Performance Tuning

How incremental computation works

GraphQL Analyzer uses Salsa for incremental computation. When a file changes, only the queries that depend on that file are recomputed. Everything else is served from cache.

Edit one file out of 100 → Only that file is re-validated
Edit schema → All documents re-validated (schema is a shared dependency)

LSP vs CLI performance characteristics

Aspect	LSP (editor)	CLI (CI/CD)
Update model	Incremental per-file	Full batch
Rule budget	Fast rules only	All rules
Typical latency	Milliseconds	Seconds

Expensive rules

Project-wide rules like noUnusedFields and noUnusedFragments analyze all documents together and can be slow in very large projects. If you experience latency in the editor, disable them in your config and run them in CI instead:

extensions:
  graphql-analyzer:
    lint:
      extends: recommended
      rules:
        noUnusedFields: off
        noUnusedFragments: off

Then in CI, enable them explicitly:

graphql check --rule noUnusedFields=error --rule noUnusedFragments=error

Structure/body separation

The analyzer separates schema structure (type definitions) from operation bodies. This means:

• Editing an operation doesn’t recompute schema analysis
• Editing the schema recomputes everything that depends on types
• Fragment indexes are stable across operation edits

This is key to keeping the IDE responsive in large codebases.

Debugging performance

Enable OpenTelemetry tracing

In VS Code settings (Preferences: Open User Settings (JSON)):

{
  "graphql-analyzer.debug.otelEnabled": true,
  "graphql-analyzer.debug.otelEndpoint": "http://localhost:4317"
}

4317 is the OTLP gRPC ingestion endpoint (the collector). View traces in Jaeger’s UI at http://localhost:16686 (requires Jaeger running locally — these are two different ports with different purposes).

Enable debug logging

In VS Code settings:

{
  "graphql-analyzer.debug.logLevel": "debug",
  "graphql-analyzer.server.env": {
    "RUST_LOG": "debug"
  }
}

For the CLI:

RUST_LOG=debug graphql validate