Built-in metadata written at document creation uses a different format, breaking time filters on upload_date / last_update_date #44217

Self Checks

• I have read the Contributing Guide and Language Policy.
• This is only for bug report, if you would like to ask a question, please head to Discussions.
• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report, otherwise it will be closed.
• 【中文用户 & Non English User】请使用英语提交，否则会被关闭 ：）
• Please do not modify this template :) and fill in all the required fields.

Dify version

1.17.1 (also present on main @ 69e9ca4)

Cloud or Self Hosted

Self Hosted (Source)

Steps to reproduce

1. Create a knowledge base and upload a document (e.g. a.pdf).
2. Open the knowledge base metadata settings and enable Built-in metadata.
3. Upload another document (e.g. b.pdf).
4. In a workflow / chatflow, add a Knowledge Retrieval node with manual metadata filtering: upload_date after any date.
5. Run it.

Root cause: built-in fields are written by three code paths, and they don't agree on the format:

time filters compile to CAST(doc_metadata ->> 'upload_date' AS FLOAT), so a single string-formatted document makes the whole query fail on PostgreSQL. I ran the SQL produced by DatasetRetrieval.process_metadata_filter_func against a minimal documents table on PostgreSQL 15 holding one row in each format:

upload_date before 1893456000 -> InvalidTextRepresentation: invalid input syntax for type double precision: "2026-10-16 08:00:00"
upload_date after 0           -> InvalidTextRepresentation: invalid input syntax for type double precision: "2026-10-16 08:00:00"
source is 'file_upload'       -> only the document written by _built_in is returned

(Verified at the SQL level with Dify's own filter function, not through a full end-to-end deployment.)

Both formats were introduced together in #15982; the knowledge pipeline later copied the build_document version in #25360.

✔️ Expected Behavior

Every document stores built-in fields in the same format: epoch seconds for the time fields and the MetadataDataSource label for source, the same label the document detail page shows (format_document_built_in_fields). Time filters and source filters then work for all documents.

❌ Actual Behavior

• Filtering on upload_date / last_update_date with before / after / < / > fails the Knowledge Retrieval node with a PostgreSQL cast error once any document has been created after built-in fields were enabled.
• Filtering source is file_upload (the value shown in the UI) silently skips those documents, because they store upload_file.

I'd like to submit a fix for this.