PowerTraces.MSSQL 3.0.0

PowerTraces.MSSQL

SQL Server provider for PowerTraces — exports OpenTelemetry traces and logs straight to SQL Server via a connection string. No collector, no EF Core, no migrations project: tables are created automatically the first time you write.

Install

dotnet add package PowerTraces.MSSQL

Export traces and logs

using OpenTelemetry.Logs;
using OpenTelemetry.Trace;
using PowerTraces.MSSQL;

builder.Services.AddOpenTelemetry()
    .WithTracing(t => t
        .AddAspNetCoreInstrumentation()
        .AddPowerTracesMsSqlExporter(connectionString))
    .WithLogging(l => l
        .AddPowerTracesMsSqlExporter(connectionString));

Optional settings (schema name, command timeout, path exclusions, export-error callback):

tracing.AddPowerTracesMsSqlExporter(connectionString, o =>
{
    o.Schema = "powertraces";              // default; tables live under this SQL schema
    o.CommandTimeoutSeconds = 30;
    o.ExcludePathPrefixes.Add("/otel"); // don't trace the dashboard's own requests
    o.OnExportError = ex => logger.LogWarning(ex, "PowerTraces export failed");
});

View what you collected

Add PowerTraces.Dashboard and register this provider's dashboard store alongside it:

builder.Services.AddPowerTracesMsSqlDashboard(connectionString);
// ...
app.MapPowerTracesDashboard("/otel", PowerTracesDatabaseProvider.SqlServer);

Connections

Unless your connection string sets its own Application Name, PowerTraces connects as PowerTraces. SqlClient pools connections per exact connection string, so this keeps the exporter and dashboard in their own pool instead of sharing your application's, and makes their sessions easy to find:

SELECT program_name, COUNT(*) AS sessions
FROM sys.dm_exec_sessions
WHERE is_user_process = 1
GROUP BY program_name;

The exporters use at most one connection each at a time; the dashboard is capped by MaxConcurrentQueries (see the PowerTraces.Dashboard README).

What gets created

Tables under the powertraces SQL schema (configurable): Resources, Spans, Logs, Traces, AttributeKeys, IgnoreRules, Settings and PinnedTraces — plus the indexes the dashboard's queries need. Creation is idempotent and lock-guarded (sp_getapplock), so it's safe for multiple instances to start at once. Spans and logs are written with SqlBulkCopy.

The trace list reads Traces (one summary row per trace, upserted as spans arrive) rather than grouping Spans on every request, so it stays in the milliseconds as the tables grow; list totals stop counting at 10,000 (shown as "10,000+"). Attribute values live only in each row's attribute JSON, which attribute filters read — there's no per-attribute index, so a filter scans the rows the other filters (time range, service, errors) leave. AttributeKeys lists every attribute key seen, for the dashboard's attribute picker. Attribute filters use JSON_VALUE with a parameterized path, which needs SQL Server 2017+ (database compatibility level 130+).

Upgrading a database written by an earlier version: missing tables are created (and the trace summaries backfilled from existing spans) once, the first time the new version starts. The SpanAttributes/LogAttributes index tables earlier versions wrote are dropped after their keys are copied into AttributeKeys.

License

Free for any use, including commercial use, under the PowerTraces Free License (LICENSE.txt in this package): you may use it and ship it unmodified inside your applications, but not modify the code or publish changed copies.