Tracing and logging in Dataverse

Server-side code in Dataverse — plug-ins, custom workflows, low-code plug-ins — runs in a sandboxed environment with limited debugger access in production. The compensation is tracing and logging: instrument your code so issues are diagnosable post-hoc without reproducing them in dev. Mature teams treat instrumentation as a first-class engineering concern.

The trace service

Plug-ins receive an ITracingService from the execution context:

var tracingService = serviceProvider.GetService<ITracingService>();
tracingService.Trace("Plug-in started for entity {0}", entity.LogicalName);

Each Trace call writes a line; traces accumulate per plug-in execution.

Where traces go.

• Plug-in trace log table in Dataverse — visible in admin UI.
• Available for a configurable retention period — typically days to weeks.
• Per-execution — one trace log record per plug-in invocation.

For most diagnostic needs, this is sufficient.

Application Insights integration

For deeper observability:

• Configure App Insights for the Dataverse environment.
• Traces flow to App Insights additionally.
• Rich query, alerting, dashboarding.

App Insights is the modern, recommended approach for production Dataverse.

What to trace.

• Method entry/exit — for execution flow.
• Key decision points — branches, conditions taken.
• External API calls — request and response (minus sensitive).
• Performance markers — operation timing.
• Errors and exceptions — full context.

A balanced level — neither too sparse (can't diagnose) nor too verbose (can't see signal in noise).

What NOT to trace.

• Passwords, API keys, tokens.
• PII — full names, emails, identifiers (without business need).
• Credit card data — PCI violation.
• Health data — HIPAA violation.

Use redaction patterns: trace customer ID redacted, not customer ID 12345.

Structured tracing

Better than free-text:

tracingService.Trace("OperationName: {0}, RecordId: {1}, Duration: {2}ms",
    operation, recordId, duration);

Structured patterns enable easier parsing and querying downstream.

Tracing in low-code plug-ins

Power Fx in low-code plug-ins:

• Trace() function adds traces.
• Log to standard plug-in trace.
• Same access patterns.

Tracing in flows.

• Each action's input/output captured automatically.
• Loop iterations tracked.
• Failures detailed.

Flow run history is the trace; no explicit Trace calls typically needed for basic flows. For complex flows, use Compose actions to log specific values.

Performance impact.

• Tracing has overhead — string formatting, write to memory.
• For high-volume code paths, minimise traces.
• Don't trace in tight loops; aggregate and trace once.

A plug-in tracing 100 lines per record × 10K records = 1M trace lines = slow.

Trace levels

Not built into Dataverse directly, but pattern:

• Error — failure.
• Warning — concerning but recoverable.
• Info — major checkpoints.
• Debug — verbose flow.

Configurable levels — Production at Warning+, Sandbox at Debug.

Application Insights query

With App Insights, you query traces in Kusto Query Language (KQL):

traces
| where customDimensions.PluginName == "MyPlugin"
| where timestamp > ago(1h)
| order by timestamp desc

Rich querying replaces digging through individual plug-in trace records.

Correlation IDs

Trace correlation across systems:

• A request entering Dataverse gets a correlation ID.
• Trace lines include it.
• External calls pass the ID.
• Receiver logs include it.

End-to-end traceability across system boundaries.

Audit vs trace

Different:

• Audit — business-relevant change records (who changed what when); for compliance.
• Trace — technical execution information; for diagnostics.

Both valuable; don't conflate.

Sensitive data handling.

• Redact before tracing — SSN: ***-**-1234.
• Hash identifiers — for correlation without exposure.
• Avoid full row dumps — trace specific fields.

Common pitfalls.

• No tracing. First production issue; nothing to look at.
• Verbose tracing. Performance hit; noise in logs.
• Sensitive data leaked. Compliance issue.
• Trace formatting expensive. Even if log not written, string formatting cost.
• Trace ignored. Logs exist but nobody looks; instrumentation wasted.

Operational rhythm.

• Daily — review error traces.
• Per incident — deep dive on relevant traces.
• Weekly — trace pattern analysis (top errors, slow operations).
• Monthly — review what's traced; adjust verbosity.

Strategic positioning

Tracing is the cheap insurance policy of server-side code. Investment in good instrumentation pays back the first time something goes wrong in production. App Insights elevates Dataverse to enterprise-grade observability. For any production deployment of meaningful scale, instrumenting plug-ins and flows is non-negotiable engineering practice. Tracing isn't optional; it's foundational. The teams that take it seriously have boring incident reviews; the teams that don't have long, painful ones.