[Bug]: RedisKVStore breaks with decode_responses=True

[chas-mafli]

Bug Description

It was fixed here #12324 but then about a month later someone reverted the fix for some reason #13201

Version

0.14.22

Steps to Reproduce

from llama_index.storage.kvstore.redis import RedisKVStore
from redis.asyncio import Redis as AsyncRedis
from redis import ConnectionPool
from llama_index.core.ingestion import DocstoreStrategy, IngestionPipeline

connection_pool = ConnectionPool(host=...,
                                 port=..., 
                                 decode_responses=True)

redis_client = Redis(connection_pool=connection_pool)

redis_kvstore = RedisKVStore(redis_client=redis_client, async_redis_client=AsyncRedis.from_pool(connection_pool))

docstore = RedisDocumentStore(redis_kvstore)

pipeline = IngestionPipeline(
    docstore=docstore,
    docstore_strategy=DocstoreStrategy.DUPLICATES_ONLY,
    vector_store=vector_store,
)

pipeline.run(documents=[...])

Relevant Logs/Tracebacks

Traceback (most recent call last):
  File "/opt/.../.../bgtask.py", line 117, in run
    func(*args, **kwargs)
    ~~~~^^^^^^^^^^^^^^^^^
  File "/opt/.../.../oden.py", line 423, in _ingest
    ingest(url, ns, fromDate)
    ~~~~~~^^^^^^^^^^^^^^^^^^^
  File "/opt/.../.../oden.py", line 241, in ingest
    ingest_documents(redis_client, vector_store, access_token, url, ns, fromDate)
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/.../.../oden.py", line 293, in ingest_documents
    pipeline.run(documents=[doc])
    ~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "/opt/.../.venv/lib/python3.13/site-packages/llama_index_instrumentation/dispatcher.py", line 413, in wrapper
    result = func(*args, **kwargs)
  File "/opt/.../.venv/lib/python3.13/site-packages/llama_index/core/ingestion/pipeline.py", line 598, in run
    nodes_to_run = self._handle_duplicates(input_nodes)
  File "/opt/.../.venv/lib/python3.13/site-packages/llama_index/core/ingestion/pipeline.py", line 458, in _handle_duplicates
    existing_hashes = self.docstore.get_all_document_hashes()
  File "/opt/.../.venv/lib/python3.13/site-packages/llama_index/core/storage/docstore/keyval_docstore.py", line 656, in get_all_document_hashes
    self._kvstore.get_all(collection=self._metadata_collection)
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/.../.venv/lib/python3.13/site-packages/llama_index/storage/kvstore/redis/base.py", line 161, in get_all
    collection_kv_dict[key.decode()] = value
                       ^^^^^^^^^^
AttributeError: 'str' object has no attribute 'decode'. Did you mean: 'encode'?

[dosubot]

Hi @chas-mafli! I'm Dosu and I'm helping the LlamaIndex team.

I can confirm the bug. The current get_all and aget_all methods in RedisKVStore unconditionally call key.decode() on lines 161 and 169, which fails when the Redis client is configured with decode_responses=True (since keys are already strings in that case) [1].

A straightforward fix would be to guard the .decode() call with an isinstance check:

collection_kv_dict[key.decode() if isinstance(key, bytes) else key] = value

This needs to be applied in both get_all (line 161) and aget_all (line 169) in base.py.

As a workaround until this is fixed, you can avoid passing decode_responses=True in your ConnectionPool configuration — the default (False) works correctly with the current implementation.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.