Home
The first screen you land on after sign-in. What every panel shows, what each colour means, what to click first when something looks wrong.
The Home dashboard answers one question: "is the system OK right now?". If everything is green you can close the tab. If anything is yellow or red, Home gives you the shortest path to "why".
Layout, top to bottom
- ① Health stripA row of four large pills at the top. Each pill is one health domain: ingestion, storage, AI, alerts. Green = fine, amber = degraded, red = something is broken.
- ② Active incidentsA horizontal carousel showing every incident currently open. Empty state ("All quiet") is the desired state.
- ③ Service health gridA grid card per service. Each card is colour-coded by its worst SLO right now.
- ④ Pulse chartA wide chart showing requests-per-second and error-rate stacked, across the whole fleet, over the time-picker range.
- ⑤ Latest deploysA timeline of the last 24 h of deploys captured from your CI webhook. Deploys correlate with anomalies — having them on Home is on purpose.
- ⑥ Top moversTwo columns: services whose error-rate spiked, services whose latency grew the most. These are the leading indicators of an unfolding incident.
① The health strip in detail
Four pills. Reading order matters — left to right is "from the apps in, to the answers out".
- IngestionAre your apps actually sending data? Shows last-second event throughput and the lag between "event happened in the app" and "event landed in storage". Green if both Collectors are up and lag is under 5 s. Amber if one Collector is unreachable. Red if no data has arrived for 60 s.
- StorageIs Doris keeping up? Green if write success rate is over 99% and BE memory is under 80%. Amber if memory is between 80–95%. Red above 95% (you're about to hit MEM_LIMIT_EXCEEDED).
- AIIs the LLM provider available, and how is the monthly budget tracking? Green if the last RCA call succeeded and budget burn is on track. Amber if budget is at 80%. Red if calls are failing.
- AlertsAre there currently firing alerts? Green = none. Amber = 1–3 firing. Red = 4+ firing or any P0-tier alert firing.
Clicking any pill jumps you to the screen that shows the full story: Ingestion → Admin → Health, Storage → Admin → Health, AI → Admin → AI, Alerts → Alerts.
② Active incidents — what you see
Each incident card has, top to bottom:
- A severity ribbon along the top edge:
P0(red),P1(amber),P2(yellow),P3(blue), info (slate). - The incident title — written by RCA when it can, by you when you opened it manually.
- A one-line summary — the AI's current best guess at the cause, with span IDs cited.
- The affected services as pills.
- A relative timestamp ("started 14 m ago"), a status badge (
open,mitigating,resolved), and an "Open" button that takes you to the incident detail page.
If the cards spill past the carousel, paginate with the arrow buttons or swipe.
③ Service health grid — colours and shapes
Every service that has reported data in the time window gets a card. Card colour:
- EmeraldAll SLOs are inside budget. No anomalies.
- AmberAt least one SLO has a burn rate above 2×. Investigate within the hour.
- RoseAn SLO is being burned faster than 14× (a "fast burn"), or an active alert is firing. Page-worthy.
- SlateNo data in the window. Either the service is silent on purpose, or it stopped reporting — click to find out.
Inside the card:
- Top line: the service name, the count of currently-firing alerts for it (if any), and a tiny pulse showing live RPS.
- Middle line: three sparklines stacked — requests/sec, error-rate %, p99 latency ms. The sparkline range matches the time picker.
- Bottom line: the worst SLO right now, written as
"<sli> · <burn-rate>x", e.g."latency<500ms · 8×".
Click anywhere in the card to open the service detail page.
④ The pulse chart — what each layer shows
A single chart, ~280 px tall. Two stacked series:
- RPS (lower, cyan area)Requests per second across every service. Tells you the size of your traffic shape.
- Errors (upper, rose area)Errors per second. Plotted on the same time axis. When this area grows visibly relative to RPS, you're in an incident.
Drag a region with the mouse to "zoom to selection" — that range becomes the new time picker. Right-click anywhere on the chart to open it in the Metrics screen for full-fidelity analysis.
The chart has deploy markers: tiny vertical lines with a build-icon badge wherever your CI told us "a deploy just happened". Hover a marker to see commit hash, author, and service. If an error spike sits right after a deploy marker — that's usually your cause.
⑤ Latest deploys
A horizontal timeline of the last 24 h of deploys. Each chip carries:
- Service name.
- Commit short hash (clickable to your VCS if you wired the integration).
- Author avatar / name.
- "OK" or "rolled back" badge.
Empty state ("No deploys") means either you're in a freeze window or the CI webhook isn't configured — set it up under Admin → Integrations → CI webhook.
⑥ Top movers
Two side-by-side lists, each capped at five rows:
- Latency movers — services whose p99 latency grew the most (in milliseconds and as a percentage of their previous-window baseline). The numeric badge on the right is the delta in ms.
- Error movers — services whose error count grew the most. The numeric badge is the delta in errors/min.
These two lists are the cheapest "what's wrong?" lookup in the whole product. If you only ever look at one part of Home, look at these.
What to click when…
- You see all green and no incidents → close the tab; the system is healthy.
- A health pill is amber/red → click the pill, it takes you to the page that explains why.
- A service grid card is rose → click it → service detail. Read the alert it's firing, then click "View traces" inside the alert.
- You see a deploy marker right before an error spike → click the marker; it opens the deploy details modal with one-click "open incident from this".
- The pulse chart shows an anomaly → drag-select the spike; the page zooms; the service grid below the chart will re-rank so the noisiest service floats to the top.