Worker aborts with V8 OOM in JSON.parse when several exec requests carry large stdin payloads; all leases lost

未关闭
  • 哪里不对？
#1,242 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
node.js, typescript
领域
api, backend

调研方向

Start at the worker and gateway handling for POST /v1/leases/:id/exec, then trace where the request body is materialized and parsed. The issue does not name files or tests, so locate the relevant entry points and existing request-size or concurrency limits first. Done means oversized or concurrent large stdin requests are rejected with a typed error instead of crashing the worker; streaming or preserving leases across restarts are broader alternatives.

由索引模型根据 Issue 内容生成。

描述

Summary

A simlock 0.5.0 worker (Node 22.23.4, macOS, gateway mode) aborts with a V8 out-of-memory error while parsing an incoming message with JSON.parse. This happens when several POST /v1/leases/<id>/exec requests carrying a large stdin (base64 payloads of tens of MB) are in flight at once. Every lease on that worker is lost: in-flight execs return 503 after about 105 s, and later calls get 404 UNKNOWN_LEASE.

Evidence

Crash report node-2026-11-03-144256.ips:

EXC_CRASH (SIGABRT) — abort() called
node::OOMErrorHandler
v8::internal::V8::FatalProcessOutOfMemory
v8::internal::Heap::CollectGarbage
v8::internal::HeapAllocator::AllocateRawWithRetryOrFailSlowPath
v8::internal::FactoryBase::NewRawStringWithMap<SeqOneByteString>
v8::internal::JsonParser<unsigned char>::MakeString
v8::internal::JsonParser<unsigned char>::Parse
v8::internal::Builtin_JsonParse
  • Worker log: the next start at 13:42:56Z logs Recovered stale daemon socket / Daemon started and re-establishes the gateway uplink. launchctl reports the last exit as -6.
  • Gateway log for one lease: device.exec → POST …/exec 503 after 105,242 ms → the client releases → POST …/exec 404 UNKNOWN_LEASE.
  • Workload: a tester-army e2e suite with 10 workers. Each worker uploads an ~89 MB fixture archive (deflated, base64 in stdin) through exec (simctl spawn <helper> untar) at the same moment. The host's load average was about 70.
  • The same burst also produced Cloudflare 524s and connection resets on other runs, because each exec holds a single long request.

Root cause

The worker (or its gateway uplink) materialises each exec request, stdin included, as one JSON string and parses it on the heap. Concurrent large stdin payloads exceed the V8 heap limit, and the process aborts instead of rejecting the request.

Suggested fix

  1. Bound the accepted exec stdin size per request, and the total in flight per worker. Reject an over-limit request with a typed error (413/PAYLOAD_TOO_LARGE) instead of crashing.
  2. Stream stdin through the uplink and into the child process instead of parsing it as a single JSON string, or offer a separate upload endpoint, so large file transfers don't sit on the JSON path.
  3. Optionally, keep lease state across a worker restart (the leases were lost even though the devices survived).

On the client side, we are capping concurrent large uploads (consumer issue linked below), but a worker should not be crashable by request size.

主要语言
TypeScript
星标
535
派生
21
平均合并
9 小时 53 分钟
30 天内合并 PR
435

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

callstackincubator/simlock 的其他 Issue

查看 callstackincubator/simlock 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。