Skip to main content

What Unified Logging Means in TeaQL

· 2 min read

A support ticket says the order search returned no results. Before asking for a database dump, find the query's business purpose, trace the SQL operation it triggered, and inspect the statement shape and result count. Those clues should mean the same thing whether the service runs on Java, Rust, or another TeaQL runtime. They do not need to look identical on screen.

Start with safe evidence​

Run one bounded query with a meaningful comment and purpose, or investigate one mutation with an audit reason. Start with the parameterized SQL shape, typed trace path, timing, and row counts. A successful query with zero rows is not a database failure; it tells you to check the filter and request context. Avoid putting customer values or secrets in the intent fields.

TeaQL has already released a shared SQL masking policy across seven runtimes. The driver receives SQL parameters as their original typed values. Rendering for operators is separate: ordinary values can be legible, policy-marked values are masked, and credentials are redacted. Once a value is masked, the displayed statement is deliberately not an exact replay. That distinction keeps logs useful without making a normal debugging session a plaintext export.

Make severity and delivery predictable​

Capability 112 proposes the next step: give each event a common name, severity, and structured record. A completed query would be INFO, a timeout with a scheduled retry would be WARN, and a final failure would be ERROR. All three can belong to the SQL category. This lets an alert look for final failures rather than treating every SQL message or retry as an incident.

The proposal uses rotating JSON Lines files on servers and the developer console in browsers. An operator could search for the same application name and event on either surface without relying on the same filesystem or output format. These delivery defaults are still a proposal, not instructions for configuring every released runtime.

Use the debugging and observability guide to pick the right evidence and check what your installed formatter supports. Keep business audit and telemetry separate: a runtime diagnostic helps explain what happened, but does not replace a governed record of who changed state or an exported trace of where time was spent.