Services¶

In DeepCube™ Web, the Services page lists every service and dependency in the grid with its health and figures, and draws the same nodes as a map of who calls whom. After reading this page you can judge a service's health at a glance, find what it depends on, and open it to see its endpoints, versions and errors.

Open Services under Signals in the navigation. The page follows the time range picker; see Getting around.

The list shows each service's health, traffic and version¶

The List view is a table with one row per service, dependency and messaging destination the grid saw in the window. Services come first, then dependencies, then messaging destinations, each sorted by name.

Select a row (or focus it and press Enter) to open the service panel beside the list. See The service panel.

Health is judged on error rate only¶

The Health column looks at the error rate and nothing else:

• Failing at an error rate of 5 % or more.
• Degraded at 2 % or more.
• Healthy below 2 %.
• No data when nothing measured the service in the window.

A service or dependency that nothing measured says No data. It is never shown as Healthy or as 0 %. Latency does not change the health word.

Some rows carry a state¶

The grid infers dependencies and expected services from the calls it sees, so a row can carry one of two notes beside its name. Hover the note to read the reason the grid gave.

Where the version comes from¶

The Version column reads service.version from the service's spans:

• A version first seen in the window (at most the last hour of the range) shows the version and a deployed badge. The badge is inferred from service.version, not from a deploy event.
• Otherwise the column shows the version the service was last seen running in the range.
• No data means the service's spans carry no service.version in the range. Set service.version in the service's OpenTelemetry resource to fill it in.

The map shows who calls whom¶

Select Map at the top right to switch views. The choice is kept in the address, so a reload or a shared link opens the same view.

Each node is a service, dependency or messaging destination:

• The value inside the node is the metric you pick at the top left: Requests (per minute), Errors (error rate) or Duration p99. A node nothing measured shows No data.
• The badge and colour show health: a check for healthy, a triangle for degraded, a cross for failing, a dash for no data.
• A dashed node is not a service that reports for itself: a dependency or messaging destination the grid inferred from the calls into it.

Each arrow is a call from one node to another seen in the window. Arrows carry no figures.

To work with the map:

• Select a node (click it, or focus it and press Enter) to open its service panel and scope the map to it. Its callers and callees stay bright and everything else dims; the selected node's arrows animate for five seconds, unless your system asks for reduced motion. The toolbar then says Near and the service name.
• Show everything clears the selection.
• Show as table replaces the drawing with two tables: the services on the map (Service, Kind, Health, Req/min, Errors, p99, Calls) and the calls between them (From, To, State).
• Zoom out, Fit and Zoom in change the drawing's size.

Limits on what the list and the map cover¶

Both views read the same grid graph, so the same limits apply to both, and the page says so when one applies.

• The last hour only. The graph covers at most the last hour of the range. When you pick a longer range, a callout reads Clamped to the last hour: followed by the window it covered.
• At most 200 nodes. When the grid has more, the page says, for example, "Showing 200 of 350 services". The ones left out are not shown. Pick a shorter range, or open the service from Traces.
• A capped span scan. When the window held more spans than the graph reads, the page is marked Sampled and says the figures were read from the first spans of the window, so they can be low.

The service panel opens beside the page¶

Selecting a service in the list or on the map opens its panel beside the page. How panels stack, close and go back is described in Getting around.

At the top of the panel:

• Endpoint narrows the panel's Requests and Errors charts to one route. It lists the service's busiest routes (up to 10) once they have been read. The figures above the charts stay the whole service's, and the panel says so. Dependencies have no Endpoint picker.
• Near this service: lists the services that call it (an arrow pointing up and left) and the services it calls (an arrow pointing down and right), each with its error rate. Select one to open its panel on top.

The panel has four tabs. The tab you are on is kept in the link.

Overview¶

• Five figures: Req/min, Errors, p95, p99 and Apdex. When nothing measured the service in the window, a No data for callout replaces them. That is missing data, not a healthy service and not zero traffic.
• A deploy callout when a new version was first seen in the window (at most the last hour of the range), for example "Deployed 1.9.0 at 11:50:02 (inferred)". It says either "Errors rose after this deploy" or "Errors did not line up with this deploy".
• Requests and Errors over the whole range: Requests per minute and Share of requests that failed (5xx).
• Duration distribution: Requests by duration, a bar per duration band from 5 ms or less to more than 10 s. The counts are estimated from the request rate in each band over the range.
• Time spent in downstream calls: the services this one calls, most total time first (up to 10), with Calls, Avg and Total. This is approximate: a call whose spans fall outside the window or the span cap is not counted, and parallel calls count in full, so the total can exceed the time the caller waited.

Endpoints¶

The service's busiest routes (up to 10), from diagnostics: Endpoint, Req/min, Errors, p95, p99, DB calls / req and Ext. calls / req. Hover an Errors value to see how many requests failed. Select a row to open its endpoint panel for this service only; see Endpoints.

• p95 and p99 are estimated from a sample of up to 1,000 requests per route.
• DB calls / req and Ext. calls / req are approximate, from the spans diagnostics read. No data in one of these columns, with the hover reason "Not counted: this read hit the span cap and found none.", cannot be told apart from zero.

Deployments¶

Every version the service ran in the range, newest first. A deployment is a new service.version first seen in the range.

When the oldest version in the range was first seen inside the range, "Same service since" above the table gives that time. When the service's spans carry no service.version, the tab says No versions seen in this range.

Errors¶

The service's errors grouped by exception type and message, the five most frequent, each with its count. Open sample trace opens the first trace the group counted, which may have aged out of retention by now. A group that kept no sample says No sample.

Related¶

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