Request rate limits for each API endpoint
To ensure service stability, each API endpoint enforces a request rate limit (QPS — queries per second). Requests exceeding the limit return HTTP status code
Rate limiting rules for knowledge retrieval (
When rate limiting is triggered, the API returns:
Recommendations:
429 Too Many Requests.
Knowledge Base Management APIs
| API | Endpoint | Max QPS |
|---|---|---|
| Retrieval | /rag/index/retrieve | 2,000 |
| Query Index Job Status | /rag/index_job/status | 20 |
| List Documents | /rag/index/files | 15 |
| Get File Details | /rag/list/index/file/details | 10 |
| Create Knowledge Base & Import | /rag/index/create_v2 | 10 |
| Submit Import Job | /rag/index/job/create | 10 |
| List Knowledge Bases | /rag/index/list | 10 |
| Update Knowledge Base | /rag/index/update | 10 |
| Delete Knowledge Base | /rag/index/delete | 10 |
| Delete Document | /rag/index/delete_file | 10 |
| List Chunks | /pipeline/{id}/chunklist | 10 |
| Update Chunk | /rag/index/chunk/update | 10 |
| Delete Chunk | /rag/index/chunk/delete | 10 |
| Retrieve Monitoring Data | /rag/index/monitor | 1 |
Data Ingestion APIs
| API | Endpoint | Max QPS |
|---|---|---|
| Get File Details | /describeFile | 10 |
| Delete File | /deleteFile | 10 |
| Apply for File Upload Lease | /applyFileUploadLease | 10 |
| Register File | /addFile | 10 |
| List Categories | /listCategory | 5 |
| Add Category | /addCategory | 5 |
| Delete Category | /deleteCategory | 5 |
| List Files | /listFile | 5 |
| Get Connector | /getConnector | 5 |
| Batch Update File Tags | /batchUpdateFileTag | 5 |
| Add Connector | /addConnector | 2 |
| Bulk Import from Authorized OSS | /addFilesFromAuthorizedOss | 5 |
Knowledge Retrieval & Question Answering
Rate limiting rules for knowledge retrieval (/knowledge/search) and knowledge Q&A (/knowledge/chat) are not yet published and will be updated later.
Handling Rate Limit Exceedance
When rate limiting is triggered, the API returns:
- Upon receiving a
429, implement exponential backoff retries (e.g., retry after 1s → 2s → 4s). - For batch operations, control concurrency to avoid traffic spikes.
- For low-QPS endpoints such as monitoring data retrieval, consider client-side caching to reduce call frequency.