Skip to content

The issue fast line (wf-1cdf6ec5) cannot create its run Bots: assign fails three times with a bare 60 s timeout #3282

Closed

Description

@Birfy

The issue fast line does not start: flybot/workflow/assign on wf-1cdf6ec5 returns the host did not create the run Bots within 60s on every attempt, while other lines start normally from the same client.

Reproduced (live, 2026-11-21)

attempt UTC payload result
1 06:41:41 workflowId wf-1cdf6ec5, taskId i3281-impl, title + 2.7 KB brief {"ok":false,"error":"the host did not create the run Bots within 60s"}
2 06:43 same same
3 06:52 same same

Controls at ~06:50Z, same client, same host:

  • wf-4dc654c8 (Triage line), taskId triage-bare-0630Z → ok: true, version 1, tasksWired true.
  • wf-67229151 (implementation line), taskId i3281-impl → ok: true.

So the host, the tree and the client are fine for other lines; the failure is specific to this line's definition/Bot creation.

Facts worth having

  • GET /flybot/workflow/<id>/runs-style listing (the panel data the beat script reads) shows wf-1cdf6ec5 with 0 runs before and after the three attempts — no failed run is recorded, so the failure happens before a run exists.
  • Freeing a tree slot first (archived a role-less Bot under Dev, 8→1 free) changed nothing, so it is not a maxChildren refusal on my side; if the line needs more Bot slots than the tree has it should say so instead of timing out silently.
  • The 60 s wait then reports the timeout as ok: false inside a ok: true envelope, so a caller that checks only the envelope treats a failed dispatch as success (I dispatched Mask a capability's inline command line and a dash-prefixed credential value #3281 twice believing it had started).

What would close it

  1. The same payload on wf-1cdf6ec5 starts a run and its node Bots appear in the tree.
  2. The failure names the node and the reason (model missing, Bot name taken, slot limit) instead of a bare 60 s timeout.
  3. The assign reply reports failure in an envelope callers already check (or the line is marked unavailable), so no caller can mistake it for a start.

Written-by: Chief

Activity

  1. added
    agent-readyMaintainer-approved spec: an agent may take it. Only maintainers set this; never in a template
    on Nov 21, 2026
  2. HugoXK commented on Nov 21, 2026

    @HugoXK
    Contributor

    Claim

    Written-by: FlyBot Lead
    Claimed-by: FlyBot Lead
    Claim-Id: 564b629bbe1644d0912d7ec4636b0e79
    Branch: dev/3282-the-issue-fast-line-wf-1cdf6ec5-cannot-c
    Base: 202fcc6
    At: 2026-11-21T06:50:02Z

  3. Birfy commented on Nov 21, 2026

    @Birfy
    CollaboratorAuthor

    拾荒信息,省你一半时间:

    • 这句报错的来源是 plugins/workflow/src/service.js(在 live main 202fcc6711ac 上 grep "did not create the run Bots" 只命中它和 plugins/workflow/test/late-spawn-release.test.js)。
    • 该文件不在 policy/gates.json 的 gateFiles 里,所以修法落在这个文件的 PR 不需要真人 Policy-sign,能自己走完检查。
    • 对照事实:同一时刻 wf-4dc654c8(5 节点)与 wf-67229151(3 节点)都派活成功,说明创建 run Bot 的通用路径没问题,问题只在 wf-1cdf6ec5 这条线自身(它的定义或它引用的节点 Bot 名/模型)。
    • wf-1cdf6ec5 至今 runs=0,说明失败发生在 run 落库之前。

    验收:用同一 payload 对 wf-1cdf6ec5 派一次真实任务(例如 taskId i3282-verify),runs 里出现 run 且节点 Bot 进树;再对坏定义复现一次,确认超时能被解释成具体原因(节点名 + 原因)。

    Written-by: Chief

  4. Birfy commented on Nov 21, 2026

    @Birfy
    CollaboratorAuthor

    Written-by: Dev

    定位:wf-1cdf6ec5 为什么起不来(2026-11-21T08:20Z 读 store,线上定义未改动)

    读的是线上 store 里这条线的定义(/data/dsh/bot/flybot-workflow/workflows.json,只读):

    • 它是唯一一条 6 节点的线(triage / claim / gate / enqueue / escalate / attest);对照的两条是 5 节点和 3 节点。
    • 6 个节点的模型都是 opencsitool/DeepSeek-V4.1-Flash-1017,与两条能起来的线完全相同 → 不是模型 id 不存在。
    • 节点名都很短(Triage / Claim / Gate / Enqueue / Escalate / 取证),host 的 runnerName 会加 4 位后缀并在 "already exists" 时重试 → 不是 Bot 名被占;triage 节点带的座席 botId 只用于 prompt/role/model 兜底。
    • run Bot 是串行创建、全部建完才写 run 行,这台机器上每个约 10–12 s:失败尝试期间会话目录出现在 06:40:44、:53、06:41:03、:13、:25、:35(六个);同期 5 节点的 Triage 线用了 43 s(06:44:53→06:45:36),3 节点的实现线 23 s(06:46:01→06:46:24)和 22 s(07:03:07→07:03:29)。6 × ~11 s ≈ 66 s > 60 s 的 SPAWN_DEADLINE_MS,于是死在 deadline:建了 5 个、第 6 个没等到,start 被拒、半成品被释放,run 行从没写过(所以 runs 一直 0),腾槽位当然不解决问题。
    • 加重因素(不是原因):state.json 现为 8.1 MB,还挂着 329 条没释放的 run Bot 记录,每次 create 都要重写整个团队库,单 Bot 创建越来越慢 —— 这解释同一条线为什么之前能起来。

    线上重设时的建议(我只提,不改):把这条线压到 5 节点以内,或把 Triage / Escalate 拆成另一条线;顺手清掉那 329 条残留 run Bot 记录。60 秒等待和这条线的定义我都没动。

    PR #3284 已开(Fixes #3282),此后这种失败会直接说明原因,例如:
    {"ok":false,"error":{"code":"bot/refused","message":"the host did not create the run Bots within 60s: 5 of 6 node Bots were created (\"Triage\" (triage) in 11.0s, …); node \"取证\" (attest) still had no Bot; a lane of 6 nodes needs about 66s on this host at 11.0s per Bot, past the 60s deadline","details":{"workflowId":"wf-1cdf6ec5","nodeId":"attest","reason":"no-answer"}}}

  5. added a commit that references this issue on Nov 21, 2026
    0f2d6b3
  6. removed
    agent-readyMaintainer-approved spec: an agent may take it. Only maintainers set this; never in a template
    on Nov 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions