Hello,

We have run into performance issues with applications taking a long time (minutes) noticing an unresponsive mongos and switching to another one (our mongos are registered in a _mongodb._tcp SRV record).

Trying to tweak socketTimeoutMS has the inconvenient of false positives on slow queries. Could there be another system in the driver that would allow proactively monitoring mongos reactivity? Thanks in advance.

We also tried setting kernel-level tcp_keepalive_time, probes, intvl much more aggressively, but this did not change the outcome. 

We're using MongoDB 8.0.20 and the Mongo Java driver 5.3.1.

Thanks in advance,