feat: add `AsyncBigtableVectorStore` Class for the async-only vector value store implementation by mic-k3y · Pull Request #186 · googleapis/langchain-google-bigtable-python

mic-k3y · 2025-08-22T22:39:23Z

This PR adds the async-only vector store class that handles the underlying data operations for the main vector store class.

Thank you for opening a Pull Request! Before submitting your PR, there are a few things you can do to make sure it goes smoothly:

Make sure to open an issue as a bug/issue before writing your code! That way we can discuss the change, evaluate designs, and agree on the general idea
Ensure the tests and linter pass
Code coverage does not decrease (if any source code was changed)
Appropriate docs were updated (if necessary)

Fixes #184 🦕

…context Added the engine class that manages the client and execution context which handles the async and sync conversion via a background event loop. Added a test file for the engine class.

…tion Removes the option to pre-instantiate and pass a BigtableDataClientAsync to the BigtableEngine. The engine now creates its own client instance, making it more self-contained.

Adjusted test cases and assertions to align with the recent modifications in the BigtableEngine class API and behavior.

This change introduces BigtableEngine.async_initialize for non-blocking asynchronous initialization.

…n LangChain's key-value store. This commit introduces the `AsyncBigtableByteStore` class. This class is designed to handle setting, getting, deleting, and yielding keys asynchronously. It will be used by the main public-facing `BigtableByteStore` for all the data operations on the table. It also includes a test suite.

The `shutdown_default_loop` method was not closing the asyncio event loop, only that class level variables were reassigned. This commit adds a line that closes the loop when the thread is successfully terminated.

# Conflicts: # src/langchain_google_bigtable/engine.py # tests/test_engine.py

The `shutdown_default_loop` method was not closing the asyncio event loop, only that class level variables were reassigned. This commit adds a line that closes the loop when the thread is successfully terminated.

…context (googleapis#163) * feat: add BigtableEngine Class for managing the client and execution context Added the engine class that manages the client and execution context which handles the async and sync conversion via a background event loop. Added a test file for the engine class. * fix(engine): update BigtableEngine setup by internalizing client creation Removes the option to pre-instantiate and pass a BigtableDataClientAsync to the BigtableEngine. The engine now creates its own client instance, making it more self-contained. * test(engine): update tests to match BigtableEngine class changes Adjusted test cases and assertions to align with the recent modifications in the BigtableEngine class API and behavior. * style(engine): update test_engine.py and engine.py format * test(engine): add tests for async_initialize workflow * feat(engine): add async_initialize factory and shared loop management This change introduces BigtableEngine.async_initialize for non-blocking asynchronous initialization. * style(engine): update test_engine.py

# Conflicts: # src/langchain_google_bigtable/async_key_value_store.py

…lementation. This class will be used for interacting with LangChain's key-value store for Bigtable. It contains methods that add, delete, get, and yield key-value pairs from the store. It's an implementation of LangChain's BaseStore(ByteStore) abstract interface. This commit also contains comprehensive test suites. It also contains a change for the table creation fixture for the engine test, simplifying it from previous approach.

… raises error. address error handling test compatibility for Python 3.9

ad548 · 2025-08-26T16:57:45Z

docs/key_value_store.ipynb

      "cell_type": "code",
      "source": [
-        "# Embedded document/text have been added to the store\n",
+        "# Embedded document/text have been added to the store. Hashed key for the raw text is used as the key.\n",


is this hashing done by the cached embedding class?

Yes. The hashing is done by the cached embedding class.

ad548 · 2025-08-26T16:59:56Z

src/langchain_google_bigtable/async_vector_store.py

+            embedding_service (Embeddings): The embedding service to use.
+            content_column (ColumnConfig): ColumnConfig for document content.
+            embedding_column (ColumnConfig): ColumnConfig for vector embeddings.
+            collection (Optional[str]): The name of the collection (optional). It is used as a prefix for row keys.


I thought we wanted to this not be optional?

