Implement Dioptra agent fleet dashboard - #1
Conversation
Vite/React UI polls status.json for k8s nodes, Cursor/Hermes agents, infra topology, and GitHub/GitLab issue backlogs (top 10 by priority). Add publish-status scraper, nginx GET-only container, and CI.
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
Serve GET /api/status from a collector cache. Require X-Dioptra-Site, allowlisted Origin/Referer, and per-IP rate limits. Wire the SPA and nginx/compose so only the Dioptra site is meant to consume the API.
Split Fleet (live status) from Setup (how agents work): Coder template desktop spaces, tools/resources, outside router vs inside vLLM, and fleet wiring diagram moved onto the Setup tab.
Show outer lifecycle plus observe/plan/act/verify cycle, tool micro-loop, CI/review re-entry, and blocked/handoff exits on the Setup tab.
Create dk-raas/dkai/tools/dioptra mirror path, GitHub→GitLab mirror workflow, Harbor publish CI, and prd-apps k8s manifests for web+api.
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Switch to reusable-gitlab-mirror-push with secrets: inherit so the shared DataKnifeAI org mirror token is used; drop per-repo secret docs.
Document shared DataKnifeAI org mirror token and call reusable-gitlab-mirror-push from ci.yml on main.
Use Node fetch for collectors (no curl/gh in API image), await shared refresh, omit HEAD bodies, count online agents consistently, use GitHub/GitLab total counts, and deploy SHA tags without applying secret.example.yaml.
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using high effort and found 4 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit ae16c89. Configure here.
| platform: "gitlab", | ||
| totalOpen: 0, | ||
| top: [], | ||
| }; |
There was a problem hiding this comment.
GitHub error blocks GitLab
Medium Severity
buildStatus() always awaits GitHub issue collection first; if that throws, the whole refresh fails and GitLab is never queried. GitLab backlog and summary counts stop updating during GitHub outages or rate limits even when GitLab credentials and the API are healthy.
Reviewed by Cursor Bugbot for commit ae16c89. Configure here.
|
|
||
| export function readJson(path, fallback) { | ||
| if (!path || !existsSync(path)) return fallback; | ||
| return JSON.parse(readFileSync(path, "utf8")); |
There was a problem hiding this comment.
Invalid JSON crashes refresh
Medium Severity
readJson() calls JSON.parse with no error handling. A malformed DIOPTRA_NODES_JSON or DIOPTRA_AGENTS_JSON file causes every buildStatus() refresh to throw, leaving cache.error set and blocking updates until the file is fixed (initial seed load at API startup can crash the process the same way).
Additional Locations (1)
Reviewed by Cursor Bugbot for commit ae16c89. Configure here.
| void refresh(); | ||
| }, REFRESH_MS); | ||
|
|
||
| server.listen(PORT, HOST, () => { |
There was a problem hiding this comment.
HTTP server starts after refresh
High Severity
The server awaits the initial refresh() call before server.listen(), preventing it from accepting connections until buildStatus() completes its external calls. This delays /healthz and /api/status availability, leading to startup probe failures and restarts, despite public/status.json being pre-loaded.
Reviewed by Cursor Bugbot for commit ae16c89. Configure here.
| kubectl apply -f k8s/web-deployment.yaml | ||
|
|
||
| kubectl -n "$NAMESPACE" rollout status deployment/dioptra-api --timeout=180s | ||
| kubectl -n "$NAMESPACE" rollout status deployment/dioptra-web --timeout=180s |
There was a problem hiding this comment.
Manual deploy resets image tag
Medium Severity
The deploy-to-prd-apps.sh script applies deployments using :latest image tags. Unlike our GitLab CI, it doesn't update images to specific commit SHAs. This can result in deploying stale images or rolling back to older versions, potentially causing mismatches between API and web components.
Reviewed by Cursor Bugbot for commit ae16c89. Configure here.
1 participant
Summary
X-Dioptra-Sitemarker, Origin allowlist, rate limitdk-raas/dkai/tools/dioptra→ Harbor → Rancherprd-appsDeploy path
GitHub
main→ CI → reusable GitLab mirror push (secrets: inherit→ orgGITLAB_TOKEN) → GitLab CI → Harbor →k8s/on prd-appsSee
docs/GITLAB_MIRROR.md. No per-repoGITLAB_TOKENneeded.Note
Medium Risk
Large greenfield change with production deploy jobs and auth that depends on shared site markers and origin allowlists; misconfigured secrets or CI kubeconfig could affect prd-apps rollouts.
Overview
Replaces the mock-only repo with a full Dioptra stack: a Vite + React SPA and a Node read-only API that serves live fleet snapshots.
The Fleet tab polls
GET /api/statuswith a baked-inX-Dioptra-Sitemarker (must matchDIOPTRA_SITE_MARKERon the API), plus an Origin/Referer allowlist and per-IP rate limiting. The API refreshes on an interval viabuildStatus(GitHub/GitLab issue collectors; nodes/agents from env JSON or seededpublic/status.json).npm run publish:statusstill writes an offline snapshot.The Setup tab adds SVG diagrams (Coder workspace provisioning, agent observe/plan/act/verify loops, fleet wiring). README and docs are updated for the new architecture; Apache-2.0 license is added.
Ops: dual Docker images (nginx SPA + API), Compose,
k8s/manifests fordioptraon prd-apps, GitHub CI (test/build + mirror to GitLab), and GitLab CI (Harbor publish + optional kubectl deploy). Collector tokens stay server-side only; the SPA marker is explicitly not a strong secret.Reviewed by Cursor Bugbot for commit ae16c89. Bugbot is set up for automated code reviews on this repo. Configure here.