Insights¶

The Insights page in DeepCube™ Web is the grid's overview: four golden signals, a short list of what needs attention, throughput and latency charts, and the busiest routes. After reading this page you can tell what each figure is computed from, which time window it covers, and where its limits are.

Open Insights from the top of the navigation, or select Open Insights or the Alerts figure on Home. The page reads the grid and the time range in the app bar; see Getting around.

Each part covers its own window¶

Most of the page covers the whole range you picked. One part does not.

The KPI strip shows four golden signals¶

Four cards sit under the title. Each shows the latest value in the range, a sparkline of the whole range, and a comparison with the window before it.

When there were no requests, Requests shows 0. The other cards show "No data" when their signal reported nothing, never 0.

The comparison line under each card¶

The line under each value compares the average over your range with the average over the window of the same length just before it (for a 45-minute range, the 45 minutes before that). It is not a comparison of the latest value.

A change too small to show reads "No change vs previous 45 min" (with your range's length). When either window has no data, the line is left out.

When the latest error rate is 5 % or more, the Errors card adds an Above 5 % badge. Hover it to see that this is a heuristic threshold.

Select a card to sort the endpoints¶

Selecting a card sorts Top endpoints by the matching figure and scrolls to the table: Requests sorts by requests, Errors by errors, Latency p95 by p95, and Apdex by p99.

What needs attention lists up to three items¶

Under the KPI strip, up to three items show what needs a look, in this order:

1. Firing alerts, up to two, critical ones first. The title is the alert's condition, for example "Error rate 4.2% exceeds 1% threshold". The line under it names the service, or "Whole grid", and ends with "computed from fixed thresholds".
2. A deploy that lined up with errors, for example "Errors on payment-service rose after 1.9.0", with "Deployed 11:50, inferred from service.version".
3. New endpoints, for example "1 new endpoint", with the route and its service, and "and 2 more" when there are several.

When there is nothing to show, the strip reads Nothing needs attention in this window.

Throughput and Latency chart the whole range¶

• Throughput shows requests per minute, split by status class (1xx to 5xx, each class that was reported). If the split cannot be read, the chart shows the total and says Not split by status.
• Latency shows p50, p95 and p99 in milliseconds.

When a chart has more than one series, it shows a legend: select a series in it to hide or show that series. Drag across either chart to set the time range to that stretch; Reset zoom in the app bar goes back. To do it from the keyboard, focus the chart, hold Shift with the arrow keys and press Enter.

Hover across charts¶

Hover across charts controls what the other chart does when you point at a time on one:

The app remembers your choice in this browser. The charts match by time, not by position, so the line lands on the same moment in both.

From the keyboard, focus a chart and use the Left and Right arrow keys to step through the points, Home and End for the first and last, and Esc to stop. The other chart follows as it does for the pointer, so with Hover across charts set to Off it does not.

vs last week¶

Select vs last week to draw last week as a dashed line: Requests last week on Throughput and p95 last week on Latency. It covers the same range exactly 7 days earlier. Select it again to turn it off.

When last week cannot be drawn, the chart says why instead of drawing a guess:

• Nothing was reported last week: the same window 7 days earlier has no data (it may be outside the grid's retention).
• Last week could not be loaded: the comparison failed to load; the chart shows this window only.

Deploy markers are labelled "(inferred)"¶

A dashed vertical line on both charts marks the time a service first reported a new version, with a label such as "payment-service 1.9.0 (inferred)". DeepCube does not receive deploy events. It infers a deploy from the moment a new service.version value first appears in the service's telemetry, and the label says so.

Markers come from the same diagnostics as What needs attention, so they cover at most the last hour of the range.

Show data as table¶

Show data as table replaces a chart with a table of the same data, a row for each point in time. A gap in the data reads "No data". Select it again to go back to the chart. The table can be focused and scrolled with the arrow keys.

Top endpoints ranks five routes¶

Top endpoints lists five routes, ranked across every route the grid served in the range. The line beside the heading says how many of how many, for example "5 of 8 routes in range, sorted by errors". The default order is by errors.

Select a column heading to sort by it (all but Share and Trend). Each sort is a new ranking of every route, not a reorder of the five shown.

Select a row (or focus it and press Enter) to open the endpoint as a panel beside the page; see Endpoints. All endpoints opens the Endpoints page with every route.

When no server spans were reported in the range, the table says so instead of showing empty rows.

When part of the page cannot load¶

Each card, chart and table loads on its own. If one fails, it says it could not load and offers to try again; the rest of the page is not affected. A KPI card whose signal reported nothing, or failed to load, reads "No data" on its own card (Requests shows 0 when there were no requests).

At the bottom of the page, a note appears when no producer, consumer or dependency traffic was reported in the window. Those signals show up once your services report messaging or outbound HTTP spans.

Related¶

• Getting around
• Endpoints
• Services
• Traces
• Metrics
• Instrument your application