The collection is forced in the main store class and is not optional there, which would force the end user to pass the collection string. I put this here as optional for flexibility in case there are any changes. Should I make the argument required here as well?

I just saw the comment about the collection filter and made the change. The collection filter is now a required argument and removed the block of code below.

ad548 · 2025-08-26T17:00:45Z

src/langchain_google_bigtable/async_vector_store.py

+            content_column (ColumnConfig): ColumnConfig for document content.
+            embedding_column (ColumnConfig): ColumnConfig for vector embeddings.
+            collection (Optional[str]): The name of the collection (optional). It is used as a prefix for row keys.
+            metadata_as_json_column (Optional[ColumnConfig]): ColumnConfig for metadata as JSON column.


can you explain this field a bit more

Yes. This puts all the metadata in one serialized JSON object and stores it in the column defined by the user here. This is helpful for accessing all the metadata for a document at once if there is any need for it and can be used to store any other metadata not defined in metadata mappings. It is used to retrieve the metadatas at once in the aget_by_ids method. It is not used for filtering or in the btql query string however.

Sg. Maybe expand the function comment a bit more.

ad548 · 2025-08-26T17:07:58Z

src/langchain_google_bigtable/async_vector_store.py

+        encoding (Encoding): The data encoding to use for the column's value.
+    """
+
+    column_family: str


the family could default to METADATA_COLUMN_FAMILY here

ad548 · 2025-08-26T17:09:17Z

src/langchain_google_bigtable/async_vector_store.py

+    def __init__(
+        self,
+        metadata_key: str,
+        encoding: Encoding,


why not take in the ColumnConfig as the parameter instead of encoding and column_qualifier separately

This is used to enforce all the metadatas to be stored under the METADATA_COLUMN_FAMILY("md"). If we allowed the user to put in the column config, we would also be giving them the option to have a different column family for the metadatas which might be different from the column family "md" internally being used in the class for filtering and retrieval.

Hmm I see. Does it need to inherit from ColumnConfig then? Not necessarily opposed to it but just wondering about the choice.

can you also expand the function comment to clarify that cf is always md

Done. I mean it doesn't need to inherit ColumnConfig, but it might be useful to let the user know that each metadata mappings are individual columns and not just regular mappings.

ad548 · 2025-08-26T17:32:38Z

src/langchain_google_bigtable/async_vector_store.py

+
+    async def asimilarity_search_with_score_by_vector(
+        self,
+        embedding: List[float],


this embedding parameter is really the query vector? Can you using that naming consistently throughout?

Yes. It is the query vector. The problem is the vector store abstract interface expects the naming to be 'embedding' for these methods. It also has the 'embedding' argument for the factory methods "from_texts, .." to be the embedding service which is misleading.

I changed all the "query_vector" arguments to "embedding" even though it might be confusing, but better than other overriding the abstract class.

Okay, just make sure to clarify it as the 'embedding query vector' in the comments.

ad548 · 2025-08-26T17:35:23Z

src/langchain_google_bigtable/async_vector_store.py

+        Args:
+            query (str): The text to find relevant and diverse documents for.
+            k (int): Number of documents to return. Defaults to 4.
+            fetch_k (int): Number of documents to fetch to pass to MMR algorithm.


maybe there should be a check for fetch_k should be greater than k

Good point. I added a check.

ad548 · 2025-08-26T18:01:59Z

src/langchain_google_bigtable/async_vector_store.py

+                f"(STARTS_WITH(_key, @{param_key})) {new_line}"
+            )
+            param_count += 1
+            if filters and "collectionFilter" in filters:


how does the user provide a row key prefix filter? something like "collection:phone#"?

Yes. It would be something like = {"collectionFilter": "mydocs:phone#"} for filtering in collection "mydocs" and row key prefix "mydocs:phone#" since every row for that collection has "my docs:" as the prefix.

