imbajin opened a new pull request, #836: URL: https://github.com/apache/hugegraph-toolchain/pull/836
Hubble could not complete the native Go Vermeer workflow because its load, task and result contracts differed from the Server proxy. An opt-in `vermeer.url` adapter now supports load/reload, nine compute forms, task status/filtering/cancel, graph cleanup and cursor-paged results. Native tasks use negative Hubble IDs; Server task IDs retain their meaning. An empty URL preserves the existing Dashboard/Server path. Native loading requires PD/Store mode; standalone mode stays disabled. Readiness uses a credential-scoped authenticated boolean endpoint rather than the global registered-worker list. Invalid native configuration retains Server tasks with a warning. Accepted loads retain their task ID per graph, including across graph switches, so queued work stays Loading until it finishes. Temporarily unavailable permitted integrations recover automatically as services start, then refresh graph metadata. Outages retain Server tasks with a retryable warning. Results preserve values across page boundaries and stop at an exactly full final page. Unsupported native history deletion is hidden. Authenticated direct access, including source-deletion cleanup, requires a super administrator because raw Store scans bypass Server label permissions. Anonymous access follows the Server mode. A protected deployment file supplies the separate Vermeer service token; user credentials and caller-supplied targets/output credentials are not persisted. Source deletion checks privately for a Vermeer copy; an existing copy requires authorized cleanup, and failures stop source deletion. Without a copy, Server's original deletion permissions apply. Standalone mode skips native cleanup entirely, and invalid drop confirmations are rejected before cleanup or deletion side effects. Validation: - Complete backend scan: **885 tests passed**; six focused frontend suites: **33 tests passed**. Checkstyle, frontend lint, i18n, TypeScript and production build passed. - Retained real PD/Store/Server/Master/Worker stack: all nine algorithms, authenticated fresh load with a separate Schema reader, 1001 results over 1000 + 1 distinct values, task cancellation, and source deletion confirmed absent from both Server and Vermeer. - Chromium: anonymous/admin/observer flows, results and pagination, outage warning, startup recovery without page reload, and an actual Worker paused during reload. Switching away and back retained the same queued task, blocked duplicate submission, and recovered without reload; page errors were zero. - Independent cross reviews verified the adapter, permissions and final lifecycle changes. Confirmed core review findings have been fixed and resolved. Repeated computations reuse a safe SHA-256 per-graph output basename instead of accumulating files per run. A real PageRank/degree sequence returned six values each with an unchanged output file set. Successful task metadata remains readable when temporary query values are unavailable; explicit result reads continue to report that condition. Checked-in development/browser helpers and six screenshots are documented in `hugegraph-hubble/docs/vermeer-development.md`. The smoke fixture repairs interrupted seeding without duplicate vertices/edges. Native task history currently reads at most 10,000 records because the upstream API has no cursor/offset. That limit is documented; its pagination follow-up thread remains open. Requires https://github.com/hugegraph/hugegraph-computer/pull/19 for PD credentials, deployment Schema credentials and initialization failure handling; the development helper pins its source revision. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> <!-- coderabbit-ai-generated:v1 --> ## Summary by CodeRabbit * **新功能** * 支持直接连接 Vermeer，进行图加载、算法计算及任务管理。 * 可查看 Vermeer 计算结果并翻页；加载中的图会显示相应状态。 * **改进** * Vermeer 暂不可用时仍可查看 Server 任务，并可重试；能力状态会自动重新检测。 * 增加 Vermeer 开发环境指南及浏览器冒烟验证工具。 <!-- end of auto-generated comment: release notes by coderabbit.ai --> Standalone native isolation also covers task history and explicit native task/result/cancel/unload operations. Graph-loading metadata updates no longer supersede a pending OLAP-mode write; graph selection and explicit retry retain mode refresh. Independent old/new regressions cover both fixes, and Chromium verifies a real mode PUT overlapping loading completion, with the fixture mode restored afterward. --- ## Upstream submission notes Source community PR: https://github.com/hugegraph/hugegraph-toolchain/pull/84. This upstream PR reuses the same source repository and branch. Validation statements above are reports from the source task; they were not independently rerun during upstream submission. At submission preflight, source-head hosted checks: 2 in_progress, 1 neutral, 1 queued, 6 skipped, 14 success. Upstream exact-head CI and current review feedback require separate confirmation; opening this PR does not establish merge or release readiness. Paired Vermeer implementation: https://github.com/apache/hugegraph-computer/pull/366. The dependency is on its implementation, not on a different base branch. Source review https://github.com/hugegraph/hugegraph-toolchain/pull/84#discussion_r4351511478 remains under investigation; no independent reproduction or repair is claimed here. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
