Production Guidelines - BizyAir-Docs

Concurrency Quotas & Rate Limiting

The platform allocates the following limits per account, based on the plan tier:

Limit	Meaning	Behavior on Exceedance
Queued tasks	Total tasks simultaneously in `Queuing` + `Preparing`	New requests return 429
Running tasks	Total tasks simultaneously in `Running`	New requests return 429
RPM	API requests per minute (including create / query)	Returns 429
WebHook callback concurrency	Concurrent callback connections made to your URL	Queued internally by the platform; no external error

Actual limits depend on your plan. If your business needs higher concurrency, submit an upgrade request via Profile → Plans, or contact sales.

Rate-limit mitigation patterns:

Use a local client queue + token bucket to control the send rate
Apply exponential backoff retry on 429
Use WebHook for long tasks to avoid occupying HTTP connections

Data Retention & Security

Data Type	Retention	Notes
Task metadata (status / request_Id / elapsed)	Long-term	Queryable anytime via `/detail`
Input files (OSS `inputs/`)	30 days	Auto-cleaned after expiry
Output files (OSS `outputs/`)	15 days	Unrecoverable after expiry
API Keys	Until manually deleted	Recommend regular rotation

Download important results to your own storage within 15 days. No recovery service is provided after expiry.

Content Moderation

The platform performs automatic content moderation on input prompts, uploaded images, and generated results:

audit_status: 1 Not reviewed (should not appear in terminal-state tasks)
audit_status: 2 Approved
audit_status: 3 Rejected; result is inaccessible
audit_status: 4 Moderation error (e.g., input inaccessible)

Tasks that fail moderation will still incur inference-time charges — filter sensitive prompts at the business layer in advance.

Billing & Balance

Billing Method

Fee = inference_cost_time × the app’s unit price. Includes:

✅ Inference time (Running stage)
❌ Does not include queueing, preparation, upload, or download time
❌ Does not include Canceled tasks
✅ Includes inference time consumed before an interruption

Querying Balance

curl -X GET "https://api.bizyair.ai/v1/wallet" \
  -H "Authorization: Bearer $BIZYAIR_API_KEY"

Response:

{
  "code": 20000, "status": true,
  "data": {
    "balance": 1250.5,
    "currency": "BZC",
    "updated_at": "2025-09-11 10:30:00"
  }
}

Low-balance Alert

It’s recommended to poll the balance periodically (e.g., hourly) at the application layer and trigger an alert when it drops below a threshold, to avoid production disruption caused by insufficient balance.

Monitoring & Observability

Recommended metrics to monitor:

Metric	Collection Method	Alert Threshold Reference
Task success rate	Compute `Success / (Success + Failed)`	< 95%
Average inference time	`inference_cost_time` P50 / P95	+50% MoM
Queue wait time	Cumulative `Queuing` time	> 5 min
429 frequency	HTTP response distribution	> 1% sustained for 5 min
WebHook callback failure rate	Your callback service logs	> 0.1%
Balance	Periodic pull from `/v1/wallet`	< 3 days of usage

Drills & Disaster Preparedness

Drill Checklist

Periodically drill a fast API Key rotation after a leak (create → deploy → delete old key)
Drill a progressive 429 degradation (reduce concurrency, apply backpressure upstream)
Drill progressive WebHook callback failure (fall back to polling)

Readiness Checklist

Confirm before going live:

API Key stored in a secrets management service, not hard-coded
Synchronous blocking call client read timeout ≥ 60s
WebHook receiver verifies the X-BizyAir-Task-Authorization header
WebHook receiver implements idempotency (guard against duplicate callbacks)
Important outputs are transferred to your own OSS / S3 storage in real time
Low-balance alerts configured
Retry and degradation logic in place

​Concurrency Quotas & Rate Limiting

​Data Retention & Security

​Content Moderation

​Billing & Balance

​Billing Method

​Querying Balance

​Low-balance Alert

​Monitoring & Observability

​Drills & Disaster Preparedness

​Drill Checklist

​Readiness Checklist

Concurrency Quotas & Rate Limiting

Data Retention & Security

Content Moderation

Billing & Balance

Billing Method

Querying Balance

Low-balance Alert

Monitoring & Observability

Drills & Disaster Preparedness

Drill Checklist

Readiness Checklist