Bingyi Sun

Results 150 comments of Bingyi Sun

Fixed in #761 . But `AddConsumerHanlder` in connection.go may have the same issue.

There are 2 problems that lead to search timeout: 1. Msgstream Close() could block because of the full buffer. I have fixed it in #17909 2. The runtime.NumCPU will get...

Datacoord failed to write segment info on etcd because of slow fdatasync. ![image](https://user-images.githubusercontent.com/9817127/216221491-b52697d2-9cae-4dec-997a-e64925cbfedf.png) ![image](https://user-images.githubusercontent.com/9817127/216221401-e43c538d-4cbd-4286-8334-fc50340105b3.png)

> So this issue is caused by a network problem? Can there be any improvement solution? Not sure. I am still finding the reason why the fdatasync is so slow....

> ``` > > [2023-02-02T04:29:47.222Z] [2023-02-02 04:29:35 - INFO - ci_test]: assert search: 0.3707082271575928 (test_all_collections_after_chaos.py:85) > > [2023-02-02T04:29:47.222Z] [2023-02-02 04:29:35 - DEBUG - ci_test]: (api_request) : [Collection.release] args: [], kwargs:...

There was a compaction happed between the insert and num_entities. log: "generate a plan for small candidates"] ["plan segment IDs"="[441500485005239355,441500485005239492,441500485005240254]"] ["target segment row"=**11525**] ["target segment size"=5560410] ["meta update: prepare for...

Prefer to use count(*) instead of num_entities to get accurate record num.