If the user didn't pass the collection filter, by default the store will filter the table by row key prefix "{collection name}:"

ad548 · 2025-08-26T18:08:51Z

src/langchain_google_bigtable/async_vector_store.py

+
+    def _process_value_filters(
+        self,
+        filter_dict: Dict[str, Any],


It would be good to provide some sort of structure/strong typing to this FilterDict. We could use a TypeAlias here. Maybe something like this

FilterType: TypeAlias = Literal[ "RowKeyPrefix",
"Qualifiers",
"ColumnQualifierPrefix",
"ColumnQualifierRegex",]

OperatorType: TypeAlias = Literal['==', '>', in', 'nin']

OperatorDict = Dict[str, [OperatorType, str]]

A forward reference (string literal "FilterDict") is used for recursion.
FilterDict: TypeAlias
FilterDict = Dict[
FilterType,
Union[
OperatorDict, # For a metadata key like "color": {"==": "blue"}
"FilterDict" # For nesting with "QualifierChainFilter" or "QualifierUnionFilter"
]
]

def _process_value_filters(
self,
filter_dict: FilterDict,
... other parameters
):
... implementation
pass

also leaning towards ColumnValueChainFilter instead of QualifierChainFilter. I think it's useful to have the term "Value there"

Done. I also changed the chain filter names to ColumnValueChainFilter and ColumnValueUnionFilter. I will change the names in the docs as well.

ad548 · 2025-08-26T18:10:45Z

src/langchain_google_bigtable/async_vector_store.py

+                # User provided a 'collectionPrefix' filter, overriding the default.
+                # Update the parameter with the user-specified prefix.
+                params[param_key] = filters["collectionFilter"].encode("utf-8")
+        if not self.collection and filters and "collectionFilter" in filters:


I think we should simplify all this by enforcing a collection during instantiation

ad548 · 2025-08-27T01:45:17Z

src/langchain_google_bigtable/async_vector_store.py

+
+# Represents the top-level filter dictionary that users can provide.
+FilterDict: TypeAlias = Dict[
+    Literal["collectionFilter", "metadataFilter"],


can we not provide both at the same time? could you update the tests to support both? Also since we want to support collection + remaining row key prefix, I'd just call this rowKeyFilter and clarify that collection is always present by default

Done. Both can be provided at the same time. It will be something like
filters={"rowKeyFilter": "...", Qualifier Filters, Value and Logic Filters}

instead of partitioning the filter into collectionFilter and metadataFilter at the top-level we had before like

filters={"collectionFilter": ...., "metadataFilter": {Qualifer Filters, Value and Logic Filters} }.

I have also updated the tests to support both at the same time and in the same dictionary.

ad548

Last few comments.

ad548 · 2025-08-27T16:29:43Z

src/langchain_google_bigtable/async_vector_store.py

+# Represents the top-level filter dictionary. It can contain a `rowKeyFilter`
+# to specify a row key prefix that follows the collection prefix by default.
+# It also supports high-level qualifier filters and nested value-based filters.
+FilterDict: TypeAlias = Dict[str, Union[str, List[str], OperatorDict, ValueFilterDict]]


could the keys here and in ValueFilterDict be constrained to some literal values?

It is possible, but we expect the user to have different metadata keys. Constraining the keys here might constrain the metadata keys the user can filter on.

ad548 · 2025-08-27T16:30:46Z

src/langchain_google_bigtable/async_vector_store.py

+    def __init__(
+        self,
+        metadata_key: str,
+        encoding: Encoding,


Hmm I see. Does it need to inherit from ColumnConfig then? Not necessarily opposed to it but just wondering about the choice.

ad548 · 2025-08-27T16:31:22Z

src/langchain_google_bigtable/async_vector_store.py

+    def __init__(
+        self,
+        metadata_key: str,
+        encoding: Encoding,


can you also expand the function comment to clarify that cf is always md

ad548 · 2025-08-27T16:31:59Z

src/langchain_google_bigtable/async_vector_store.py

+            content_column (ColumnConfig): ColumnConfig for document content.
+            embedding_column (ColumnConfig): ColumnConfig for vector embeddings.
+            collection (Optional[str]): The name of the collection (optional). It is used as a prefix for row keys.
+            metadata_as_json_column (Optional[ColumnConfig]): ColumnConfig for metadata as JSON column.


