1 Cause Analysis
No MCS route is found (MCS exception): MCS algorithm problems (not grouped, not activated, plug-in problems, and license problems), and MCS connection problems (connecting to other MCSs, ZooKeeper, Kafka, and CMU).
MCS response timeout: The MCS session is full and the thread is suspended.
If reverse image search is normal but no image is found in the blocklist static library, the reverse image search process is normal, but no image is found in the blocklist static library. This indicates that the fault occurs in the blocklist static library, the possible cause is that the feature extraction of the blocklist static library is not persisted (operation problem). Try the feature extraction again or extract the feature extraction again.
If data is found in structured search but the reverse image search result is 0, the DPC and Mongo are normal. If the reverse image search result is 0, the MCS connection is abnormal. The Kafka data consumption may be abnormal, and residual groups exist in the ZooKeeper, the camera is not connected to the camera in the residual group.
l If no result is found during vehicle search by image, vehicle search by image includes attribute search (structured attribute value, with similar vehicle license plates) and image retrieval (feature value in the image and similarity). If no result is found during attribute search, the structured feature extraction may fail. The image used for search may be blurry and features may be unclear. If the value is image retrieval, the image features are directly obtained, regardless of the definition. Therefore, the corresponding analysis task may be incorrect or the feature extraction function is not enabled.
Query failure: The SDK obtains the target ID from the MCS and then obtains the image from the Mongo. The failure may be caused by an SDK fault or an exception in accessing the Mongo.
l feature extraction failure or no available GPU: The resident GPU is not enabled or the image quality is poor.
2. MCS group creation and activation
Box: Choose Resource Management > MCS Algorithm Management > Select the algorithm corresponding to the algorithm root directory to create a group, add an MCS, and activate the algorithm.
CloudIVS 3000: Choose Application Configuration > Configuration > Application Name > Algorithm Management, and click the MCS Algorithm Management tab. Click the target service to automatically calculate groups or add groups. Choose Add Group > Add MCS > Activate Algorithm.
Cloud Center: Choose Resource Management > Node Management > MCS Algorithm Management. Select the algorithm corresponding to the algorithm root directory to create a group, add an MCS, and activate the algorithm.
If the plug-in does not exist, upload the plug-in on the Plug-in Management page. If the green tick indicates that the plug-in has been uploaded, the plug-in has been uploaded.
Note: Check whether the plug-in version matches.
3. The third-party algorithm license has expired or is not imported.
Apply for a third-party algorithm license (not the same as the application license).
CloudIVS 3000: Log in to the CSP and choose Application Configuration > Application Configuration > Algorithm Management > Third-Party License Management to generate a machine code.
Box: Choose Resource Management > Device Management > Intelligent Analysis License Management/License Service Configuration in the SEM to generate a machine code.
Cloud Center: Log in to the SEM, choose Resource Management > Device Management > Intelligent Analysis License Management/License Service Configuration, and generate a machine code.
4. The connection between the MCS and ZooKeeper or Kafka is abnormal, or the CMU is faulty.
After the MCS is started, node information is written to the ZooKeeper for the CMU to query. If the MCS is disconnected from the ZooKeeper, the CMU cannot be used properly. When the MCS is started, the Kafka needs to be connected. If the connection fails, the Kafka cannot be used. The CMU obtains MCS routes for upper-layer devices. If the CMU is faulty, this fault also occurs. Solution: Restart the MCS service and CMU.
Restart the MCS. For details, see "Restarting the MCS" in the VCM Module Restart Operation.
Restart the CMU. For details, see "Restarting the CMU" in the VCM Module Restart Operation.
5 Installing the MCS in Labeling Mode
Check whether the MCS group name contains the faceTag field to determine whether the labeling mode is used.
Uninstall and delete the labeled MCS, reinstall the MCS plug-in, and create a group activation algorithm in non-labeled mode.
STB: Check whether the MCS group name contains the faceTag field in the SEM.
CloudIVS 3000: Check whether the MCS group name contains the faceTag field in the CSP.
Cloud Center: Check whether the MCS group name contains the faceTag field in the SEM.
6 MCS response timeout
MCS session full: If the MCS internal session is full, the retrieval result summary is suspended, causing response timeout.
