feat: Added IBM Db2 vector store integration by priyanshu-krishnan1 · Pull Request #3518 · deepset-ai/haystack-core-integrations

priyanshu-krishnan1 · 2026-07-01T14:31:13Z

Related Issues

Adds IBM Db2 vector store integration for Haystack

Proposed Changes:

Added a new integration for IBM Db2 database with vector search capabilities:

Db2DocumentStore: Document store with vector similarity search using DB2's native VECTOR type
Db2EmbeddingRetriever: Retriever component for semantic search with metadata filtering
FilterTranslator: Converts Haystack filters to DB2 SQL with support for complex logical operators

How did you test it?

Unit tests for document store operations, filter translation, and connection handling
Integration tests using Docker Compose with IBM Db2 Community Edition
Haystack document store mixin tests
Manual verification with local Db2 instance

Notes for the reviewer

Follows standard Haystack document store patterns (similar to pgvector, oracle)

Checklist

I have read the contributors guidelines and the code of conduct
I have updated the related issue with new insights and changes
I added unit tests and updated the docstrings
I've used one of the conventional commit types for my PR title: feat: Add IBM Db2 vector store integration

socket-security · 2026-07-01T14:31:35Z

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	pypi/ibm-db@3.2.9

View full report

sjrl · 2026-07-01T15:38:31Z

Hey @bogdankostic I was already reviewing this in #3458 (comment) so I'll take this over

sjrl · 2026-07-02T07:41:53Z

+    Use inside a Haystack pipeline after a text embedder::
+
+        pipeline.add_component("embedder", SentenceTransformersTextEmbedder())
+        pipeline.add_component("retriever", Db2EmbeddingRetriever(
+            document_store=store, top_k=5
+        ))
+        pipeline.connect("embedder.embedding", "retriever.query_embedding")


Please wrap python code in code blocks.

Done! The example in the class docstring is now wrapped in a .. code-block:: python directive.

Sorry I didn't provide more context in my earlier comment but could we enclose with triple-backticks instead of .. code-block::. Our API doc strings don't support the .. code-block:: syntax.

…ct.toml

…ove obsolete test_from_dict_without_filter_policy test

CLAassistant · 2026-07-02T12:06:38Z

All committers have signed the CLA.

sjrl · 2026-07-02T12:14:42Z

Hey @priyanshu-krishnan1 and @GeetikaChughIBM thanks for the updates! @GeetikaChughIBM would it be possible for you to sign the CLA agreement as well? #3518 (comment)

* test: replace live DB fixture with mocked Db2DocumentStore in serialization tests * refactor: move util method unit tests to test_document_store.py and delete test_document_store_unit.py * Remove dead code: unreachable return in count_unique_metadata_by_filter_async * Fix typo: ____cause__ -> __cause__ in hasattr check * Move empty documents check after isinstance check in write_documents * Add missing LICENSE.txt (Apache 2.0) to ibm_db integration * Wrap Python example in proper code-block directive in Db2EmbeddingRetriever docstring * Convert run() docstring to Haystack :param:/:returns: style in Db2EmbeddingRetriever * Add missing __init__ docstring with :param:/:raises: to Db2EmbeddingRetriever * Add full :param:/:returns: docstring to run_async() in Db2EmbeddingRetriever * Simplify README to lightweight format following the conventions * Fix line length in run_async() docstring to pass fmt-check --------- Co-authored-by: priyanshu-krishnan1 <priyanshu.krishnan1@ibm.com>

…ecret for credentials * refactor(ibm_db): inline connection params and use Secret for credentials Remove the Db2ConnectionConfig dataclass and pass connection parameters directly to Db2DocumentStore.__init__(). username and password are now required Secret values (oracle-style) and are serialized as env-var references, never as plaintext. Tests use Secret.from_token in the live fixture; serialization tests build their own env-var-backed store. * refactor(ibm_db): rename Db2DocumentStore/Db2EmbeddingRetriever to IBMDb2 prefix Renames the public classes to IBMDb2DocumentStore and IBMDb2EmbeddingRetriever (and their test classes), updating imports, __all__ exports, serialization type strings, and docstrings.

…G.md (#10)

GeetikaChughIBM · 2026-07-03T13:02:33Z

Hey @priyanshu-krishnan1 thanks for opening the new PR! I've left an initial set of comments.

Also I noticed that the LICENSE.txt file is missing. Please add one. You can copy the one at the top-level of the repo which is here

Added LICENSE.txt copied from the top-level repo license (Apache 2.0).