Sg. Maybe expand the function comment a bit more.

ad548 · 2025-08-27T16:33:46Z

src/langchain_google_bigtable/async_vector_store.py

+        param_count += 1
+
+        # Process top-level qualifier filters from the flattened structure.
+        if "Qualifiers" in remaining_filters:


maybe ColumnQualifiers to keep it consistent?

ad548 · 2025-08-27T16:34:46Z

src/langchain_google_bigtable/async_vector_store.py

+        row_key_prefix = f"{self.collection}:"
+
+        # If a 'rowKeyFilter' is provided, append it to the collection prefix.
+        if "rowKeyFilter" in remaining_filters:


nit: "RowKeyFiler" - pascal case to keep it consistent with others

ad548 · 2025-08-27T16:36:54Z

src/langchain_google_bigtable/async_vector_store.py

+
+    async def asimilarity_search_with_score_by_vector(
+        self,
+        embedding: List[float],


Okay, just make sure to clarify it as the 'embedding query vector' in the comments.

ad548 · 2025-08-27T16:40:26Z

src/langchain_google_bigtable/async_vector_store.py

+        if final_query_params.distance_strategy == DistanceStrategy.COSINE:
+            relevance_score_fn = self._cosine_relevance_score_fn
+        else:  # If distance to use not declared as COSINE, uses EUCLIDEAN by default
+            relevance_score_fn = self._euclidean_relevance_score_fn


where is self._euclidean_relevance_score_fn defined?

The class inherits those methods from the abstract VectorStore class.

ad548 · 2025-08-27T16:48:56Z

tests/test_btql_query_builder.py

+
+    def test_filter_equal(self, store: AsyncBigtableVectorStore) -> None:
+        """Tests the '==' (equal) operator and verifies UTF8 encoding."""
+        params = QueryParameters(filters={"color": {"==": "red"}})


I wonder if these standalone column value filters should be in there own "ColumnValueFilter" dict? Maybe that's what's preventing us from having the dict keys be constrained to certain literals? Let me know if that makes it more complex or something.

Yeah. It's those standalone column value filters preventing us from having the dict keys be constrained to certain literals. I delegated the ColumnValueFilter dict translation task to the method we already have for value filters(_process_value_filters) and rewrote the tests for this change. Now, I have literals for the top-level filters dict.

This works!

ad548 · 2025-08-27T19:48:26Z

tests/test_async_vector_store_core.py

+        assert len(results_cosine) == 1
+        doc, score = results_cosine[0]
+        assert isinstance(doc, Document)
+        assert doc.page_content == "a document about cats"


maybe add some test cases where text is not exactly the same as the text

I can't add some test cases where text is not exactly the same as the text unless I use a real embedding model. I am using the DeterministicFakeEmbedding class from LangChain used for testing that generates a random embedding and hashes it for that same text.

ad548 · 2025-08-27T20:00:19Z

tests/test_async_vector_store_core.py

+        self, store: AsyncBigtableVectorStore
+    ) -> None:
+        """Tests amax_marginal_relevance_search returns the correct top result."""
+        texts = ["foo", "bar", "baz", "boo"]


same here but maybe test data with some redundant values and make sure top results are varied. like foo, foo2, foo3, bar, baz etc

