bug(agents): agent launcher fails --publish-image on transient registry push timeouts
維護者通常 1 天內回覆
還沒有人認領這個 Issue。
評估
研究方向
從 scripts/agents/image.sh 開始，檢查 deliver_agent_image；目前 ce push 遇到任何錯誤都會導致啟動失敗。選擇有界重試方案前，先檢查 launcher 現有的記錄和錯誤處理。完成條件是：暫時性故障會記錄每次嘗試並能夠復原，同時仍會列印以 digest 固定的映像參照；持續性故障仍可採取行動，且 sandbox 和設定維持不變。
由索引模型根據 Issue 內容生成。
描述
Related to #3469. Found while running the registry delivery path from #4744.
User Story
I run gator against a remote Kubernetes gateway, publishing the tooling image with scripts/agents/run.sh --publish-image. I directly hit push failures that a retry would have absorbed, so that one slow layer upload doesn't throw away a completed build.
Problem Statement
--publish-image runs one ce push and fails the launch on any error (scripts/agents/image.sh, deliver_agent_image: ce push "$image_tag" || fail "image push failed"). Large layer uploads to Artifact Registry intermittently fail with net/http: timeout awaiting response headers. Each failure aborts the launcher after the image is already built and some layers are already uploaded.
Impact / Why This Matters
The gator tooling image is about 450 MB compressed. Two consecutive --publish-image --build-only runs failed on a different large layer each time. A plain docker push of the same local tag then succeeded on its second attempt. That means rerunning the launcher by hand until it works, or pushing outside the launcher and resolving the digest manually. Doing it by hand also bypasses the launcher's digest check.
Acceptance Criteria
- A transient push failure is retried a bounded number of times before the launcher fails.
- Each retry is logged with the attempt number and the underlying error.
- After a successful retry, the launcher still resolves and prints the digest-pinned
Tooling image:reference. - A persistent failure (for example an authentication error) still fails with an actionable error and no sandbox, provider, or settings changes.
Reproduction Steps
- Configure Docker auth for an Artifact Registry repository the user can push to.
- On an arm64 Mac, run
DOCKER_DEFAULT_PLATFORM=linux/amd64 ./scripts/agents/run.sh --agent gator --gateway <gateway> --publish-image <region>-docker.pkg.dev/<project>/<repo>/gator:<tag> --build-only. - Observe intermittent
failed commit on ref "layer-sha256:…": … net/http: timeout awaiting response headersfollowed byerror: image push failed.
Environment
- OpenShell: launcher from #4744 (
c8afb7321); gateway0.1.6-dev.26+g229382e51 - OS: macOS (arm64), Docker with buildx emulating linux/amd64
- Runtime, deployment, or integration: Kubernetes compute driver on a managed cloud cluster; Artifact Registry
Suggested UX (if applicable)
openshell-agent-launcher: Publishing '…/gator:payload-13-7fa522fae'.
openshell-agent-launcher: push attempt 1/3 failed: timeout awaiting response headers; retrying in 5s
openshell-agent-launcher: Tooling image: …/gator@sha256:e88f5a66…
Logs
failed commit on ref "layer-sha256:eda61e0e…": failed to do request: Put "https://<region>-docker.pkg.dev/artifacts-uploads/…": net/http: timeout awaiting response headers
error: image push failed
- 主要語言
- Rust
- 星號
- 17.1k
- 分支
- 1.8k
- 平均合併
- 2 天 9 小時
- 30 天內合併 PR
- 355
環境準備
- 沒有 Dockerfile 或 Docker Compose 檔案
- 有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue，再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫，在一個分支上完成修改。
- 送出 Pull Request，並在描述裡引用這個 Issue 編號。
NVIDIA/OpenShell 的其他 Issue
-
bug: workspace access hint suggests a --subject command that fails for hyphen-leading OIDC subjects未關閉
難度 2/5 1-3 小時 新手友好度 78/100
維護者通常 1 天內回覆
-
state:triage-needed
難度 2/5 1-3 小時 新手友好度 82/100
維護者通常 1 天內回覆
-
bug(install): install.sh prints gawk "escape sequence \. treated as plain ." warnings on RPM hosts可能已有人在做 @ericcurtin 於 1 天前認領。 未關閉state:triage-needed
難度 2/5 1-3 小時 新手友好度 82/100
維護者通常 1 天內回覆
-
bug(sdk/go): browser login callback binds 127.0.0.1 but redirect_uri uses localhost可能重新可做 關聯的 PR 已關閉且未合併。 未關閉state:triage-needed
難度 2/5 1-3 小時 新手友好度 65/100
NVIDIA/OpenShell#4518 · 1 則留言 ·
維護者通常 1 天內回覆
-
state:triage-needed
難度 2/5 1-3 小時 新手友好度 62/100
維護者通常 1 天內回覆
相似的 Issue
-
checkpoint/verify/rebuild_indexes on a thread holding a WriteTransaction deadlock instead of INT-001未關閉bug known-issue
難度 2/5 1-3 小時 新手友好度 74/100
project-minigraf/minigraf#734 ·
維護者通常 1 天內回覆
-
難度 1/5 1 小時以內 新手友好度 88/100
維護者通常 4 天內回覆
-
難度 2/5 1-3 小時 新手友好度 70/100
維護者通常 1 天內回覆
-
bug CLI documentation
難度 2/5 1-3 小時 新手友好度 67/100
openai/codex#56630 · 3 則留言 · 3 個 reaction ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 64/100
DayuanJiang/PecoFence#139 ·
維護者通常 1 天內回覆