch of the PR and tried my script and it worked like a charm! So yeah, that'd work for me 🙂

[julian-risch]

Great to hear that. Thank you for trying it out. 👍