Getting Started
The BYOC Ops Console is a web app for monitoring and triaging the health of CelerData's BYOC StarRocks fleet.
Accessing the App​
- Open the app URL (
http://<bastion-ip>:8501— ask an admin, or use the wake-up link if the server is asleep; it boots in ~10 seconds). - Log in with your email and password. New accounts arrive by email invite — follow the activation link to set your password.
Roles​
What you can see and do depends on your role:
| Role | Access |
|---|---|
superadmin | Everything, including Settings, agent controls, LLM usage, Ask-AI |
admin | All operational pages and full data |
support | Operational pages; data may be limited to assigned customers |
viewer | Read-only operational pages; data may be limited to assigned customers |
If a page is missing from your sidebar, your role doesn't include it. If a customer's clusters don't appear anywhere, you may be scoped to specific customers — ask a superadmin to check Settings → Customer Access.
The Sidebar​
| Page | What it's for |
|---|---|
| Fleet Sentinel | Fleet-wide health at a glance — scores, KPIs, critical clusters |
| Investigator Agent | AI investigation reports + ask the AI your own question |
| AI Triage | The issues inbox — grouped alerts with a triage workflow |
| Fleet Patrol Agent | Twice-daily AI fleet sweep reports |
| Breaking Thresholds | Clusters approaching resource limits |
| Issues Pattern / Alerts Pattern | Looker analytics on issues and raw alerts |
| Usage Patterns | Cluster utilization trends |
| LLM Usage | AI token/cost tracking (superadmin) |
| AI Chat | Ask ad-hoc questions about any cluster |
| Settings | Users, access, rules, agent controls |
| Help | In-app quick reference |
Sidebar entries are ordinary links — middle-click or Cmd/Ctrl-click opens a page in a new tab.
The Data Pipeline Banner​
The app checks every few minutes that the data behind its pages is still arriving. When something has stopped, a one-line strip appears above the header on every page:
| Strip | Means |
|---|---|
| Red — "Data pipeline down: numbers on these pages may be stale." | At least one feed has stopped |
| Amber — "Data pipeline lagging: numbers on these pages may be stale." | A feed is late but still moving |
After the headline it names each feed with a problem and how old its data is. Hover over the strip for the full text. The feeds:
| Feed | Affects |
|---|---|
| Metrics MV | The hourly fleet metrics behind Breaking Thresholds, Usage Patterns, the evidence on AI Triage issues, and the metrics AI Chat quotes. They refresh about every 6 hours, so a few hours of lag is normal |
| Raw metrics feed | Everything built on metrics, a little later |
| Health scorer (jumper) | Fleet Sentinel scores and the Breaking Thresholds cards: they keep showing the last run until it recovers |
| Slack alert ingest (jumper) | New issues on AI Triage |
While the strip is up, read scores and metrics as "as of" their own time, not as the cluster's state right now. When the Metrics MV is down, AI Chat is told so as well: it gives the time of every metric it quotes and says the metrics are stale. Nothing to do on your side: tell an admin if the strip stays for more than an hour. When everything is healthy, the strip is hidden.
A Typical Workflow​
- Start on Fleet Sentinel — anything red?
- Open AI Triage — new issues are grouped and summarized automatically from Slack alerts (within ~5 minutes of firing).
- Assign an issue to yourself, read the AI summary and evidence, and click Ask AI Assistant for a deep investigation.
- Track fixes via triage status, and create a JIRA ticket from the issue if it needs engineering work.
- Got an error, a crash stack or odd log lines? Check whether it's a known issue, from the issue's Known Issues panel or by pasting it into AI Chat. See Known Issues.
- Tuning a POC or benchmark? Ask AI Chat to evaluate the cluster with the Configuration Advisor. See Configuration Advisor for what to prepare.
This is a new change I am just make in got see how the edit works.