…lementation (googleapis#176) This class will be used for interacting with LangChain's key-value store for Bigtable. It contains methods that add, delete, get, and yield key-value pairs from the store. It's an implementation of LangChain's BaseStore(ByteStore) abstract interface. This commit also contains comprehensive test suites. It also contains a change for the table creation fixture for the engine test, simplifying it from previous approach. chore(main): release 0.5.0 (googleapis#167) Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com> feat: add playground notebook for key-value store (googleapis#179) * feat: add playground notebook for key-value store This commit adds a notebook for how to use key-value stores. * fix: update embedding model used in key_value_store notebook This commit also updates the embedding model used for playground notebook. * fix: update readme.rst file * fix: address PR comments for the notebook chore(main): release 0.6.0 (googleapis#182) Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com> initial get_table and execute_query implementation feat: add `AsyncBigtableVectorStore` Class for the async-only vector value store implementation (googleapis#186) * feat: add AsyncBigtableVectorStore Class for the async-only vector value store implementation This commit adds the async-only vector store class that handles the underlying data operations for the main vector store class along with its test files. * fix: include comments on the key-value store usage cells Included comments explaining how the embedding cache key-value store works. * fix: include comments on the key-value store usage cells Included comments explaining how the embedding cache key-value store works. * fix: header date for async_vector_store.py fix: include comments on the key-value store usage cells Included comments explaining how the embedding cache key-value store works. * fix: typing for tests and async vector store class fix: add os environment function for the tests * fix: isort for imports for test files * fix: revert removed test_async_key_value_store.py file. * fix: address PR comments * fix: address PR comments * fix: address PR comments * fix: address PR comments * fix: address PR comments * fix: address PR comments * fix: address PR comments * fix: add check to make sure each document has an ID. feat: add `BigtableVectorStore` Class for LangChain Vector Store integration (googleapis#189) * feat: add `BigtableVectorStore` class for Bigtable <> LangChain Vector Store integration. * feat: add `BigtableVectorStore` class for Bigtable <> LangChain Vector Store integration. * fix: isort imports for test file. * fix: fixture return types. * chore: add more test cases * fix: remove await from sync store test suite * fix: reformat test_vector_store.py * fix: address PR comments. chore(main): release 0.7.0 (googleapis#187) Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com> feat: add playground notebook for the vector store (googleapis#191) * feat: add BigtableVectorStore Playground notebook. * feat: add BigtableVectorStore Playground notebook. fix: update vector_store.ipynb * fix: update vector_store.ipynb * fix: update vector_store.ipynb * fix: update vector_store.ipynb * feat: change documentation for the library feat: add BigtableVectorStore Playground notebook. fix: make sure the typing_extensions is supported across all python versions supported by this library feat: change documentation for the library fix: update vector_store.ipynb fix: update vector_store.ipynb fix: update vector_store.ipynb feat: add BigtableVectorStore Playground notebook. fix: update vector_store.ipynb * feat: update documentation for the library feat: update documentation for the library * feat: update documentation for the library feat: update documentation for the library feat: update documentation for the library * feat: update documentation for the library * fix: reformat vector_store.ipynb * fix: address PR comments * fix: address PR comments. * fix: address PR comments. chore(main): release 0.8.0 (googleapis#192) Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com> resolve merge conflict with upstream

mic-k3y and others added 30 commits July 17, 2025 16:51

feat: add BigtableEngine Class for managing the client and execution …

cd06327

…context Added the engine class that manages the client and execution context which handles the async and sync conversion via a background event loop. Added a test file for the engine class.

Merge branch 'main' into main

1608536

fix(engine): update BigtableEngine setup by internalizing client crea…

8d5ba9e

…tion Removes the option to pre-instantiate and pass a BigtableDataClientAsync to the BigtableEngine. The engine now creates its own client instance, making it more self-contained.

test(engine): update tests to match BigtableEngine class changes

4c96875

Adjusted test cases and assertions to align with the recent modifications in the BigtableEngine class API and behavior.

style(engine): update test_engine.py and engine.py format

99aca75

test(engine): add tests for async_initialize workflow

bd5391a

feat(engine): add async_initialize factory and shared loop management

f905d91

This change introduces BigtableEngine.async_initialize for non-blocking asynchronous initialization.

style(engine): update test_engine.py

9df570c

fix: Ensure the loop is closed in BigtableEngine.shutdown_default_loop

cea2777

The `shutdown_default_loop` method was not closing the asyncio event loop, only that class level variables were reassigned. This commit adds a line that closes the loop when the thread is successfully terminated.

