-
Notifications
You must be signed in to change notification settings - Fork 3.6k
Hybrid search on multi struct array field,is it possiple #45779
Replies: 4 comments · 10 replies
-
|
I think you might need to create index on image_vector field and future_vector field. @alwayslove2013 we probably need to reject search on non indexed multi vector field. and also support flat index. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
you mean autoindex index is not an index? image_vector and feature_vector field have an autoindex already! |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
can you offer full log of all your querynode? it seems to be right but from the error stack it falls into SearchBruteForce.cpp which means it is running on bruteforce mode |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
[2025/11/24 01:04:32.869 +00:00] [INFO] [observers/target_observer.go:526] ["Update readable segment version"] [collectionID=462141493833216217] [channelName=by-dev-rootcoord-dml_8_462141493833216217v0] [nodeID=3] [oldVersion=1763946252873048737] [newVersion=1763946262870787995] I20251124 01:04:42.458734 8658 thread_pool.h:67] [SERVER][operator()][knowhere_search][]Successfully set priority of knowhere thread. [2025/11/24 01:04:42.665 +00:00] [WARN] [segments/segment.go:611] ["Search failed"] [clientRequestUnixmsec=1763946282453] [traceID=a95d52cd40a296a4d98fb3ac4a0ad929] [mvcc=462407934305894412] [collectionID=462141493836954625] [segmentID=462141493829685674] [segmentType=Growing] [withIndex=false] [2025/11/24 01:04:43.070 +00:00] [WARN] [segments/segment.go:611] ["Search failed"] [clientRequestUnixmsec=1763946282453] [traceID=a95d52cd40a296a4d98fb3ac4a0ad929] [mvcc=462407934410752016] [collectionID=462141493836954625] [segmentID=462141493829685674] [segmentType=Growing] [withIndex=false] [2025/11/24 01:04:43.875 +00:00] [WARN] [segments/segment.go:611] ["Search failed"] [clientRequestUnixmsec=1763946282453] [traceID=a95d52cd40a296a4d98fb3ac4a0ad929] [mvcc=462407934620467216] [collectionID=462141493836954625] [segmentID=462141493829685674] [segmentType=Growing] [withIndex=false] [2025/11/24 01:04:45.480 +00:00] [WARN] [segments/segment.go:611] ["Search failed"] [clientRequestUnixmsec=1763946282453] [traceID=a95d52cd40a296a4d98fb3ac4a0ad929] [mvcc=462407935039897615] [collectionID=462141493836954625] [segmentID=462141493829685674] [segmentType=Growing] [withIndex=false] |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
I run milvus with docker,image is milvus:v2.6.4. The log is docker logs XX --since 1m after i run the code above. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
see attachfile. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
seems to be you don't have any data in the field? but agree this is a issue, we should return emply result instead of throw error. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
yes. in my test datas,features array may have zero element. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
it is not as expected. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Is this reasonable? In practical scenarios, it is quite common for some fields to lack data |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
milvus don't support vector field to be null. |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
this might be change in the future |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
Try this example, hybrid search on two struct fields: It works fine: |
Beta Was this translation helpful? Give feedback.
All reactions
-
|
@avayayu After thorough inspection, we have identified an issue with Knowhere's brute force algorithm when handling empty embedding lists. This problem will be fixed in this PR and will be included in the next release. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
my schema like on the picture above.,and my milvus is v2.6.4.
then I do hybrid search like this:
but i always get errors like this:
pymilvus.exceptions.MilvusException: <MilvusException: (code=65535, message=fail to search on QueryNode 3: worker(3) query failed: N6milvus21ExecOperatorExceptionE :Operator::GetOutput failed for [Operator:PhyVectorSearchNode, plan node id: 2314] : => Brute force search fail: brute_force inner error at /workspace/source/internal/core/src/query/SearchBruteForce.cpp:266
can I do hybrid search like this?
Beta Was this translation helpful? Give feedback.
All reactions