Dwara

Appearance

429 Too Many Requests ​

A 429 response means the request was rejected by a rate limiter, quota, or AI token budget. The error envelope includes the code field to distinguish which.

Symptoms ​

  • Client receives 429 Too Many Requests with {"error":{"code":"...","message":"...","request_id":"..."}}.
  • The Retry-After header indicates when to retry.
  • Metrics show rate limiter or budget counters increasing.

Likely causes ​

Rate limiting ​

The consumer has exceeded its rate limit. The error code is rate_limited.

Diagnose:

sh
# Check rate limiter metrics
curl -k https://127.0.0.1:2019/metrics | grep dwara_rate_limited

Fix: Increase the rate limit in the consumer's policy or wait for the window to reset. Rate limits are configured per-policy:

yaml
policies:
  - name: standard
    rate_limit:
      requests_per_second: 100
      burst: 50

Quota exceeded ​

The consumer has exceeded its daily or monthly request quota. The error code is quota_exceeded.

Diagnose:

sh
# Check quota usage from the admin API
curl -k https://127.0.0.1:2019/analytics/usage?consumer=my-consumer

Fix: Increase the quota in the consumer's policy or wait for the quota window to reset (daily at midnight, monthly on the 1st):

yaml
policies:
  - name: standard
    quotas:
      daily_requests: 100000
      monthly_requests: 1000000

AI token budget exhausted ​

The consumer has exhausted its AI token or cost budget. The error code is ai_budget_exceeded.

Diagnose:

sh
# Check AI budget metrics
curl -k https://127.0.0.1:2019/metrics | grep dwara_ai_budget

Fix: Increase the token budget in the consumer's policy or wait for the budget window to reset:

yaml
policies:
  - name: ai-standard
    token_budget:
      tokens_per_min: 100000
      tokens_per_day: 1000000
      cost_per_day_micros: 5000000  # $5/day

Admission queue full ​

The gateway's concurrency cap has been reached and the admission queue is full. The error code is queue_full.

Diagnose:

sh
# Check admission queue metrics
curl -k https://127.0.0.1:2019/metrics | grep dwara_admission

Fix: Increase the concurrency cap or queue size in the listener config, or scale out the gateway:

yaml
listeners:
  - name: main
    concurrency:
      max_connections: 10000
      queue:
        max_queued: 1000
        timeout_ms: 5000

See also ​