nav/Dirigent
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: wait for gateway ready before post-install model validation #2

Merged
hzhang merged 2 commits from :feat/whispergate-mvp into feat/whispergate-mvp 2026-02-26 11:15:14 +00:00
Conversation 0 Commits 2 Files Changed 4 +61 -37
zhi commented 2026-02-26 08:46:03 +00:00
Collaborator
Copy Link

问题

安装脚本在 gateway restart 后立即执行 model 可见性校验，但 gateway 重启是异步的（systemd），导致校验时新进程还没初始化完成，models list 看不到刚注册的 whisper-gateway/no-reply，触发 rollback。

这是 HANDOFF 文档中 P0/P1 问题的根因。

修复

  1. waitForGatewayReady() — 重启后轮询 openclaw gateway status，等 RPC probe ok 再继续（最多30s）
  2. validateNoReplyModelAvailable() 加重试 — 5次尝试，每次间隔2s，应对 model catalog 刷新延迟
  3. CONFIG.example.json — contextWindow: 4096 → 200000，maxTokens: 64 → 8192（OpenClaw 要求最低 16000）

测试

  • node --check scripts/install-whispergate-openclaw.mjs ✅
  • 手动验证：写入 provider 配置后 models list 确实需要 gateway 重启才刷新，证实竞态条件存在
## 问题 安装脚本在 `gateway restart` 后立即执行 model 可见性校验，但 gateway 重启是异步的（systemd），导致校验时新进程还没初始化完成，`models list` 看不到刚注册的 `whisper-gateway/no-reply`，触发 rollback。 这是 HANDOFF 文档中 P0/P1 问题的根因。 ## 修复 1. **`waitForGatewayReady()`** — 重启后轮询 `openclaw gateway status`，等 RPC probe ok 再继续（最多30s） 2. **`validateNoReplyModelAvailable()` 加重试** — 5次尝试，每次间隔2s，应对 model catalog 刷新延迟 3. **`CONFIG.example.json`** — `contextWindow: 4096 → 200000`，`maxTokens: 64 → 8192`（OpenClaw 要求最低 16000） ## 测试 - `node --check scripts/install-whispergate-openclaw.mjs` ✅ - 手动验证：写入 provider 配置后 `models list` 确实需要 gateway 重启才刷新，证实竞态条件存在
zhi added 1 commit 2026-02-26 08:46:03 +00:00
fix: wait for gateway ready before post-install model validation de04e21aa1 
Root cause: gateway restart is async (systemd), but validateNoReplyModelAvailable()
ran immediately after, hitting a race condition where the new gateway process
hadn't finished initializing yet. This caused 'model not listed' validation
failures, triggering config rollback even though the config was correct.

Changes:
- Add waitForGatewayReady() that polls 'openclaw gateway status' for RPC probe
- Add retry loop (5 attempts, 2s interval) to validateNoReplyModelAvailable()
- Fix CONFIG.example.json: contextWindow 4096->200000, maxTokens 64->8192
  (OpenClaw requires minimum 16000 contextWindow)
zhi force-pushed feat/whispergate-mvp from de04e21aa1 to fd6c4dd3a2 2026-02-26 08:47:48 +00:00 Compare
zhi added 1 commit 2026-02-26 08:56:29 +00:00
fix: extract senderId from event.prompt instead of ctx in hooks f33dc13af4 
Root cause: PluginHookAgentContext in before_model_resolve only has
agentId, sessionKey, sessionId, workspaceDir, messageProvider.
senderId, channelId, input are NOT available in this hook phase.

The plugin was reading ctx.senderId (undefined) -> inHumanList=false
for ALL Discord sessions -> shouldUseNoReply=true -> all silenced.

Fix: use event.prompt which contains the full user message including
the 'Conversation info (untrusted metadata)' JSON block, and extract
sender_id from there. Same fix applied to before_prompt_build.
hzhang merged commit 6b3d89634a into feat/whispergate-mvp 2026-02-26 11:15:14 +00:00
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
No items
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
orion zhi nav
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: nav/Dirigent#2
Delete Branch ":feat/whispergate-mvp"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?