MCS exception suspension: The internal retrieval thread of the MCS is suspended, resulting in response timeout.
Restart the MCS. For details, see "Restarting the MCS" in the VCM Module Restart Operation.
7. Reverse image search is normal, but no image is found in the blocklist static library.
On the VCMAPP page, retrieve the target image for the second time.
Attempt to automatically re-extract the person library (restart feature extraction). In the secondary feature extraction scenario, if the extraction is successful, the secondary feature extraction will not be manually triggered again. The operation items are as follows:
1) Delete the feature data of the face_feature_xxxxx person list from the MongoDB.
2) Delete the face library feature data from the MCS.
3) Change the face_state field from 0 to 4 in the face_feature_info table and delete the featureId field. For details, see http://3ms.huawei.com/km/blogs/details/6050681.
8. Data is available in structured search, but the result of reverse image search is 0.
If data is found in structured search, the DPC and Mongo are normal. The Kafka push message is normal, and the MCS is faulty.
Residual groups exist in the ZooKeeper. (The MCS group is deleted on the CSP, but residual groups may exist in the background.)
Data varies according to MCS groups. If residual groups exist in the ZooKeeper, the started MCS does not retrieve and save feature values of some cameras. Solution: Delete the residual group, go to ZooKeeper, and delete the unknown group node on the algorithm page corresponding to the /MCS-Cluster node.
Box and center cloud:
Method for deleting an MCS group in the background: http://3ms.huawei.com/km/blogs/details/6441029
CloudIVS 3000: (This issue is not found in CloudIVS 3000.)
Run the docker ps -a | grep zk command to access the container.
docker exec -ti xxx (Container number) bash
Access the ZooKeeper: cd /usr/local/iGET/SRE/software/zookeeper/bin/
IP address: cat /usr/local/iGET/config/cluster/clusterConfig.properties|grep zookeeper
Run the ./zkCli.sh -server IP address queried just now, for example, 3.17.249.115:21816.
Adding the account password: addauth ivsdigest Change_Me
Run the ls /MCS-Cluster/0208000100_1vN command to view the group information.
Check whether there are unknown Group nodes in the corresponding group. If yes, delete them.
For example, if the MCS_GROUP_2 group is abnormal: delete /MCS-Cluster/0305000100_1vN/MCS_GROUP_2
An exception occurs when the MCS consumes Kafka data (because the connection between the MCS and Kafka is abnormal):
The MCS obtains feature values from the Kafka. If the MCS consumes abnormal data in the Kafka, this fault occurs.
Restart the MCS. For details, see "Restarting the MCS" in the VCM Module Restart Operation.
9 No result is found during vehicle search by image.
Vehicle search by image includes attribute search and image retrieval. Attribute extraction fails because the image is blurry.
The image retrieval matches the similarity based on the features in the image. (This function is valid only for specific functions. For vehicle analysis tasks with feature value extraction enabled, this function is not available for license plate analysis and pedestrian and vehicle hybrid analysis.)
10 Query failed.
If the query fails, the SDK returns the query failure. The possible cause is that the SDK is faulty or the SDK has received the result ID returned by the MCS but an exception occurs when accessing the Mongo. Solution: Restart the SDK or Mongo.
Restart the SDK. For details, see section "SDK Restarting" in the VCM Module Restart Operation.
Restart the Mongo. For details, see "Restarting the Mongo" in the VCM Module Restart Operation.
11 GPU resident enabled
Box and center cloud:
Choose SEM Resource Management > Service Management > Plug-in Management, click Installed Node of the face plug-in, and select a P4 card to enable resident.
CloudIVS 3000:
After the plug-in is installed, choose Resident Memory Configuration and enable Resident Mode in Custom Configuration.
12 If the error message " feature extraction failed" or "No available GPU" is displayed, the CMU is disconnected from the subCMU on the resident GPU.
Restart the CMU or subcMU on the resident GPU. For details about the IP address of the resident GPU, see section 10 in the Reverse Image Search Failure. For details about how to restart the CMU and subcMU, see section "CMU Restart" and "SubcMU Restart" in the VCM Module Restart Operation.
