Intelligent Retrieval FAQ
This topic provides answers to frequently asked questions about Intelligent Retrieval in Intelligent Media Services (IMS). The following sections cover billing, capacity and performance, instance management, and troubleshooting.
Billing
Why am I still being charged daily even though I am not using Intelligent Retrieval?
Intelligent Retrieval capabilities are billed based on enabled indexes. As long as an index remains enabled, fees are continuously incurred regardless of whether you have initiated any search requests or stored any data in the library.
To stop billing, disable or delete the corresponding indexes in the console, including Face Search, Large Model Search, and others. Make sure the status shows as unconfigured before billing stops.
If the Delete button is unavailable in the console, the index is still enabled. Disable the index first, and then delete it.
Why does an empty face library still incur fees?
Face Search incurs subscription-based fees as soon as it is activated, regardless of whether any data has been stored in the face library. To stop billing, disable the Face Search feature and complete the unsubscription process.
How is the number of stored faces calculated?
Billing is based on the number of stored faces. The system analyzes each video frame, and every face detected in a frame counts as one storage unit. For example, if the same face is detected once per second in a 30-second video, that counts as 30 storage units.
Capacity and performance
Is queries per second (QPS) independent across different search libraries?
The QPS of each search library is independent and additive. For example, if you have three search libraries that each support 30 queries per second, the overall throughput can reach approximately 90 queries per second. For higher stability and capacity limits, use a dedicated instance.
Why does hybrid search require two calls to retrieve clip-level information?
This is by design as a two-stage recall mechanism. The first call returns a list of matched media assets. After you specify a media asset identifier, a second call returns clip-level information for that asset. Billing is based only on the number of search requests and vector storage duration. The two-stage calls do not incur duplicate charges.
Instance management
Can a shared instance be directly upgraded to a dedicated instance?
Direct upgrades are not supported. You must first disable the existing instance, then re-enable it and select the dedicated instance type. Indexes created on the previous instance must be rebuilt after the switch. Perform the switch during off-peak hours.
Troubleshooting
What should I do if the namespace parameter does not take effect during face search?
Make sure that the namespace used during the search is consistent with the one used when you created and populated the library. If the parameter is confirmed correct but still does not take effect, submit a ticket and provide the request ID for troubleshooting.
How do I handle an index analysis failure?
On the Media Management page in the console, select the media assets that need to be re-analyzed and initiate a new index analysis task. If failures persist, check whether the media asset file is complete, whether the format is supported, and whether the storage region matches the service region.
What is the daily fixed deduction? Will it affect other functions after closing?
Cost composition: The daily fixed generation is mainly vector feature storage costs. After intelligent retrieval is activated, the system will pre-allocate storage space for the index, and billing will start from the moment it is activated, regardless of whether the retrieval has been initiated or whether there is data in the database.
What functions will use it: text search, graph search, mixed search, face search and other capabilities all rely on vector indexes, which will not be available after the index is closed.
What will not be affected: Disable intelligent search will not affect the capabilities of other products (such as the Internet search provided by the large model service), nor will it affect the storage, editing, transcoding, and other processing capabilities of media capital.
Close Path: Enter the media asset management in the console, find the relevant configuration of intelligent retrieval, close the corresponding vector index and confirm that the status is unconfigured. Billing will stop only when the unconfigured status is reached. Same day fees already incurred will not be refunded as a result of the closure.