APPLICATION TRACING
Application Performance Monitoring with OpenTelemetry
Follow instrumented requests from service-level observations into span waterfalls and related database or log evidence. Spekra helps engineering teams examine where request time is spent and coordinate the next investigation step.
Guided pilot scope · Reviewed 2026-10-16. No production AI diagnosis, automatic infrastructure/DDL remediation or compliance certification is offered.
How traces reach the workspace
Configure an instrumented application or OpenTelemetry Collector to export HTTP JSON to /v1/traces, using the organization's X-Pulse-Api-Key header. /v1/metrics accepts supported metric exports. Do not put integration credentials in URLs.
A trace ID groups operations; span IDs and parent IDs describe individual operations and nesting. Spekra groups incoming spans, converts their times to milliseconds and renders relative offsets in a waterfall. Instrumentation determines which operations exist in that evidence.
What the trace can tell you
Inspect request duration and the exported HTTP, database or internal spans. Database statement attributes can be enriched with table names and SQL operation. This association is not a verified execution plan, complete SQL parser or proof of an index problem.
For example, a 251 ms sample request with a 1 ms database span suggests examining the other request work. Child spans may overlap; adding all durations is not always wall-clock latency. Validate delayed and split-batch exports rather than assuming complete cross-batch reconstruction.
Supported metrics and observed statistics
Metric storage supports gauges, delta/cumulative sums and explicit histograms. Exponential histograms and summaries are rejected. Selected 64-bit values and timestamps remain decimal strings to preserve precision. Queries are bounded latest-point views, not a general aggregation engine.
Production service statistics use observed trace samples within a trailing 60-second window, capped at 10,000 observations. Export sampling and missing instrumentation affect the displayed rates/percentiles. An empty window does not prove the application had no traffic.
Validate acceptance and data handling
The metrics receiver can return HTTP 200 with partialSuccess.rejectedDataPoints. Check that field instead of assuming every point was accepted. Identical series/timestamp exports deduplicate; conflicting values reject rather than overwriting evidence.
The receiver retains supplied trace attributes, including SQL statements when present. The database collector's normalizer does not universally redact this path. Remove credentials, sensitive SQL literals, user data and unnecessary attributes before export. The adapter accepts JSON, not OTLP protobuf or gRPC.
Frequently asked questions
Does an SDK exporter work without configuration?
Not necessarily. Many exporters default to a transport this adapter does not accept. Configure HTTP JSON and test endpoint paths, headers, service identity, timestamps and the actual payload.
Is the sample demo proof of automatic instrumentation for every language?
No. Retained acceptance uses actual HTTP/SQL requests with manual instrumentation. Each customer's SDK/exporter combination needs its own validation.
Does Spekra provide a complete service dependency graph?
Production shows observed services and trace waterfalls. A complete automatically discovered distributed dependency graph is not implemented.
Evaluate your own workflow
Start with supported sample or staging infrastructure and agree on a success measure. Contact the operator to arrange a pilot. This static website does not create accounts or send invitations automatically.
Request a guided pilot →