GeetikaChughIBM · 2026-07-03T13:04:21Z

Hi @sjrl, we have addressed all the review comments, please provide your feedback.
Thanks!

sjrl · 2026-07-06T06:29:20Z

+    def test_write_documents(self, document_store: IBMDb2DocumentStore):
+        """Test basic write with duplicate handling - default policy is NONE."""
+        doc = Document(content="test doc")
+        assert document_store.write_documents([doc]) == 1
+        # Default policy is NONE — a second write of the same doc raises DuplicateDocumentError
+        with pytest.raises(DuplicateDocumentError):
+            document_store.write_documents([doc])


test_write_documents currently shadows WriteDocumentsTest.test_write_documents
from the mixin.

Could you rename it (e.g. test_write_documents_default_policy_none) so the mixin's version still runs?

I've renamed the original test_write_documents to test_write_documents_default_policy_none as suggested, and added a separate test_write_documents override to satisfy the mixin's required hook - it just asserts that a basic write returns 1. I think now the mixin's version runs, and the duplicate-rejection behaviour (DuplicateDocumentError) is now tested independently in test_write_documents_default_policy_none.

sjrl · 2026-07-06T06:55:47Z

+        database: str,
+        hostname: str,
+        username: Secret,
+        password: Secret,


It could be nice to set default values for these like

Suggested change

password: Secret,

username: Secret = Secret.from_env_var("DB2_USERNAME"),

password: Secret = Secret.from_env_var("DB2_PASSWORD"),

Both username and password now default to Secret.from_env_var("DB2_USERNAME") and Secret.from_env_var("DB2_PASSWORD") respectively.

sjrl · 2026-07-06T07:00:32Z

Hey @GeetikaChughIBM thanks for addressing all of the comments! I have a few minor ones left and then this is good to go!

sjrl · 2026-07-06T07:02:22Z

+
+    Use inside a Haystack pipeline after a text embedder:
+
+    .. code-block:: python


Our api docstrings don't support this type of markdown .. code-block:: python so could we switch to using the triple-backticks with the python tag?

Okay, switched to triple-backtick fenced code block (```python) in the docstring.

sjrl · 2026-07-06T07:02:52Z

+        :param query_embedding: Dense float vector from an embedder component.
+        :param filters: Runtime filters, merged with constructor filters according to filter_policy.
+        :param top_k: Override the constructor top_k for this call.
+        :returns: A dictionary with key ``documents`` containing a list of matching :class:`Document` objects.


Suggested change

:returns: A dictionary with key ``documents`` containing a list of matching :class:`Document` objects.

:returns: A dictionary with key `documents` containing a list of matching :class:`Document` objects.

Done.. changed to documents in the :returns: docstring.

sjrl · 2026-07-06T07:03:37Z

+    Booleans are stored in JSON as ``true``/``false`` and read back by ``JSON_VALUE`` as the
+    lowercase strings ``'true'``/``'false'``. ibm_db would otherwise bind a Python bool as the
+    integer ``1``/``0``, so the comparison would never match. Convert it explicitly.


Please only use single backticks for inline code comments

Suggested change

Booleans are stored in JSON as ``true``/``false`` and read back by ``JSON_VALUE`` as the

lowercase strings ``'true'``/``'false'``. ibm_db would otherwise bind a Python bool as the

integer ``1``/``0``, so the comparison would never match. Convert it explicitly.

Booleans are stored in JSON as `true`/`false` and read back by `JSON_VALUE` as the

lowercase strings `'true'`/`'false'`. ibm_db would otherwise bind a Python bool as the

integer `1`/`0`, so the comparison would never match. Convert it explicitly.

Done! now using single backticks for inline code comments

* fix: replace placeholder copyright with 'Copyright 2023-present deepset GmbH' in LICENSE.txt * fix: remove CHANGELOG.md * fix: rename test_write_documents to test_write_documents_default_policy_none to avoid shadowing mixin * fix: add default Secret env var values for username and password params * fix: replace code-block directive with triple-backtick markdown in docstring * fix: use single backticks for 'documents' key in :returns: docstring * fix: replace double backticks with single backticks in inline code comments in filters.py * fix: add test_write_documents mixin override and separate test_write_documents_default_policy_none

Added IBM Db2 Vector Store

840d23d

priyanshu-krishnan1 requested a review from a team as a code owner July 1, 2026 14:31

priyanshu-krishnan1 requested review from bogdankostic and removed request for a team July 1, 2026 14:31

github-actions Bot added topic:CI type:documentation Improvements or additions to documentation labels Jul 1, 2026

priyanshu-krishnan1 mentioned this pull request Jul 1, 2026

feat: add IBM Db2 Vector Store (ibm-haystack) #3458

Closed

sjrl requested review from sjrl and removed request for bogdankostic July 1, 2026 15:37

sjrl self-assigned this Jul 1, 2026