Repository navigation
fix(director): tolerate slow GPU startup and recover lost devices - #136
Conversation
Keep float capability checks and the full aperture model unchanged. Record the initialization failures and bounded device-selection comparison. Refs: MF-106, MF-62
Keep resource objects unchanged and retain prior release history. Refs: MF-106
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: 9da97d9e-078c-414d-8380-a8c79da445e3) |
Refs: MF-106, PR-136
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: c830eae9-953c-437a-a73b-43f9a1036aac) |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f54794bb53
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Refs: MF-106, MF-62 Validation: 44 Director tests, type check, native slow-request cancellation and GPU-loss recovery
Refs: MF-106, MF-62 Validation: release mirrors and provenance witness
Refs: MF-106, MF-62 Validation: release contract and all resource objects unchanged
Shuang-su
changed the title
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: 705a2046-86c6-47e2-ac05-a165c2651c4a) |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 150c307ab8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: 54c5a537-dd17-45e1-82cc-939a096dab19) |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 43a2adc6b0
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
浏览器在电源或驱动状态变化后可能迟迟不返回 WebGPU 设备。Director 原来在20秒直接终止，设备丢失后也缺少保留编辑的恢复入口。此修改采用浏览器默认适配器策略，把20秒改为慢提示，继续等待同一请求；90秒无响应时提供明确错误、重试和对应资源Viewer入口。取消后迟到的设备会释放。
GPU丢失立即停止预览、输出和异步缩略图，显示“画面已暂停”。重新创建同一资源的设备保留工作机位、光圈、视频机位、播放位置、精细度及撤销历史；失败结束后清除过期的等待提示。WebGPU/RGBA32F能力、模型点数、分辨率、孔径和最终128／256／512采样要求保持。
验证：
Refs: MF-106，MF-62。唯一Spec/Plan/证据为docs/changes/106-acg-director/。
来源：完整产品候选3aecaba6b0a7242570c0b03c38fad6a331672b86；初始62d5936由rc1保留，慢启动b708555由rc2保留，最终完整候选由viewer-v5.21.9-rc3保留。5.21.9尚未生产发布；预览及正式SHA、部署和输出将在通过后回填。回退为5.21.8部署6ac87f543e88e78932eb4e4e。
Note
Medium Risk
Changes Director WebGPU initialization, cancellation, and GPU-loss recovery paths—user-visible if mishandled—but scope is limited to photography startup/runtime and does not alter Viewer SDK or final render contracts.
Overview
5.21.9 hardens Director WebGPU startup and recovery without changing photography quality (RGBA32F, aperture, sampling targets).
Graphics creation now uses the browser-default adapter (
powerPreference: "default") and a newwaitForGraphicsDeviceflow: 20s is only a slow-start hint with retry/Viewer actions, 90s is the real failure boundary, and cancel/retry aborts in-flight work and destroys late devices so stale GPUs cannot update the UI. Model loading and display callbacks honor the activeAbortSignal.On GPU loss, a shared
observeFailurehub stops preview/export work immediately; the app can recreate the device for the same resource and restore camera, project/shots, video timeline state, selection, and dirty flag. Error UI distinguishes “retry frame” vs “recreate graphics device,” and controls pause while the device is unusable.Release metadata (
data/index.json, version history, ledger, MF-106 docs) and metaflow-viewer 5.21.9 version mirrors are updated; new unit tests cover startup timers and device-loss observation.Reviewed by Cursor Bugbot for commit 43a2adc. Bugbot is set up for automated code reviews on this repo. Configure here.