The deep‑dive for a single service — golden‑signal health plus panels for operations, dependencies, infrastructure, logs, and alerts. Use it to understand when a service degraded and why.
Get here: click a service on the Services list.
Service Health (tiles)
Seven tiles, each with a sparkline and % change vs. previous period (all trace metrics, exact):
- Latency — average response time (ms).
- Error Rate — % of operations that errored (per Error Code Series, default
5.*). - Throughput — requests per minute.
- SL Success —
100 − error rate %. - SL Latency — % of requests within the Latency Threshold (default 500 ms).
- Apdex Score — 0–1 against the Apdex Threshold (default 500 ms; min 1, max 300000).
- Total Count — total operations in the window.
Category Summary (cards)
Below the health tiles, four cards break the service’s traffic down by transaction category, so you can see which kind of work is healthy or degraded before opening a panel. Each card shows Throughput (RPM) · % Failed · Latency · Request count, and — except Others — a View link that jumps straight to that category’s panel.
- Web Operations — inbound HTTP/RPC server requests (
HTTP/RPC Systems,kind=server), the requests this service answers. View → Web Operations. Use it to see whether the service’s own endpoints are slow or erroring. - DB Operations — calls this service makes to databases (
Database). View → Databases. Use it to tell a slow query apart from slow application code. - Messaging Systems — producer/consumer queue work (
Messaging queues). View → Messaging Systems. Use it to spot consumer lag or failed publishes. - Others — everything not in the three panels above: outbound client calls to external services, internal spans, and any other category. There is no dedicated panel, so no View link. Use it to notice when the pain is outside the named categories (e.g. a third‑party dependency); then investigate via Distributed Tracing / Trace Explorer filtered by category or kind.
How to use it for debugging
Read the four cards together to localize the problem to a layer in one glance: if Web Operations error rate is up while DB and Messaging are clean, the fault is in the service’s own request handling; if DB latency spikes while Web is flat, chase the query; a healthy Web/DB/Messaging set with a bad Others number points at an external dependency or background work. From the worst card, click View (or open Distributed Tracing for Others) and drill into a trace. No data available simply means the service emitted no spans of that category in the window.
Alongside the cards, Top HTTP Error Operations by Status Code lists Operation · Status Code · Count — the quickest way to see which endpoints are throwing 5xx/4xx and how often.
Panels (navigation tabs)
Tabs along the top switch which panel you’re looking at, each a deeper cut of the same service: Overview, Web Operations, Databases, Messaging Systems, Distributed Tracing, Infrastructure, Topology (Service Maps), Network, Alerts, and Logs.
Drill‑down flow: from a Top 20 operation to the waterfall
- Pick the operation in a Top‑ops table (Most Time Consuming / Highest Error Rate / Highest Throughput) — or click the Slowest Span value to jump straight to one span.
- Operation detail — tiles (P99/P95 Latency, Error Rate, Total Error Count, Throughput, Total Operations) + Segment Breakdown (which downstream call dominates) + error breakdown, built from a representative sampled span/segments. View Traces and Share in the header.
- View Traces → the full Root/All Spans list for that operation; toggle Worst Spans; a row’s … → View details opens the span.
- Waterfall → the Trace Detail Timeline; follow the longest/errored span, read Exclusive Duration, and inspect the Span Detail.
Controls that change the numbers
- Compare with previous period overlays the prior window.
- Error Code Series (regex, default
5.*), Apdex Threshold, Latency Threshold, Range/Step.
Limits
Ranked operation tables show the Top 20; the Alerts panel shows the latest 50. Use Share to copy a permalink with the full drill‑down context.