Task 56 — Rate limits и квоты

Phase: 6 Initiative: 35 Status: done Owner: — Slug: rate-limits-quotas

Goal

Per-agent, per-skill, per-user, per-tenant лимиты concurrency, calls, токенов, оценочной стоимости, storage, message volume.

Acceptance criteria

• Add QuotasConfig section to AppConfig.cs (per-agent, per-skill, per-user, per-tenant limits)
• Create src/agent/Quotas/Models.cs — QuotaLimitType enum, QuotaScope enum, QuotaStatus record, QuotaViolation record
• Create src/agent/Quotas/IQuotaService.cs interface with CheckLimit, RecordUsage, GetStatus, Reset methods
• Implement QuotaService.cs — in-memory quota tracking with sliding window rate limiting
• Create src/agent/Quotas/QuotaGuard.cs — enforcement helper with graceful degradation
• Integrate quota checks into AgentCore (pre-request validation)
• Add quota tracking to skill execution path (concurrency tracking)
• Add quota tracking to mesh delegation path
• Implement rate limit headers (RateLimitInfo for HTTP response headers)
• Add QuotasController — GET /api/quotas, GET /api/quotas/{scope}/{id}, GET /api/quotas/rate-limit
• Register quota services in DI (CLI and WebAPI Program.cs)
• Unit tests for quota enforcement logic (27 tests: QuotaServiceTests + QuotaGuardTests)
• dotnet build + dotnet test pass (1496 tests pass, 8 pre-existing failures)

Scope / Likely files

src/agent/Mesh/Quotas/

Dependencies

• блокирует / опирается на: task_012 — budget-guardrails
• блокирует / опирается на: task_048 — delegation-boundaries

Risks / Rollback

Многоуровневые лимиты сложно отлаживать; явный report при отказе.

Links

• Backlog: ../backlog.md
• Roadmap (EN): ../../ROADMAP-EN.md
• Roadmap (RU): ../../ROADMAP-RU.md

Implementation notes

• Created src/agent/Quotas/ with 4 files (Models.cs, IQuotaService.cs, QuotaService.cs, QuotaGuard.cs)
• QuotaService: In-memory quota tracking with ConcurrentDictionary, sliding window rate limiting via ConcurrentBag
• QuotaGuard: Enforcement helper with CheckAndGetDegradationMessage and LogSoftWarnings
• QuotaLimitType enum: 12 limit types covering agent, skill, user, and tenant scopes
• QuotaScope enum: Agent, Skill, User, Tenant
• AgentCore integration: pre-check on HandleAsync, Begin/End concurrency tracking, usage recording after LLM calls
• QuotasController: 4 API endpoints for quota status and rate limit info
• DI registration: CLI and WebAPI Program.cs
• Tests: 27 unit tests (QuotaServiceTests + QuotaGuardTests)

• Build: succeeded

  Tests: 1496 passed, 8 pre-existing failures (NumericValidatorTests)