Merge remote-tracking branch 'origin/main'

75355ed

# Conflicts: # src/langchain_google_bigtable/engine.py # tests/test_engine.py

fix: ensure the loop is closed in BigtableEngine.shutdown_default_loop

2f670fd

The `shutdown_default_loop` method was not closing the asyncio event loop, only that class level variables were reassigned. This commit adds a line that closes the loop when the thread is successfully terminated.

Merge remote-tracking branch 'origin/main'

a5a13d0

fix: import MutationsExceptionGroup and FailedMutationEntryError

f096c4e

fix: import MutationsExceptionGroup and FailedMutationEntryError

5471c4b

Merge remote-tracking branch 'origin/main'

b5c6356

# Conflicts: # src/langchain_google_bigtable/async_key_value_store.py

style: remove type from variable loop

17366b1

fix: add test case for testing mutation errors

c32f600

fix: add test case for testing mutation errors

d6a49b6

fix: add test case for testing mutation errors

bc6fa2e

fix: address comments left in this PR

ca65302

Merge branch 'main' into main

1eef0c2

Merge remote-tracking branch 'origin/main'

ba2bfbb

fix: move the key_value_store.py file into the src directory.

27aacef

Merge branch 'googleapis:main' into main

5a0d628

chore: add key value store classes and functions to __init__.py

c096afd

Merge remote-tracking branch 'origin/main'

3ea516b

feat(test): add test case for testing operating on non-existent table…

e8c12a2

… raises error. address error handling test compatibility for Python 3.9

mic-k3y requested a review from a team August 22, 2025 22:50

blunderbuss-gcf bot assigned georgecma Aug 22, 2025

mic-k3y requested review from ad548 and georgecma August 22, 2025 22:50

mic-k3y assigned mic-k3y and unassigned georgecma Aug 22, 2025

fix: revert removed test_async_key_value_store.py file.

dcb6ff0

mic-k3y requested review from ad548 and removed request for ad548 August 26, 2025 15:19

ad548 reviewed Aug 26, 2025

View reviewed changes

mic-k3y added 2 commits August 26, 2025 21:58

fix: address PR comments

3732d3e

fix: address PR comments

016939f

mic-k3y requested a review from ad548 August 26, 2025 22:32

mic-k3y added 2 commits August 26, 2025 22:41

fix: address PR comments

ec37b27

fix: address PR comments

647bcf6

ad548 reviewed Aug 27, 2025

View reviewed changes

mic-k3y added 2 commits August 27, 2025 12:06

fix: address PR comments

d918063

fix: address PR comments

6830d4e

mic-k3y requested a review from ad548 August 27, 2025 12:30

ad548 reviewed Aug 27, 2025

View reviewed changes

fix: address PR comments

12282f4

mic-k3y requested a review from ad548 August 27, 2025 19:41

ad548 previously approved these changes Aug 27, 2025

View reviewed changes

fix: add check to make sure each document has an ID.

feddc5d

mic-k3y dismissed ad548’s stale review via feddc5d August 27, 2025 20:02

mic-k3y requested a review from ad548 August 27, 2025 20:09

ad548 approved these changes Aug 27, 2025

View reviewed changes

mic-k3y merged commit 18d4a33 into googleapis:main Aug 27, 2025
11 checks passed

release-please bot mentioned this pull request Aug 27, 2025

chore(main): release 0.7.0 #187

Merged

Conversation

mic-k3y commented Aug 22, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mic-k3y Aug 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mic-k3y Aug 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mic-k3y Aug 26, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mic-k3y Aug 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ad548 left a comment

Choose a reason for hiding this comment

Uh oh!

mic-k3y Aug 26, 2025 •

edited

Loading

mic-k3y Aug 26, 2025 •

edited

Loading

mic-k3y Aug 26, 2025 •

edited

Loading

mic-k3y Aug 27, 2025 •

edited

Loading