Daniel (dB.) Doubrovkine
Daniel (dB.) Doubrovkine
@andrewparmet Want to finish this?
> @dblock What is the current state of the exception API? Are you all happy with the implementation as it stands? In that case is the remainder of the work...
Are you spawning threads and creating instances of the client? Can you please post a repro? Is your actual usage of the client sync or async? I do think that...
Are you using a sync client or an async client? My guess is that you're using a sync client. I [wrote a benchmark](https://gist.github.com/dblock/ca2ad6ac1150b6680cd03d47aef09eef) that compares synchronous, threaded and asynchronous I/O...
Here's the sync/async benchmark profile, it's actually stuck doing synchronous I/O.
> This is strange, why it is struck in synchronous I/O. My understanding was I/O should be done in async fashion, even when it is a sync client. Unfortunately not,...
> @dblock can we use some other implementation of HttpConnection. Generally we use `RequestsHttpConnection `, does that also have the same problem? Yes it does. That one does allow changing...
> @dblock here is my thinking, as a user of the OpenSearch client I feel that OpenSearch sync client is easy to use. Most of the services have the same...
> @dblock can we check milvus. Seems to be [using grpc](https://github.com/milvus-io/pymilvus/blob/master/pymilvus/orm/connections.py#L439), whole other ball game.
Thanks @navneet1v. I do think the problem is clear, we want the client to not block on I/O across requests.