Add rate limiting to gRPC and HTTP APIs
Problem
The Knowledge Graph has no rate limiting on any API surface. One misbehaving client can starve the cluster for everyone else.
Affected entry points:
- gRPC API (
crates/gkg-server/src/grpc/) - query endpoints, redaction exchange - HTTP webserver (
crates/gkg-server/src/webserver/) - REST API, health endpoints
gRPC layer
- Token bucket or sliding window limiter as a gRPC interceptor
- Per-namespace limits scoped by
organization_idor namespace from JWT claims - Global fallback limit for unauthenticated or unscoped requests
- Return
RESOURCE_EXHAUSTEDwhen limits are hit
HTTP layer
- Rate limiting middleware on REST API endpoints
- Return
429 Too Many RequestswithRetry-Afterheader
Query engine
- Max concurrent queries per namespace
- Query result size limits (currently unbounded - no cap on rows returned)
- Query timeout enforcement with resource cleanup on abort
Configuration
- Configurable via
config/default.yamland environment variables - Separate limits for read vs write operations
- Separate limits for API queries vs indexer operations
Context
No rate limiting exists anywhere in the codebase. The worker pool uses semaphore-based concurrency control (crates/indexer/src/worker_pool.rs) but that only covers indexer workers, not API requests.
PREP assessment: https://gitlab.com/gitlab-org/architecture/readiness/-/merge_requests/67
PREP template: operational_lifecycle/error_handling_and_degraded_state_behavior.md (related)
Edited by Michael Angelo Rivera