Skip to main content

Feature Overview

APM is organized as Home, Services, Explore, Events, and Integration.

1. Home​

Home shows service health, SLOs, alerts, and top metrics for the current organization. A section with no data stays empty or degraded. It does not invent metrics.

  • Health and SLO summary: See which services are healthy and whether availability or latency goals are met.
  • Alert entry: Open application alerts that are still being handled, then drill into the service.
  • Section degradation: If the trace store, collection, or notification dependency fails, that block degrades without hiding the rest.

2. Services​

2.1 Service directory​

The directory is the list of services discovered by agents.

  • Organization scope: Only authorized services in the current organization are listed. A failed organization check rejects the request instead of falling back to global visibility.
  • Archive and restore: Archive services you no longer watch. The active window is about 15 minutes, and the directory reconciles about once a minute.
  • Unassigned repair: A superuser can filter unassigned services and set an organization. Unassigned services stay out of the default current-organization list and do not open trace or metric queries.

2.2 Service topology​

Shows call dependencies. The window cannot exceed 7 days. Redis, PostgreSQL, and similar dependencies without an agent can appear when they are part of a call. That inference is topology-only. It is not written into the service directory, application list, or CMDB.

2.3 SLO​

The target rate is greater than 0 and at most 100. The form accepts 0.001 as the minimum and three decimal places. SLI types are Availability (share of non-error requests), Latency (P95 below threshold), and Latency (P99 below threshold). Latency goals require a positive millisecond threshold. The evaluation window is required: Rolling 7 days, Rolling 30 days, or Calendar month. Environment is required. An empty endpoint means the whole service. Home and the service page read the same evaluation.

3. Explore​

3.1 Traces​

Search by time, service, and environment. The UI offers 15m, 1h, 4h, 1d, and 7d. The longest window is 7 days. There is no free query language. Detail supports a waterfall, a flame graph, and a span list. Sensitive attributes are masked. The list is a bounded sample and is not the error rate for the whole window.

3.2 Endpoints​

Search spans and endpoints with an allow-listed parameter set inside the organization. Open an endpoint from a service, then return to a single trace.

3.3 Errors​

Group failures by exception or issue fingerprint. On the service detail page, entry-failure counts and all error spans for that service are labeled separately. The explore list shows the loaded error kinds. It does not present the current page as the window total.

4. Events​

4.1 Alerts​

Alert states are Firing, Recovered, and Closed. Events themselves are immutable. An unassigned alert can be claimed or assigned. The current assignee can reassign. Manual assign and reassign notify the new person through system user channels. My alerts filters by assignee, and the list and distribution use the same filter. The distribution does not count claim, assign, or reassign actions. Evidence is the scan-point trend and the original event. It may be pending or expired. Do not rebuild history from the live store. A manual close does not append a new scan point. Policy edits affect later evaluations only.

4.2 Policies​

Policies do not use Monitoring expressions or log search. Metrics are error rate, P95, P99, throughput, and no-traffic. Choose the service in Service. The environment is already part of that option, and the page says the environment is included in the service scope and the version is fixed to all versions. An endpoint can be left empty, which monitors the whole service. Required fields are policy name, alert name, evaluation interval (1 to 60 minutes), organization, aggregation window (1 to 1440 minutes), and aggregation method: avg_over_time, last_over_time, max_over_time, or min_over_time. Critical, Error, and Warning each have a comparator and a threshold. Comparators are >, ≥, <, and ≤. An empty threshold leaves that severity off. Error rate is entered as a percentage. An alert fires after 1 to 60 consecutive windows meet the threshold, and recovers after consecutive windows no longer meet it. The no-data alert is separate from the no-traffic metric. Assignees are optional. When set, they must belong to the policy organization and must not be disabled. Delivery uses System Management channels and can be retried. An optional copy can be sent to Alert Center, and Alert Center cannot write it back.

5. Integration​

5.1 Applications​

Create an application and assign an organization. If the organization check fails, the save is rejected. A superuser can filter unassigned applications and assign an organization. Unassigned applications are not in the default current-organization list.

5.2 Add integration​

Languages are Java, Python, Node.js, Go, and .NET. Instructions point at the in-product download and the region's internal endpoint. Public tokens are not supported. The optional sample rate is written only into this install script, defaults to full sampling, is not stored, and does not change a running application. The drawer lists the Web and RPC frameworks covered by the pinned agent version and the client-span component types. That matrix is static. Unknown agent artifacts are rejected. The download itself does not require a login.

5.3 Reporting instances​

Shows reported instances and their organization. A superuser can filter unassigned instances. An unassigned instance does not open its traces or metrics.

Note: APM does not include monitoring plugins, log search, a free query language, tail sampling, a separate edge gateway, or cross-tenant public OTLP.