[Bug]: when milvus service is down, search will retry many times. It will result the client wail for a long time, but got a timeout exception in the end. about pymilvus HOT 3 CLOSED

I think this is caused by the retry behaviour implemented in https://github.com/milvus-io/pymilvus/blob/d91038798b02c1cf7708d672bf9ac94789467334/pymilvus/decorators.py#L48C13-L48C101
which only applies the supplied timeout to retries if the timeout is an int variable.

I don't fully understand why the supplied timeout has to be an int to be applicable for retries, given that the public APIs (search, insert, upsert...) specify the timeout as a float parameter. This means that the retry behaviour changes depending on the provided datatype (int or float).

Can anyone shed some light on this issue please?

from pymilvus.

Could we have a retry_times parameter to specify no retry in vector search?

from pymilvus.

close this for retry_times was supported by #1826

@micronetboy feel free to reopen this issue

from pymilvus.